CMOS-MEMS Chemiresistive and Chemicapacitive Chemical Sensor System Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Electrical and Computer Engineering

Nathan S. Lazarus

B.S.E., Electrical Engineering, University of Pennsylvania M.S., Electrical Engineering, Carnegie Mellon University

Carnegie Mellon University Pittsburgh, PA

April, 2012

i

Acknowledgments I would like to begin by thanking my adviser, Gary Fedder, both for welcoming me into his lab and supporting me throughout my Ph.D. I am particularly grateful that he has allowed and encouraged me to explore numerous tangents that I have proposed in the past few years that in many respects turned into my final thesis, and which I feel have strengthened me greatly as an independent researcher.

I

would also like to thank my thesis committee members, Rongchao Jin, Yi Luo, and Jeyanandh Paramesh, for giving their time and helpful advice on my research direction. I am also thankful to the NIOSH research group that has worked with me on many aspects of the end-of-service-life indicator project. Jay Snyder has been an active and helpful project manager, providing a great deal of helpful technical suggestions and guidance. I would like to thank Suresh Santhanam for his assistance throughout, from guidance on microfabrication processes, teaching me to use many of the pieces of equipment in the cleanroom, and doing the standard CMOS-MEMS processing that much of my work depended on. Lee Weiss and Larry Schultz have provided invaluable assistance with inkjetting and other aspects of the project.

Niti Garg and Huifeng Qian synthesized the

nanoparticles and nanoclusters without which many portions of this thesis would not be possible. I would also like to thank the other students and professors who have been involved in the project over the years, including David Lambeth, Jeremy Greenblatt, Tony Rozzi, John Wu, Zhaotao Xu, Aakriti Gupta and Kelly Frank. I have had the good fortune to work with a wonderful group of people in the CMU MEMS Lab. Tamal Mukherjee gave advice and help, as well as guiding through numerous tapeouts. I am thankful to the senior graduate students who gave me important assistance starting out. Sarah Bedair and Chiung Lo spent many hours helping me with my first CMOS design and tapeout. Peter Gilgunn has been an ii

important source on cleanroom fabrication. I would also like to thank the many members of the MEMSLab who have been both helpful in the lab and provided important distractions outside of it. From running after work with Dylan Fang, discussing politics and international finance with Leon Wang, and playing on the ECE departmental softball team with Kristen Dorsey and Louis Draghi and disc golf with John Reinke, to being taught about the nuances of mechanical modeling by Congzhong Guo and international cricket by Ashwati Krishnan, I have enjoyed my time at CMU and for that I will be eternally grateful. I would also like to thank those other students that I have been able to work with in my time here, including Philip Bergeron, Jonathan Rotner, Andy Zhang, Chenyang Wu, Erdinc Tatar, Amy Wung, Jingwei Liu, Abishek Jajoo, Jason Guo, Jie Wang, Sean Yen, Lionel Wong and Fernando Alfaro. I would also like to thank the professors at the University of Pennsylvania who were so important in helping me choose to seek my Ph.D. I am grateful to David Pope, my freshman adviser, for originally suggesting that I go on to graduate school and who provided a great deal of friendship, helpful advice and opportunities. I am thankful to Jan Van der Spiegel for giving me my first opportunity to do research in his lab, and Viktor Gruev, who, as a busy post-doc, took many long hours of his time to guide me and help me through many of the pitfalls of being a young researcher. I would also like to thank Gianluca Piazza for sparking my interest in MEMS. Most of all, I would like to express gratitude to my family for their love and support. My parents and my brother Greg and sister Brenda have sustained me throughout, and I would never have been able to complete this work without them. This research was funded by the National Institute of Occupational Safety and Health under contract 200-2002-00528 and 254-2011-M39538 and by the US Air Force Office of Scientific Research under grant FA9550-07-1-0245.

iii

Abstract Integrating chemical sensors with testing electronics is a powerful technique with the potential to lower power and cost and allow for lower system limits of detection. This thesis explores the possibility of creating an integrated sensor system intended to be embedded within respirator cartridges to notify the user that hazardous chemicals will soon leak into the face mask. For a chemical sensor designer, this application is particularly challenging due to the need for a very sensitive and cheap sensor that will be exposed to widely varying environmental conditions during use.

An octanethiol-coated gold nanoparticle chemiresistor to detect

industrial solvents is developed, focusing on characterizing the environmental stability and limits of detection of the sensor. Since the chemiresistor was found to be highly sensitive to water vapor, a series of highly sensitive humidity sensor topologies were developed, with sensitivities several times previous integrated capacitive humidity sensors achieved. Circuit techniques were then explored to reduce the humidity sensor limits of detection, including the analysis of noise, charge injection, jitter and clock feedthrough in a charge-based capacitance measurement (CBCM) circuit and the design of a low noise Colpitts LC oscillator.

The

characterization of high resistance gold nanoclusters for capacitive chemical sensing was also performed. In the final section, a preconcentrator, a heater element intended to release a brief concentrated pulse of analate, was developed and tested for the purposes of lowering the system limit of detection.

iv

Contents Contents .................................................................................................................................................... v Introduction .................................................................................................................................................. 1 1.1 End-of-Service-Life Indicators ............................................................................................................. 1 1.2 Integrated chemical sensor system .................................................................................................... 3 1.3 Thesis contributions ............................................................................................................................ 7 1.4 Thesis outline ...................................................................................................................................... 7 C8 Nanoparticle Resistive Chemical Sensors ................................................................................................ 9 2.1 Introduction ........................................................................................................................................ 9 2.2 Prior Work ......................................................................................................................................... 10 2.2.1 Metal-oxide sensors ................................................................................................................... 10 2.2.2 Intrinsically conducting polymers .............................................................................................. 10 2.2.3 Carbon black loaded polymer .................................................................................................... 11 2.2.4 Coated nanoparticle chemiresistors .......................................................................................... 12 2.3 C8 Gold Nanoparticle Sensors........................................................................................................... 13 2.3.1 Sensor Fabrication ..................................................................................................................... 13 2.3.2 Sensor Characterization ............................................................................................................. 16 2.3.2.1 Analyte Response ................................................................................................................ 16 2.3.2.2 Temperature Sensitivity ...................................................................................................... 18 2.3.2.3 Humidity Response ............................................................................................................. 19 2.3.2.4 Electrical Drift...................................................................................................................... 20 2.3.2.5 Noise Characterization ........................................................................................................ 22 2.3.3 Integrated Nanoparticle Sensors ............................................................................................... 27 2.4 Discussion.......................................................................................................................................... 32 CMOS-MEMS Capacitive Chemical Sensors ................................................................................................ 34 v

3.1 Introduction ...................................................................................................................................... 34 3.2 Prior Work ......................................................................................................................................... 35 3.2.1 Sensor topologies....................................................................................................................... 35 3.2.2 Integrated sensors ..................................................................................................................... 37 3.3 Released Interdigitated Sensor ......................................................................................................... 38 3.3.1 Sensor Fabrication ..................................................................................................................... 38 3.3.2 Capacitive wicking channel sensor............................................................................................. 40 3.3.2.1 Device.................................................................................................................................. 40 3.3.2.2 Device Modeling ................................................................................................................. 41 3.3.2.3 Response Time .................................................................................................................... 45 3.3.3 Direct Coated Sensor ................................................................................................................. 47 3.3.3.1 Design.................................................................................................................................. 47 3.3.3.2 Thermal Characterization.................................................................................................... 49 3.3.3.2 Cross Sensitivity .................................................................................................................. 50 3.3.4 Encapsulated Electrode Sensor.................................................................................................. 51 3.3.4.1 Design.................................................................................................................................. 51 3.3.4.2 Experimental Results .......................................................................................................... 54 3.4 Integrated Vertical Parallel-Plate Sensor .......................................................................................... 56 3.4.1 Device ......................................................................................................................................... 56 3.4.2 Fabrication ................................................................................................................................. 58 3.4.3 Experimental Results.................................................................................................................. 61 3.4.3.1 Humidity Response ............................................................................................................. 61 3.4.3.2 Response Time .................................................................................................................... 62 3.4.3.3 Thermal Characterization.................................................................................................... 64 3.4.3.4 Cross Sensitivity .................................................................................................................. 67 3.4.3.5 Oxide pillar density ............................................................................................................. 68 3.4.4 Polymer Etched Vertical Parallel Plate Design ........................................................................... 71 vi

3.4.4.1 Design.................................................................................................................................. 71 3.4.4.2 Fabrication .......................................................................................................................... 72 3.4.5 Polysilicon heater ....................................................................................................................... 76 3.5 Discussion.......................................................................................................................................... 78 Capacitive Sensing Electronics .................................................................................................................... 80 4.1 Introduction ...................................................................................................................................... 80 4.2 Charge-Based Capacitance Measurement Circuit ............................................................................ 82 4.2.1 Circuit ......................................................................................................................................... 82 4.2.2 Noise .......................................................................................................................................... 83 4.2.3 Charge injection ......................................................................................................................... 87 4.2.4 Clock feedthrough ...................................................................................................................... 92 4.2.5 Clocking jitter ............................................................................................................................. 93 4.2.6 Noise measurement ................................................................................................................... 95 4.2.7 Discussion................................................................................................................................... 96 4.3 Low phase noise Colpitts oscillator ................................................................................................... 97 4.3.1 LC oscillator ................................................................................................................................ 97 4.3.2 Oscillator topology ................................................................................................................... 100 4.3.3 Integrated Sensor System ........................................................................................................ 102 4.3.4 System Measured Results ........................................................................................................ 106 3.3.4.2 Colpitts Oscillator Measurements .................................................................................... 108 4.3 Analysis and future work ................................................................................................................ 111 Gold Nanocluster Capacitive Sensors ....................................................................................................... 112 5.1 Introduction .................................................................................................................................... 112 5.2 Gold Nanoclusters ........................................................................................................................... 113 5.2.1 Au25........................................................................................................................................... 114 vii

5.2.1.2 Chemical sensitivity........................................................................................................... 114 5.2.1.3 Environmental sensitivity.................................................................................................. 116 5.2.1.4 Manufacturing repeatability ............................................................................................. 118 5.2.2 Au38 and Au144 ........................................................................................................................... 122 5.3 Analysis and future work ................................................................................................................ 125 Preconcentration ...................................................................................................................................... 127 6.1 Introduction .................................................................................................................................... 127 6.2 Background ..................................................................................................................................... 127 6.2.1 Concentration factor ................................................................................................................ 127 6.2.2 Past preconcentrators.............................................................................................................. 129 6.3 Microhotplate preconcentrator ...................................................................................................... 132 6.3.1 Design....................................................................................................................................... 132 6.3.2 Fabrication ............................................................................................................................... 134 6.3.3 Characterization ....................................................................................................................... 137 6.4 Vertical flow preconcentrator......................................................................................................... 138 6.4.1 Fabrication ............................................................................................................................... 138 6.4.2 Thermal modeling .................................................................................................................... 141 6.4.3 Flow system ............................................................................................................................. 146 6.4.4 Preconcentrator coating .......................................................................................................... 148 6.4.5 Analyte testing ......................................................................................................................... 150 6.4.6 Low flow testing ....................................................................................................................... 155 6.5 Discussion and Future Work ........................................................................................................... 161 Conclusions and Future Work ................................................................................................................... 163 7.1 Contributions .................................................................................................................................. 163 7.2 Conclusions ..................................................................................................................................... 165 viii

7.3 Future work ..................................................................................................................................... 166 References: ........................................................................................................................................... 168

ix

List of Tables Table 1.1: Recommended exposure limits.................................................................................................... 1 Table 1.2: Past chemical sensors integrated with electronics ...................................................................... 4 Table 3.1: Simulated and measured capacitance of sensor ....................................................................... 42 Table 3.2 Material properties and sensitivity of chemicals in cross sensitivity test ................................... 51 Table 3.3 Sensitivities for different capacitive sensor topologies .............................................................. 62 Table 3.4 Sensitivity of sensor to organic solvents with analyte properties .............................................. 67 Table 4.2: Parameters used for calculating RLC Allan variance .................................................................. 99 Table 4.1: Common conversions between phase noise power spectral density and Allan variance ......... 99 Table 5.1: Table of analyte properties and sensitivity .............................................................................. 116 Table 5.2: Sensitivity comparison table .................................................................................................... 120

x

List of Figures Figure 1.1: End-of-service-life indicator........................................................................................................ 2 Figure 1.2: Integrated sensor system............................................................................................................ 3 Figure 2.1: C8 gold nanoparticles................................................................................................................ 12 Figure 2.2: Chemiresistor process flow (a) initial wafer (b) gold deposition (c) gold patterning (d) bondpad deposition (e) SU-8 patterning (f) deposition of nanoparticles .................................................. 14 Figure 2.3: Spiral electrode (a) before and (b) after inkjetting of nanoparticles........................................ 15 Figure 2.4 AFM of the outer ring of an inkjetted device............................................................................. 16 Figure 2.5: I-V characteristic of C8 gold nanoparticle sensor ..................................................................... 17 Figure 2.6: Sensor response to toluene and acetone ................................................................................. 17 Figure 2.7: Temperature response ............................................................................................................. 18 Figure 2.8: Resistance temperature detector ............................................................................................. 19 Figure 2.9: RTD temperature sensitivity ..................................................................................................... 19 Figure 2.10: Humidity response of C8 nanoparticle sensor ........................................................................ 20 Figure 2.11: (a) Test of turn-on transient and (b) magnified view of single transient ............................... 21 Figure 2.12: Voltage dependence test of turn on transient ....................................................................... 22 Figure 2.13: On/off test in nitrogen ............................................................................................................ 22 Figure 2.14: Noise spectrum of C8 nanoparticle sensor ............................................................................. 24 Figure 2.15: Chemiresistor modulation testing circuit ............................................................................... 25 Figure 2.16: Allan deviation for different modulation frequencies ............................................................ 25 Figure 2.17: Modulation of a system (a-c) with input independent voltage noise and (d-f) a resistive sensor whose resistance varies with a 1/f noise relationship .................................................................... 27 Figure 2.18: Chemiresistor process flow: (a) CMOS chip, (b) timed oxide etch, (c) wet aluminum etch, (d) platinum deposition, (e)platinum patterning and (f) nanoparticle deposition .......................................... 28 Figure 2.19: Platinum electrode chemiresistor (a) before and (b) after inkjetting .................................... 29 Figure 2.20: (a) Via array after aluminum etch and (b) close up of vias ..................................................... 30 xi

Figure 2.21: (a) Via array after aluminum etch and (b) close up of vias for a shorter etch time ............... 31 Figure 2.22: Aluminum electrode sensor .................................................................................................... 32 Figure 2.23: Sensor response to toluene .................................................................................................... 32 Figure 3.1 (a) Vertical parallel-plate and (b) interdigitated capacitive chemical sensors .......................... 35 Figure 3.2 (a) Planar ETH Zurich sensor [60] (b) structured sensor [61] and (c) oxide etched sensor [63] 37 Figure 3.3 Released interdigitated sensor .................................................................................................. 39 Figure 3.4 CMOS-MEMS fabrication process .............................................................................................. 39 Figure 3.5: SEM of single channel capacitive humidity sensor ................................................................... 41 Figure 3.6: SEM of wicking channel before (a) and after (b) polymer deposition ...................................... 41 Figure 3.7 Humidity response of single channel device.............................................................................. 44 Figure 3.8 Five channel sensor after inkjetting ........................................................................................... 44 Figure 3.9 Humidity response of five channel device ................................................................................. 45 Fig. 3.10 Simulated transient response of water vapor diffusion in (a) an unreleased and (b) a released structure and (c) average concentration in each device ............................................................................ 46 Figure 3.11 Eighty-seven channel coated device (a) before and (b) after inkjetting of polyimide ............ 48 Figure 3.12 Humidity responses of 87 channel coated and 5 channel wicked devices .............................. 49 Figure 3.13 Temperature response of coated sensor ................................................................................. 50 Figure 3.14 Coated sensor response to industrial solvents ........................................................................ 50 Figure 3.15 Encapsulated sensor cross section (a) before and (b) after jetting of polymer ...................... 52 Figure 3.16 COMSOL simulation of encapsulated sensor ........................................................................... 53 Figure 3.17 (a) Encapsulated device after inkjetting of polymer and (b) magnified view of one of the capacitive channels ..................................................................................................................................... 54 Figure 3.18 (a) Encapsulated device after syringe deposition of polymer and (b) magnified view of one of the capacitive channels............................................................................................................................... 54 Figure 3.19 Humidity response of encapsulated sensor............................................................................. 56 Figure 3.20 Techniques for creating a vertical parallel-plate sensor in CMOS ........................................... 57 xii

Figure 3.21 Integrated vertical parallel plate sensor fabrication process .................................................. 59 Figure 3.22 (a) Cross section of sensor (b) cross section that passes through oxide pillars and (c) top view .................................................................................................................................................................... 60 Fig. 3.23: SEM image of sensor (a) before and (b) after inkjetting of polyimide with inset picture of filled release holes ............................................................................................................................................... 61 Fig. 3.24: Humidity response of vertical parallel-plate sensor ................................................................... 62 Fig. 3.25: (a) Simulated response time of sensor and (b) diagram of response model .............................. 63 Fig. 3.26 (a) Absorption and (b) desorption transient measured from sensor ........................................... 64 Fig. 3.27 Temperature response of vertical parallel plate sensor .............................................................. 65 Fig. 3.28: FIB cross sections (a) near anchor and (b) at point of maximum displacement ......................... 66 Fig. 3.29 Simulated vertical deflection for different fill percentages; 0 μm is at the outer circumference, 110 μm is the edge of the center hole ........................................................................................................ 66 Figure 3.30: Optical interferometer measurement of original design (a) after aluminum release (b) after polyimide deposition. Revised design with additional pillars (c) after etching and (d) after deposition in design with pillars ....................................................................................................................................... 69 Figure 3.31: Humidity response for higher pillar density sensor ................................................................ 70 Fig. 3.32 Vertical parallel-plate sensor (a) before and (b) after vertical polymer etch .............................. 71 Fig. 3.33 Process flow for sensor with vertical polymer etch ..................................................................... 72 Fig. 3.34 Device (a) after spin coating polyimide and (b) after polyimide etch .......................................... 73 Figure 3.35: Humidity response for etched sensors; lines are COMSOL simulated responses, individual data points are measured data ................................................................................................................... 74 Figure 3.36: Simulated response time for etched polymer sensors ........................................................... 75 Figure 3.37: Difference in (a) rising and (b) falling time constants between etched sensors and reference humidity sensor .......................................................................................................................................... 76 Figure 3.38: Device with added polysilicon heater ..................................................................................... 77 Figure 3.39: Humidity response at different temperatures ....................................................................... 77 Figure 3.40: Difference in (a) rising and (b) falling time constants between sensors at different temperatures and the reference ................................................................................................................ 78 xiii

Figure 4.1: (a) Relaxation oscillator, (b) ring oscillator and (c) switched capacitor amplifier .................... 81 Figure 4.2: (a) CBCM circuit and timing diagrams for (b) first measurement and (c) second measurement. .................................................................................................................................................................... 82 Figure 4.3: (a) CBCM circuit and noise equivalent circuits (b) when NMOS is turned on and (c) when PMOS is turned on. ..................................................................................................................................... 84 Figure 4.4: Equivalent model for NMOS charge injection .......................................................................... 88 Figure 4.5: Charge injection voltage resulting from NMOS transistor........................................................ 90 Figure 4.6: (a) CBCM circuit including parasitic overlap capacitances and capacitive divider circuits (b) formed by Φ2 input and (c) formed by Φ1 input ......................................................................................... 92 Figure 4.7: Normalized frequency Allan deviation for phase locked Agilent 34401 function generators.. 94 Figure 4.8: Measured Allan deviation. ........................................................................................................ 95 Figure 4.9: RLC circuit.................................................................................................................................. 97 Figure 4.10: NPN Colpitts Oscillator.......................................................................................................... 101 Figure 4.11: Simulated phase noise spectral density of NPN Colpitts oscillator ...................................... 101 Figure 4.12: Frequency counter diagram .................................................................................................. 102 Figure 4.13: PTAT circuit ........................................................................................................................... 104 Figure 4.14: Simulated temperature sensitivity of PTAT circuit ............................................................... 105 Figure 4.15: Integrated sensor chip .......................................................................................................... 106 Figure 4.16: Capacitive sensors after inkjet deposition of polymer ......................................................... 106 Figure 4.18: Monte Carlo simulation of PTAT output voltage at room temperature ............................... 107 Figure 4.17: PTAT temperature sensitivity ............................................................................................... 107 Figure 4.19: Buffer output of Sensor Colpitts oscillator ........................................................................... 108 Figure 4.20 Humidity testing with Colpitts oscillator; frequency is clock divided counter output .......... 109 Figure 4.21 Stability test of Colpitts oscillator .......................................................................................... 110 Fig. 5.1: Chemicapacitive sensor after inkjetting of Au25 .......................................................................... 114 Fig. 5.2: Au25 response to ethanol............................................................................................................. 115 xiv

Fig. 5.3: Au25 response to water vapor ..................................................................................................... 117 Fig. 5.4: Au25 temperature response......................................................................................................... 117 Fig. 5.5: Au25 sensor jetted with 50:50 mixture of toluene and DMSO .................................................... 118 Fig. 5.6: Au25 sensor jetted in toluene ...................................................................................................... 120 Fig. 5.7: (a) image showing location of FIB cut and (b) FIB cross section ................................................. 121 Fig. 5.8: (a) Au38 and (b) Au144 sensors after jetting .................................................................................. 122 Fig. 5.9: Au144 chemical response .............................................................................................................. 123 Fig. 5.10: Au144 humidity response............................................................................................................ 124 Fig. 5.11: Au38 humidity response ............................................................................................................. 124 Figure 6.1 Tube preconcentrator .............................................................................................................. 129 Figure 6.2 Sandia membrane preconcentrator......................................................................................... 130 Figure 6.3 Top view of University of Michigan packed microheater ........................................................ 132 Figure 6.4 (a) Top view and (b) flow channel cross section of first generation preconcentrator design . 133 Figure 6.5 First generation preconcentrator process flow ....................................................................... 135 Figure 6.6 Bottom of preconcentrator (a) after vertical silicon etch and (b) after isotropic etch; (c) shows the initial layout ........................................................................................................................................ 136 Figure 6.7 Preconcentrator design (a) with released membrane and (b) with silicon bridges to the substrate ................................................................................................................................................... 136 Figure 6.8 Temperature response (a) with released membrane and (b) with silicon bridges to the substrate ................................................................................................................................................... 137 Figure 6.9 (a) Heater viewed through shadow mask and (b) preconcentrator after deposition of activated carbon through mask ................................................................................................................................ 138 Figure 6.10 Fabrication process flow for vertical flow preconcentrator .................................................. 139 Figure 6.11 (a) Diagram of vertical flow preconcentrator and (b) picture of final structure ................... 141 Figure 6.12 Lumped thermal model of preconcentrator .......................................................................... 142 Figure 6.13 (a) Steady state temperature response and (b) transient response to 1.4 W input power .. 143 xv

Figure 6.14 Finite element model of preconcentrator heating ................................................................ 144 Figure 6.15 (a) Backside picture of preconcentrator and (b) silicon bridge at higher magnification ....... 145 Figure 6.16 Diagram of preconcentrator mount ...................................................................................... 146 Figure 6.17 Flow system for preconcentrator testing .............................................................................. 147 Figure 6.18 Flow system dilution test ....................................................................................................... 148 Figure 6.19 (a) Preconcentrator after deposition of Tenax TA and (b) magnified image of individual holes .................................................................................................................................................................. 149 Figure 6.20 (a) Preconcentrator after deposition of Q Bond and (b) magnified image of individual holes .................................................................................................................................................................. 150 Figure 6.21 Preconcentrator analyte testing with approximately (a) 10% (b) 30% and (c) 90% of pores clogged with Tenax TA .............................................................................................................................. 152 Figure 6.22 (a) Preconcentrator pulse for Q Bond preconcentrator and (b) comparison with analyte pulse from 90% clogged Tenax preconcentrator ............................................................................................... 154 Figure 6.23 Analyte pulses for different input powers ............................................................................. 155 Figure 6.24 Preconcentrator test (a) with and (b) without preceding 300 ppb toluene pulse ................ 156 Figure 6.25(a) Test system with nitrogen around testing box and (b) test demonstrating reduction in relative humidity inside test box .............................................................................................................. 157 Figure 6.26 Preconcentrator pulses with nitrogen around testing box .................................................... 158 Figure 6.27 Finite element model of thermal behavior of flow system; vertical and horizontal axes are in meters, temperature in degrees Kelvin. The preconcentrator is at the 5 cm horizontal mark. .............. 159 Figure 6.28 Preconcentrator testing (a) without and (b) with metal tube added between test box and preconcentrator. (c) and (d) show the corresponding preconcentrator inputs ...................................... 160 Figure 6.29 Preconcentrator pulses with metal tube (a) without and (b) with exposure to toluene ...... 161

xvi

Chapter 1 Introduction 1.1 End-of-Service-Life Indicators Determining when a respirator cartridge is fully loaded with chemical and is about to leak into the respirator face mask is an important problem in industry. An end-of-service-life indicator (ESLI) is a sensor embedded within the respirator cartridge bed that notifies the wearer when the concentration of chemical has reached a specified threshold. This threshold is set by NIOSH in their guide to chemical hazards [1] as recommended exposure limits (REL), the average chemical level deemed to be hazardous upon exposure for an eight hour period. Table 1.1 shows the recommended exposure limits for a number of common solvents and industrial chemicals. These exposure limits range from hundreds or thousands of parts per million for relatively safe solvents such as IPA and ethanol to a few parts per million or even hundreds of parts per billion for particularly hazardous chemicals such as benzene and Freon.

Chemical Acetone Benzene Ethanol Freon (Carbon Tetrachloride) Toluene IPA Table 1.1: Recommended exposure limits

NIOSH Eight Hour Recommended Exposure Limits (ppm) [1] 250 0.1 1000 2 100 400

1

For an ESLI, the sensor must be located within the cartridge rather than in the respirator facemask to avoid exposing the user unnecessarily; a cross section of the proposed system is shown in Fig. 1.1.

Face mask

Cartridge

Indicator in face mask

PCB Sensor

Carbon absorbent

Air flow Figure 1.1: End-of-service-life indicator OSHA regulations on respirators state that an air-purifying respirator used to filter out a chemical must be “equipped with an end-of-service-life indicator (ESLI) certified by NIOSH for the contaminant; or if there is no ESLI appropriate for conditions in the employer’s workplace, the employer [must implement] a change schedule for canisters and cartridges that is based on objective information of data that will ensure that canisters and cartridges are changed before the end of their service life.” [1] There is currently no ESLI available that is able to sense the broad range of analytes necessary for most respirator cartridges [3]. Passive sensors based on color change in an absorbent layer have been developed ([4][5]). However, indicators relying on the user seeing a change in color on the cartridge do not work well in situations with low lighting or poor

2

visibility. A number of alternative techniques depending on active sensing of the chemical change are also being investigated, including surface acoustic wave sensing [6], optical detection of reflected light [7], and calorimetric sensing of absorption of chemical analyte [8]. These techniques have not yet been demonstrated as practical for most respirator cartridges. Change-out schedules are generally used instead, resulting in unnecessary replacement costs if a cartridge is changed out sooner than necessary or compromising safety if it is changed too late.

1.2 Integrated chemical sensor system In this work, an integrated system containing several types of chemical sensors on a single CMOS die is envisioned (Fig. 1.2). Having a number of types of chemical sensors in a system allows better differentiation between different analytes, since different forms of chemical sensors preferentially detect different aspects of the absorbed molecules [9]. Creating a sensor system directly on a CMOS die allows amplification and digitization of the output signal on-chip, improving system performance [10]. The ability to share process steps, such as using the metal layers from CMOS to create chemical sensor electrodes, allows a significant cost reduction that is important for a disposable application such as a sensor inside

Chemicapacitors

Chemiresistors

Interface Electronics

Temperature sensor

Figure 1.2: Integrated sensor system

3

of a respirator cartridge. A single chip system would also reduce the size of the sensor, reducing the impact on airflow through the respirator cartridge. Most of the common forms of chemical sensors have been successfully integrated with testing electronics. Table 1.2 shows a number of the notable examples in the literature of sensors and electronics on a single chip. Systems with a single chemical sensor or arrays of a single modality of chemical sensor integrated with testing electronics are common, and the examples in the table are a small subset of the total work in the area.

Affiliation

Mode of Chemical Sensor

ETH Zurich Univ. of Warwick National Chung Hsing University CalTech NIST HKUST University of Pisa

Capacitive/Calorimetric/Gravimetric ChemFET/Resistive

ETH Zurich

Static bending

Capacitive Resistive Microhotplate resistive Microhotplate resistive ChemFET

Circuitry Various (Amplification/ADC/Control) Amplification/Filtering Ring Oscillator Multiplexor/Current Mirror Opamp/Multiplexor Multiplexor/Current Mirror Transresistive amplifier Various (Amplification/ADC/Filtering)

Reference Number [11] [12] [13] [14] [15] [16] [17] [18]

Table 1.2: Past chemical sensors integrated with electronics

However, systems containing multiple types of sensor on a single die are uncommon, and there are only a handful that have been demonstrated in the literature. Adding additional types of chemical sensor requires additional processing, adding expense and lowering potential yield, and the fabrication processes are frequently incompatible. Certain combinations of chemical sensors can be easily made on the same die using identical or nearly identical structures and fabrication processes. The chemFET and chemiresistors demonstrated in [12] 4

were made on the same chip because both types of device require noble metal electrodes for low contact resistance. Both sensors can also be created with identical electrodes by having a third terminal underneath the sensitive layer to control the FET behavior if necessary. An impressive example of a sensor system with multiple different chemical sensor structures and circuitry integrated into a single chip was developed at ETH Zurich [11]. In their sensor system, three different types of chemical sensor, a resonant cantilever gravimetric sensor, a capacitive sensor and a calorimetric sensor were integrated with sensor interface circuitry that amplifies the sensor response and converts it to a digital output, which in addition to serving as the final output provided feedback control for the sensors in the system. The researchers were able to create a shared fabrication process based on a backside KOH silicon etch followed by frontside reactive ion etching (RIE) steps to release the devices. The same group has also created variations on their system, such as the capacitive, gravimetric and microhotplate-based system created in 2007 ([19]). For use in an end-of-service-life indicator system, the target analytes are various industrial solvents, particularly various volatile organic carbons (VOCs).

Due to the low

recommended exposure limits in the hundreds of parts per billion range for particularly hazardous industrial chemicals, the limits of detection of the sensors used must reach these levels.

Since a respirator cartridge will not be used in a controlled environment, the

temperature and humidity must also be measured to avoid false positive readings. These sensors must also be highly sensitive, since even a very small change in humidity or temperature can overwhelm analyte changes of only a few hundreds of parts per billion. None 5

of the currently available integrated sensors are able to meet these constraints. The ETH Zurich sensor system [11] is the only technology with a broad array of chemical sensors on-chip, allowing analytes to be differentiated, but due to the constraints of their fabrication technology the sensors used were significantly less effective than similar non-integrated systems. The capacitive chemical sensor uses a topology with a very large parasitic capacitance that results in a much lower sensitivity than the state of the art in chemicapacitive sensors. The other sensors have significant drawbacks; for instance, the thermocouples in the calorimetric sensor use the aluminum-polysilicon junction available in CMOS rather than junctions with a larger Seebeck effect used in other thermocouples, resulting in a smaller output signal. As a result, their system, with limits of detection demonstrated of 1 to 5 parts per million for the volatile organic carbons tested [11], would not be able to reach the limits of detection necessary for an ESLI system. This work is focused on developing such a system by creating higher sensitivity chemical sensors in a CMOS process.

One promising technology is the use of thiol-coated gold

nanoparticles in chemiresistive sensing. Gold nanoparticle chemiresistors are sensitive to many VOCs, including many common industrial solvents such as toluene and IPA; limits of detection have also been demonstrated under a part per million for some analytes [20]. A highly sensitive capacitive humidity sensor is another important part of an ESLI system, since a respirator cartridge will be exposed to the surrounding environmental conditions.

This work

demonstrates important progress toward a practical end-or-service-life indicator by integrating gold nanoparticle chemiresistor sensors for measuring VOCs and the development of high 6

performance MEMS capacitive chemical sensors for measuring both humidity and other analytes. The contributions of the thesis are summarized in the following section.

1.3 Thesis contributions Demonstration of integration of gold nanoparticle chemiresistors on a CMOS die using patterned platinum electrodes connecting to the tungsten vias in a CMOS process Theoretical analysis and experimental verification that modulating gold nanoparticle chemiresistors to higher frequencies does not improve the limit of detection of the sensors Development of parallel-plate micromachined chemicapacitive sensors with sensitivity equal to the fundamental material limit of the polymer used Investigation of high resistance gold-nanoclusters as chemicapacitive materials Detailed analysis of noise, charge injection and clock feedthrough for a charge-based capacitance measurement (CBCM) circuit in capacitive sensing Design of a high performance capacitive sensing system including a low phase noise Colpitts oscillator to lower the limit of detection of a chemicapacitive sensor

1.4 Thesis outline The thesis is organized as follows. Chapter 2 is a discussion of the gold nanoparticle chemiresistive sensor, including characterization of the sensor, measurement and analysis of modulated sensor testing, and integration onto a CMOS chip. A series of topologies of high sensitivity MEMS capacitive chemical sensors are presented in Chapter 3. Chapter 4 focuses on the chemicapacitive electronics, including the analysis and characterization of the CBCM circuit 7

and the development of the Colpitts oscillator system.

Chapter 5 discusses the gold

nanocluster chemicapacitive material, and a means of further lowering the limit of detection using a preconcentrator are addressed in Chapter 6. Chapter 7 concludes the thesis and presents possibilities for future work.

8

Chapter 2 C8 Nanoparticle Resistive Chemical Sensors 2.1 Introduction Resistive chemical sensors are based on measuring a change in conductivity in a material when chemical molecules are absorbed. Although a large number of different types of materials have been demonstrated as chemiresistors, most are based either on the expansion or deformation in the material when analyte is absorbed, or on the absorbed analyte serving as dopants in a semiconducting material, providing charge carriers to change the material conductivity. Chemiresistors can be used to measure a very wide range of different analytes, depending on the absorption characteristics of the specific material used, including carbon monoxide[21], hydrogen[22], volatile organic carbons ([23]-[24]), water vapor [25], and chemical warfare agents and explosives detection [26].

However, resistive chemical

sensors are typically not selective to a specific analyte, requiring an array containing a large number ofchemiresistors with different materials, such as the array of carbon black chemiresistors in [24] where a series of carbon black composite sensors with different polymer components were used to differentiate between chemicals. In this work, a discussion of the most common methods for creating chemiresistors will be discussed, as well as motivation for selecting the C8 nanoparticle material for use in an ESLI sensor. Several topologies for the sensor will then be discussed, as well as a method for integrating a chemiresistor on a CMOS die.

9

2.2 Prior Work 2.2.1 Metal-oxide sensors Certain semiconducting metal oxides, such as tin oxide (SnO2), zinc oxide (ZnO) and indium oxide (In2O3), when heated to several hundred degrees Celsius, experience a change in resistance when exposed to chemical analytes. This is believed to result from charge carriers donated or released by analyte on the surface of the semiconductor [27]. This effect has been known and been successfully used in macroscale commercial products for decades in applications such as carbon monoxide and natural gas leak detectors [27]. As MEMS and silicon micromachining technology has advanced, there have been successful efforts to miniaturize the sensors, creating structures known as microhotplates, a suspended heater element coated with a thin sensitive film. The primary advantages of a microhotplate based sensor are a reduction in power consumption due to a smaller mass and the possibility of multiple different sensors on a single chip. CMOS-based tin oxide microhotplate sensors were first demonstrated in 1993 [28], with subsequent development of multiple sensor arrays and integrated circuitry such as in 0 and [29]. Since metal oxide sensing is based on surface adsorption of analyte, there have also been efforts to create a higher surface area to volume ratio using nanowires [30]. Although metal oxide sensors are a mature and successful technology, this type of sensor does have several notable disadvantages. Due to the necessity to heat the sensor up to high temperatures, the sensors are relatively high power, even with a microhotplate structure. The devices are also complicated, typically incorporating numerous layers including a heater, membrane, sensor electrodes and sensitive film as in [28].

2.2.2 Intrinsically conducting polymers Although most polymers are insulating, there are some forms of polymers that, when appropriately doped, can act as semiconductors or conductors, with charge carriers able to travel down 10

a polymer chain and hop between different chains [31]. As chemical sensors, the conductivity response results from changes in the conductivity along the chain, the probability of hopping between chains, or in the physical structure of the polymer upon absorption [31], such as a change between amorphous and crystalline structure [32]. Common intrinsically conductive polymers that have been used include polypyrrole [33], polythiophenes [34], and polyaniline [35]. Although there have been commercial products based on conductive polymers [32], this type of sensor also has a number of limitations including high sensitivity to humidity, significant baseline drift, and limited sensor lifetime [31].

2.2.3 Carbon black loaded polymer An insulating polymer filled with a high concentration of conductive carbon black particles can also be used for chemical sensing [36].

Carbon black loaded polymers experience percolation

conduction [32], where above a volume threshold of conductive materials electrons are able to follow a conductive path through the length of the polymer, forming a conductor. Below this threshold, electrons are unable to pass through the material and the material is an insulator [37]. Since the conductivity is based on the volume percentage of the conductive material, the conductivity will drop when analyte is absorbed and the polymer matrix expands. Carbon black sensors are particularly sensitive close to the percolation threshold, the boundary between conductive and insulating behavior [32]. The conduction behavior of a carbon black loaded polymer is based primarily on volume percentage of conductive material, rather than the type of polymer containing the carbon black particles. Different polymers will tend to absorb different amounts of analyte, changing the volume expansion; as a result, the selectivity can be varied by changing the insulating polymer [24]. Arrays of slightly different carbon-black loaded polymers can therefore be used to differentiate between different analytes, as in [24] and [38]. There have been successful commercial products with carbon black, such

11

as [38], although as with intrinsically conducting polymers, carbon black sensors are sensitive to aging related drift and insensitivity to some analytes [31].

2.2.4 Coated nanoparticle chemiresistors Due to the success of carbon black loaded polymers, other materials with conductive particles suspended in a dielectric matrix have also been investigated for chemical sensing.

Conductive

nanoparticles coated in a dielectric film were first demonstrated for chemical sensing in 1998 using octanethiol (SC8H17) monolayers on 2 nm diameter gold nanoparticles [39], a material referred to as C8 after the number of carbon molecules in the octanethiol groups. Figure 2.1 shows a simplified diagram of the material; since the nanoparticles are nanometers in scale, the real particles have a much larger number of thiol groups in the coating.

HH HHHHH H

S-C-C-C-C-C-C-C-C-H

Au core

HH HHHHH H

Au core

Au core

Figure 2.1: C8 gold nanoparticles Other nanoparticle-based materials have since been reported, including gold nanoparticles with varied coating thickness by changing the carbon chain length in the thiol group such as hexanethiol (SC6H13) (denoted C6) [41] and dodecanethiol (SC12H25) (denoted C12) [42]. Sensor arrays including more complex thiol groups to add additional selectivity have also been demonstrated, such as the array 12

in [43] with separate sensors with C8, 4-(phenylethynyl)-benenethiol (DPA), 6-phenoxyhexane-1-thiol (OPH) and methyl-6-mercaptohexanoate (HME) coated gold nanoparticles. The primary conduction mechanism in gold nanoparticle films has been demonstrated to be electrons tunneling between neighboring gold cores [44]. Since the tunneling probability is inversely related to distance, nanoparticles that have thicker coatings have higher resistance; for instance, C12 has a resistivity about 100 times greater than C8, due to the longer thiol chain length and thus core to core spacing [44]. When the film is exposed to chemical analyte, the material expands, resulting in an increase in core to core spacing and an increase in resistance [23]. This effect is the dominant chemical sensing mechanism in most cases, although the change in dielectric constant of the nanoparticle coating upon absorption also changes the activation energy necessary to charge a neighboring particle, and can result in a reduction in resistance for some analytes [23].

2.3 C8 Gold Nanoparticle Sensors 2.3.1 Sensor Fabrication As discussed in the previous section, a wide range of materials have been used successfully for resistive chemical sensing. Past work at CMU focused on semiconducting polythiophenes [34]; however, polythiophenes have very high resistance that in some cases was difficult to measure, and were found to have unacceptably high drift for an ESLI application. As a result, the decision was made to switch to gold nanoparticle based materials, primarily C8 sensors due to the moderate resistivity, sensitivity to a broad range of analytes, and extensive modeling available in the literature. The C8 used was synthesized by Niti Garg and Rongchao Jin from Carnegie Mellon’s chemistry department using a procedure based on the technique from [46]. The C8 gold nanoparticles were found to have a size distribution centered at 3 nm in diameter.

13

Au

SU8 SiO2

Ti Si

(a)

C8

(d)

(b)

(e)

(c)

(f)

Figure 2.2: Chemiresistor process flow (a) initial wafer (b) gold deposition (c) gold patterning (d) bondpad deposition (e) SU-8 patterning (f) deposition of nanoparticles Gold spiral interdigitated electrodes were fabricated on a 1 μm thermal silicon oxide layer in a fabrication process developed by Suresh Santhanam of CMU’s Electrical and Computer Engineering department. Beginning with an oxide coated wafer (Fig. 2.2 (a)), a 2 nm thick titanium adhesion layer followed by a 50 nm thick gold layer is first sputtered on the chip (Fig. 2.2 (b)) at 200 W and 50 W power and 11 seconds and 2 minutes deposition times for the titanium and gold respectively using a Perkin Elmer 6J sputtering system. After patterning a gold wet etch followed by ion milling of the titanium with a Commonwealth Scientific Ion Milling System at 500 V for 1.8 minutes forms the electrode (Fig. 2.2 (c)). A 0.5 μm gold layer deposited at 50 W and 10 minutes deposition time using a Perkin Elmer 6J sputtering system is patterned on the pads using a lift-off process to allow successful wirebonding (Fig. 2.2 (d)). A 50 μm thick layer of SU8 is then patterned to allow reference capping and to minimize the

14

exposure of the wiring to the external environment (Fig. 2.2 (e)). Gold nanoparticles are deposited on the electrode using inkjet deposition in 1,2,4-trichlorobenzene to form the final sensor (Fig. 2.2 (f)). The spiral electrode before and after inkjetting of nanoparticles is shown in Fig. 2.3. The spiral pattern was chosen to better conform to the circular inkjet drop. The outermost four traces are grounded and and the thicker coffee ring deposits occur outside the active electrode area. The intent was to have the variable coffee rings not included as part of the device, which would consist of the more uniform center of the drop; for the thicker coatings used in much of this work, the coffee ring does extend into the active area as shown. The gold nanoparticles were inkjetted by Larry Schultz and Lee Weiss of the Carnegie Mellon Robotics Institute with a 30 μm inkjet nozzle as 16 iterations of 21 drops with a short delay between iterations, giving a total of 224 drops. The film thickness measured using atomic force microscopy (AFM) varies from roughly 500 nm in the outer “coffee ring” deposits down to approximately 100 nm in the thinner center area.

The AFM taken near the outside edge of a

nanoparticle sensor is shown in Fig. 2.4.

(b) (a) Figure 2.3: Spiral electrode (a) before and (b) after inkjetting of nanoparticles

15

Figure 2.4 AFM of the outer ring of an inkjetted device

2.3.2 Sensor Characterization 2.3.2.1 Analyte Response The I-V characteristic of a gold nanoparticle film is modeled as [44]:

I

K (e

V

e

V

)

(2.1)

where I is the current, V the voltage, and K and α are agglomerations of constants given in [44] that incorporate the effects of the core-to-core spacing, available energy states, resistor channel area and activation energy. The I-V curve of the C8 chemiresistor was measured and fitted to the theoretical model with an R2 value of 0.9996 (Fig. 2.5) with K equal to 0.00457 and α equal to 0.08026. The response is linear for voltages up to 12 V, the resistive region that is typically measured, with a significant curvature present for higher voltages. The sensor response for the common solvents toluene and acetone is shown in Fig. 2.6, giving sensitivities of 0.0091% and 0.0013% change in resistance per ppm of toluene and acetone respectively. [23] gives a more detailed study of the response for C8 nanoparticles to a wider set of different analytes. Since the gold nanoparticles are particularly sensitive 16

to toluene, toluene tends to be tested in most C8 sensors in the literature, and can be used to compare the response of this device to other demonstrated sensors. The University of Michigan study in [23] gives a sensitivity of 0.037%/ppm of toluene, while the sensor in the original Naval Research Laboratory paper had a sensitivity of approximately 0.1%/ppm [39]. Although both of these examples have sensitivities much larger than the sensor in this work, by a factor of 4 for the Michigan sensor and a factor of 11 for the NRL sensor, they also vary widely from each other, even though all three sensors are made with nominally identical C8 gold nanoparticles.

One likely reason for the discrepancy is

differences in layer thicknesses and film structure.

In [45], an analysis was performed on

dodecanedithiol (C12), finding that the sensitivity to toluene varied widely by a factor of three or more for different thicknesses of films due to the resulting differences in film structure.

Figure 2.5: I-V characteristic of C8 gold nanoparticle sensor

17

Figure 2.6: Sensor response to toluene and acetone

2.3.2.2 Temperature Sensitivity For a sensor that could be exposed to a wide range of environmental conditions, the temperature response is an important criterion. The temperature response was measured using a temperature controlled heater block in a nitrogen flowstream . The temperature sensitivity of the sensor (Fig. 2.7) appears linear over the temperature range of 20°C to 45°C and the slope is 0.37%/°C. In order to compensate for the temperature sensitivity, a resistive temperature sensor, shown in Fig. 2.8, is included on the sensor chip. The sensor is a resistance temperature detector (RTD) created using a serpentine gold trace; since the resistivity of gold varies linearly over the range of temperatures near room temperature, the gold trace can be used to measure the temperature of the chip. A coating of SU8 is used to eliminate any effects from chemical exposure on the RTD.

Figure 2.7: Temperature response The RTD was characterized by heating the chip to a series of controlled temperatures and measuring the resistance; the sensitivity is shown in Fig. 2.9. The resistance was found to be linear with temperature, with a sensitivity of 15.8 Ω per degree C.

18

Figure 2.8: Resistance temperature detector

Resistance (Ohms)

5500 y = 15.82x + 4832.3

5450 5400 5350 5300 5250

5200 20

25

30

35

40

45

Temperature (degrees C)

Figure 2.9: RTD temperature sensitivity

2.3.2.3 Humidity Response As an end-of-service-life indicator, the C8 gold nanoparticle sensor is primarily to detect solvents in an industry setting. However, as with most forms of chemical sensors, the chemiresistor is not perfectly selective, and will respond to other vapors, called interferents, present in the air as well. For any sensor that is exposed to an uncontrolled environment, such as a respirator cartridge, the most important interferent is water vapor, since water is typically present at concentrations hundreds of times larger than the concentrations of solvent being detected. The humidity response of the sensor was measured

19

by injecting water vapor using the solvent pump setup used for analyte testing; a Honeywell HIH-4000 humidity sensor was used to measure the relative humidity. The response was approximately linear (Fig. 2.10) , with a sensitivity of 0.10% change in resistance per percent relative humidity. Since the sensor could be exposed to a wide range of humidity during the use of a single respirator cartridge, this response must be compensated to isolate the chemical response, requiring a reference humidity sensor

Change in Resistance (%)

to be part of the ESLI system.

5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0

10

20 30 Relative Humidity (%)

40

50

Figure 2.10: Humidity response of C8 nanoparticle sensor

2.3.2.4 Electrical Drift In addition to the responses to environmental factors such as temperature and humidity, the gold nanoparticle chemiresistors were found to experience a significant electrical transient when turned on in air, with a time constant of approximately two minutes. Fig. 2.11 shows a sensor repeatedly turned on with 2 V for 20 minutes then off for 20 minutes. After each turn off period, the transient is clearly apparent.

20

730

Resistance (KΩ)

Resistance (KΩ)

730 720

710 700

720 710

700 690 680

690

40

680 0

20

40 60 Time(Minutes)

80

100

42 44 Time(Minutes)

Figure 2.11: (a) Test of turn-on transient and (b) magnified view of single transient

To better understand the voltage transient, an additional test was run with different voltages applied to the sensor (Fig. 2.12). The transient was found to be heavily voltage dependent, with no obvious transient for applied voltages below 300 mV, and progressively worse for higher voltages. The voltage threshold suggests that an energy barrier must be overcome to trigger the transient effect. One possibility is that the voltage transient results from a chemical reaction with a reactive component of air such as oxygen or water vapor. To test this theory, a further test was run with the sensor in a nitrogen environment (Fig. 2.13). The sensor is again repeatedly turned on with 2 V for 20 minutes then off for 20 minutes. Although some drift is visible in the nitrogen test, likely due to temperature, the turn-on electrical transient is no longer present. In a practical end-of-service life indicator, the sensor would be exposed to air. To minimize the electrical transient, the sensor should be measured at low voltages, or run for several minutes before taking measurements to minimize the transient effect.

21

5

760

4

750

3

740

2

730

1

720

0

710

Voltage(V)

Resistance (kΩ)

770

Resistance Voltage

-1 0

100 200 Time(Minutes)

300

Resistance (Ohms)

Figure 2.12: Voltage dependence test of turn on transient 900 880 860 840 820 800 780 760 740 720 700 0

50

100

150

Time(Minutes)

Figure 2.13: On/off test in nitrogen

2.3.2.5 Noise Characterization An important specification for any chemical sensor system is the limit of detection, or the minimum concentration of analyte that can be successfully differentiated from noise. The limit of detection can be expressed as [47]:

LOD

Rmin S

(2.2)

where ΔRmin is the minimum detectable change in resistance, set by the noise level, and S is the concentration sensitivity in terms of resistance per ppm of analyte. In order to estimate the noise 22

spectrum for a gold nanoparticle sensor, time domain data was first taken with a Keithley 6485 picoammeter to measure the current through the sensor with 1 V across it. The power spectral density of a signal is defined as [40]:

Px ( )

rx (k )e

j k

(2.3)

k

where rx is the autocorrelation function of the signal x, given by the expression rx(k)=E(x(n)x(n+k)) and E is the expected value. Since the signal over all time is not available, the autocorrelation function cannot be determined with perfect precision. Instead, the power spectral density can be estimated from the available data. One common technique used to estimate the power spectral density is the periodogram method. The periodogram estimate is defined as:

Pest ( )

1 N 1 | x ( n )e N 0

j n 2

|

(2.4)

A periodogram estimates the power spectral density by taking a finite windowed function, and performing a discrete fourier transform. The noise spectrum was estimated for a gold nanoparticle sensor using the periodogram function available in MATLAB (Fig. 2.14), and found to experience significant 1/f noise, which has also been demonstrated in the literature ([42],[48]). The expected thermal noise floor of the chemiresistor with 1.6 MΩ resistance is 2.6x10-26 A2/Hz, giving an expected

noise corner of approximately 10 MHz.

23

Power spectral density (A2/Hz)

-10

10

-15

10

-20

10

-25

10

-2

10

0

10 Frequency(Hz)

2

10

Figure 2.14: Noise spectrum of C8 nanoparticle sensor

Due to the 1/f noise characteristic, the C8 chemiresistor has significantly lower noise at higher frequencies. As a result, another group has suggested the possibility that measuring the sensors at higher frequencies would reduce the fundamental limit of detection of the sensor [41]. To test this theory, the setup shown in Fig. 2.15 was built to measure the sensor at higher frequencies. A sinusoidal voltage is applied to the input of an inverting amplifier operational amplifier circuit with gain of approximately one to convert the resistance into an ac voltage signal. A Stanford Research SR830 lock in amplifier is then used to measure the output signal. A lock in amplifier uses a mixer to demodulate the signal and bring it back to DC, where a 1 Hz bandwidth low pass filter removes any interference and noise at other frequencies.

The SR830 has a maximum frequency of 102 kHz, which is significantly

below the corner frequency of the chemiresistor (10 MHz); however, a clear frequency dependence should be visible even at this lower frequency.

24

R1 Rs Vin

ADC

+ Lock In Amplifier

Figure 2.15: Chemiresistor modulation testing circuit

The Allan deviation, which will be discussed in more detail in chapter 4, is a statistical measure of the sample to sample variation in a signal. Since the sample to sample variation sets the smallest change that can be successfully detected, the Allan deviation is a measure of the minimum detectable change in resistance. The Allan deviation for four modulation frequencies for a C8 chemiresistor are shown in Fig. 2.16. No clear frequency dependence was found between the modulation frequency and the limit of detection as specified by the Allan deviation. The flattening at the leftmost part of the curve is characteristic of a 1/f noise spectrum; since this flat portion is not changing significantly, this suggests that the modulation is not shifting the noise spectrum along the 1/f curve.

Allan Deviation (Ω/Ω)

0.001

0.0001 50 Hz 500 Hz 5 kHz

0.00001

50 kHz

0.000001 1

10

100

1000

Integration Time (s)

Figure 2.16: Allan deviation for different modulation frequencies 25

Frequency modulation does not appear to improve the limit of detection of the chemiresistor. Voltage modulation works by assuming that the input noise is of the form: v

(

(2.5)

n , t)

η(ωn,t) is a function with zero mean and 1/f frequency characteristic. For an AC voltage input, the totalvoltage can be expressed as:

Vtotal

vin,mag sin t

(

(2.6)

n , t)

where vin,mag is the magnitude of the input voltage and ω is the modulation angular frequency. When this signal is filtered with an ideal bandpass filter of width dω centered around ω, the resulting output voltage will be:

v filtered

vin,mag sin t

(2.7)

( , t )d

assuming an approximately flat noise characteristic near ω. At high ω, η will be small due to the 1/f noise characteristic, resulting in a large reduction in noise compared with a measurement near DC. For resistive sensors that have an intrinsic 1/f noise that is fundamentally a resistance fluctuation, modulation does not improve the noise performance [49]. In this case, a resistive sensor can be modeled as [49]: R

R0 [1

(

n , t )]

(2.7)

where R0 is the average resistance. The output voltage from the inverting amplifier in Fig. 2.13 will be of the form [49]: Vo

R0 [1

R1 (

n , t )]

(2.8)

vin,mag sin t

Using a Taylor approximation, this output voltage is roughly equal to [49]: vout

R1 vin,mag sin( t )(1 R0

(

n , t ))

R1 vin,mag sin( t ) R0

R1 vin,mag sin( t ) ( R0

26 n , t)

(2.9) The second component of the above equation is the noise component of the signal. Since the noise component is multiplied by the sinusoid at the input frequency, the noise will also be modulated up to the frequency of the input voltage. This effect is shown graphically in Fig. 2.17. For a voltage noise independent of the input (Fig. 2.17(a)), the noise voltage is unchanged when the input voltage is applied (Fig. 2.17 (b)). This means that, when bandpass filtered around the modulation frequency, the noise will be lower than if the input had been at low frequency. For a resistive sensor (Fig. 2.17 (d)), the noise is modulated up to the input frequency (Fig. 2.17(e)), resulting in significant 1/f noise remaining after bandpass filtering (2.17(f)).

vn

Rn 1/f noise

1/f noise

ω

(a)

v

(d)

v Input voltage

ω v

(b)

v

Input voltage

(e)

ω (c)

ω

(f)

ω

ω

Figure 2.17: Modulation of a system (a-c) with input independent voltage noise and (d-f) a resistive sensor whose resistance varies with a 1/f noise relationship

2.3.3 Integrated Nanoparticle Sensors In order to integrate a gold nanoparticle sensor by post-processing on top of a foundry CMOS chip, one option is to use electroless gold plating to form a gold contact layer on top of an aluminum 27

electrode ([50][51]). However, gold cannot be used in most fabrication equipment due to the risk of contamination of other people’s processes, and is therefore restricted in the Carnegie Mellon cleanroom. This risk of contamination makes a gold plated chip unsuited for the additional processing necessary to include other sensor modalities.

To avoid this contamination, a process flow was

developed to create electrodes using platinum, another noble metal (Fig. 2.18). Si SiO2 Photoresist

(a)

(b)

(c)

Al Pt W

C8

(d)

(e)

(f)

Figure 2.18: Chemiresistor process flow: (a) CMOS chip, (b) timed oxide etch, (c) wet aluminum etch, (d) platinum deposition, (e)platinum patterning and (f) nanoparticle deposition The process begins with a standard Jazz 0.35 μm CMOS chip (Fig. 2.18 (a)). An oxide etch is performed down to the third lowest metal layer in the CMOS stack (Fig. 2.18 (b)) using a Plasma-Therm 790 RIE system using CHF3 and O2 at 22.5 and 16 cc/min flow rates respectively and 100 W power. Photoresist is then deposited to protect the bondpads after which an aluminum etch (Transene aluminum etchant A for 5 hours at room temperature) is performed to remove the exposed aluminum (Fig. 2.17 (c)), leaving the TiW adhesion layer. Platinum is then sputtered onto the chip using a Perkin Elmer 6J sputtering system at 50 W for 4 minutes, after which the photoresist is removed; the TiW serves as an adhesion layer for the platinum (Fig. 2.18 (d)). Photoresist is patterned for the electrodes using a Heidelberg direct write laser (DWL) 66 system, and the platinum is ion milled for 10 minutes at 28

500 V, followed by a 3 hour room temperature wet etch using Transene TiW etchant to remove the exposed adhesion layer (Fig. 2.18 (e)). C8 nanoparticles are then dissolved in trichlorobenzene (TCB) and inkjetted onto the electrode (Fig. 2.18 (f)) using the same number of drops and timing used for the non-integrated sensor. Fig. 2.19 shows the integrated chemiresistor before and after inkjet deposition of polymer. To obtain electrical contact with the metal layers in CMOS, a large square of platinum is patterned above a large array of tungsten vias to the lower metal layer; this square is visible in Fig. 2.19 (b).

Square of platinum above array of vias

(a)

(b)

Figure 2.19: Platinum electrode chemiresistor (a) before and (b) after inkjetting

The array of vias is necessary because a large number of vias are damaged during the aluminum etch. The aluminum etch etches aluminum more quickly than the TiW adhesion layer; however, for a long etch the thin TiW has been seen to etch away, exposing the underlying vias to the etch. Fig. 2.20 shows the array after the removal of the aluminum layer. Two different types of vias, white and dark, are apparent in the SEM image. The vias were analyzed using energy-dispersive x-ray spectroscopy (EDX), and the white vias were found to be entirely tungsten, while the dark vias were measured to be primarily silicon dioxide, with a smaller amount of aluminum. This suggests that the tungsten in the darker vias was removed, and the EDX is measuring the underlying silicon and aluminum layers. Since no titanium was seen in the EDX measurements, the TiW layer appears to have been completely etched 29

through over the vias, leaving the tungsten exposed. Based on the image, close to half of the vias are destroyed by the etch; however, due to the large number of vias in the array, electrical contact is successfully made between the chemiresistor and the underlying metal layer. A longer etch would likely destroy additional vias, jeopardizing the electrical contact between the layers.

(a)

(b)

Figure 2.20: (a) Via array after aluminum etch and (b) close up of vias

The five hour aluminum etch time was chosen for compatibility with the etch used for the aluminum etched capacitive sensor discussed in the following chapter. For that sensor a distance of several microns of aluminum has to be etched under oxide to release a structure; for the chemiresistive sensor an etch this long is unnecessary, with only 400 nm of exposed aluminum being etched. The experiment was repeated using a shorter aluminum etch of 15 minutes to verify that the aluminum etch was attacking the tungsten vias; the via array after this shorter etch was examined (Fig. 2.21), and no vias were found to be visibly damaged.

30

(a)

(b)

Figure 2.21: (a) Via array after aluminum etch and (b) close up of vias for a shorter etch time For comparison, a sensor was created on a CMOS chip by inkjet deposition directly on an aluminum metal electrode. An octagonal spiral electrode shape was used, with a square in a higher layer metal used to contain the deposited polymer. The sensor after deposition of polymer is shown in Fig. 2.22. The responses to toluene vapor with C8 gold nanoparticles for non-integrated and integrated platinum electrodes and the aluminum electrode sensor were measured and compared (Fig. 2.23). The two platinum electrodes have similar sensitivities of 0.009%/ppm, while the aluminum electrode had a much lower sensitivity of 0.0036%/ppm.

31

Figure 2.22: Aluminum electrode sensor

Figure 2.23: Sensor response to toluene

Since there is no significant drop in sensitivity in the integrated chemiresistor, this integration technique would be a practical method of incorporating a gold nanoparticle chemiresistor onto a CMOS die. The fabrication technique, based on an anisotropic oxide etch followed by a wet etch of aluminum, is similar to that used for the vertical parallel plate capacitive sensor discussed in chapter 3, suggesting that the two sensors could be successfully combined into a single sensor system with several shared processing steps.

2.4 Discussion A C8 gold nanoparticle chemiresistive sensor intended to be inserted in a respirator cartridge as an ESLI was characterized, with a focus on the environmental drift and noise performance. Modulating 32

the sensor input voltage in an attempt to lower the limit of detection was also investigated, with no improvement in the noise performance found for the sensor. A theoretical analysis was also performed demonstrating that modulation is not an effective way of removing 1/f noise for a resistive sensor since the noise is modulated to a higher frequency by the input voltage. A fabrication method was then developed for integrating a chemiresistive sensor onto platinum electrodes patterned on a CMOS die. The next chapter will focus on creating a highly sensitive chemicapacitive sensor to allow the effects of humidity, one of the major environmental chapters discussed, to be removed from the chemiresistive response.

33

Chapter 3 CMOS-MEMS Capacitive Chemical Sensors 3.1 Introduction Capacitive chemical sensing measures the change in capacitance between two electrodes when a chemical analyte is absorbed. Capacitive sensors are typically made using a low dielectric constant polymer layer; when a high dielectric constant analyte is absorbed into the layer, there will be a large increase in capacitance. Since water vapor has a high dielectric constant and is present in high concentrations in most environmental conditions, the humidity response tends to dominate. As a result, capacitive chemical sensors are primarily used for humidity sensing, and this type of sensor is used in the majority of commercial humidity sensors [52]. In this chapter, past techniques for fabricating capacitive chemical sensors and integrating the sensors with testing electronics will be discussed. Several improved integrated chemical sensor topologies will then be presented, with related modeling and experimental results.

34

3.2 Prior Work 3.2.1 Sensor topologies There are two primary categories of capacitive chemical sensors, vertical parallel-plate and planar interdigitated sensors. A vertical parallel-plate sensor consists of an absorbent layer, typically polymer, sandwiched between two electrode layers (Fig. 3.1(a)). In an interdigitated sensor, the electrodes are typically planar, and the measured capacitance is measured laterally across the gaps (Fig. 3.1(b)).

Electrode 1

Electrode 2 (a)

Metal

Dielectric

Absorbent Layer

Substrate

Electrode 1

Electrode 2 (b)

Figure 3.1 (a) Vertical parallel-plate and (b) interdigitated capacitive chemical sensors In a vertical parallel-plate sensor, the top electrode is typically patterned as shown to allow analyte to absorb into the sensor ([53], [54]), although very thin or porous electrodes permeable to analyte can also be used ([55], [56]). There have also been demonstrations of sensors where both electrodes are permeable, allowing analyte to absorb into the material from the top and bottom ([55], [57]. This type of sensor gives the highest fractional sensitivity, or percent change in capacitance, for a given absorbent, since the absorbent material is measured directly with minimal parasitic capacitance in parallel. However, vertical parallel35

plate sensors do have several disadvantages. Response time is typically relatively long for a non-permeable electrode, since analyte must diffuse underneath the electrode to reach the measured area. The diffusion length can be minimized by reducing the electrode width, but this width is typically limited by the lithography used. A standard vertical parallel-plate sensor also requires a layer of metal to be patterned after the deposition of the absorption layer, typically polymer. Since polymer is easily etched and cannot withstand higher temperatures, this requirement complicates the fabrication process, particularly if a final release etch must be performed for other MEMS devices on the same chip. The fabrication of an interdigitated sensor eliminates this problem by depositing the polymer layer as the final step. This type of sensor can be made using a simpler two step process, the deposition and patterning of a metal electrode followed by the deposition of polymer ([58], [59]). An interdigitated sensor is also significantly faster than a vertical parallelplate sensor, since analyte can absorb directly into the polymer layer. For a nanometer-scale film thickness, the response can be almost instantaneous. However, the electrodes of standard interdigitated sensors such as those in [58] and [59] rest on a non-absorbent dielectric substrate, typically silicon dioxide, resulting in a parasitic capacitance directly in parallel to the capacitance through the polymer layer, resulting in a total capacitance given by:

Csensor

C poly

Cox

(3.1)

where Cpoly is the capacitance through the polymer and Cox is the parallel capacitance through the oxide. Only the polymer capacitance will change when the sensor is exposed to analyte, resulting in a fractional sensitivity given by: 36

C sensor C sensor

C poly C poly

(3.2)

C ox

The parallel oxide capacitance results in a significant degradation of the fractional sensitivity, since the oxide capacitance can be as large as or larger than the polymer capacitance, particularly for a very thin sensitive layer.

Aluminum

SiO2

Polymer

Si

Electrode 2

Electrode 1 (a)

Electrode 2

Electrode 2

Electrode 1 Electrode 1

(b)

(c)

Figure 3.2 (a) Planar ETH Zurich sensor [60] (b) structured sensor [61] and (c) oxide etched sensor [63]

3.2.2 Integrated sensors Past integrated capacitive chemical sensors have been interdigitated designs to simplify fabrication and allow the use of the metal layers already available in the CMOS stack. An extensive study of integrated sensors was performed at ETH Zurich in the late 1980s and early 1990s. The initial ETH Zurich design [60] (Fig. 3.2(a)) used a single electrode layer formed from the top metal electrode exposed by the pad etch in a standard CMOS process. This layer is then coated with polymer which is patterned using standard lithography. By minimizing CMOS postprocessing to a single deposition and patterning step, the group was able to minimize sensor cost. However, the oxide remaining directly between the electrodes forms an even larger parallel oxide capacitance than the standard interdigitated design. Later work from ETH Zurich 37

[61] reduced this effect slightly by creating a multi-layer set of electrodes in a non-planarized CMOS process (Fig. 3.2 (b)). The integrated humidity sensors from the company Sensirion are based on this technology. The stacked electrode creates a gap between the top layer and bottom layer electrodes that can be filled with polymer, increasing the polymer capacitance. A large parallel capacitance remains with this design; according to [62], the ETH Zurich sensor has a sensing capacitance of 1.4 pF in parallel with a parasitic oxide capacitance of 6.4 pF. Since only 18% of the measured capacitance is through the polymer, the sensitivity would be 18% of that of the polymer used. The alternative sensor topology shown in Fig. 3.2 (c) has also been successfully demonstrated [63]. After CMOS fabrication, lithography is performed for an anistropic oxide etch to remove the oxide directly between the electrodes. The polymer is then patterned to form the final sensor. Although this sensor requires more CMOS post processing than the ETH Zurich designs, the parallel oxide capacitance is reduced to only the capacitance vertically down to the substrate, resulting in a higher sensitivity.

3.3 Released Interdigitated Sensor 3.3.1 Sensor Fabrication A method for improving the sensitivity of an integrated capacitive chemical sensor was developed [64]. By etching away the oxide and the underlying substrate, forming the structure shown in Fig. 3.3, the parasitic parallel capacitance can be limited to only the air capacitance above and below the structure. The structure partly inspired by a structure from Seacoast Science Inc. [66] that uses short posts in the Multi-User MEMS process (MUMPs) to elevate the 38

Electrode 1 Electrode 2

SiO2

Al Polymer

Figure 3.3 Released interdigitated sensor electrodes a short distance above the silicon substrate. This work demonstrates the first successful integration of a released interdigitated sensor with CMOS electronics.

Al Si

CMOS electronics

(a)

(c)

(b)

(d)

SiO2 Polymer

Figure 3.4 CMOS-MEMS fabrication process The sensor was fabricated using on the Carnegie Mellon CMOS-MEMS fabrication process [65], which uses the metal layers available in the CMOS stack as etch masks for the creation of MEMS structures. As a result, the CMOS-MEMS process requires no lithography after the CMOS process. The CMOS-MEMS process begins with a standard CMOS chip (Fig. 3.4 (a)). An anisotropic oxide is etch is performed down to the silicon substrate (Fig. 3.4 (b)) while 39

the metal layers protect the CMOS circuitry. A 50 μm anistropic silicon etch is then performed, followed by a short isotropic silicon etch to release the structures (Fig. 3.4 (c)). Finally polymer is deposited in solution into the structure using inkjet deposition (Fig. 3.4(d)).

3.3.2 Capacitive wicking channel sensor 3.3.2.1 Device Past work at Carnegie Mellon developed a technique for depositing polymer into a MEMS device using capillary wicking ([67], [68]). Instead of depositing directly on the MEMS device, the inkjet drops were deposited in an anchored cavity beside the device called an inkjet well. For humidity sensing, the polymer used was a formulation of polyimide in solution (HD Microsystems PI 2556) that was further diluted by a factor of 24:1 with a 50:50 solvent mixture of n-methyl-2-pyrrolidone and methoxy propanol (HD Microsystems T-9039) to obtain a low enough viscosity to deposit with the inkjet. Capillary forces then pull the solution into a narrow wicking channel in the MEMS device, filling the device with polymer. The capacitance between the sides of the wicking channel forms the capacitive chemical sensor. Fig. 3.5 shows a scanning electron microscope (SEM) image of a device consisting of a single wicking channel; Fig. 3.6 shows the wicking channel before and after deposition of polymer. The capacitor is created using the lowest three metal layers in the CMOS stack. The inkjet well is a created by surrounding an area consisting of the lowest metal layer in the CMOS stack with a wall of the top metal layer. The grooves in the well are intended both to guide the solution to the wicking channel and to slot the metal layer forming the well. To prevent the two sides of the wicking

40

channel from pulling together from surface tension, periodic trusses hold the electrodes apart as shown in Fig. 3.6 (a). Capacitor (wicking channel)

Well (inkjet target)

Figure 3.5: SEM of single channel capacitive humidity sensor

Polymer in channel

Open Truss to hold channel electrodes apart (b) (a) Figure 3.6: SEM of wicking channel before (a) and after (b) polymer deposition

3.3.2.2 Device Modeling Polyimide was chosen as a sensing material for the capacitive sensor. Polyimide is commonly used for humidity sensing due to its linearity, stability and compatibility with other 41

fabrication processes [69]. For this work, it also allows comparisons with numerous other sensors in the literature. As a common sensing material, its behavior upon exposure to water vapor has also been extensively studied. A finite element model of the device at 0% relative humidity was made using the computer program COMSOL [70]. A dielectric constant of 3 was used for polyimide in dry air [60]. The simulated capacitance of the single channel device is shown in Table 3.1, with the results measured from the sensor. The simulated results were higher than those measured from the device. This model fails to account for a number of second order effects such as the thin layer of sidewall polymer from the CMOS-MEMS processing and the precise geometry of the electrode sidewalls [71]. After refining the simulation to add 150 nm thick sidewall polymer and with a more accurate gap measurement, the sensor was re-simulated, giving capacitances closer to the measured values for the sensor. First Order Simulation

Refined Simulation

Measured

Pre-deposition Capacitance

3.94 fF

4.93 fF

6.15 fF

Post-deposition Capacitance

8.25 fF

9.09 fF

10.3 fF

Table3.1: 3.1:Simulated Simulatedand andMeasured measuredCapacitance capacitanceofofSensor sensor Table

A theoretical model was created to model the sensor behavior upon water vapor absorption. The device was modeled as a parallel-plate capacitor using the expression: C

p

0

d

A

(3.3) 42

where εp and ε0 are the dielectric constants of polyimide and vacuum, A is the area of the capacitor and d is the gap between the electrodes. A fixed capacitance value was added to this value to account for the capacitance through the air above and below polymer; this value was fitted from the measured capacitance value in dry air. The dielectric constant of water vapor in polyimide can be modeled using the semi-empirical Looyenga’s equation [52]:

p

[ (

1/ 3 H 2O

1/ 3 p0

)

1/ 3 3 p0

]

(3.4)

where γ is the volume fraction of water in the film, and εH20 and εp0 are the dielectric constants of water and dry polyimide respectively. Looyenga’s model is based on the assumption that the water vapor absorbed into the polymer acts as a finely dispersed mixture of isotropic particles with identical permittivity [72]. Polyimide has a volume fraction of water approximately given by [54]:

c1 (% RH ) c2

(3.5)

where c1 and c2 are temperature dependent constants that were fitted and %RH is the percent relative humidity. Although polyimide expansion upon water vapor absorption also results in a change in capacitance, this expansion is disregarded because the sides of the channel are fixed by trusses and the volume expansion coefficient for polyimide is relatively small, only 60-75 ppm/%RH [73]. The capacitance of the sensor was measured for different relative humidity values and compared to the theoretical model (Fig. 3.7), matching the model closely. The sensitivity of the sensor was measured to be 0.16% change in capacitance per percent relative humidity. 43

Figure 3.7 Humidity response of single channel device

Channel Inlet

Wicking Channels

Inkjet Well

Electrical Routing Figure 3.8 Five channel sensor after inkjetting For many applications a larger total capacitance is more desirable to create a larger total change in capacitance relative to the noise floor of the measuring instrument or interface electronics. One method for filling a larger device is to use several wicking channels filled from the same inkjet well. These channels are then electrically connected in parallel to obtain a larger capacitance. Fig. 3.8 shows a 5 wicking channel device made using this method. The electrical routing between the parallel capacitive beams was placed over the etch pit to avoid 44

blocking the inlets to the wicking channels. The humidity response for the five channel device was measured and compared to the theoretical model (Fig. 3.9). The capacitance of the fivechannel device is roughly seven times that of the single-channel device, possibly due to variations in the polymer fill. The fitted sensitivity of the device is 0.16% change in capacitance per percent relative humidity. As expected, the one channel and five channel devices have comparable sensitivities.

Figure 3.9 Humidity response of five channel device

3.3.2.3 Response Time Another important advantage of a released interdigitated sensor is that the top and bottom of the polymer layer are exposed to the air, allowing analyte to enter from both sides of the device. As a result, this type of sensor will react more quickly to changes in analyte concentration than a non-released sensor. At low concentrations, water absorption into a polymer layer has been demonstrated to be governed by diffusion of water molecules into the material[74]. As a result, the absorption of water vapor into polymer can be modeled using Fick’s second law of diffusion[74]:

45

c t

D

2

(3.6)

c

where c is the concentration of water vapor and D is the diffusion coefficient of water vapor for the specific polymer. The diffusion coefficient for water vapor in polyimide is 15x10-14 m2/s [75]. The response times to a 3% step in water vapor concentration for the sensor and a comparable unreleased structure were simulated using finite element modeling (Fig. 3.10).

(a)

Average Concentration (mol/m 3 )

Released

Unreleased

0.045 0.040 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000

0

50

100

Time (s)

(b) (c) Fig. 3.10 Simulated transient response of water vapor diffusion in (a) an unreleased and (b) a released structure and (c) average concentration in each device

46

The average concentration in each device is shown in Fig. 3.10 (c). As shown, the released sensor is able to respond more quickly to the step in concentration, with a time constant of approximately 12 seconds compared with a time constant of over 40 seconds for the unreleased sensor.

3.3.3 Direct Coated Sensor 3.3.3.1 Design To obtain even larger capacitance values, a structure consisting of 87 parallel wicking channels was designed (Fig. 3.11). However, when polyimide in solution was deposited the liquid was not contained within the inkjet well, since the gap between the sides of the inkjet well (350 μm) was much larger than the size of the inkjet drops. Instead, the solution spread over the capacitor, coating the electrodes. Since the electrodes for this device are anchored at both ends, the sensor was also found to be mechanically robust enough to survive direct inkjet drops.

47

Electrodes

Well (a)

(b)

Figure 3.11 Eighty-seven channel coated device (a) before and (b) after inkjetting of polyimide This structure provides several important advantages over sensor made with polymer wicking. With a capacitor filled using capillary wicking, polymer only fills the gap directly between the electrodes, resulting in parallel capacitance through the air above and below the polymer. When coated, polymer will wick down between the beams, as with the capillary channel; however, a significant thickness of polymer can also be left on the top surface of the device, minimizing any parallel capacitance above the electrodes. This reduction in parallel capacitance will result in a higher sensitivity. A coated structure also removes the need to devote area for a large well structure, allowing a larger total capacitance per area than a comparable wicked structure.

48

The humidity response of the 87 channel coated structure was measured, and compared to that of the 5 channel wicked device (Fig. 3.12). The coated sensor was found to have a slightly larger humidity sensitivity (0.18% change in capacitance per %RH), although this difference is within the measurement error of the reference humidity sensor used.

Figure 3.12 Humidity responses of 87 channel coated and 5 channel wicked devices

3.3.3.2 Thermal Characterization In the released interdigitated structure, the polymer is less constrained than in a traditional interdigitated structure. Expansion of the polymer could result in a larger thermal response. To investigate this possibility, the temperature response of the 87 channel coated sensor was measured by using a tape heater to heat the chip (Fig 3.13). A thermocouple attached to the package with thermal grease was used to monitor the temperature. The test was run in nitrogen to remove the effect of water vapor desorbing from the polymer as the temperature increases. capacitance per °C.

The measured sensitivity to temperature is 0.07% change in

Since any practical chemical sensor system would likely require a

temperature sensor due to the thermal response of most chemical sensors, this effect could be compensated. 49

Figure 3.13 Temperature response of coated sensor

3.3.3.2 Cross Sensitivity As part of a sensing system for chemicals such as industrial solvents, the humidity sensor will be exposed to significant concentrations of these solvents. The sensitivity of the sensor to those chemicals is thus an important specification of the sensor. The change in capacitance of the large coated sensor was measured for several different concentrations of four organic solvents; the results are shown in Fig. 3.14. Table 3.2 contains the measured sensitivities for the four solvents, as well as their dielectric constants and vapor pressures. The highest solvent sensitivity of the humidity sensor is to toluene; since toluene has a dielectric constant of less than polyimide, this implies that swelling rather than dielectric constant change is the dominant mechanism for toluene. Although swelling that causes expansion of the plate would cause a reduction in capacitance, swelling that results in a more complete fill of the gap with polymer would result in a rise in capacitance.

50

Figure 3.14 Coated sensor response to industrial solvents

Dielectric Constant Acetone Toluene Ethanol IPA

20.7 [77] 2.3 [62] 24.3 [62] 19.92 [80]

Vapor Pressure (Pa) (calculated at 300 K from [76]) 33272 4180 6755 8812

Sensitivity (%/%) 0.9 2.4 1.3 0.7

Dipole Moment (Debyes) 2.9[78] 0.53[78] 1.7[78] 1.6[79]

Molecular Mass (g/mol) 58.08 92.14 46.07 60.01

Table 3.2 Material properties and sensitivity of chemicals in cross sensitivity test

3.3.4 Encapsulated Electrode Sensor 3.3.4.1 Design As the coated sensor demonstrated, an improvement in sensitivity can be obtained by coating the top surface of the capacitive sensor in addition to filling in the gap directly between the electrodes. Although coating the sensor removes the parasitic capacitance above the sensor, there will be capacitance through the air below the sensor that will lower the sensitivity of the structure. If the electrodes can be completely surrounded by polyimide, the sensitivity can be further increased. This approach has been successfully demonstrated in the nonintegrated sensor in [82]. In that sensor, a layer of polyimide was first spin-coated on the substrate, followed by the patterning of an aluminum electrode layer. Two further layers of polyimide were then deposited above the electrode layer, resulting in electrodes completely surrounded by polyimide. This fabrication technique is not ideal for an integrated sensor; as with a standard vertical parallel-plate sensor, it requires a layer of electrodes to be patterned above a polymer layer rather than taking advantage of the metal layers already available in the CMOS stack.

51

Inkjet hole

Al

SiO2

Si

Polymer

(a)

(b) Figure 3.15 Encapsulated sensor cross section (a) before and (b) after jetting of polymer

To create a similar structure on a CMOS chip, it is necessary to be able to deposit polymer on the backside of the electrode. One possible method would be to etch away the backside completely and deposit inkjet drops from the back as well as the front of the chip. This method would require a significant amount of additional processing, and would also result in a large hole through the chip that might complicate packaging. An alternative possibility is to fill the silicon etch hole left by the CMOS-MEMS process with polymer in solution until it completely fills and rises above the electrode layer. Fig. 3.15 shows the cross section of the proposed device before and after jetting of polymer. A large hole big enough to incorporate an inkjet drop must be left beside the sensor electrodes as shown in Fig. 3.15 (a). The five channel wicking topology from Fig. 3.8 was chosen for the encapsulated device because the large open areas can be used for inkjet drops and the coating of the device can be easily observed. Prior to

52

encapsulation, polymer was wicked into the channel to ensure polymer in the most direct path between the electrodes. The effects of encapsulating the sensor in polyimide were simulated in COMSOL and compared to similar simulations for both a sensor with polyimide on the top surface, and a sensor with polyimide only directly between the electrodes (Fig. 3.16).

Based on the

simulation, the encapsulated sensor is expected to have a sensitivity 27% higher than that of the wicked sensor with polyimide directly between the electrodes, and 7% higher than the

Change in Capacitance (%)

coated sensor.

18.00 16.00 14.00 12.00 10.00 8.00 6.00 4.00 2.00 0.00

Polyimide Only Between Electrodes Added Polyimide Above Sensor Encapsulated

0

20

40

60

80

100

Relative Humidity (%)

Figure 3.16 COMSOL simulation of encapsulated sensor

Fig. 3.17 shows the device after the cavity below the device was repeatedly filled to overflow with inkjetted polymer in solution. The device was not completely encapsulated, due to the low concentration of polymer in the inkjetted solution that is required to successfully inkjet. To ensure encapsulation, higher concentration drops were added using a syringe; the

53

result is shown in Fig. 3.18.

All five channels were clearly heavily coated by polymer after

syringe deposition.

(a)

(b)

Figure 3.17 (a) Encapsulated device after inkjetting of polymer and (b) magnified view of one of the capacitive channels

(a)

(b)

Figure 3.18 (a) Encapsulated device after syringe deposition of polymer and (b) magnified view of one of the capacitive channels

3.3.4.2 Experimental Results The humidity sensitivity of the encapsulated sensor was measured (Fig. 3.19). The measured sensitivity was 0.21% change in capacitance per %RH, a 40% and 17% improvement in 54

sensitivity over the wicked and coated sensors respectively. However, the electrical connection to the encapsulated sensor was irrevocably broken above 40% relative humidity. Since the polyimide underneath the device expands when water vapor absorbs into the material, the capacitive beams experience a large strain, resulting in the destruction of the device. The stresses in the sensor were simulated using COMSOL, and found to reach a maximum stress of approximately 230 MPa for 40% relative humidity. The yield stress of thin film aluminum has been found to be thickness dependent, varying between 200 MPa for a 0.5 μm film and 136 MPa for a 1.0 μm film [83]. Since the metal layers in the device are approximately 600 nm thick, the yield stress is expected to fall in that range. The sensor could be redesigned to lower the stresses on the anchor to make the design more practical. One possible method would be to have the capacitor suspended on flexible springs some distance from the anchor, allowing the capacitor beams to deflect. Another possibility would be to reduce the etch depth of the silicon; a thinner layer of polyimide underneath the device than the 50 μm currently used would lower the stress on the anchor and is likely why this problem did not happen to the device in [82]. The beam could also be widened near the anchor to lower the stress in the beam at that point.

55

10

%Change in Capacitance

9

8 7 6 5 4 3 2 1 0 0

5

10

15

20

25

30

35

40

%RH

Figure 3.19 Humidity response of encapsulated sensor

3.4 Integrated Vertical Parallel-Plate Sensor 3.4.1 Device Although an interdigitated sensor can approach the sensitivity of the polymer using the methods discussed in the previous section, a vertical parallel-plate sensor is fundamentally more sensitive because it measures only the polymer layer without any significant parallel parasitic capacitances.. Since there are no parasitic capacitances in parallel, the sensitivity of a vertical parallel-plate sensor should thus approximately equal that of the polymer used. The fabrication process for a vertical parallel-plate sensor commonly consists of depositing metal for the bottom electrode, spin-coating a polyimide layer, followed by depositing and patterning the top metal layer (for instance, in [53], [54], and [84]). The polymer deposition would be a serious problem if the device was to be integrated with other released structures. Spin coating polymer after structure release might destroy the other structures on the chip, and depositing polymer before release might cause the polymer to be etched away or destroyed by high temperatures during the release etch. 56

There is however an alternative method that has been demonstrated for creating a vertical parallel-plate sensor. Seacoast Science Inc. successfully created a vertical parallel-plate sensor using the MUMPs technology [85]. In the MUMPs process, a sacrificial oxide layer is used to create a vertical gap between two polysilicon layers. Seacoast then deposited polymer in solution using inkjet printing through a hole in the center of the structure, with capillary forces pulling polymer underneath the capacitor plates. A similar device can be created on a CMOS chip by etching away one of the layers and filling the resulting gap with polymer. To create a similar structure in CMOS, a conductive layer must be obtained on either side of a vertical gap. One possibility is to etch away one of the metal layers (metal 2 in Fig. 3.20 (a)) and measure the capacitance between the metal layers above and below (metal layers 3 and 1). In this technique, the series oxide capacitances above and below the polymer layer would result in a significant degradation in sensitivity. However, in most CMOS processes the metal layers consist of adhesion layers above and below a core metal layer. For example, the CMOS process used in this work has an aluminum layer between two TiW adhesion layers. By using an etch

Metal

SiO2

Si

Polymer

Metal 3 Adhesion layers

Metal 2 (etched) Metal 1

57

(a)

(b)

Figure 3.20 Techniques for creating a vertical parallel-plate sensor in CMOS

that selectively attacks the aluminum and leaves the TiW layers intact, these layers can be used as the top and bottom electrodes of the vertical parallel-plate sensor (Fig. 3.20 (b)).

3.4.2 Fabrication A fabrication process was developed to integrate a vertical parallel-plate sensor in CMOS [86]. Beginning with a standard CMOS chip (Fig. 3.21 (a)), an anisotropic oxide etch is performed down to the second lowest metal layer (Fig. 3.21 (b)). The aluminum part of this layer is then wet etched using the etchant Transene Aluminum Etchant Type A, leaving the TiW layers intact (Fig. 3.21 (c)). Photoresist is used to protect the aluminum bondpads and the topmost metal layer is used as a mask to protect the first and third metal layers from the aluminum etch. The chip is then rinsed first in de-ionized (DI) water, followed by successive rinses in acetone and methanol. These rinses in low surface tension solvents are necessary to avoid surface tension during the final drying step pulling the cavity closed. Polymer is then inkjetted into the structure through an access hole in the center of the structure, with capillary forces pulling the polymer solution between the two capacitor plates (Fig. 3.21 (d)).

58

Metal

SiO2

Si

Polymer

(a)

(c)

(b)

(d)

Figure 3.21 Integrated vertical parallel plate sensor fabrication process A 300 μm diameter circular structure was fabricated using the process from Fig. 3.20. An 80 μm hole in the center serves as a target for drops from a 60 μm inkjet nozzle. Smaller holes serve both as release holes for the aluminum etch and as access holes for water vapor to absorb into the polyimide layer. By using gaps in the etched metal layer, oxide pillars were created around the hole at the center of the device, holding the top and bottom plates apart. Fig. 3.22 shows two simplified cross sections of the device. In this design, the top plate is anchored at the outer edge of the sensor, as well as with four pillars near the center of the device as shown in Fig. 3.22 (c)

59

Metal Si

Hole for inkjet Release holes deposition

SiO2 Polymer

Adhesion layers Oxide Pillars

A

(a) Oxide pillars

B

A’

Anchored outer edge

A

B

(b)

B’

A’

B’

(c) Figure 3.22 (a) Cross section of sensor (b) cross section that passes through oxide pillars and (c) top view SEM images of the sensor before and after inkjetting are shown in Fig. 3.23. Since the polymer fill could not be seen optically during jetting, the sensor was deliberately overloaded, resulting in a polymer residue on the top surface of the device. Although a reproducibility study was not performed for this device, [85] presents a study of the comparable MUMPS sensor, testing the capacitance of 24 sensors on 6 different chips (4 sensors per chip) filled with siloxanefluoro alcohol (SXFA). The variation on a single chip averaged 3.2% at 20°C, and chip-to-chip variation averaged 9% at 20°C.

60

Unfilled hole (a)

Filled hole

(b) Fig. 3.23: SEM image of sensor (a) before and (b) after inkjetting of polyimide with inset picture of filled release holes

3.4.3 Experimental Results 3.4.3.1 Humidity Response The response of the vertical parallel-plate sensor was measured from 0% to 40% relative humidity (Fig. 3.24) using the flow system presented in [64]. The theoretical model for polyimide was fit to the measured data with an adjusted R2 value of 0.9964. The sensitivity in the linear region was 0.31% change in capacitance per percent relative humidity. A second sensor was tested using a Miller-Nelson HCS-401 flow-temperature-humidity control system over the range from 30% to 75% relative humidity, giving a comparable linear response with a sensitivity of 0.30%/%RH. Table 3.3 summarizes the sensitivities of the different sensor topologies discussed in the preceding sections, as well as the ETH 61

Zurich sensor and a non-integrated polyimide vertical parallel plate sensor, showing a clear increase in sensitivity over the previous integrated topologies. The integrated vertical parallel plate sensor had a comparable sensitivity to that of the comparable non-integrated sensor.

Fig. 3.24: Humidity response of vertical parallel-plate sensor Sensitivity (% Change/%RH) Non-Integrated Vertical Parallel-Plate Sensor(Dokmeci and Najafi, UMichigan) [53]

0.31

ETH Zurich Integrated Sensor (Sensirion) [62]

0.048

CMU Released Interdigitated Sensor (Wicked)

0.15

CMU Released Interdigitated Sensor (Coated)

0.18

CMU Encapsulated Sensor CMU Integrated Vertical Parallel-Plate Sensor

0.21 0.31

Table 3.3 Sensitivities for different capacitive sensor topologies

3.4.3.2 Response Time The response time of a vertical parallel plate sensor is generally longer than that of an interdigitated sensor, due to the need for analyte to diffuse a significant length underneath the electrodes. For this sensor, this length is particularly long because the analyte must diffuse all the way from the top of the CMOS stack before reaching the active electrodes. Any additional residue left on the 62

top surface of the electrode will increase this length even further.

In order to study this effect, the

concentration directly between the plates of the sensor was simulated in COMSOL (Fig. 3.25(a)) for a 3% step in relative humidity using the diffusion model from section 3.3.2.33.3.2.3 Response Time. Polymer was assumed to reach to the top of the release holes, as shown in Fig. 3.25(b). The simulated response time constant for the sensor was 68 seconds.

(a)

(b)

Fig. 3.25: (a) Simulated response time of sensor and (b) diagram of response model A test was run with low concentration pulses to determine the actual response time of the sensor. Fig. 3.26 shows the absorption and desorption responses to a pulse of water vapor, along with the response of a reference Honeywell HIH-4000 humidity sensor. Unlike the simulated response, the input humidity pulse is not a sudden step; the flow system used includes a 250 mL mixing volume and 32 inch length of quarter inch tubing, a total volume of tubing equal to 25.7 mL, that will result in a more gradual change in addition to a delay of approximately 16 seconds before the plug of analyte enters the test box. The response time can however be compared to the reference sensor, which has a time constant to changes in relative humidity of 15 s [87].

63

The measured rising and falling time constants of the measured sensor were both 3.16 minutes, while the reference sensor had a rising time constant of 2.25 minutes and a falling time constant of 2.0 minutes. The response time constant of the integrated sensor is 55 s longer than the response of the reference sensor for a rising pulse and 70 s longer for a falling pulse, compared with a simulated response that is 53 s longer (the 68 s response time constant for the sensor simulation compared with a response of 15 s from the datasheet for the Honeywell sensor). Polymer residue on the top surface of the device, which was not included in the simulation, could account for a slightly longer measured response time.

(a)

(b)

Fig. 3.26 (a) Absorption and (b) desorption transient measured from sensor

3.4.3.3 Thermal Characterization The temperature response of the integrated vertical parallel-plate sensor was measured using the experimental setup discussed in 3.3.3.2. Fig. 3.27 shows the temperature response of the sensor; the temperature sensitivity was measured to be 0.19%/°C. This sensitivity is large, 2.7 times larger than the earlier released interdigitated sensor with the same polymer.

64

Fig. 3.27 Temperature response of vertical parallel plate sensor

A possible reason for the higher than expected thermal sensitivity is that too few oxide pillars were used to hold the plates apart during the polymer deposition, resulting in the top plate of the capacitor deflecting downward during polymer deposition due to surface tension. Fig. 3.28 shows two focused ion beam (FIB) cross sections, one close to the anchored edge of the plate (Fig. 3.28 (a)) and one far from the anchors, near the point of maximum displacement (Fig. 3.28 (b)). Far from the anchors, the plates come close to contact; based on the capacitance value, the average gap is 191 nm, compared with an expected gap of 450 nm if the plates were properly held apart. Since the structure is held apart at the anchors, the gap likely closes to significantly less than 191 nm far from the anchors, likely a few tens of nanometers. Since the center of the top electrode is pulled down during the polymer deposition, the different rates of thermal expansion of the materials in the plate result in the top plate deflecting downward. Fig. 3.29 shows the deflection for different amounts of polymer fill. If the capacitor is completely filled with polymer, the polymer will resist the deflection, resulting in little change in gap. However, if some of the space between the electrodes is filled with air, due to voids or an incomplete fill, the top plate can deflect. Although the sensitivity and total capacitance value suggest that the gap is mostly filled, it is 65

possible that the top and bottom electrode are only heavily coated in a significant portion of the capacitor. A higher pillar density will hold the gap apart, reducing the thermal sensitivity.

M3

M3

Etched M2

Etched M2

M1

M1

(a)

M3

M3 Etched M2

M2

M1

M1

(b)

Fig. 3.28: FIB cross sections (a) near anchor and (b) at point of maximum displacement

Fig. 3.29 Simulated vertical deflection for different fill percentages; 0 μm is at the outer circumference, 110 μm is the edge of the center hole

66

3.4.3.4 Cross Sensitivity As with the interdigitated sensor, the cross sensitivity of this sensor to other analytes is an important specification, since the sensor will be exposed to organic solvents as part of an ESLI system. Table 3.4 shows the response of the sensor to five common organic vapors. Since a capacitive sensor is sensitive to the dielectric constant of absorbed chemical analyte, three high dielectric constant alcohols (IPA, ethanol and methanol) were chosen, as well as acetone and toluene. The largest measured response was for methanol, the highest dielectric constant material, with a sensitivity of 1.35x10 -4 % change in capacitance per ppm. Although the response is significant, the sensor is designed to be used in an array of different chemical sensors with much higher sensitivity, allowing techniques such as principal component analysis to be used to differentiate between different chemicals. The sensitivity is also smaller than that measured for the interdigitated sensor for the analytes measured.

Analyte

Dielectric Constant

Vapor Pressure at 20 ° C (Pa) calculated from [76]

Dipole Molecular Sensitivity Moment Mass (%/ppm) (Debyes) (g/mol)

Toluene

2.39 [62]

2900

1.60x10-5

0.53[78]

92.14

Acetone

20.7 [77]

24500

1.80x10-5

2.9[78]

58.08

IPA

19.92 [80]

4340

7.50x10-6

1.6[79]

60.01

Ethanol

24.51 [62]

5850

5.70x10-5

1.7[78]

46.07

Methanol

32.65 [62]

12800

1.35x10-4

1.68[78]

32.04

Table 3.4 Sensitivity of sensor to organic solvents with analyte properties

67

3.4.3.5 Oxide pillar density In addition to the thermal effects, the pulling down of the top plate of the capacitive sensor can also result in electrical contact between the two plates, resulting in a short circuit of the capacitance that makes the device unusable. In the released design, a high percentage, five out of the eight sensors that were fabricated were found to have the top and bottom electrodes shorted. This yield would be much too low for a successful commercial product. To address this problem, a revised sensor with a higher density of oxide pillars was designed to create a more robust sensor. The design was fabricated with slots in the etched aluminum layer with a spacing of 25 μm between neighboring pillars, a reduction of over a factor of four from the previous design. Wyko interferometer images of the devices without added pillars after the aluminum etch and after polymer deposition are shown in Fig. 3.30 (a) and (b) respectively. The plate is flat after aluminum etching, but has deflected approximately 400 nm after inkjet deposition for an expected gap of 450 nm. For the design with pillars, there was again no noticeable deflection after aluminum etching (Fig. 3.30 (c)), but after deposition a smaller deflection of approximately 100 nm is visible; the pillar locations are also clearly apparent in the Wyko image as periodic spots without deflection. The yield has been demonstrated to be much higher than the previous design, with eleven sensors successfully fabricated without a case of shorting between the electrode plates.

68

μm

μm 1.0

200 0.5

150

100 0.0 50 -0.5 0

0

50

100

150

200

250

(a)

μm

300μm

(c)

μm

μm

μm 0.7

1.5 200

200 1.0

150

150

0.4

0.5 100

100 0.0 50

50 -0.5 0

0

50

100

150

200

250

(b)

300 μm

0

0

50

100

(c)

150

0 200

250

300 μm

(d)

Figure 3.30: Optical interferometer measurement of original design (a) after aluminum release (b) after polyimide deposition. Revised design with additional pillars (c) after etching and (d) after deposition in design with pillars The measured capacitance of the device without exposure to water vapor was found to be significantly lower than the previous design, dropping from 8.6 pF to approximately 3.6 pF in the higher pillar density design, a reduction of more than a factor of two. This reduction in the capacitance results primarily from the increase in the gap between the electrodes. The humidity sensitivity of both sensor topologies was measured (Fig. 3.31).

69

Change in Capacitance (%)

With Added Pillars

Original Design

18 16 14 12 10 8 6 4 2 0

0

20 Relative Humidity (%)

40

Figure 3.31: Humidity response for higher pillar density sensor

The original design shows a nonlinear response at low humidity larger than the response in the design with higher pillar density. In the linear region above 10% relative humidity, the fractional sensitivity was modestly lower, 0.27% change in capacitance per percent relative humidity for the revised design compared with 0.31%/%RH for the original design. This reduction in sensitivity may result from voids or air pockets left by the polymer fill. The fractional sensitivity is given by: S

C polymer C polymer

C air

(3.7)

where Cpolymer and Cair are the capacitance of the polymer filled and unfilled areas respectively. When the top plate is free to move, the areas with large deflection will have the smallest gap, since the distance between the plates will be smaller as the top plate moves downward. These areas are also likely mostly filled with polymer, since the polymer is holding the plate in place, resulting in an increase in Cpolymer due to the reduction in gap in those regions. The most likely area not to be completely filled is the outermost edge of the device, the area furthest from the inkjet target. Since this area is close to the 70

anchor at the outer edge, the gap will not close significantly in this region, so Cair will not have a corresponding increase due to the gap reduction. As a result, Cpolymer will be larger relative to Cair, resulting in a higher sensitivity according to equation 3.7. In the design with a fixed gap, the capacitance is smaller closer to the target due to the increase in the gap, while the outer rim capacitance remains unchanged, so the polymer capacitance is a smaller percentage of the total and the fractional sensitivity will be lower.

3.4.4 Polymer Etched Vertical Parallel Plate Design 3.4.4.1 Design Although the vertical parallel-plate design is significantly more sensitive than the interdigitated design, there was also an undesirable increase in the response time. As discussed in section 3.4.3.2, the polymer fills the vertical release holes as well as the active region of the capacitor, as shown in Fig. 3.32(a), so water vapor must diffuse a significant distance before reaching the sensor. By performing a vertical polymer etch, as shown in Fig. 3.32 (b), the diffusion distance can be reduced, improving the response time. This technique was demonstrated in the non-integrated sensor in [69]. There will also be a reduction in sensitivity, due to the additional air capacitance in parallel to the sensor, but by timing the length of the etch this reduction can be minimized.

Metal

SiO2

Si

Polymer

(a) (b) Fig. 3.32 Vertical parallel-plate sensor (a) before and (b) after vertical polymer etch 71

3.4.4.2 Fabrication Fig. 3.33 shows the process flow for the etched polymer sensor. As with the standard vertical parallel plate sensor, a vertical oxide etch using a Plasma-Therm 790 RIE system using CHF3 and O2 at 22.5 and 16 cc/min flow rates respectively and 100 W power (Fig. 3.33 (b)) is followed by a five hour etch in Transene aluminum etchant A (Fig. 3.33(c)). Since a vertical polymer etch will remove any polymer on the top surface of the chip, inkjet deposition is no longer necessary; instead, polyimide in solution, diluted similarly to the inkjet ink, was spun coat 15 times on the top surface of the chip (Fig. 3.33 (d)). Since inkjet deposition is a serial process that cannot be easily scaled to large numbers of chips, while spin coating could be performed over a full wafer if necessary, this refinement is an important advantage.

Inkjetting also creates a “coffee ring”, thicker deposits around the edge of the

drying drop, as shown in Fig. 3.23 (b). Since the thickness varies, the desired polymer etch time would vary across the sensor, resulting in either overetch or underetch in parts of the device. Since spin

(a)

(c)

(e)

(b)

(d)

(f)

Fig. 3.33 Process flow for sensor with vertical polymer etch coating spreads polymer over the entire chip, the coffee ring is far from the device and this effect is minimized. In inkjetting, a series of iterations of inkjet drops are placed on the device, with a delay between iterations to allow solvent to evaporate. For spincoating, the polyimide was heavily diluted to 72

the same concentration used for inkjetting. The polyimide was then spin coated on the device ten times, with a delay of a minute between depositions to match the delay between iterations in inkjetting. An O2 plasma etch was then performed in a Trion RIE system to remove the polymer in the release holes (Fig. 3.33 (e)). Since the oxygen plasma etch is not completely anisotropic, there is also the possibility of removing a significant percentage of the polymer under the plate as well as in the release holes, as shown in Fig. 3.33 (f). This overetched sensor would have a short diffusion length, and should be very fast; however, the sensor would also be expected to have a very low sensitivity. After deposition, three different sensors were etched in O2 plasma at 50 W and 14 cc/min oxygen flow rates for etch times of 4.4, 8.8 and 17.5 minutes respectively. Fig. 3.34 shows one device before and after etching of polymer.

(a)

(b)

Fig. 3.34 Device (a) after spin coating polyimide and (b) after polyimide etch

As expected, there is significantly less coffee ring on the top surface of the device than in the inkjetted sensor. Some variation is apparent in the polymer on the top surface of the chip, but the thick coffee ring deposits are eliminated. After etching, there is minimal polymer on the top surface of the chip for all three etch times. The sensitivity response of each sensor was measured and compared with an unetched inkjetted sensor (Fig. 3.35) A longer etch time is expected to lower the sensitivity if 73

polymer is etched under the adhesion layer. Fig. 3.35 also includes COMSOL simulated results for an unetched sensor, a sensor with a purely vertical etch, and a sensor with only 15% of polymer remaining underneath the adhesion layer.

Figure 3.35: Humidity response for etched sensors; lines are COMSOL simulated responses, individual data points are measured data

The sensors with etch times of 4.4 minutes and 8.8 minutes had comparable sensitivities of 0.21%/%RH, while the sensor with an etch time of 17.5 minutes had a lower sensitivity of 0.09%/%RH. The unetched sensor had a sensitivity of 0.27%/%RH, while the simulated sensitivity of a sensor with a purely vertical etch was 0.24%/%RH. Both the shorter etches appear to have had a small amount of polymer etching underneath the adhesion layer. The sensor with the longest etch time has had most of the polymer removed, matching the simulated sensor with only 15% of the polymer under the adhesion layer remaining. The response time of the sensor to a 3% step of relative humidity was modeled in COMSOL using the diffusion model. Fig. 3.36 below shows the average concentration underneath the top 74

electrode as a function of time. The simulated response time constants were 70 s, 4 s and 0.15 s for the unetched sensor, vertically etched sensor and sensor with 15% of polymer remaining respectively.

Figure 3.36: Simulated response time for etched polymer sensors

Fig. 3.37 shows the rising and falling response time constants for different etch times relative to the reference Honeywell sensor. A reduction of several minutes in both the rising and falling response times was found between the unetched sensor and the sensor etched for 4.4 minutes.

As with the

sensitivities, the difference between the sensors etched for 4.4 and 8.8 minutes was small, within the limit of error of the measurement, with a response time for both similar to that of the Honeywell reference. The sensor etched for 17.5 minutes was significantly faster than the reference for both rising and falling transients.

75

Difference in Time Constant (Min)

6 5 4

3 2 1 0 -1 -2 0

5

10

15

20

15

20

Etch Time (Min) Difference in Time Constant (Min)

(a)

2

1.5 1

0.5 0

-0.5 0

5

10 Etch Time (Min)

(b)

Figure 3.37: Difference in (a) rising and (b) falling time constants between etched sensors and reference humidity sensor

3.4.5 Polysilicon heater Another method that can be used to reduce the response time of a humidity sensor is to heat the sensor to an elevated temperature, since the diffusion coefficient in polymers increases with temperature [88]. A small on-chip heater, as in [89], can be used to locally heat the sensor to minimize power consumption. On-chip heaters have also been demonstrated for thermally resetting sensors and allowing fast measurement of very high humidity without condensation [90]. Since a polysilicon layer is available in many CMOS processes, a heating resistor can be added without any additional processing; Fig. 3.38 shows a cross section of the device with added heater.

76

Metal Si

SiO2 Polymer

Release holes

Inkjet well

Adhesion layers Polysilicon heater

Figure 3.38: Device with added polysilicon heater At higher temperatures, a smaller quantity of water vapor leaves the vapor phase to enter the absorbent layer, resulting in a lower sensitivity. Fig. 3.38 shows the humidity response for a sensor run with the heater at room temperature, 45°C and 65°C for an unetched inkjetted sensor. An unetched sensor was used to simplify measurement of the response time constant, due to the slower response. The sensitivity at 45°C is 0.11%/%RH, a drop of over a factor of two over the sensitivity at room temperature. A further increase in the temperature to 65°C results in a drop in the humidity sensitivity

Change in Capacitance (%)

to 0.05%/%RH. 12 10 8 6

22° C

4

45° C

2

65° C

0 0

20

40

Relative Humidity (%)

Figure 3.39: Humidity response at different temperatures The sensor was exposed to a small pulse of humidity at room temperature and with the temperature elevated to 45°C and 65°C with the heater (Fig. 3.40). The rising transient drops from over five minutes at room temperature to approximately two minutes at 45°C; a smaller drop to 1.5 minutes was found

77

when the sensor temperature was raised further, to 65°C (Fig. 3.40(a). Similar drops were found for the

Difference in Time Constant(Min)

falling transients (Fig. 3.40(b)). 6

5 4 3 2 1 0 20

30

40

50

60

70

Difference in Time Constant(Min)

Temperature (° C)

(a)

2 1.5 1 0.5 0 -0.5 -1 20

40

60

80

Temperature (° C)

(b) Figure 3.40: Difference in (a) rising and (b) falling time constants between sensors at different temperatures and the reference

3.5 Discussion A series of different topologies of chemicapacitive sensors were investigated for creating a high sensitivity chemicapacitive chemical sensor. The initial interdigitated designs were found to be more sensitive than past integrated sensors, but still lower than the levels of many non-integrated sensors. An integrated vertical parallel plate sensor with sensitivity approximately equal to that of the absorbent polymer layer was then demonstrated. Finally, a series of techniques were investigated to improve the robustness and speed of the vertical parallel plate sensor while minimizing losses in sensitivity. Since the sensitivity of this topology is primarily material limited, further refinement of the structure of the

78

sensor itself would have limited benefit; the next chapter focuses on techniques for improving the limit of detection by instead lowering the noise level of the sensor measurement circuitry.

79

Chapter 4 Capacitive Sensing Electronics 4.1 Introduction On-chip measurement electronics are necessary to fully realize the benefits of integrating a capacitive chemical sensor on a CMOS die. There are two primary methods that are commonly used for measuring chemicapacitive sensors. One method is to use the sensing capacitor to set the frequency of an oscillator circuit, such as the relaxation oscillator in [91] (Figure 4.1 (a)) and the digital ring oscillator in [92] (Figure 4.1 (b)). The main advantage of oscillator circuits is the direct generation of a digital output without having to create a separate analog-to-digital converter. A comparable relaxation oscillator in [93] found a resolution of 1.4 x 10-6, or 3 aF out of a 2.2 pF capacitance. Chemicapacitive sensors can also be measured with switched capacitor circuits, which use the amount of charge accumulating on the sensor to measure the capacitance. Examples of this type of interface include the switched capacitor amplifiers used in [[94]-[95]] and [96] (Figure 4.1 (c)). The limits of detection for switched capacitor circuits can also be a few attoFarads out of a picoFarad scale capacitance [97].

80

Csensor

-

vout

-

+

+

-

Csensor

+ (a) Φ1’

V1

Φ2’

C1 CSensor

C2 Φ2

-

Φ1’

Φ2’

V2

+

Φ1

Cref

(b)

Φ1’

Φ2’

+

-

Vout

(c)

Figure 4.1: (a) Relaxation oscillator, (b) ring oscillator and (c) switched capacitor amplifier In this chapter, two specific capacitance measurement circuits are evaluated. A simple switched capacitor circuit used for initial sensor testing, the charge-based capacitance measurement (CBCM) circuit, is analyzed. A Colpitts oscillator design having theoretically lower limit of detection and including digital circuitry to measure the oscillator frequency and output a low frequency digital output is then presented.

81

4.2 Charge-Based Capacitance Measurement Circuit 4.2.1 Circuit Vdd A Φ2 C Vc

Vs Cp

Φ1

(a)

Vs

Vs

Φ1

Φ1

Φ2

Φ2

Qin

Q1=Vdd (Cp+C)

Qin Q2=Vdd (Cp)

(b)

(c)

Figure 4.2: (a) CBCM circuit and timing diagrams for (b) first measurement and (c) second measurement. The CBCM circuit shown in Fig. 4.2(a) was developed to measure attoFarad scale interconnect capacitance in CMOS processes [98]. By using two separate measurements, the sensing capacitance C can be isolated from any parasitic capacitance Cp [99]. In the first measurement (Fig. 4.2(b)), the Vs input is grounded while the center node Vc is first discharged 82

to ground, then charged to Vdd, resulting in both capacitances C and Cp being charged. This cycle is repeated at frequency fs and measured with an ammeter, resulting in the current:

I1

f s (C C p )Vdd

(4.1)

In the second measurement (Fig. 4.2 (c)), the Vs input is grounded while the center node is discharged, but brought to Vdd when the center node is charged, resulting in no voltage drop across the sensing capacitor. Charge is only stored on the parasitic capacitance Cp, resulting in total current: I2

f s C pVdd

(4.2)

C

I1 I 2 f sVdd

(4.3)

The sensing capacitance is calculated as:

4.2.2 Noise A CBCM circuit can be analyzed as a simple switched capacitor circuit with two clock phases, corresponding to when each of the two transistors is turned on. One technique that can be used to analyze the thermal noise of a switched capacitor circuit is to treat each phase of the clock separately, with each switch opening resulting in the sampling of the system noise [100]]. A more detailed discussion of the noise of a sample-and-hold circuit, which has similar behavior, is presented in [101]. The 1/f noise can be considered to be zero for a transistor in triode when the drain to source voltage is 0 V [102]. Since the center node has been completely charged or discharged when the transistor switch is closed and the noise spectrum is sampled, the noise will not have a significant 1/f component from the transistors.

83

During the first clock phase, when Φ1 is high, the n-channel transistor is turned on and the center node is brought to ground through a small switch resistance Ron, (Fig. 4.3 (b)). The off resistance of the p-channel transistor is assumed to be large enough to be neglected. During the second phase, when Φ2 is low, the p-channel transistor is turned on, creating the circuit shown in Fig. 4.3(c). Since both circuits are purely resistive networks, the total noise power integrated over all frequencies during each phase is: v n21,tot

v n22,tot

kT C

(4.3)

where k is Boltzmann’s contant and T is the absolute temperature.

Ron,n

C

2

vn2

C

(b) vn2

C

1

Ron,p (a)

(c)

Figure 4.3: (a) CBCM circuit and noise equivalent circuits (b) when NMOS is turned on and (c) when PMOS is turned on. The noise for the first clock phase will be sampled when the NMOS transistor turns off. Before being sampled, the noise is not uniformly distributed in frequency, since the circuit acts as an RC filter. However, the cutoff frequency of the filter is much higher than the switching frequency, resulting in significant aliasing that creates effectively a white noise spectrum.

84

The total sampled charge noise measured by the CBCM for the first clock phase is: q n2,tot

v n21,tot C 2

kTC

(4.4)

After sampling, the charge noise power is uniformly spread between the discrete time frequencies -1/2 to +1/2 (f/fs, where fs is the clocking frequency) with discrete time power spectral density of

Qn2

kTC

(4.5) The noise from the PMOS switch is measured directly, without aliasing, but will only contribute noise current during the portion of the cycle that is turned on; as a result, the effective noise spectral density averaged over one cycle for the PMOS transistor is [101]: v n22 f

(4.6)

4mkTRon, p

where m is the fraction of the cycle that the transistor is on. The noise is filtered by the RC network formed by the circuit, so the frequency dependent noise contribution on the center node is: v n22, filtered ( f ) f

4mkTRon, p |1

j 2 fRon, p C | 2

4mkTRon, p 1 4

2

2 2 f 2 Ron , pC

(4.7)

The charge noise measured at each cycle for clock phase 2 is: q n22 ( f ) f

4mkTRon, p C 2 1 4

2

2 2 f 2 Ron , pC

(4.8)

The actual measurement is performed at low frequencies with a dc picoammeter, so the charge noise from a large number of individual cycles is averaged. In sensor testing, the 85

switches were clocked at 100 kHz, with the picoammeter obtaining a value every 0.1 s, resulting in 10,000 cycles being averaged. Due to aliasing, the sampled NMOS noise is much larger than the unaliased noise from the PMOS transistor. For a sensing capacitance of 8.8 pF, the charge noise amplitude from the NMOS transistor after filtering with a brickwall (sinc) filter with bandwidth 10 Hz is 1.86 x 10-18 C, corresponding to a capacitance of 5.65 x 10-19 F. The charge amplitude from the PMOS transistor after filtering is several orders of magnitude lower; for an on resistance of 200 Ω and with the transistor on one quarter of the cycle, the charge amplitude is 5.7 x 10-22 C.

4.2.2.1 Noise characterization One technique for characterizing the noise and the limit of detection of a system is the Allan variance: 2 C

1 M

1 (C k C k 1 ) 2 C 02 12

M k

1 2M

M k 1

~ (C k

~ Ck 1 ) 2

(4.9)

~

where M is the number of samples and C k is the kth capacitance value, normalized by the initial capacitance C0. The Allan deviation, the square root of the Allan variance, is a measure of the sample-to-sample variation of a sensor, and indicates the minimum charge that can be detected. The minimum detectable capacitance is thus given by

C

C0

2 C

(4.10)

The Allan deviation of a system is dependent on the sampling period. Longer sampling times result in additional averaging, reducing higher frequency noise. However, for long

86

averaging times, sensor baseline variation due low frequency drift from effects such as sensor aging results in an increase in Allan deviation.

4.2.3 Charge injection The analysis of the CBCM circuit in section 4.2.1 was based on the assumption that the transistors behave as ideal switches. In a real switched capacitor circuit, charge within the transistor channel will be released onto the center node capacitance, a phenomenon known as charge injection [103]. To estimate the magnitude of the charge injection, an analysis based on the model in [103] was performed. For the CBCM circuit, charge will be injected into the center node when either of the two transistor switches is opened. The charge injected by the NMOS transistor causes an initial voltage at the center node when the PMOS transistor is closed, rather than the 0 V assumed by the ideal model. The charge injected by the PMOS transistor results in excess charge being stored on the capacitor, resulting in the fully charged voltage varying from Vdd. The model makes the assumption that the transistor gate voltage is turned off with a constant slope α given by [103]: VG ,on VTE t fall

(4.11)

where VG,on is the gate voltage while the transistor is on, VTE is the threshold voltage of the transistor and tfall is the time necessary for the gate voltage to drop from VG,on to VTE. The total charge in the transistor channel, Cg(VG-VTE), where Cg is the gate capacitance, exits equally on

87

either side of the channel; this charge is assumed to flow at a constant rate as the gate voltage drops, and can be treated as a constant current of magnitude [103]:

Ic

1 C g (VG VTE ) 2 t fall

(4.12)

Cg 2

The transistor channel is treated as a time variable conductance given by [103]: g[VG (t )]

(VG (t ) VTE )

(VG ,on

(4.13)

t VTE )

where β is a constant equal to (W/L)μCox of the transistor. The charge injection model for the NMOS transistor is shown in Fig. 4.4; the PMOS will behave similarly, aside from sign changes.

NMOS transistor Vc αCg/2 g(VG(t))

C I1

I2

Figure 4.4: Equivalent model for NMOS charge injection

To solve the model, the total current entering the center node must be equal to 0, giving the expression: Cg 2

I1

I2

(4.13)

88

The current into the capacitor and through the channel in terms of the voltage Vc was then substituted for I1 and I2: Cg 2

C

dVc dt

Vc g

C

dVc dt

Vc (VG,on

t VTE )

(4.14)

The above expression is a first order differential equation that can be re-written: dVc dt

Vc C

(VG,on

t VTE )

Cg 2C

(4.15)

This differential equation was then solved numerically using Euler’s method. Fig. 4.5 shows the voltage resulting from charge injection from the NMOS transistor for a sensor capacitance of 8.8 pF using the transistor characteristics from the Jazz 0.35 μm design kit. The fall time used was the worst case fall time, 20 ns, for a square wave for the Agilent 33120A function generator used for testing, taken from the datasheet. The model assumes that the transistor channel will be open once the gate voltage reaches the threshold voltage, at 16.9 ns, so only the simulated results to this point are shown. For the NMOS transistor, VG,on is 3.3 V, VTE is 0.57 V, and α is 1.65x108 V/s.

89

-0.1 Voltage(mV)

-0.3 -0.5 -0.7

Theoretical Model

-0.9

Simulated

-1.1 -1.3 0

5

10 Time(ns)

15

Figure 4.5: Charge injection voltage resulting from NMOS transistor

Fig. 4.5 shows the charge injection voltage from the theoretical model and simulated in Cadence using the Jazz 0.35 μm design kit. The charge injection goes through three distinct phases. In the first phase, the first few nanoseconds of Fig. 4.5, the voltage Vc remains relatively low, and most of the charge injected remains on the capacitor. As the voltage continues to drop, a larger amount of charge flows back through the channel conductance; this effect results in the slower drop in voltage seen in the middle period of the plot. In the last few nanoseconds of the plot, the transistor gate voltage is nearing the threshold and the transistor is mostly off, resulting in most of the injected charge remaining on the capacitor and little charge flowing back through the channel. The final injection voltage for the NMOS transistor from the theoretical model is -1.3 mV, corresponding to a total charge injected of -11.4 fC, compared with a simulated response of -0.86 mV, an injected charge of -7.6 fC. The PMOS transistor used had a width double that of the NMOS transistor, resulting in a larger injected 90

charge. The injection voltage for the PMOS transistor in the theoretical model is 3.0 mV, corresponding to injected charge of 26.4 fC; the simulated response is 2.0 mV, corresponding to injected charge of 17.6 fC. The total error in the measurement charge will be the sum of the NMOS and PMOS charge injected, or 15 fC for the theoretical model and 10 fC for the Cadence simulations. As was noted in [104], both the parasitic and total capacitance measurements will experience the same charge injection, since the sensor capacitance is still present in both cases, despite the change in voltage on the other side of the capacitor.

Since the parasitic

measurement is then subtracted from the total measurement, the charge injected will be subtracted away, resulting in no measurement error neglecting noise variations. For the fixed capacitances in [104], this analysis is correct; however, for a time-varying capacitive sensor, the sensor capacitance may not be identical between the two measurements. In the current setup, a single parasitic measurement is taken at the beginning of the test, and the parasitic is assumed to be constant. The potential error of this approach can be estimated in the model. To test how a change in capacitance affected the charge injection, the model was repeated for a sensor capacitance of 10.56 pF, 20% higher than the original capacitance. The NMOS charge injected was found to be -12.7 fC, and the PMOS charge injected was 28.5 fC; the net charge error is 15.8 fC, an increase of 0.8 fC from the error with a sensing capacitance of 8.8 pF. This increase is small, although it is significantly above the theoretical noise level of the CBCM circuit of 5.65 x 10-19 F. Since the approximate sensor capacitance is known, this effect could also be compensated to correct the parasitic measurement. 91

4.2.4 Clock feedthrough During the period when the clocking signals are transitioning, but are below the transistor threshold voltage, the clock signal will couple from the transistor gate through the overlap capacitances to the center node of the CBCM circuit, a phenomenon known as clock feedthrough [105].

Fig. 4.6(a) shows the CBCM circuit including the parasitic overlap

Cgd,overlap p Φ2

Cgs overlap,p

C+Cp

Φ2 Cgd overlap,p

C+Cp (b)

Cgd overlap,n

Cgd,overlap n

Φ1

Φ1

Cgs overlap,n

C+Cp

(a)

(c)

Figure 4.6: (a) CBCM circuit including parasitic overlap capacitances and capacitive divider circuits (b) formed by Φ2 input and (c) formed by Φ1 input capacitances. When the Φ1 and Φ2 inputs transition, these overlap capacitances form the capacitive divider circuits shown in Fig. 4.6 (b) and (c). Each of the clock transitions of Φ1 and Φ2 will result in changes in the voltage at the center node from clock feedthrough; however, not all the feedthrough will be detected by the picoammeter on the output. The positive transition of Φ2, which turns off the PMOS transistor, and the positive transition of Φ1, which turns on the NMOS transistor, will both be removed by the NMOS transistor bringing the center node to ground. The remaining two transitions will 92

affect the voltage at the center node before the picoammeter measurement. During the negative transition of Φ1, clock feedthrough will occur during the period between the threshold voltage of the NMOS transistor, VTE,n, and ground, giving a step in voltage at the center node of: C gd ,overlap,n

Vc, NMOS

C gd ,overlap,n

C

Cp

VTE ,n

(4.16)

The total charge will be given by: Qc, NMOS

C gd ,overlap,n C gd ,overlap,n

C

Cp

VTE ,n (C gd ,overlap,n

C

Cp)

C gd ,overlap,nVTE ,n

(4.17)

During the negative transition of Φ2, feedthrough will occur between the beginning of the transition and the voltage Vdd-|VTE,p|, when the PMOS transistor turns on, causing a step in voltage at the center node of: C gd ,overlap, p

Vc, PMOS

C gd ,overlap, p

C

Cp

| VTE , p |

(4.18)

giving a total charge of: Qc, PMOS

C gd ,overlap, p C gd ,overlap, p

C

Cp

| VTE , p | (C gd ,overlap, p

C

Cp)

C gd ,overlap, p | VTE , p | (4.19)

Using the values for overlap capacitance and threshold from the design kit, the clock feedthrough charge is -2.33 fC for the NMOS transistor and 6.77 fC for the PMOS transistor. Unlike the charge injection, the clock feedthrough charge in both transitions is independent of the sensor capacitance, and will be canceled completely by the parasitic measurement.

4.2.5 Clocking jitter The capacitance measurements taken by the CBCM circuit assume ideal clocking signals at fixed frequency fs. Real clocks experience variation in the spacing between transitions, a 93

phenomenon known as jitter [106]. For the CBCM circuit, this non-ideality will be reflected as a change in the number of charge pulses that are measured by the ammeter, resulting in variation in the effective frequency of the clock. The clocking signals used for the capacitance testing were 1 MHz square waves generated by phase locked Agilent 34401 function generators.

Allan deviation

1.00E-08

1.00E-09 0.1

1

10

100

1000

Integration time (s)

Figure 4.7: Normalized frequency Allan deviation for phase locked Agilent 34401 function generators A frequency counter was used to estimate the frequency stability of the clocking signals [107]; the resulting Allan deviation is shown in Fig. 4.7. The minimum Allan deviation is 1.04x10-9 at 1.2 seconds integration time, corresponding to a minimum frequency variation of 0.0010 Hz out of the 1 MHz clock. This frequency variation is several orders of magnitude below the Allan deviation measured with the picoammeter in Fig. 4.3.

The variation is also an order of 94

magnitude lower than the theoretical noise limit of the CBCM circuit calculated in section 4.2.2, which estimated a noise level of 5.65x10-19 F out of an 8.8x10-12 F capacitance, which would correspond to an Allan deviation of 6.5x10-8.

4.2.6 Noise measurement The Allan deviation for the CBCM circuit measuring an 8.8 pF polyimide vertical parallel plate sensor was measured in a dry nitrogen environment; Fig. 4.8 shows the Allan deviation after temperature compensation based on the output voltage of an AD22100 temperature sensor mounted on the chip package with thermal grease. The sensor was measured with 1 MHz clocking frequency and an input voltage of 3.3 V.

Integration time (s)

Allan Deviation

1.E-03 0.1

1

10

100

1000

1.E-04 1.E-05 1.E-06 Resistor

Sensor

Figure 4.8: Measured Allan deviation.

The Allan deviation of a Keithley 6485 picoammeter measurement of a comparable current through a 33 kΩ resistor is also shown, and is an estimate of the noise level of the picoammeter. The expected Allan deviation for the resistor assuming white thermal noise is 5.23x10-8 at 0.1 s integration time. The Allan deviation is relatively flat for low integration times, approximately equal to that of the resistor; for higher integration times, sensor drift due to 95

imperfectly compensated temperature and aging dominates, resulting in a rising Allan deviation.

The lowest Allan deviation was 6.49 x 10-6 for an averaging time of 0.5 s,

corresponding to a minimum detectable signal of 0.06 fF and a limit of detection of 0.0023% change in relative humidity. This limit is significantly higher than the estimated noise level of the CBCM circuit, 5.65 x 10-19 F; the system is currently limited by the noise limit of the picoammeter being used to measure the sensor current. The typical RMS noise datasheet specification for the Keithley 6485 picoammeter is given to be 100 pA in for measurements between 2 μA and 20 μA [108], corresponding to a minimum of 1x10-5 fraction of the total signal. This is comparable to the measured Allan deviation.

4.2.7 Discussion Charge-based capacitance measurement is a useful, simple to design circuit for measuring capacitance. The technique is also resistant to both clock feedthrough and charge injection, which can be problematic for many types of switched capacitor circuits. However, there are several notable disadvantages. Taking a small current off-chip for measurement by external equipment may be problematic for many applications. The noise performance, a limit of detection of 60 aF out of a capacitance of 8.8 pF, is also not as low as the limit of detection of the relaxation oscillator in [93] or the switched capacitor circuit in [97], both a few attoFarads out of a picoFarad scale capacitance.

96

4.3 Low phase noise Colpitts oscillator 4.3.1 LC oscillator To improve the noise performance, an alternative method for measuring the sensing capacitance is considered.

Low phase noise LC voltage controlled oscillators (VCOs) are

commonly used to create stable clock signals in RF integrated circuits. By replacing the fixed capacitance with a capacitive sensor, these circuits can be used to convert the capacitance signal into a digital clock frequency signal.

R

L

C

Figure 4.9: RLC circuit To determine the theoretical limit of detection of an LC oscillator, the double sided phase noise of a small frequency offset from the resonant frequency f0 for the ideal parallel RLC oscillator shown in Fig. 4.9 was estimated based on the Leeson equation [109]: v n2, L( f ) 10 log(

f 2 v sig

) 10 log(

kTf 02 R 2Q 2 ( f ) 2 Psig

)

(4.21)

where vn,φ and vsig are the phase components of the noise spectral density and of the signal voltage, respectively, Q is the quality factor of the tank, and Psig is the signal power. The expression is derived by multiplying the noise current from the resistor by the impedance of the

97

tank at a small frequency offset from resonance [109]. The single sided phase noise power spectral density Sφ is related to the phase noise by [110]: 1 L( f ) 10 log( S ( f )) 2

(4.22)

resulting in a phase noise spectral density of: S ( f)

kTf 02 R Q 2 ( f ) 2 Psig

(4.23)

The Allan variance is related to the phase noise power spectral density by [111]: 2 f

( )

2 ( f0 )

S ( f ) sin 4 ( f )df

2 0

(4.24)

Table 4.1, based on the table in [111], shows the relationship between various forms of phase noise and the Allan variance. The variables hn are constants with respect to frequency, and f h is the maximum noise cutoff frequency. From the table, the phase noise power spectral density is white frequency noise, and the Allan variance can be written: 2 f

( )

kTR 2Q 2 Psig

(4.25)

98

Table 4.1: Common conversions between phase noise power spectral density and Allan variance The Allan variance was calculated for the parameters shown in Table 4.2, giving a value of 1.4x10-21, corresponding to an Allan deviation of 3.7x10-11 and a minimum detectable frequency change of 0.04 Hz out of a resonant frequency of 1.07 GHz.

T(°K) R(Ω) Q L (nH) C (pF) Psig (V2) τ (s)

300 338 5.6 9.3 2.4 1 1

Table 4.2: Parameters used for calculating RLC Allan variance To obtain the minimum detectable capacitance, the frequency sensitivity to capacitance must be calculated. For an RLC circuit, the resonant frequency f 0 is related to the circuit parameters by the expression: f0

1 2

1 LC

(4.26)

99

The frequency sensitivity with respect to capacitance was found by taking the derivative with respect to C, giving: f C

1 C 2 (

1 LC

1

) 2

1 ( C L 2

3/ 2

)

(4.27)

For the circuit parameters in Table 4.2, the frequency sensitivity to capacitance is 2.2x10 20 Hz/F, resulting in a minimum detectable capacitance of 1.8x10-22 F. Although in a practical oscillator the active devices will add additional noise [109], an LC oscillator appears capable of a significant improvement in the limit of detection over the performance of past capacitive chemical sensor testing electronics. The magnitude of the additional noise from the active devices will be determined using Cadence simulation in the following section.

4.3.2 Oscillator topology The single ended Colpitts oscillator shown in Fig. 4.10 was chosen to implement the LC oscillator. For capacitive sensing, a differential output is unnecessary and the Colpitts oscillator has superior phase noise characteristics than the more common differential cross-coupled oscillators [112]. An NPN implementation was chosen because bipolar transistors are available in the Jazz SiGe technology and have superior 1/f noise characteristics over MOSFETs [113]. The 1/f noise will appear as 1/f3 phase noise; since 1/f3 phase noise corresponds to a flat Allan variance (Table 4.1), this noise ultimately sets the minimum detectable frequency.

The

resonant frequency of the Colpitts oscillator is given by equation 4.28; the circuit parameters C1, C2 and L were set to be 3.5 pF, 7.7 pF and 9.3 nH respectively, corresponding to a resonant frequency of 1.07 GHz.

100

f

1 2

1 L

(4.28)

C1C 2 C1 C 2

The circuit was simulated using the Jazz 0.18 μm design kit and found to have a simulated resonant frequency of 0.99 GHz and output amplitude of 0.69 V pp. The Colpitts oscillator was operated at 4.29 mW, excluding the power in the bias network.

L

Vbias

Q1

Vmirror

C1

Q2

C2

Power spectral density (rad2/Hz)

Figure 4.10: NPN Colpitts Oscillator 1000 100 10 1 0.1 0.01 0.001 0.0001 0.00001 0.000001 0.0000001

Simulated Fitted flicker frequency

Fitted white frequency

1

10

100

1000

10000

Relative frequency(Hz)

Figure 4.11: Simulated phase noise spectral density of NPN Colpitts oscillator

101

Fig. 4.11 shows the simulated phase noise power spectral density and the fitted 1/f3 flicker frequency and 1/f2 white frequency noise components. The Allan variance set by the 1/f3 phase noise is 4.2x10-16, corresponding to an Allan deviation of 2.0x10-8 and a minimum detectable frequency of 19.8 Hz.

This is larger than the theoretical result because the

theoretical model did not incorporate noise from the transistors in the circuit. The resulting minimum detectable capacitance, using the frequency sensitivity calculated from (4.27), is 9x10-20 F.

4.3.3 Integrated Sensor System An integrated sensor system based on the Colpitts oscillator design was laid out and fabricated in the Jazz 0.18 μm BiCMOS process. The following sections discuss the other components of the sensor system, a digital frequency counter circuit used to convert the oscillator frequency to a low frequency digital signal, and a PTAT temperature sensor circuit to allow on-chip measurement of the chip temperature.

4.3.3.1 Frequency Counter A digital frequency counter circuit was designed to measure the oscillator frequency. One technique for creating a frequency counter is to measure the number of clock pulses occurring in a fixed time interval [115]. Fig. 4.12 shows a diagram of the full system. The C1Out

Buffer

24 bit counter

CE RST

Sensor oscillator

Output register

C2Out

Reference oscillator

Figure 4.12: Frequency counter diagram

RST

24 bit counter

Control Logic

CE RST LatchIn

102

buffered output of the sensor NPN Colpitts and the output of a reference oscillator made using a comparable Colpitts with standard process capacitors are each fed into 24-bit digital counters. The outputs of the reference counter are combined with logic gates to create timing signals for the sensor counter. The sensor counter is first stopped by lowering a clock enable input, allowing the counter output to stabilize; the time between turning on the counter and lowering the clock enable signal is the fixed time interval for the measurement. The output from the sensor counter, corresponding to the number of pulses during the interval, is then latched into an output register. Finally, both counters are reset and the next measurement cycle begins.

4.3.3.2 PTAT Temperature Sensor Since the capacitive chemical sensor and most other types of chemical sensors are highly sensitive to temperature, an on-chip chemical sensor is an important component of any sensor system. A common technique for measuring the temperature on an integrated chip is known as a proportional-to-absolute-temperature (PTAT) circuit. The PTAT circuit is based on the relationship between the current through a p-n junction and absolute temperature. In the on state, the current through a p-n diode can be simplified to:

I

AKe

qV kT

(4.29)

where K is a constant combining the non-area dependent parameters.

103

The PTAT circuit takes advantage of the dependence to convert the temperature to a voltage that can then be measured. Fig. 4.13 shows the PTAT circuit used; the design is a modified version of the circuit presented in [117]. D3 and D4 have areas n times larger than D1 and D2.

+ V1

D1

-

+ D3

V3 -

D2

D4

I1

I2 R1

R2

-

Vout

+

M3

M2

M1

Figure 4.13: PTAT circuit In this circuit, currents I1 and I2 are equal since both are driven by identical current mirror transistors M1 and M2, and are given by the expression:

I1

I2

A3

qV3 Ke kT

A1

qV1 Ke kT

(4.30)

The expression can then be rearranged to give:

A3 A1

q (V1 V3 ) e kT

(4.31)

104

After taking the natural logarithm and rearranging, the difference in voltage drop between D 1 and D3 is: V1 V3

A kT ln( 3 ) q A1

kT ln(n) q

(4.32)

In the circuit used, the two diodes in series result in a doubling of the resulting voltage difference.

The differential amplifier forces the drain voltages of M 1 and M2 to be

approximately equal. The current through resistor R1 will thus be: I2

2

kT ln(n) qR1

(4.33)

and the final output voltage can be found from the ratio of the two resistors to be: Vout

V DD

2kTR2 G ln(n) qR1

(4.34)

where G is the current gain of the current mirror, giving a linear relationship with the absolute temperature on the chip. The circuit was modeled in the Jazz 0.18 μm design kit, giving the temperature response shown in Fig. 4.14. The simulated temperature sensitivity is 5.7 mV/°C, compared with a theoretical value of 4.1 mV/°C. The noise in the circuit was also simulated, finding an unnormalized Allan deviation of 2.8x10-4 V corresponding to a limit of detection of

Temperature Sensitivity

0.05°C.

2.2 2.1 Voltage(V)

2 1.9 1.8 1.7

1.6 1.5 1.4 0

20

40

60 Temperature (°C)

Figure 4.14: Simulated temperature sensitivity of PTAT circuit

80

100

120 105

4.3.4 System Measured Results The sensor system was successfully taped out with a die area of 1.5 mm by 2 mm; Fig. 4.15 shows a picture of the chip before aluminum etching after it was returned from the semiconductor foundry. Fig. 4.16 shows the capacitive sensors in the sensor Colpitts oscillator after inkjet deposition of polyimide.

Colpitts Oscillators

Digital Circuitry

Capacitive Sensors

Figure 4.15: Integrated sensor chip

2.5 mm

Figure 4.16: Capacitive sensors after inkjet deposition of polymer 106

4.3.4.1 PTAT Measurements The PTAT circuit was measured for a range of temperatures; Fig. 4.17 shows the temperature response as well as the simulated response. The measured sensitivity was larger than expected, 12.6 mV/°C, and the baseline output voltage also dropped from 1.92 V in simulation to 1.4 V in measurement. The Allan variance was measured to be 7.4x10-7, giving a limit of detection of 1.0 mV

2 1.9 1.8

Voltage (V)

1.7

1.6 1.5 Measured

1.4

Simulated

1.3 1.2 1.1 1

20

30

40 50 Temperature (°C)

60

Figure 4.17: PTAT temperature sensitivity corresponding to approximately 0.08°C, similar to the simulated value of 0.05°C. One possible reason for the drop in bias voltage and increase in sensitivity for the PTAT circuit would

Number in Bin

PTAT Output Voltage

20 18 16 14 12 10 8 6 4 2 0

107 1.4 1.475 1.55 1.625 1.7 1.775 1.85 1.925 2 2.075 2.15 2.225 2.3 Voltage (V)

Figure 4.18: Monte Carlo simulation of PTAT output voltage at room temperature

be a significant mismatch between the transistor M3 in the output stage and the transistors connected to the diodes. This mismatch could create a larger current gain in the current mirror, which in turn would result in a drop both in the bias point at room temperature as well as a greater sensitivity. To test this hypothesis, a Monte Carlo simulation incorporating both process variation and transistor mismatch was run in the Cadence design kit to see the effects of variation in the transistor sizes and process parameters on the bias point of the circuit; Fig. 4.18 shows the resulting histogram. Based on the simulation, a large change in the output voltage at 25°C is possible for the PTAT circuit due to mismatch and process variations. In 100 trials, the output voltage was found to vary from under 1.5 V to as high as 2.2 V. Other effects such as transistor corners and variation in the opamp bias current were also investigated, but were found to have significantly smaller effects.

3.3.4.2 Colpitts Oscillator Measurements The sensor Colpitts oscillator was designed with a common emitter output buffer stage able to drive a 50 Ω load, allowing the signal at this stage to be measured without using the digital circuitry. The output was measured with an Agilent E4440A spectrum analyzer using an RF probe station, giving the output signal shown in Fig. 4.19. A spread out peak was measured at 2.44 GHz, with a maximum

-40.00 -50.00

Amplitude (dBm)

-60.00 -70.00 -80.00 -90.00 -100.00 -110.00 2.2

2.3

2.4

2.5

2.6

2.7

Frequency (GHz)

Figure 4.19: Buffer output of Sensor Colpitts oscillator

108

magnitude of -48 dBm, which corresponds to a peak to peak output voltage of approximately 1 mV; the integrated voltage over the peak was also less than 10 mV, significantly smaller than the designed voltage of 1 Vpp. The peak was also found to be unstable, and the oscillator periodically stopped oscillating. The digital frequency counter includes a debugging output, a frequency divided output 29 times lower than the oscillator frequency, allowing the counter operation to be verified. Since the oscillator peak is over a factor of two higher than the designed frequency of 1 GHz, the digital circuitry was in some cases found to fail due to the inability of the input flip flops to accurately register the oscillator output. When stable oscillation did occur, the frequency was not consistent, tending to primarily to occur approximately in multiples of powers of two times 1.8 MHz; based on simulation in cadence, this behavior most likely occurs due to the first few flip flops in the counter failing to divide the frequency, passing the undivided oscillation instead. 1.8 MHz would correspond to an oscillator input of 922 MHz for the counter, close to the designed frequency.

The stability was very poor, generally falling out of

oscillation in a few seconds after turning on the system, although in some cases oscillation was maintained for several hours, allowing analyte testing. Using an off-chip frequency counter, the humidity sensor was measured using for the sensor

1830000

16

1828000

14

Frequency (Hz)

1826000

12

1824000 1822000

10

1820000

8

1818000

6

1816000

4

1814000

Relative Humidity (%)

system. Due to the problems with oscillator stability, extended tests could not be run, requiring shorter

Frequency (Hz) Relative Humidity (%)

2

1812000 1810000

0 0

5

10

Time (Min)

Figure 4.20 Humidity testing with Colpitts oscillator; frequency is clock divided counter output

109

tests of one or two pulses of analyte. Fig. 4.20 shows one test with a pulse of water vapor. In a series of tests with analyte, the results were inconsistent; several showed clear responses as with the test in Fig. 4.20, but for other tests no response was visible. The frequencies of stable oscillation also varied, suggesting the possibility that in some cases the Colpitts is oscillating correctly with the sensor capacitor, and in others some combination of parasitic elements are oscillating instead independently of analyte. In addition to difficulties with obtaining oscillation, the oscillator output was not found to be as stable as predicted in the cadence modeling. The Allan deviation of the oscillator was measured for one of the modes of oscillation (Fig. 4.21), giving a minimum Allan deviation of 0.0002, rather than the expected value of 2x10-8, a factor of 104 larger. If pulses are being missed periodically by the counter output, which is likely if the oscillator is oscillating at a higher frequency than expected, this could

Allan Deviation

0.01

0.001

0.0001 1

10

100

Integration Time (s)

Figure 4.21 Stability test of Colpitts oscillator

110

explain the Allan deviation.

4.3 Analysis and future work In this section, a detailed analysis of the charge-based capacitance circuit used in testing the chemicapacitive sensors was performed. The circuit found to be unable to reach limits of detection demonstrated by other groups for capacitive chemical sensors. Based on this result, a low phase noise Colpitts LC oscillator was designed along with a digital frequency counter and PTAT temperature sensor, with the designed limit of detection over an order of magnitude lower than past demonstrated chemicapacitive sensor electronics. Although many of the elements of the sensor system worked as designed, the Colpitts oscillator was found to oscillate only intermittently and frequently at incorrect frequencies. Further work exploring this behavior and changing the design would likely give behavior closer to simulation. Other types of LC oscillators could also be used that might allow comparable limits of detection with better startup stability.

111

Chapter 5

Gold Nanocluster Capacitive Sensors 5.1 Introduction Thiol-coated gold nanoparticles are a promising material for resistive chemical sensing, and have been shown to be sensitive to a number of analytes such as volatile organic carbons [118]. In chemiresistive sensing, the change in resistance results primarily from the changing distance between conductive gold cores as the material expands [118]. The benefits of nanoparticles in a capacitive chemical sensor have been less clear. Since chemicapacitive sensing is not based on conduction, changing core distance due to a swelling film would have no effect on the behavior. However, conductive particles in a capacitive chemical sensor would have two potentially beneficial effects. A conductive particle would effectively short portions of the dielectric, resulting in a smaller effective gap between the capacitive electrodes. This smaller gap would allow a larger capacitance to be created without using more chip area. A nanoparticle film might also have more bonding sites available for analyte, allowing higher absorption and thus a larger chemical response. Several efforts have been made to use larger nanoparticles for capacitive humidity sensors. In [119], gold nanoparticles 10 nm in size were encapsulated in thick coatings of polyvinyl alchohol (PVA) to create particles 700 nm in diameter that were deposited on an interdigitated capacitive sensor. Since the gold core is only a very small fraction of the total 112

volume, the response is likely dominated by the PVA in this sensor, although this possibility was not tested in [119]. In [120], humidity sensors were created using silica nanoparticle aerogels as a dielectric material. The 30 nm silica nanoparticles were used to obtain a higher pore density to improve the sensitivity of the sensor. Since these nanoparticles are dielectric, rather than conductive, and the sensing mechanism is based on adsorption of water on the surface of the particles, rather than absorption into a film, the sensing behavior is likely to be significantly different from thiol-coated nanoparticles. There has not been a past demonstration of smaller conductive nanoparticles in chemicapacitive sensors.

5.2 Gold Nanoclusters For capacitive sensing, an insulating class of nanoparticles is necessary.

Large gold

nanoparticles used for chemiresistive sensing, such as C8, are much too conductive. Instead a class of materials known as gold nanoclusters, small particles consisting of a handful of gold atoms stabilized by a thiol coating, was investigated. Certain numbers of gold atoms have been demonstrated to be chemically stable. The materials Au25(SCH2CH2Ph)18 [121], Au38(SCH2CH2Ph)18

[122]and Au144(SCH2CH2Ph)60 [123], contain 25, 38 and 144 gold atoms in the nanocluster

core and are referred to as Au25, Au38, and Au144 respectively. Due to the smaller size of the nanocluster cores, a larger number of dielectric gaps occur between the electrodes, resulting in a very large resistivity several orders of magnitude higher than the C8 material. Since the thiol coatings of all three materials are similar, the variation of the behavior due to the difference in core size can also be studied.

The materials used were synthesized by collaborators in

Rongchao Jin’s group in the Carnegie Mellon Chemistry department. 113

5.2.1 Au25 5.2.1.2 Chemical sensitivity Au25 has a conductive core consisting of 25 gold atoms that are then coated with SCH 2CH2Ph groups. The particle was synthesized and measured to have a core of 1 nm with a capping agent length of 0.2-0.3 nm. Au25 was dissolved in dimethyl sulfoxide (DMSO) and deposited into a vertical parallel plate sensor; the device is shown in Fig. 5.1.

Fig. 5.1: Chemicapacitive sensor after inkjetting of Au25 Unlike the polyimide sensors, the visible Au25 on this sensor tended to agglomerate as particulate on the top surface of the device, instead of as a continuous film. The most likely reason for this behavior is that some fraction of the nanoclusters became unstable and agglomerated in solution over time. To verify that material was successfully deposited between the capacitor plates, the dry capacitance was measured and found to be approximately 13.8 pF. For a comparable device, polyimide had a capacitance of 8.8 pF; based on this measurement, the dielectric constant between the plates is 4.5, suggesting that the gap was filled with

114

nanoclusters. This device was made with the older topology without a higher density of pillars, so differences in plate deflection could also have affected the capacitance. The chemical response of the sensor was measured for a number of different analytes; the sensor response to different concentrations of ethanol is shown in Fig. 5.2. Table 5.1 shows the sensitivities to a number of different chemical analytes. The measured sensitivity of Au 25 to volatile organic solvents is more than a factor of 10 higher than the response for the polyimide sensor in Table 3.2; for instance, the sensitivity to ethanol increased from 0.00013%/ppm to 0.0070%/ppm.

1050 ppm

15.4

14.4

14.2

900 ppm

600 ppm

300 ppm

14.6

450 ppm

14.8 150 ppm

Capacitance (pF)

15

750 ppm

15.2

14 13.8

13.6 0

20

40

60 Time (Min)

80

100

Fig. 5.2: Au25 response to ethanol

115

Dielectric Constant 20.7 Acetone [125] 2.39 Toluene [125] 19.92 IPA [126] 24.51 Ethanol [125] 32.65 Methanol [125] 1.95 Octane [125] 1.84 Pentane [125] Analyte

Vapor Pressure at 20°C (Pa) (calculated from [124])

Chemical family

Sensitivity (% change in capacitance per ppm)

2.45E+04

Ketone

0.002065

2.90E+03

Aromatic

0.00616

4.34E+03

Alcohol

0.006655

5.85E+03

Alcohol

0.00699

1.28E+04

Alcohol

0.0033

1.39E+03

Alkane

0.0038

5.63E+04

Alkane

0.0002

Dipole moment (Debyes)

Molecular mass (g/mol)

2.9[78]

58.08

0.53[78]

92.14

1.6[79]

60.01

1.70[78]

46.07

1.68[78]

32.04

0[78]

114.23

0[78]

72.15

Table 5.1: Table of analyte properties and sensitivity

5.2.1.3 Environmental sensitivity One of the largest drawbacks of capacitive chemical sensing for measuring analytes other than water vapor is that capacitive sensors generally respond strongly to changes in humidity. Although this effect can be compensated using other sensors, such as the polyimide humidity sensor, the humidity response remains an important specification. Fig. 5.3 shows the response of the Au25 sensor to a range of relative humidity. The humidity sensitivity is 2.1%/%RH; this sensitivity is very large, close to seven times that of the comparable polyimide sensor. This high sensitivity suggests that the material might be a suitable candidate for humidity sensing, but would definitely require compensation for use to measure other analytes.

116

Fig. 5.3: Au25 response to water vapor In testing the Au25 sensors, a very large amount of baseline drift was found to be present in the sensor. To understand the behavior, the temperature response of the sensor was measured (Fig. 5.4), since gradual heating or cooling of the measurement chamber could result in changes in the capacitance.

Fig. 5.4: Au25 temperature response

The thermal response was found to be extremely large, a change in capacitance of close to 600% over a temperature range of 20°C.

The comparable change for the polyimide sensor 117

was only 6%, a factor of 100 lower; the reason for the dramatic increase in capacitance with temperature is unknown. A large temperature response does not necessarily prevent the use of the sensor for chemical sensing, since temperature can be measured and compensated for with an on-chip temperature sensor; however, a thermal response this large would require an extremely accurate temperature sensor to allow use for measuring small changes in chemical concentration.

5.2.1.4 Manufacturing repeatability The first Au25 sensor fabricated had very high sensitivity to chemical analyte, as shown in the preceding sections. This first sensor was fabricated in a first generation vertical parallel plate sensor, the topology with too few pillars, allowing the top plate to deflect and close much of the sensor gap. Since other attempts to create further sensors in that technology failed, due to the problems with shorting across the gap previously mentioned, a later set of sensors were fabricated with a different batch of material using the later high pillar density design. Unlike the first batch of Au25 nanoclusters, this batch failed to dissolve completely in the solvent DMSO; toluene had to be added before the nanoclusters were dissolved completely. A sensor was jetted using 12 mg of nanoclusters dissolved in 600 μL of DMSO mixed with 600 μL of toluene (Fig. 5.5).

Fig. 5.5: Au25 sensor jetted with 50:50 mixture of toluene and DMSO

118

With previous sensors jetted with a single solvent, deposits of polymer appear gradually as the solvent evaporates, particularly when the solvent is nearly completely evaporated. For the toluene-DMSO mixture, the nanocluster deposits were visible almost immediately after jetting, while the solvent continued to evaporate for several minutes.

The most likely

explanation for this is the difference in the boiling points between the two solvents. Toluene has a boiling point of 110.6°C, while DMSO has a higher boiling point of 189.0°C [127]. As a result, the toluene evaporates quickly, while the DMSO remains for several minutes. Since the nanoclusters have been demonstrated to not stay in solution in DMSO, the nanocluster deposits form after the toluene evaporates. The sensor was measured and found to have a capacitance approximately the same as that of an air gap, suggesting that the nanoclusters are not filling the gap between the plates. The sensitivity to humidity was also measured and found to be very small, with a 1% change in capacitance for a 40% step in relative humidity. One possible reason for the failure to fill the gap completely is the solvent mixture; the nanoparticle deposits forming while solvent remains in between the plates could result in an incomplete fill. To address this possibility, another sensor was jetted using 15 mg of Au 25 nanoclusters in 600 μL toluene. The resulting sensor is shown in Fig. 5.6. Since toluene has a lower boiling point than most of the solvents inkjetted, the solvent evaporates quickly, allowing more drops to be deposited and resulting in the thick dark regions apparent in the image.This effect could also be achieved by drop casting or spin coating at higher viscosity to obtain a thicker film. 119

Fig. 5.6: Au25 sensor jetted in toluene

Analyte Acetone IPA Methanol Toluene Water

First Generation Sensor Sensitivity (% change in capacitance per ppm) 0.002065 0.006655 0.0033 0.00616 2.1

Sensor jetted in toluene (% change in capacitance/ppm) 0.00042 0.001381 0.0008 0.00064 0.17

Table 5.2: Sensitivity comparison table The sensitivity was measured for the sensor jetted in toluene, and compared to the original sensor (Table 5.2). The sensitivity dropped by a approximately a factor of 5 to 10 for the different analytes tested compared with the original sensor. As discussed in chapter 3, some reduction in sensitivity is expected for the higher pillar density design; however, with polyimide this reduction was less than 20%. Since the toluene evaporates much faster than DMSO, it is possible that the dissolved nanoclusters do not deposit effectively under the plate as the solvent dries, preventing a complete fill. To investigate this possibility, a sensor jetted 120

with toluene was etched using a focused ion beam (FIB) system to obtain a cross section; the FIB cuts are shown in Fig. 5.7.

Metal 3

Location of cut

Metal 1

Fig. 5.7: (a) image showing location of FIB cut and (b) FIB cross section

Metal 2: etched layer (flled with Au25)

The FIB cut was performed near the edge of the device, as shown in Fig. 5.7(a), away from the thick deposits covering the surface of the chip; as a result, this area would be a likely location if part of the gap did not fill with polymer. However, the FIB cross section showed a uniform layer in the etched metal 2 region; for the angled image shown, an air gap would be expected to show the top of the lower vias or some variation due to air bubbles. The area appears to be completely filled with nanoclusters; a second cut gave a similar result. Although the cross section could not be taken in every location, so an area of incomplete fill remains concievable, the most likely reason for the lower sensitivity is an unknown variation in the nanocluster material. The inability to dissolve the current batch of nanoclusters in DMSO supports this possbility, suggesting a chemical difference in the nanoclusters themselves.

121

5.2.2 Au38 and Au144 To test the effect of the size of the nanocluster on the sensing performance, sensors were fabricated with larger gold nanoclusters with similar dielectric coatings, Au38 and Au144. Due to the initial manufacturing yield issues with the vertical parallel plate sensor discussed in Chapter 3, both materials were instead deposited on the coated released interdigitated sensor topology. The material behavior can still be compared with the Au25 measurement, after correcting for the corresponding drop in sensitivity for the different topology; the polyimide sensor sensitivity was 42% lower for the coated released interdigitated design, and a similar drop is likely for the nanocluster materials. The sensors after jetting are shown in Fig. 5.8; both Au38 and Au144 were jetted in trichlorobenzene (TCB) .

(a)

(b)

Fig. 5.8: (a) Au38 and (b) Au144 sensors after jetting

The dry capacitance of each sensor was measured, with a capacitance of 1.39 pF and 820 fF for the Au38 and Au144 respectively. The dielectric constant for the comparable polyimide sensor was 1.0 pF, giving a dielectric constant of 4.0 for Au38 and 2.3 for Au144. Since Au25 had a

122

dielectric constant of 4.5, the dielectric constant appears to be inversely proportional to the size of the nanocluster used in the sensor. The sensitivity of the Au144 sensor was measured for several analytes (Fig. 5.9) with sensitivities of 0.00072%/ppm, 0.00052%/ppm and 0.00024%/ppm for ethanol, acetone and IPA respectively. Compared with the sensitivities measured for the original Au25 sensor, this is a reduction of more than a factor of 10. Due to the released interdigitated design for this sensor, a factor of two is likely due to the different topology, but the sensitivity appears to be lower than that of the Au25 sensor by at least a factor of five. This result is however closer to the measurements from the later batch of Au25. The humidity response was also measured (Fig. 5.10) and found to be 0.11%/%RH, 19 times lower than that of the original Au25 sensor.

0.9

Change in Capacitance (%)

0.8 0.7 0.6 0.5

Acetone

0.4

Ethanol

0.3

IPA

0.2

0.1 0 0

500

1000

1500

2000

Concentration (ppm)

Fig. 5.9: Au144 chemical response

123

%Change in Capacitance

6 5 4 3 2 1 0 0

20

40

60

%RH Fig. 5.10: Au144 humidity response The Au38 sensor was also measured for several analytes, but with no obvious response for large concentrations of industrial solvents. A small response to pulses of water vapor was measured (Fig. 5.11), with a sensitivity of 0.01%/%RH. This response was consistent for several different sensors containing Au38, with none showing responses except to large pulses of water vapor.

Change in Capacitance (%)

0.6 0.5 0.4 0.3

0.2 0.1 0 0

10

20 30 Relative Humidity (%)

40

50

Fig. 5.11: Au38 humidity response

124

5.3 Analysis and future work Since gold nanoclusters have not been previously tested for capacitive chemical sensing, the behavior was not known before testing.

The coating material was similar in all three

materials, so the main difference between the materials was expected to be the size of the gold core. The anticipated results were a dependence of both dielectric constant and sensitivity on the core size. The dielectric constant was expected to vary based on the assumption that the nanocluster core would act to short out portions of the dielectric material, resulting in a narrower effective gap. Individual large nanoclusters would have a larger effective dielectric constant by creating a conductive path through a larger distance, although smaller nanoclusters could have a higher packing density that might also result in a higher dielectric constant. A 2D simulation in COMSOL of closely packed circular particles in a dielectric matrix was performed. The inclusion of gold nanoclusters 1 nm in diameter with coating thickness of 0.3 nm was found to increase the effective dielectric constant by approximately a factor of two. An increase in the diameter conductive core from 1 nm to 1.4 nm, with the coating thickness held fixed, increased the effective dielectric constant of a material by a further 24.6%. Based on the measurement of the dielectric constant in the three devices, the dielectric constant was inversely proportional to the size of the nanocluster, suggesting that the packing density might vary significantly with the size of the particle. A study of the film structure would be necessary to confirm this hypothesis.. The results suggest that smaller nanoclusters had higher dielectric constants and would be more desirable to achieve a larger capacitance per unit area.

125

The sensitivity was expected to be higher the smaller the nanocluster, due to the larger number of potential bonding sites for absorbed chemical. However, this was not consistently shown in the measured results. Although all three nanocluster materials had similar dielectric coating layers, the chemical sensing behavior has been demonstrated to be dramatically different. Au25 shows a large response to most analytes tested, at least for the original sensor, Au144 shows a more modest response, and Au38 had almost no response to chemical analyte. A possible reason for this behavior is that, despite the similar coating layers, the different nanoclusters do not react identically to different chemicals. For instance, in the synthesis paper [128], both Au144 and Au25 are synthesized; the resulting products are then dissolved in acetone. Since Au25 is acetone soluble, and Au144 is not, the two nanocluster materials can be isolated. However, this difference in solubility also implies that the two nanocluster materials, with similar coating layers, are likely to absorb solvents differently. To truly understand the nanocluster materials, a more complete study should be performed, including a careful analysis of the morphology of the materials and experimentation with a broader range of nanocluster core sizes and coating materials. This was beyond the scope of this work, but would be a promising avenue for future research.

126

Chapter 6 Preconcentration 6.1 Introduction An end-of-service-life indicator must be able to accurately measure extremely small concentrations to be accepted as a practical system. The recommended exposure limit (REL) for a particularly dangerous chemical can be in the parts per billion range; for instance, benzene, a common industrial solvent and component of gasoline, has an exposure limit of 100 ppb [1]. Measuring chemicals in this range is difficult for most types of chemical sensors due to sensor and electrical noise limits, as well as low frequency sensor drift due to temperature and other factors. Previous chapters have discussed lowering the detectable limit by increasing sensor sensitivity and lowering system noise. An alternative method for lowering the minimum detectable concentration is to use a separate element called a preconcentrator to absorb chemical which then can be released in a concentrated pulse that can be more easily measured. This chapter will focus on efforts to add a preconcentrator to the chemical sensor system.

6.2 Background 6.2.1 Concentration factor In gas chemical sensing, a preconcentrator typically consists of a high surface area adsorbent material and a heater element. The heater element is used to abruptly heat the preconcentrator to an elevated temperature, causing the adsorbed analyte to be desorbed into the airflow. Since a preconcentrator is intended to amplify the concentration of analyte in a flow stream, the most important figure of merit is the amount of amplification, called the 127

concentration factor.

The concentration factor is defined as the ratio of the output

concentration to the concentration of the airflow before concentration [129].

The

concentration factor is in turn related to the amount of analyte that can be absorbed into the material, setting the total number of molecules that can be released in the pulse. The speed at which the element can be heated and the analyte desorbed determines the width of the preconcentration pulse. The concentration will vary since the analyte is not desorbed at a constant rate, but the average output concentration is: c precon

N precon wQ flow

(6.1)

where Nprecon is the number of atoms of analyte desorbed, w is the width of the pulse, and Qflow is the flow rate. The concentration factor is thus approximately: f

c precon

N precon

c flow

wQ flowc flow

(6.2)

where cflow is the concentration of analyte in the airflow before preconcentration. In practice the peak, rather than average, concentration is used, since that will be the maximum signal measured at the chemical sensor, assuming a sensor with instantaneous response. Since the amplification is dependent both on the quantity of analyte and how quickly it can be released, most preconcentrators are intended to maximize both speed and total adsorption. Surface adsorption of chemical is used rather than absorption into the material to allow faster desorption. The speed of the sensor does set an upper bound on preconcentrator speed, however. Although a preconcentrator able to release its load of analyte in less than a millisecond can achieve extremely high concentration factors, most types of chemical sensors 128

have absorption time constants on the order of seconds, so the preconcentrator pulse will not be detected.

6.2.2 Past preconcentrators 6.2.2.1 Tube preconcentrators Before modern micromachining techniques were developed, preconcentrators were commonly made by using a tube packed with adsorbent particles such as activated carbon, and wrapped with wire to form a resistive heater (Fig. 6.1). Due to the simplicity of the design and ease of fabrication, this type of preconcentrator remains in common usage, with numerous recent examples in the literature ([130]-[133]). This approach is particularly popular in gas chromatography due to the need to prevent breakthrough, analyte leaking through the preconcentrator and sensor exposure before the preconcentrator is pulsed.

In a tube

preconcentrator, additional adsorbent can be easily added to lengthen the time before breakthrough. Adsorbent particles Inlet

Wire (resistive heater)

Figure 6.1 Tube preconcentrator

129

A tube preconcentrator has several disadvantages over more modern micromachined devices. Wrapping a tube with wire and packing it with adsorbent particles is not a batch process, so manufacturing this type of preconcentrator in quantity will be more expensive. The tube is also a large thermal mass, requiring a large amount of power to heat the tube to an elevated temperature and release the absorbed analyte.

As a result, this type of

preconcentrator is primarily used for large benchtop gas chromatography systems where the added cost is small relative to the cost of the instrument and power is not a major concern.

6.2.2.2 Coated preconcentrators The first micromachined alternative to the tube preconcentrator was a large coated membrane preconcentrator developed by Sandia National Laboratory [129]. Fig. 6.2 shows a cross section of the preconcentrator design; the device consists of a 2.5 mm on a side nitride membrane with a patterned resistive heater coated by adsorbent sol gel. The airflow then passes across the top surface of the membrane. A variation on this technique using suspended hotplates with airflow vertically through the chip has also been demonstrated [134].

Si Nitride

Resistor Adsorbent

Figure 6.2 Sandia membrane preconcentrator

130

A hotplate preconcentrator has the advantage of a small, thermally isolated mass, allowing quick low power heating. This topology also has low adsorption volume due to the limited area coated with adsorbent material, limiting the maximum concentration factor for this type of preconcentrator. Due to this problem, a simple coated hotplate is not commonly used in more recent work; instead, more complicated 3D coated structures are used to maximize the surface area, as in the coated micropillars in [135] and [136] and the coated channels in [137]. Sandia also demonstrated a structure called a 3D-PC where a large number of vertical flow channels are etched all the way through a silicon chip and coated with adsorbent to maximize the surface area [137].

6.2.2.3 Packed microheater For applications such as gas chromatography where analyte breakthrough is an important concern, an alternative technology was developed at the University of Michigan based on packed microheaters [138]. The Michigan design consists of set of suspended silicon walls etched into a silicon chip. Adsorbent particles are then packed into the structure; the silicon walls are used both to restrain the adsorbent particles and as heater elements. A simplified top view of the structure is shown in Fig. 6.3. The primary advantage of this structure is that the high number of adsorbent particles allows a large amount of chemical analyte to be desorbed, allowing for a large analyte pulse and minimizing the possibility of significant concentrations of analyte breaking through.

131

Inlet

Adsorbent Silicon walls particles

Outlet

Figure 6.3 Top view of University of Michigan packed microheater The Michigan group has also demonstrated a multiple stage preconcentrator [139], where the flow passes through multiple stages of heater, each with a different adsorbent designed to target different analytes, to further reduce the risk of breakthrough. Due to the greater mass, a packed microheater design is typically higher power than the various microhotplate designs. Packing of the microheater can also add to the fabrication cost. In [138], the adsorbent was “manually transferred to the microheater structure and carefully packed between the heating elements,” which is likely to be much higher cost than more automatic coating methods used in the coated preconcentrator designs.

6.3 Microhotplate preconcentrator 6.3.1 Design Since the concentration factor is inversely related to the flow rate, as shown in (6.2), preconcentrators are mostly used with a controlled airflow generated at very low flow rates using flow controllers or a pump. For an ESLI system, the added cost and complexity of 132

incorporating flow control would likely make the system impractical. Forced flow does pass through the respirator regularly due to the breathing of the user. Human breathing rates are activity dependent, but are on the order of tens of liters per minute [140]. In order to maintain a reasonable concentration factor, the preconcentrator pulse must either be timed to target the period when the breathing flow rate is nearly zero or the sensor and preconcentrator must be placed in a high flow resistance channel such as a small capillary to lower the flow rate. A first generation preconcentrator was designed to have a small flow channel patterned using the thick permanent photoresist Microchem SU-8 on a silicon chip. Flow passes over a microhotplate preconcentrator followed by a gold nanoparticle chemiresistor. Diagrams of the preconcentrator system are shown in Fig. 6.4.

Unlike the Sandia microhotplate

preconcentrator [129], which was made using a large silicon nitride membrane, this design uses a silicon oxide membrane. Their membrane could be made hundreds of microns on a side due to the tensile residual stresses present in the nitride layer. Silicon oxide has compressive residual stresses, resulting in buckling for large membranes; to prevent this buckling, the preconcentrator was designed with a large thermally isolated silicon membrane intentionally left underneath the membrane, a technique demonstrated in [141] for a tin oxide microhotplate sensor system. SU-8

Cap

Adsorbent

Flow

Preconcentrator

Chemical sensor (a)

Silicon Island (b)

Figure 6.4 (a) Top view and (b) flow channel cross section of first generation preconcentrator design

Sensor 133

6.3.2 Fabrication Fig. 6.5 shows the fabrication process used to fabricate the first generation preconcentrator. The process begins with a 400 μm thick silicon wafer with 1 μm thickness oxide on both sides (Fig. 6.5(a)). A 50 nm thick gold electrode is then patterned on a 2.8 nm thick titanium adhesion layer by wet gold etching followed by ion milling of the titanium (Fig. 6.5(b)); a 0.5 μm thick gold layer is subsequently deposited on the wirebonding pads using a liftoff process. The process for the gold and titanium patterning was identical to those used for the non-integrated chemiresistor, and are described in section 2.3.1. The backside is then patterned for a silicon deep reactive ion etch (DRIE). To leave silicon underneath most of the membrane, thinner ring shaped openings were used over the center of the preconcentrator, while a wider ring was used for the outer edge. This technique, known as aspect ratio dependent etch (ARDE) modulation [142], relies on the tendency in plasma etching for larger openings to result in a higher etch rate. The etch recipe used is described in [142], and the total etch length was 5 hours and 10 minutes. This results in the cross section shown in Fig. 6.5(c). Next an isotropic silicon etch is performed to remove the vertical silicon left behind by the anisotropic silicon etch (Fig. 6.5 (d)). 50 μm thick SU-8 is then patterned on the top surface to form the flow channels (Fig. 6.5(e)).

134

(a)

(b)

(c)

SiO2

Au

Si

Photoresist SU-8

Ti

(d)

(e)

Figure 6.5 First generation preconcentrator process flow Fig. 6.6(a) shows the underside of a preconcentrator after the anisotropic etch. In addition to the desired larger vertical etch rate, the larger outer ring shaped opening also experienced a larger lateral etch rate, resulting in the gradual removal of the outer silicon rings. Although this effect was not anticipated, a silicon island will still be left in the desired area with proper timing, although with a slightly wider oxide membrane than expected. Fig. 6.6(b) shows a preconcentrator after the isotropic etch, resulting in the removal of the silicon rings and leaving behind the silicon island. The initial layout is shown in Fig. 6.6(c). The remnants of the rings on the backside have no effect on the behavior aside from the addition of a small amount of additional mass.

135

(b)

(a)

(c)

Figure 6.6 Bottom of preconcentrator (a) after vertical silicon etch and (b) after isotropic etch; (c) shows the initial layout In fabrication, buckling was evident in most of the etched membranes, primarily due to the wider than anticipated membrane width; this can be seen in Fig. 6.7 (a). In many devices, the lead to the device was broken by the buckling, resulting in an unusable heater. An alternative design used silicon bridges across the membrane (Fig. 6.7(b)), creating a more robust device; this structure will also require more power to heat, due to the higher thermal conductivity to the substrate.

(a)

(b)

Figure 6.7 Preconcentrator design (a) with released membrane and (b) with silicon bridges to the substrate

136

6.3.3 Characterization The temperature response for each of the two preconcentrator designs was measured for different applied heating power (Fig. 6.8). As expected, the released preconcentrator had a larger response to applied power, rising to 250°C for 300 mW applied power with a slope of 590°C/W, while the device with preconcentrator bridges heated to 60°C for the same power level, with a slope of 125°C/W.

90 300

Temperature (degrees C)

80

Temperature (degrees C)

250 200 150

100 50 0 0

100

200 Power (mW)

(a)

300

400

70 60 50 40

30 20 10

0 0

100

200 300 Power (mW)

400

500

(b)

Figure 6.8 Temperature response (a) with released membrane and (b) with silicon bridges to the substrate To make a functional preconcentrator, adsorbent material must be added to the device. Since a microhotplate does not have a very large surface area, coating the plate would not give a very large concentration factor. An alternative possibility is to deposit high surface area particles such as activated carbon on the top surface of the microhotplate. These particles can then be heated to release the adsorbed analyte. Shadow masking was used to selectively deposit FHJ 400 (Nichem) activated carbon on the microhotplate without carbon landing on the sensor (Fig. 6.9).

137

The shadow mask consisted of a through etched silicon chip with a hole comparable in size to the preconcentrator; this hole was then aligned with the heater using a device bonder. The mask was bonded to heat release tape on top of silicon chips comparable in size with the preconcentrator chip, resulting in a gap of at least a few hundred microns between the mask and the heater during deposition. As shown, the activated carbon was primarily on the device, but carbon was also deposited in other areas as the mask was removed, including on top of the SU-8 layer. Since carbon on top of the SU-8 layer would make sealing the flow channel difficult, this method of deposition would likely not be practical without better control of the activated carbon deposition.

(a)

(b)

Figure 6.9 (a) Heater viewed through shadow mask and (b) preconcentrator after deposition of activated carbon through mask

6.4 Vertical flow preconcentrator 6.4.1 Fabrication Due to the difficulties in depositing material and sealing the chip, an alternative topology based on the Sandia 3D preconcentrator was devised. By using vertically etched 138

channels through a silicon chip, Sandia National Labs was able to create a high surface area silicon structure that could then be coated with adsorbent material and used as a preconcentrator.

This method has the advantage of simplified fabrication and low flow

resistance due to the large number of parallel flow paths. The two mask fabrication process flow for the second-generation preconcentrator is shown in Fig. 6.10. The process begins with a bare 300 μm thick silicon wafer (Fig. 6.10 (a)). A layer of platinum 100 nm thick is patterned on a 2.5 nm titanium adhesion layer using ion milling (Fig. 6.10(b)). The wafer is then patterned for the through holes, mounted on a carrier wafer with heat release tape and etched using a standard DRIE silicon etch (Fig. 6.10(c)). The preconcentrator wafer was then removed from the heat release tape by heating to 120°C for several minutes.

The preconcentrator was then coated by adding Tenax TA (Scientific

Instrument Services), a polymer based on 2.6-diphenylene oxide [143], in methylene chloride either by drop casting or dip coating (Fig. 6.10(d)). The concentrations of methylene chloride used are discussed in more detail in section 6.44.

Si Ti Pt Tenax TA

(a)

(b)

(c)

Air flow

(d)

Figure 6.10 Fabrication process flow for vertical flow preconcentrator

139

The second-generation preconcentrator consisted of an array of holes etched through a silicon disk (Fig. 6.11 (a)). The disk is isolated from the substrate by 8 silicon microbridges and heated using a platinum ring heater. The holes used were hexagons 33 microns in diameter, allowing over 8000 to be densely packed in the structure (Fig. 6.11(b)). The hexagonally close packed holes are expected to give a surface area of 225 mm2 for the 4 mm diameter circular preconcentrator, a factor of 18 increase in surface area over a flat hotplate structure. Since the structure is made primarily from single crystal silicon, it is very mechanically robust; despite having the thermally isolated region attached to heat release tape and dicing tape during the process, none of the 55 devices on the wafer were damaged during processing.

140

Ring Heater

Silicon Microbridge

Pad

(a)

(b)

Figure 6.11 (a) Diagram of vertical flow preconcentrator and (b) picture of final structure

6.4.2 Thermal modeling

The preconcentrator was modeled using the lumped thermal capacitance model shown in Fig. 6.12, modeling the heat capacity of the central mass as a thermal capacitance and the heat loss to the substrate as lumped thermal conductances. The loss to the substrate is primarily through the eight silicon bridges and the air in the gaps.

141

Pheater

Csi

RBridges

RAir

Figure 6.12 Lumped thermal model of preconcentrator The thermal capacitance is [144]:

CSi

VSi

Si cSi

(6.4)

where VSi, ρSi and csi are the volume, density and specific heat capacity of the silicon disk. For the fabricated preconcentrator, only approximately 15% of the silicon mass in the disk remains, giving a total thermal capacity of 9.4x10-4 J/°C. The thermal conductance of a silicon bridge is [144]: G

Si hw

(6.5)

L

where h, w, and L are the height, width and length of the bridge, and κsi is the thermal conductance of silicon. The thermal conductance of a single silicon bridge is 0.074 W/°C, giving a combined thermal conductance of 0.592 W/°C. The thermal conductance of the air gap was calculated to be 0.026 W/°C, a small portion of the total loss to the substrate. The temperature at time t after power is turned on is given by the expression [144]:

T

T

(Ti

T )e

tGtotal / C

(6.6)

142

where Ti and T∞ are the initial temperature and the temperature at time infinity respectively and Gtotal is the total thermal conductance incorporating the conductances of the eight bridges and the air gap. C/Gtotal is the time constant τ of the thermal system. The temperature where all the input power flows out to the substrate is:

T

Ti

Pheater Gtotal

(6.7)

For a 1.4 W input power, corresponding to 12 V applied to the heater resistor, the calculated temperature rise is small, only 4° C, with a time constant of 1.5 ms. The steady state temperature response of the preconcentrator was measured and the device was found to give a linear response to input power (Fig. 6.13(a)), with a slope of 90.2°C/W and a maximum temperature of 154°C for 1.4 W input power. The silicon bridges to the substrate result in power loss to the substrate and the smaller response compared to the original microhotplate preconcentrators. The turn on transient was also measured for 1.4 W input power (Fig. 6.13 (b)), with a time constant of 20 seconds.

(a)

(b)

Figure 6.13 (a) Steady state temperature response and (b) transient response to 1.4 W input power 143

Both of the these results are different from the responses predicted by the theoretical model. One possible reason for this discrepancy is that, in the real preconcentrator, the thermal mass is not a single point; instead, it is distributed over a large area and linked to the silicon bridges with other silicon beams that will add a large thermal resistance in series with the estimate in the theoretical model. To more accurately model the device, a 3D heat transfer simulation was run in COMSOL; the silicon disk was modeled as a material with thermal conductivity and heat capacity 18% of silicon to incorporate the effects of the holes on the thermal behavior. The more accurate distributed finite element model of the silicon disk results in a large increase in the temperature increase, with a maximum temperature of 383°K, or 110°C. The thermal time constant was approximately 70 ms, which is larger than the theoretical model but still faster than the measured behavior.

Figure 6.14 Finite element model of preconcentrator heating

144

Although the COMSOL model is much closer to the behavior measured for the preconcentrator, there remains a significant 32% difference in the final temperature. In making the model, the amount of silicon remaining was judged based on the silicon visible on the top surface. However, the DRIE etch is not perfectly anisotropic; silicon may also be removed underneath the visible area. One possible way this might occur is if the etch was not stopped exactly when the bottom of the wafer was reached; plasma could then reflect and etch the bottom of the wafer. To verify if this had occurred, the backside of the preconcentrator was imaged using an SEM (Fig. 6.15). Based on the pictures, it is clear that at least 50 μm of silicon was etched vertically from the backside, a sizeable fraction of the total thickness of the 300 μm thick silicon wafer. Some lateral etching is also evident on the bottom of the wafer, which would reduce the amount of silicon even further.

(a)

(b)

Figure 6.15 (a) Backside picture of preconcentrator and (b) silicon bridge at higher magnification Adding the 50 μm etch of the backside to the COMSOL model increases the maximum temperature to 120°C. The lateral etching is more difficult to estimate accurately, but likely 145

accounts for some of the remaining discrepancy. The measured time constant remains slower than predicted by the model; due to the large amount of power lost to the substrate, the most likely explanation is that the entire chip is heating as well, and this larger thermal mass results in the larger time constant.

6.4.3 Flow system For flow testing, a custom mount (Fig. 6.16) was designed to allow the preconcentrator to be placed in the chemical sensor flow system. The preconcentrator is mounted on a printed circuit board (PCB) with a drill hole slightly larger than the silicon disk using conductive epoxy; the pads are several millimeters in size, allowing manual positioning. The printed circuit board is then sandwiched between machined Teflon spacers and stainless steel blocks containing a threaded pipe fitting, creating a small sealed chamber for the preconcentrator.

Pipe fitting

Stainless steel Teflon Preconcentrator PCB

(a)

Figure 6.16 Diagram of preconcentrator mount

146

In previous testing of chemical sensors, the flow system was run with a 1 L/min nitrogen flow. This flow rate is very high for a preconcentrator; most micromachined preconcentrators are run at flow rates of a few mL/min to obtain larger concentration factors.

For

preconcentrator testing, the flow system was modified to allow testing at relatively small concentrations and lower flow rates (Fig. 6.17). A low concentration is desirable because preconcentrators can saturate, reach a maximum volume of chemical analyte, at higher concentrations, lowering the concentration factor. The M6 solvent pump is used to pump chemical analyte into a heater block with a high nitrogen flow rate of 1 to 2 L/min controlled by one flowmeter. Most of this flow is then passed to an exhaust where an adjustable valve is used to control the flow resistance to dilute the flow into a larger 1 to 2 L/min flow, generating a low concentration signal. For testing at low flow rates, another flowmeter directs a few mL/min into the preconcentrator and test box, while the remaining flow passes to a second exhaust. 10 mL/min

1 to 2 L/min

Flowmeter

Flowmeter 1 to 2 L/min

Heater Block

Test Box/ Preconcentrator

Exhaust

Flowmeter Exhaust Adjustable Analyte source (M6 Valve pump)

Figure 6.17 Flow system for preconcentrator testing

147

A test was run using the adjustable valve to control the dilution of 500 ppm toluene in the system (Fig. 6.18); a C8 gold nanoparticle sensor was used to measure the concentration. By varying the flow impedance of the exhaust using the adjustable valve, the rate of flow into the other flow stream could be adjusted successfully, controlling the concentration of analyte. For the lowest setting shown, the concentration was reduced by a factor of 30 compared with the concentration with the valve completely off.

Figure 6.18 Flow system dilution test

6.4.4 Preconcentrator coating For preconcentrators aimed at volatile organic carbons with 5 to 12 length carbon chains, the most commonly used adsorbents are high surface area materials such as Tenax and various activated carbons [135]. A more complete discussion of common preconcentrator materials used in gas chromatography for a wide range of analytes is also given in [145]. The activated carbon materials are typically deposited as particulate, since the structure of the carbon sets the surface area and would be lost if dissolved. The polymer Tenax TA has been 148

demonstrated to maintain its adsorption properties after deposition from solution, and in [135] a successful preconcentrator was demonstrated with Tenax TA deposited using inkjet deposition in methylene chloride. Based on this result, Tenax TA was chosen to coat the vertical flow preconcentrator. Several different preconcentrators were coated using drop casting with different concentrations of Tenax TA in methylene chloride.

The highest concentration used was

27mg/mL, and resulted in many of the pores being completely clogged with Tenax TA. At lower concentrations, down to a few mg/mL, very few pores were completely clogged. Fig. 6.19 shows a preconcentrator after deposition of Tenax TA; this preconcentrator was deposited at higher concentration, resulting in a large number of clogged pores. As shown in Fig. 6.19 (b), little or no coating is visible in the unclogged pores; Tenax appears to either deposit completely or with little visible coating.

(a)

(b)

Figure 6.19 (a) Preconcentrator after deposition of Tenax TA and (b) magnified image of individual holes

149

A second set of preconcentrators were coated by the company Restek with the material Q Bond, another polymer used as a stationary phase in gas chromatography columns. Unlike the Tenax TA preconcentrators, Q Bond was not dissolved in solution; instead, it was deposited in the form of sub-micron particles on the surface of the silicon by pulling a particle laden liquid through the device under pressure. Fig. 6.20 shows a preconcentrator after Q Bond deposition. Q Bond appears as small dust-like particles covering much of the holes; the coating of particles is small in most cases, with layers on the order of a few hundred nanometers typical.

(a)

(b)

Figure 6.20 (a) Preconcentrator after deposition of Q Bond and (b) magnified image of individual holes

6.4.5 Analyte testing Initial analyte tests were performed on the preconcentrator at 100 mL/min nitrogen flow rates. Fig. 6.21 shows tests for three different Tenax coated preconcentrators with different percentages of pores completely clogged; the same C8 nanoparticle sensor was used for all four tests, allowing for direct comparison.

In the case of the mostly clogged

preconcentrator, the preconcentrator was exposed to 15 minutes at 1000 ppm of toluene; the 150

other preconcentrators were exposed to 20 minute pulse of 1500 ppm toluene to load the preconcentrator before being pulsed. Due to the high concentrations and long exposure times, the preconcentrator is expected to be completely loaded with analyte in all three tests. In most tests with the preconcentrators, the analyte pulse took the form of a two distinct pulses, seen most clearly in Fig. 6.21 (c). The first pulse is believed to be analyte desorbing into the flow stream as the preconcentrator is quickly heated; the concentration of analyte then drops when most of the analyte has been removed. The second pulse is the temperature increase in the test box as the system heats up due to the large amount of power in the preconcentrator. This pulse will increase indefinitely if the preconcentrator is left on, and has also been confirmed with the on-chip temperature sensor.

151

Precon

585000

2.5 2 1.5 1 0.5 0 -0.5

583000 581000 579000

577000 75

80

85

Power (W)

Resistance (Ohms)

Sensor

90

Time (minutes)

640000 639000 638000 637000 636000 635000 634000 633000 632000 631000 630000

2.5 2 1.5

1 0.5

Power (W)

Resistance (Ohms)

(a)

0 -0.5 70

75

80

85

90

(b)

484000 483000 482000 481000 480000 479000 478000 477000 476000

2.5 2 1.5 1 0.5

Power (W)

Resistance (Ohms)

Time (minutes)

0 -0.5 32

33

34

35

36

Time (Minutes)

(c)

Figure 6.21 Preconcentrator analyte testing with approximately (a) 10% (b) 30% and (c) 90% of pores clogged with Tenax TA

As shown in Fig. 6.21, the largest analyte peak occurs for the preconcentrator with almost all of the holes completely clogged with Tenax TA; the pulse is approximately 3.2 kΩ in height, corresponding to a maximum concentration of 60 ppm. For the preconcentrator with 152

only 30% of the holes clogged, a 1.2 kΩ pulse occurs, corresponding to a concentration of 30 ppm. For the final preconcentrator, with very few holes completely clogged with Tenax TA, no analyte pulse is visible, with only a small temperature transient apparent. The tests were also repeated with a second preconcentrator with similar coatings, and the results were comparable in each case; a larger number of clogged pores corresponds to a higher analyte pulse in testing. Since the SEM also showed little or no coating apparent in the unclogged pores, this further suggests that pores are either completely clogged or have minimal coating. The test was repeated for a Q Bond preconcentrator (Fig. 6.22(a)), and found to have a similar two peak response, with a peak concentration of toluene measured to be 40 ppm based on the resistance change and the sensitivity measurement in Fig. 2.6. Fig. 6.22(b) shows the analyte peak for a 90% clogged preconcentrator and the Q Bond preconcentrator on a single plot. The Tenax TA preconcentrator has a slightly larger peak concentration; however, the peak is also much shorter than the Q Bond preconcentrator, with most of the analye appearing over one minute, while the Q Bond peak is spread over more than three minutes. Even though the peak concentration is lower, this result suggests that the total volume of chemical adsorbed on the surface is higher for the Q Bond preconcentrator; the slower desorption is likely due to a stronger bond with the surface than in the Tenax case.

153

Precon

1070000

2.5

1065000

2

1060000

1.5

1055000

1

1050000

0.5

1045000

0

1040000

Power (W)

Resistance (Ohms)

Sensor

-0.5 65

70

75

80

Time(Min)

(a)

Concentration Toluene (ppm)

70

60 50 40

QBond

30

Tenex

20

10 0

-0.5

0.5

1.5

2.5

Time (Min)

(b)

Figure 6.22 (a) Preconcentrator pulse for Q Bond preconcentrator and (b) comparison with analyte pulse from 90% clogged Tenax preconcentrator A further test was run to determine the effects of the turn on power on the analyte pulse from the preconcentrator. Higher power and thus higher temperature is expected to result in a larger but shorter preconcentrator pulse. Fig. 6.23 shows the analyte peaks from a 90% clogged Tenax preconcentrator run at 0.5, 1.4 and 2.2 W input power. The 2.2 W peak is the largest, as expected, with a peak concentration of approximately 75 ppm of toluene; the peak in the 1.4 W case is lower, 60 ppm, while peak in the the 0.5 W case is significantly 154

smaller, at approximately 20 ppm. The pulse width is comparable for both the higher power peaks, but the 0.5 W peak is distinctly slower and wider than both of the other two, as expected. The total volumes of toluene vapor, calculated based on the area under the curve and assuming the concentration is uniformly distributed in the cross section of the flow tube at any given moment, were 0.6, 1.0 and 1.3 mL for powers of 0.5 W, 1.4 W and 2.2 W respectively.

90

Concentration (ppm)

80

70 60 50

2.2W

40

30

1.4W

20

0.5W

10 0 0

0.2

0.4

0.6

0.8

1

Time (Min)

Figure 6.23 Analyte pulses for different input powers

6.4.6 Low flow testing Although analyte has been successfully detected from the preconcentrator, in these tests the concentration factor is much less than one; the pulse of analyte measured from the preconcentrator is much smaller than the concentration the preconcentrator is exposed to before pulsing. The high flow rate is a large factor in this result; the preceding tests were run at 100 mL/min flow, compared with a flow rate of a few mL/min more commonly used in preconcentrator testing.

This higher flow rate results in an immediate reduction in 155

concentration, based on (6.1). If the preceding test was run at a flow rate of 6 mL/min, the maximum analyte pulse in Fig. 6.23 would be 1250 ppm instead of 75 ppm. The preceding tests have also been run under the assumption that the preconcentrator was completely saturated; exposing the preconcentrator to a lower concentration for a longer time period could also be used to obtain a larger concentration factor. A 90% clogged Tenax preconcentrator was run at 6 mL/min with a 300 ppb pulse of toluene (Fig. 6.24(a)). The 300 ppb pulse of analyte was turned on for 20 minutes, followed by turning on the preconcentrator at 2.2 W input power. The shape of the output analyte pulse was similar to the previous tests; however, the pulse is larger than expected, a concentration of 24 ppm or a concentration factor of 80. A second test, again with the preconcentrator at 2.2 W

Sensor

2.5

1.5 1 0.5

0 -0.5 90

110

130

Time(Min)

(a)

150

Resistance (Ohms)

2

Precon

1880000

2.5

1860000

2

1840000

1.5

1820000

1

1800000

0.5

1780000

0

1760000

Power (W)

1980000 1970000 1960000 1950000 1940000 1930000 1920000 1910000 1900000

Power (W)

Resistance (Ohms)

input power, was run with no analyte (Fig. 6.24(b)), giving a similar analyte pulse.

-0.5 0

20

40

60

Time(Min)

(b)

Figure 6.24 Preconcentrator test (a) with and (b) without preceding 300 ppb toluene pulse One possible reason for this behavior would be leakage of water vapor into the flow system during the test. The flow system and test chamber depend on positive pressure to 156

prevent leakage into the system; at high flow rates, the hundreds of mL/min normally used in the flow system, nitrogen leaking out minimizes the flow of water vapor or other analytes into the system. However, at very low flow rates, the positive pressure is insufficient to prevent leakage, resulting in significant concentrations of water vapor in the system. The relative humidity was measured in the flow system at 6 mL/min nitrogen flow and found to be greater than 10%. To address this problem, the low flow portion of the flow chamber was sealed in a 13 gallon plastic bag, with several liters per minute of nitrogen flow into the bag (Fig. 6.25). A test of the relative humidity in the system after the nitrogen flow was turned on was also performed (Fig. 6.25 (b)), with the relative humidity dropping below 5% after 100 minutes and dropping below 1% humidity several hours later.

Relative Humidity (%)

50 40 30 20 10 0 0

20

40

60

80

100

Time (Min)

(a)

(b)

Figure 6.25(a) Test system with nitrogen around testing box and (b) test demonstrating reduction in relative humidity inside test box The preconcentrator was also tested at 6 mL/min with nitrogen around the test box (Fig. 6.26). The first pulse shows a significant analyte pulse; this is common, and results from residual water vapor and chemical on the device from being previously exposed to the outside environment. The second and third pulses show no analyte pulse, demonstrating that running the system with nitrogen around the test box eliminates the undesired pulse with no analyte. 157

2.5 2

1.5 1 0.5

Power (W)

Sensor Resistance (Ohms)

7920000 7900000 7880000 7860000 7840000 7820000 7800000 7780000 7760000 7740000 7720000

Sensor Precon

0 -0.5 0

50

100

150

Time(Min)

Figure 6.26 Preconcentrator pulses with nitrogen around testing box

The temperature transient is another undesirable effect that tends to interfere with the analyte response. Based on the response of the on-chip RTD, the steady-state temperature rise (the temperature after the preconcentrator is turned on for an extended period) was estimated to be 1.5°C at the sensor, which in the test box is 5 cm from the preconcentrator heater. The flow system was modeled using COMSOL, using the measured temperature to estimate the heat loss along the length of the flow system (Fig. 6.27). The temperature was found to fall off very rapidly as the length of the flow tube increased. If the distance between the sensor was doubled to 10 cm from the current length of 5 cm, the temperature rise would drop to 0.3°C. The addition of a further 5 cm would drop the increase to 0.1°C.

158

Figure 6.27 Finite element model of thermal behavior of flow system; vertical and horizontal axes are in meters, temperature in degrees Kelvin. The preconcentrator is at the 5 cm horizontal mark. Based on this result, a length of metal tubing approximately 10 cm long was added to the flow system between the test box and the preconcentrator mount. A test was run with the system to verify the effects of the metal tubing (Fig. 6.28). Fig. 6.28 (a) shows the response to three preconcentrator pulses of both the sensor and the on-chip temperature sensor before the addition of the metal tubing without analyte. Fig. 6.28 (b) shows a similar test after the metal tube was added. The preconcentrator power in both tests was 2.2 W. After the addition of the metal tube, the temperature response is clearly reduced, with the variation in the RTD output dropping from over 100 ohms over the course of the original test to less than 20 ohms after the tube was added.

159

Sensor Resistance 9650000

36525

9600000

36505

9550000

36485 36465

9500000

36445

9450000

36425

9400000

36405

9350000

36385

9300000

RTD resistance (Ohms)

Resistance (Ohms)

RTD Resistance

36365 0

100

200

300

Time (Min)

(b)

9050000

2 1.5

9000000

1

8950000

0.5

8900000

0

8850000

-0.5

0

100

200

300

Precon

9620000 9600000 9580000 9560000 9540000 9520000 9500000 9480000 9460000 9440000

30 25

20 15 10

Power (W)

2.5

Resistance (Ohms)

9100000

Sensor

Power (W)

Sensor Resistance (Ohms)

(a)

5 0 0

100

200

Time(Min)

Time (Min)

(c)

(d)

300

400

Figure 6.28 Preconcentrator testing (a) without and (b) with metal tube added between test box and preconcentrator. (c) and (d) show the corresponding preconcentrator inputs The preconcentrator was tested with the additional metal tube in the system. Two tests were run, the first with no analyte in the system (Fig. 6.29 (a)), and the second with a 30 minute pulse of 30 ppm of toluene in the system before the preconcentrator was pulsed (Fig. 6.29(b)). The larger concentration of toluene was used instead of 300 ppb to obtain a visible analyte pulse before the preconcentrator pulse to verify that the preconcentrator was successfully exposed. In each test, the preconcentrator was pulsed twice, each time for 5 minutes, leading to the two peaks shown. In both tests the pulse from the preconcentrator was small, on the 160

order of 10 to 15 ohms; the test with toluene appears marginally larger, but no clear preconcentration is evident. Since there was little response to the higher concentration, the 300 ppb test was not repeated.

2.5

9390

2 1.5

9380

1 9370

0.5

9360

0

9350

-0.5 0

20

40

Resistance (Ohms)

9400

Precon Power

9000

2.5

8990

2 1.5

8980

1 8970

0.5

8960

Power (W)

Sensor

Precon Power

Power (W)

Resistance (Ohms)

Sensor

0

8950

-0.5 340

350

360

370

380

Time (Min)

Time (Min)

(a)

(b)

Figure 6.29 Preconcentrator pulses with metal tube (a) without and (b) with exposure to toluene

6.5 Discussion and Future Work Two designs of preconcentrator intended to amplify the concentration of analyte were designed. Despite extensive testing and experimentation, neither design was successfully demonstrated with a preconcentration factor greater than 1. The first design was found to be impractical for deposition of preconcentrator material, and would have had a very small surface area due to the microhotplate design. The second design addressed these problems by creating a large surface area by etching numerous holes through a silicon chip. Although these improvements did allow deposition of adsorbent material and testing, the preconcentration factor remained less than one for all of the tests performed, meaning that the preconcentrator did not amplify the concentration. 161

This failure most likely comes from a combination of several factors. A uniform, thick deposition of adsorbent material was never demonstrated, despite a number of attempts, with the preconcentrator holes tending to either clog completely or have a very thin layer. By creating a thicker, higher surface area coating on the preconcentrator, a larger amount of analyte would be released upon heating the device. A further study of coating techniques, as well as a modified design with larger holes less likely to clog, would be a potential avenue of future research. The current design also does not heat uniformly due to the heater design; it is possible that only the regions with the heater are heating to the levels estimated, so only a small portion of the adsorbed analyte is being released when the preconcentrator is pulsed, or the preconcentrator pulse is very spread out due to gradual heating of the device.

A

redesigned device with faster more uniform heating might also create a more concentrated analyte pulse.

162

Chapter 7 Conclusions and Future Work 7.1 Contributions Creating a successful end-of-service-life indicator creates numerous challenges for a sensor designer, with tight restrictions in limits of detection, stability, size, power and cost. This thesis explored a series of approaches for creating and improving an integrated chemical sensor system intended for use in a respirator cartridge. A C8 gold nanoparticle chemiresistor was characterized for use to detect solvents and other volatile organic carbons that might be used in an industrial setting. Since a sensor within a respirator cartridge cannot be maintained in a controlled environment, an emphasis was placed on the response to environmental factors such as temperature and humidity. As with many types of chemical sensors, the chemiresistor responded strongly to these factors as well as to chemical analyte, requiring the inclusion of temperature and humidity sensors in the final system to account for these effects.

A technique

proposed in the literature for lowering the limit of detection by modulating the input voltage was also attempted, without a clear improvement in the response, and a theoretical analysis explaining this result was developed. The next section of the thesis focused on creating an integrated capacitive humidity sensor to compensate for the humidity response of the chemiresistor. Due to the low limit of detections necessary for certain industrial chemicals, the humidity sensor must have very high sensitivity as well to remove even small transitions in water vapor. Initially work was focused on creating an improved interdigitated sensor by using the standardized CMU CMOS-MEMS fabrication process, with sensors several times more sensitive than past integrated sensors and roughly half of the material limit of the 163

polyimide used. An alternative fabrication method was then developed based on filling with polyimide a void created by an etched aluminum layer, with sensitivities approximately equal to that of the material limit. The circuitry was analyzed to attempt to further reduce the limit of detection of the system. In initial testing, a charge-based capacitance measurement circuit was used to measure the capacitance by charging and discharging the sensor capacitance and using an off-chip picoammeter to measure the resulting current. The noise, charge injection and clock feedthrough were analyzed to develop a theoretical understanding of the circuit, with a final determination that the sensor setup was limited by the noise levels of the external picoammeter. To reduce the noise level, a Colpitts LC oscillator was designed with theoretical limits of detection several orders of magnitude lower than the limits previously demonstrated in capacitive chemical sensing. A frequency counter was used to convert the oscillator output to a low frequency digital output, with an integrated PTAT temperature sensor circuit used to measure the output temperature. In addition to measuring humidity, chemicapacitive sensors can be used to measure other types of analytes, including volatile organic carbons.

The possibility of using the high sensitivity

chemicapacitive topology to measure industrial solvents was explored in the next section by examining high resistivity gold nanocluster materials for use for capacitive sensing. One material, Au25, was found to have very high sensitivity to most analytes tested, with responses several times larger than polyimide, but with difficulty obtaining repeatable performance in later sensors. Two other nanocluster materials, Au38 and Au144, were also tested, with several times lower sensitivity measured for Au144 and negligible response for Au38 for most analytes tested.

164

In the final portion of the thesis, a method for lowering the limit of detection by adding a preconcentrator to the system was evaluated. A preconcentrator would amplify the concentration of analyte, allowing a sensor with a higher limit of detection to be used. Several designs were fabricated for use as a heater in a preconcentrator system.

An initial design, a planar microhotplate

preconcentrator, was found to be difficult to coat with adsorbent material and test for preconcentration, leading to a second design with a larger structure consisting of numerous vertical holes etched through a silicon chip. The second generation preconcentrator was successfully coated with several preconcentrator materials and tested, although analyte amplification was not demonstrated, likely due to problems with obtaining uniform thick coatings of adsorbent material on the silicon heater.

7.2 Conclusions The end-of-service-life indicator problem is sufficiently complicated and difficult that it is not possible to make a definite statement whether the integrated system proposed would be better than the alternatives that have been demonstrated such as color changing strips. Both the resistive and capacitive gold nanoparticle films tested for detecting industrial solvents showed high sensitivities, but also were highly sensitive to environmental factors; these can be corrected for to some extent, but remain important problems for any sensor in this application. A detailed study by an expert on techniques such as principal component analysis and pattern recognition would be necessary to truly determine whether an array of these types of sensors could practically differentiate between a comparatively safe analyte such as toluene and a deeply hazardous one such as benzene. There is a significant possibility that a far more complicated system such as a micro gas chromatograph, with additions such as controlled flow and gas chromatograph channels, might be necessary to achieve the desired limits and selectivity, which would be more costly than the system proposed. This becomes 165

particularly important in an application such as occupational safety where a false negative could result in very serious consequences for the user. If the more costly and complicated system was necessary, a regulatory decision would have to be made about whether the additional cost in what is now a cheap disposable cartridge would truly justify the benefits. The capacitive sensor work is the most complete and mature, and is expected to have the greatest impact. The chemicapacitive sensors developed here demonstrate that, with a handful of postCMOS processing steps, a large improvement in sensitivity can be achieved. The sensor topologies developed are close to the fundamental sensitivities of the polyimide layer used, so the benefit of further changes to the sensor is expected to be small. Since there have been a number of successful commercial products with standalone integrated capacitive humidity sensors, notably from the company Sensirion, the more sensitive sensors shown here could be envisioned as practical products with little further refinement.

7.3 Future work This work develops a number of important improvements to an integrated chemical sensor system, particularly with respect to integrated chemicapacitive sensors. However, several important research areas in the thesis could not be evaluated in their entirety, and remain worthy of further study; these are summarized in the following sections. Colpitts LC oscillator: Although the Colpitts oscillator designed was tested, and oscillation was demonstrated, the performance and chemical sensing behavior was not as successful as expected in the simulated design. Further work exploring the causes of this behavior and refining the design could allow limits of detection closer to that expected in the initial simulations.

Possible changes include

redesigning the sensor chip to more carefully decouple the on-chip oscillators and optimizing the testing

166

board to minimize the effects of external capacitance and inductance. Other types of LC oscillators may also allow similar limits of detection with less stringent startup conditions. Gold nanoclusters: The preliminary data presented in this thesis demonstrates that gold nanoclusters based materials are capable of significant responses to several chemical analytes. However, there are a number of different areas that could be studied in evaluating the behavior. The three nanocluster materials tested in this thesis had similar coating materials; varying the coating material and testing the chemical response could provide a better understanding of the chemical response. The development of a model of the sensing behavior, as was done for the C8 nanoparticles in [23], by testing a wider range of chemical analytes would also add valuable insight into the materials. Preconcentrator: In fabricating the preconcentrator used, difficulty was found in obtaining a coating over the full surface area of the device. Further research in this area could focus on both redesigning the vertical etch holes to allow easier deposition, as well as trying other materials and deposition techniques. There is also potential for improving the design of the heater to obtain faster more uniform heating, which would also increase the size of the analyte peak and improve the preconcentration.

167

References: [1] NIOSH Pocket Guide to Chemical Hazards, 2011. [Online]. Available: www.cdc.gov/niosh/npg [2] “Respiratory protection,” OSHA Occupational Safety and Health Standards, Standard Number 1910.134. [3] G. Favas, “End of service life indicator for respirator cartridges, part I: Literature Review,” Australian DoD, DSTO-TN-0657, 2005. [4] G. L. Eian, “Cartridge respirator with service life indicator,” U.S. patent 4,326,513, Apr. 27, 1982. [5] K. Leichnitz, “Colorimetric indicator for the indication of the exhaustion of gas filters,” U.S. patent 4,684,380, Aug. 4, 1987. [6] D. D. Dominguez, R. Chung, V. Nguyen, D. Tevault, and R. McGill, “Evaluation of SAW chemical sensors for air filter lifetime and performance monitoring,” Sensors and Actuators B, Vol. 53, pp. 186-190, 1998. [7] B. H. King, A. M. Ruminski, J. L. Snyder, and M. J. Sailor, “Optical-fiber-mounted porous silicon photonic crystals for sensing organic vapor breakthrough in activated carbon”, Advanced Materials, Vol. 19, pp. 4530-4534, 2007. [8] P. C. Magnante, “Respirator cartridge end-of-service-life indicator”, U.S. Patent 4,146,887, Mar. 27 1979. [9] A. Hierlemann and R. Gutierrez-Osuna, “Higher-order chemical sensing,” Chem. Rev. 2008, vol. 108, pp. 563-613. [10] C. Hagleitner, A. Hierlemann, D. Lange, A. Kummer, N. Kerness, O. Brand and H. Baltes, “Smart single-chip gas sensor microsystem,” Nature, vol. 414, Nov. 2001 pp. 293-296. [11] C. Hagleitner, D. Lange, A. Hierlemann, O. Brand and H. Baltes, “CMOS single-chip gas detection system comprising capacitive, calorimetric and mass-sensitive microsensors,” IEEE J. Solid-State Circuits, Vol. 37, no. 12, Dec. 2002 pp. 1867-1878. [12] J. A. Covington, S. L. Tan, J. W. Gardner, A. Hamilton, T. Koickal, T. Pearce, “Combined smart chemFET/resistive sensor array,” Proc. IEEE Sensors 2003, pp. 1120-1123. [13] C. Dai, “A capacitive humidity sensor integrated with micro heater and ring oscillator circuit fabricated by CMOS-MEMS technique,” Sensors and Actuators B 122 (2007) pp. 375-380. [14] J. A. Dickson and R. M. Goodman, “Integrated chemical sensors based on carbon black and polymer films using a standard CMOS process and post-processing,” Proc. ISCAS 2000, pp.341-344. [15] M. Y. Afridi, J. S. Suehle, M. E. Zaghloul, D. W. Berning, A. R. Hefner, R. E. Cavicchi, S. Semancik, C. B. Montgomery, and C. J. Taylor, “A monolithic CMOS microhotplate-based gas sensor system,” IEEE Sensors J. vol. 2, no. 6, Dec. 2002 pp. 644-655. [16] B. Guo, A. Bermak, P. C. H. Chan, G. Yan, “A monolithic integrated 4 x 4 tin oxide sensor array with on-chip multiplexing and differential readout circuits,” Solid-State Electronics vol. 51 (2007) pp. 69-76.

168

[17] G. Barillaro and L. M. Strambini, “An integrated CMOS sensing chip for NO2 detection,” Sensors and Actuators B 134 (2008) pp. 585-590. [18] M. Zimmermann, T. Volden, K.-U. Kirstein, S. Hafizovic, J. Lichtenberg, O. Brand and A. Hierlemann, “A CMOS-based integrated-system architecture for a static cantilever array,” Sensors and Actuators B 131 (2008) pp. 254-264. [19] Y. Li, C. Vancura, D. Barrettino, M. Graf, C. Hagleitner, A. Kummer, M. Zimmermann, K.-U. Kirstein, and A. Hierlemann, “Monolithic CMOS multi-transducer gas sensor microsystem for organic and inorganic analytes,” Sensors and Actuators B 126 (2007) pp. 431-440. [20] W. H. Steinecker, M. P. Rowe, and E. T. Zellers, “Model of vapor-induced resistivity changes in gold-thiolate monolayer-protected nanoparticle sensor films,” Anal. Chem., 2007, vol. 79, pp. 4977-4986. [21] D. Barrettino, M. Graf, W. Song, K. Kirstein, A. Hierlemann, and H. Baltes, “Hotplate-based monolithic CMOS Microsystems for gas detection and material characterization for operating temperatures up to 500°C,” IEEE J. Solid State Circuits, Vol. 39, No. 7, July 2004. [22] L. Sheng, Z. Tang, J. Wu, P. C. H. Chan, J. K. O. Sin, “A low power CMOS compatible integrated gas sensor using maskless tin oxide sputtering,” Proc. 1997 International Conference on Solid State Sensors and Actuatos (Transducers), June 16-19 1997, pp. 939-942 [23] W. H. Steinecker, M. P. Rowe, and E. T. Zellers, “Model of Vapor Induced Resistivity Changes in Gold-Thiolate Monolayer-Protected Nanoparticle Sensor Films,” Anal. Chem. 2007 vol. 79, pp. 49774986 [24] B. J. Doleman, M. C. Lonergan, E. J. Severin, T. P. Vaid and N. S. Lewis, “Quantitative Study of the Resolving Power of Arrays of Carbon Black-Polymer Composites in Various Vapor-Sensing Tasks,” Anal. Chem 1998 vol. 70, pp. 4177-4190

[25] P. K. Kannan, R. Saraswathi, J. B. B. Rayappan, “A highly sensitive humidity sensor based on DC reactive magnetron sputtered zinc oxide thin film,” Sensors and Actuators A 164 (2010) pp. 8-14 [26] R. R. Smardzewski, N. L. Jarvis, A. W. Snow, H. Wohltjen, “Nanoelectronic chemical sensors for chemical agent and explosives detection,” Proc. Army Science Conference 2005 [27] J. Watson, “The tin oxide gas sensor and its applications,” Sensors and Actuators 5 (1984) 29-42 [28] J. S. Suehle, R. E. Cavicchi, M. Gaitan, and S. Semancik, “Tin oxide gas sensor fabricated using CMOS micro-hotplates and in-situ processing,” IEEE Electron Device Letters, Vol. 14, No. 3, March 1993 [29] S. Semancik, R. E. Cavicchi, M. C. Wheeler, J. E. Tiffany, G. E. Poirier, R. M. Walton, J. S. Suehle, B. Panchapakesan, and D. L. Devoe, “Microhotplate platforms for chemical sensor research,” Sensors and Actuators B 77 (2001) 579-591 [30] K. Ryu, D. Zhang, and C. Zhou, “High-performance metal oxide nanowire chemical sensors with integrated micromachined hotplates,” Applied Physics Letters 92 093111 (2008) [31] K. Arshak, E. Moore, G. M. Lyons, J. Harris and S. Clifford, “A review of gas sensors employed in electronic nose applications,” Sensor Review Vol. 24 No. 2 (2004) pp. 181-198 [32] K. J. Albert, N. S. Lewis, C. L. Schauer, G. A. Sotzing, S. E. Stitzel, T. P. Vaid, and D. R. Walt, “Cross-reactive chemical sensor arrays,” Chem. Rev. 2000, 100, 2595-2626 169

[33] A. C. Partridge, M. L. Jansen, W. M. Arnold, “Conducting polymer-based sensors,” Mat. Sci. and Eng. C, 12 (2000) pp. 37-42 [34] B. Li, S. Santhanam, L. Schultz, M. Jeffries-EL, M. C. Iovu, G. Sauve, J. Cooper, R. Zhang, J. C. Revelli, A. G. Kusne, J. L. Snyder, T. Kowalewski, L. E. Weiss, R. D. McCullough, G. K. Fedder, D. N. Lambeth, “Inkjet printed chemical sensor array based on polythiophene conductive polymers,” Sensors and Actuators B 123 (2007) 651-660

[35] N. E. Agbor, M. C. Petty, and A. P. Monkman, “Polyaniline thin films for gas sensing,” Sensors and Actuators B 28 (1995) pp. 173-179 [36] B. Lundberg and B. Sundqvist, “Resistivity of a composite conducting polymer as a function of temperature, pressure, and environment: applications as a pressure and gas concentration transducer,” J. Appl. Phys. 60 (3), 1 August 1986 pp. 1074-1079 [37] A. Bunde and J. W. Kantelhardt, “Diffusion and conduction in percolation systems- theory and applications,” in Diffusion in Condensed Matter, part IV, P. Heitjans and J. Karger, Ed. New York: Springer, 2005, pp. 895-914 [38] B. C. Munoz, G. Steinthal and S. Sunshine, “Conductive polymer-carbon black compositesbased sensor arrays for use in an electronic nose,” Sensor Review Vol. 19 No. 4, 1999, pp. 300-305 [39] H. Wohltjen and A. W. Snow, “Colloidal metal-insulator-metal ensemble chemiresistor sensor,” Anal. Chem. Vol. 70, 1998, pp. 2856-2859 [40] S. Kay, “Spectral estimation,” in Advanced Topics in Signal Processing, Editors J. S. Lim and A. V. Oppenheim, New Jersey: Prentice Hall, 1988. pp. 58-122 [41] M. G. Ancona, A. W. Snow, E. E. Foos, W. Kruppa and R. Bass, “Scaling properties of gold nanocluster chemiresistor sensors,” IEEE Sensors J. Vol. 6, No. 6, Dec. 2006 pp. 1403-1414 [42] C. Kurdak, J. Kim, A. Kuo, J. J. Lucido, L. A. Farina, X. Bai, M. P. Rowe and A. J. Matzger, “1/f noise in gold nanoparticle chemosensors, Applied Physics Letters, vol. 86, 2005, 073506 [43] K. Scholten, F. I. Bohrer, E. Dattoli, W. Lu and E. T. Zellers, “Organic vapor discrimination with chemiresistor arrays of temperature modulated tin-oxide nanowires and thiolatemonolayer-protected gold nanoparticles,” Nanotechnology, vol. 22, 2011 125501 (7pp) [44] R. H. Terrill, T. A. Postlethwaite, C. Chen, C. Poon, A. Terzis, A. Chen, J. E. Hutchison, M. R. Clark, G. Wignall, J. D. Londono, R. Superfine, M. Falvo, C. Johnson, E. T. Samulski, and R. W. Murray, “Monolayers in three dimensions: NMR, SAXS, thermal and electron hopping studies of alkanethiol stabilized gold clusters,” J. Am. Chem. Soc., vol. 117, 1995 pp. 1253712548 [45] Y. Joseph, B. Guse, T. Vossmeyer and A. Yasuda, “Gold nanoparticle/organic networks as chemiresistor coatings: the effect of film morphology on vapor sensitivity,” J. Phys. Chem. C, 2008, 112, pp. 12507-12514 [46] M. P. Rowe, K. E. Plass, K. Kibum, C. Kurdak, E. T. Zellers and A. J. Matzger, “Single-phase synthesis of functionalized gold nanoparticles,” J. Chem. Mater., vol. 16, pp. 3513-3517, 2004 [47] S. Bedair, “Sub-nanogram mass loading CMOS-MEMS cantilever resonator oscillators for gas detection,” Ph.D. dissertation, Dept. of Elect. and Comp. Eng., Carnegie Mellon University, 2008 170

[48] E. L. Covington, R. W. Turner, C. Kurdak, M. P. Rowe, C. Xu, and E. T. Zellers, “Electrical noise in gold nanoparticle chemiresistors,” Proc. IEEE Sensors 2008, pp. 102-105, 2008 [49] S. D. Senturia, Microsystem Design. Norwell, Massachusetts: Kluwer Academic Publishing, 2001 [50] J. Dickson and R. Goodman, “Integrated chemical sensors based on carbon black and polymer films using a standard CMOS process and post-processing,” Proc. ISCAS 2000, May 2000, pp. 341-344 [51] S. S. Bedair, B. Li, J. R. Cooper, S. Santhanam, R. D. McCullough, D. N. Lambeth, G. K. Fedder, “A CMOS MEMS gold plated electrode array for chemical vapor detection,” Proc. IEEE Sensors 2006, Oct. 2006, pp. 1074-1077 [52] R. Fenner and E. Zdankiewicz,”Micromachined water vapor sensors: A review of sensing technologies,” IEEE Sensors J. vol. 1, pp. 309-317, Dec. 2001. [53] M. Dokmeci and K. Najafi, “A high-sensitivity polyimide capacitive relative humidity sensor for monitoring anodically bonded hermetic micropackages,” J. Micromech. Syst., Vol. 10, No. 2, pp. 197-204, June 2001. [54] H. Shibata, M. Ito, M. Asakura, and K. Watanabe, “A digital hygrometer using a capacitance-tofrequency converter,” in Proc. IMTC 1995, pp. 100-106, 1995.

[55] K. W. Misevich, “Capacitive humidity transducer,” IEEE Trans. on Industrial Electronics and Control Instrumentation, vol. IECI-16, no. 1, pp. 6-12, July 1969. [56] B. Patissier, “Humidity sensors for automotive, appliances and consumer applications,” Sensors and Actuators B vol. 59, pp. 231-234, 1999. [57] P. M. Harrey, B. J. Ramsey, P. S. A. Evans, D. J. Harrison, “Capacitive-type humidity sensors fabricated using the offset lithographic printing process,” Sensors and Actuators B vol. 87 pp. 226-232, 2002. [58] M. Kitsara, D. Goustouridis, S. Chatzandroulis, M. Chatzichristidi, I. Raptis, Th. Ganetsos, R. Igreja and C. J. Dias, “Single chip interdigitated electrode capacitive chemical sensor arrays,” Sensors and Actuators B vol. 127 pp. 186-192, 2007. [59] H. Endres, R. Hartinger, M. Schwaiger, G. Gmelch, and M. Roth, “A capacitive CO2 sensor system with suppression of the humidity interference,” Sensors and Actuators B vol. 57 pp. 83-87, 1999. [60] L. Chandran, H. Baltes and J. Korvink, “Three-dimensional modeling of capacitive humidity sensors, Sensors and Actuators A, vol. 25-27, pp. 243-247, 1991. [61] T. Boltshauser, C. Leme, and H. Baltes, “High sensitivity CMOS humidity sensors with onchip absolute capacitance measurement system,” Sensors and Actuators B, vol. 15-16 pp. 75-80, 1993. [62] A. Hierlemann, Integrated Chemical Microsensor Systems in CMOS Technology. Berlin, Germany: Springer-Verlag, 2005. [63] C. Dai, “A capacitive humidity sensor integrated with micro heater and ring oscillator circuit fabricated by CMOS-MEMS technique,” Sensors and Actuators B, vol. 122, pp. 375-380, 2007. [64] N. Lazarus, S. S. Bedair, C.-C. Lo, and G. K. Fedder, “CMOS-MEMS Capacitive Humidity Sensor,” J. Micromech. Syst., vol. 19, no. 1, pp. 183-191, Feb. 2010. 171

[65] G. K. Fedder, S. Santhanam, M. L. Reed, S. C. Eagle, D. F. Guillou, M. S.-C. Lu, L. R. Carley, “Laminated high-aspect-ratio microstructures in a conventional CMOS process,” Sensors and Actuators A vol. 57 pp. 103-110, 1996. [66] T. E. Mlsna, S. Cemalovic, M. Warburton, S. T. Hobson, D. A. Mlsna, and S. V. Patel, “Chemicapacitive microsensors for chemical warfare agent and toxic industrial chemical detection,” Sensors and Actuators B, vol. 116 pp. 192-201, 2006. [67] S. S. Bedair, “Sub-nanogram mass loading CMOS-MEMS cantilever resonator oscillators for gas detection,” Ph.D. dissertation, Dept. of Elec. and Comp. Eng., Carnegie Mellon University, 2008. [68] S. S. Bedair and G. K. Fedder, “Picogram material dosing of microstructures,” J. of Applied Physics, vol. 106, no. 10, pp. 104913 (7 pp.), 2009. [69] J. Kim, S. Hong, B. Moon and K. Kim, “High-performance capacitive humidity sensor with novel electrode and polyimide layer based on MEMS technology,” Microsyst. Technol., vol. 16, no. 12, pp. 2017-2021, Dec. 2010. [70] COMSOL Multiphysics, version 3.5a, COMSOL AB, Burlington MA. [71] C.-C. Lo, “CMOS-MEMS Resonators for Mixer-Filter Applications,” Ph.D. dissertation, Dept. of Elec. and Comp. Eng., Carnegie Mellon University, 2008. [72] L. D. Landau, E. M. Lifshitz and L. P. Pitaevskii, Electrodynamics of Continuous Media. New York: Pergamon, 1984. [73] K. Sager, A. Schronth, and G. Gerlach, “Humidity-dependant mechanical properties of polyimide films and their use for IC-compatible humidity sensors,” in Proc. 8th Transducers, Stockholm, Sweden, pp. 736-739, 1995. [74] A. Tetelin and C. Pellet, “Accurate model of the dynamic response of a capacitive humidity sensor,” in Proc. IEEE Sensors 2003, Toronto, ON, Canada, pp. 378-383, 2003. [75] J. Seo and H. Han, “Water diffusion studies in polyimide thin films,” J. Appl. Polym. Sci., vol. 82, no. 3, pp. 731-737, Oct. 2001. [76] D. W. Green and R. H. Perry, Perry’s Chemical Engineers’ Handbook (8th edition) McGraw Hill, New York, USA 2008. [77] Kaye and Laby Tables of Physical Constants, National Physical Laboratory http://www.kayelaby.npl.co.uk/2009 [78] R. J. W. LeFevre, Dipole Moments: Their Measurement and Application in Chemistry, John Wiley and Sons Inc., 1953 [79] H. A. Rizk and I. M. Elanwar, “Dipole moments of glycerol, isopropyl alcohol, and isobutyl alcohol,” Canadian Journal of Chemistry, 46, 507 (1968) [80] C. Cornila, A. Hierlemann, R. Langgenhager, P. Malcovati, H. Baltes, G. Noetzel, U. Weimer, and W. Gopel, “Capacitive Sensors in CMOS Technology with Polymer Coating,” Sensors and Actuators B vol. 24-25, pp. 357-361, 1995. [81] NIOSH Pocket Guide to Chemical Hazards, www.cdc.gov/niosh/npg, 2009. [82] J. Laconte, V. Wilmart, D. Flandre, and J.-P. Raskin, “High sensitivity capacitive humidity sensor using 3-layer patterned polyimide sensing film,” Proc. of IEEE Sensors 2003, pp. 372377, 2003.

172

[83] D. Son, Y. Lee, J. Ahn, and D. Kwon, “Evaluation of Young’s modulus and yield strength of thin film structural material using nanoindentation technique,” Proc. MRS Symposium, vol. 562, 1999, pp. 201-206. [84] P. J. Schubert and J. H. Nevin, “A polyimide-based capacitive humidity sensor,” IEEE Transaction on Electron Devices, Vol. ED-32, no.7, pp. 1220-1223, July 1985. [85] S. V. Patel, T. E. Mlsna, B. Fruhberger, E. Klaassen, S. Cemalovic, and D. R. Baselt, “Chemicapacitive microsensors for volatile organic compound detection,” Sensors and Actuators B, vol. 96, pp. 541-553, 2003. [86] N. Lazarus and G. K. Fedder, “Integrated Vertical Parallel Plate Capacitive Humidity Sensor,” Proc. Solid-State Actuators and Microsystems Workshop (Hilton Head) 2010, pp. 242-245, 2010. [87] Honeywell, “HIH-4000-01,” HIH-4000 datasheet, July 2006. [88] A. Tetelin, C. Pellet, C. Laville and G. N’Kaoua, “Fast response humidity sensors for a medical microsystem,” Sensors and Actuators B, vol. 91, 2003, pp. 211-218 [89] C. Zhao, M. Qin, and Q. Huang, “A fully packaged CMOS interdigital capacitive humidity sensor with polysilicon heaters,” IEEE Sensors J., vol. 11, no. 11, 2011, pp. 2986-2992. [90] U. Kang and K. D. Wise, “A high speed capacitive humidity sensor with on-chip thermal reset,” IEEE Trans. on Electron Devices, vol. 47, no. 4, Apr. 2000, pp. 702-710. [91] H. Shibata, M. Ito, M. Asakura, and K. Watanabe, “A digital hygrometer using a capacitance-to-frequency converter,” In Proc. IMTC 1995, pp. 100-106, 1995. [92] C. Dai, “A capacitive humidity sensor integrated with micro heater and ring oscillator circuit fabricated by CMOS-MEMS technique,” Sensors and Actuators B 122 (2007) pp. 375-380. [93] M. Gasulla, X. Li and G. Meijer, “The noise performance of a high-speed capacitive-sensor interface based on a relaxation oscillator and a fast counter, IEEE Trans. Instrum. Meas. 54 (2005) pp. 1934-1940. [94] C. Hagleitner, D. Lange, A. Hierlemann, O. Brand, and H. Baltes, “CMOS single-chip gas detection system comprising capacitive, calorimetric and mass-sensitive microsensors,” IEEE J. Solid-State Circuits, 37 (2002), 1867-1878. [95] A. Hierlemann, Integrated Chemical Microsensor Systems in CMOS Technology Berlin Germany: Springer-Verlag, 2005. [96] Y. Y. Qiu, C. Azeredo-Leme, L. R. Alcacer, and J. E. Franca, “A CMOS humidity sensor with on-chip calibration,” Sensors and Actuators A 92 (2001) pp. 80-87. [97] C. Hagleitner, A. Hierlemann, and H. Baltes, “CMOS single-chip gas detection systems: part II,” Sensors Update 12 (2003) pp. 51-120. [98] J. C. Chen, D. Sylvester, and C. Hu, “An on-chip, interconnect capacitance characterization method with sub-femto-Farad resolution,” IEEE Trans. on Semiconductor Manufacturing, vol. 11, no. 2, May 1998 pp. 204-210. [99] Y. Chang, H. Chang, T. Lu, Y. King, W. Ting, Y. Ku and C. Lu, “Charge-based capacitance measurement for bias-dependent capacitance,” IEEE Electron Device Letters, Vol. 27, No. 5, May 2006, pp. 390-392. [100] R. Schreier, J. Silva, J. Steensgaard, and G. Temes, “Design-oriented estimation of thermal noise in switched-capacitor circuits, IEEE Trans. Circuits Syst. I, vol. 52, 2005 pp. 2358-2368. 173

[101] K. Kundert, “Simulating switched-capacitor filters with spectre RF,” 2006, available at: http://www.designers-guide.com/Analysis//sc-filters.pdf. [102] E. Klumperink, S. Gierkink, A. Van der Wel, and B. Nauta, “Reducing MOSFET 1/f noise and power consumption by switched biasing,” IEEE J. Solid-State Circuits, vol. 35, 2000, pp. 9941001. [103] G. Wegmann, E. Vittoz, and F. Rahali, “Charge injection in analog MOS switches,” IEEE J. Solid State Circuits, vol. sc-22, no. 6, Dec. 1987, pp. 1091-1097. [104] Y. Chang, H. Chang, C. Hsieh, H. Lai, T. Lu, W. Ting, J. Ku, and C. Lu, “A novel simple CBCM method free from charge injection-induced errors,” IEEE Electron Device Letters, vol. 25, no. 5, May 2004, pp. 262-264. [105] P. E. Allen and E. Sanchez-Sinencio, Switched Capacitor Circuits, New York, NY: Van Nostrand Reinhold Company, 1984. [106] A. Hajimiri, S. Limotyrakis, and T. H. Lee, “Jitter and phase noise in ring oscillators,” IEEE J. Solid State Circuits, vol. 34, no. 6, June 1999 pp. 790-804. [107] S. T. Dawkins, J. J. McFerran, and A. N. Luiten, “Considerations on the measurement of the stability of oscillators with frequency counters,” IEEE Trans. on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 54, no. 5, May 2007 pp. 918-925. [108] “6485 Picoammeter”, Keithley Instruments Inc., 2008. [109] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, Cambridge UK: Cambridge University Press, 1998. [110] P. C. Chang, H. M. Peng, and S. Y. Lin, “Allan variance estimated by phase noise measurements,” Proc. 36th Annual PTTI Meeting, 2004, pp. 165-172. [111] S. R. Stein, “Frequency and time - their measurement and characterization,” Precision Frequency Control, Vol. 2, Chap. 12, 1985. [112] R. Aparicio and A. Hajimiri, “A noise shifting differential Colpitts VCO,” IEEE J. Solid State Circuits, vol. 37, no. 12, Dec. 2002 pp. 1728-1736. [113] I. Nam, Y. Kim, K. Lee, “Low 1/f noise and DC offset RF mixer for direct conversion receiver using parasitic vertical NPN bipolar transistor in deep n-well CMOS technology, Proc. 2003 Symp. on VLSI Circuits, 2003, pp. 223-226. [114] A. Hajimiri and T. H. Lee, “A general theory of phase noise in electrical oscillators,” IEEE J. Solid State Circuits, vol. 33, no. 2, Feb. 1998, pp. 179-194. [115] G. L. Samuelson and D. C. Ailion, “Low cost integrated circuit versatile pulse and frequency counter,” Rev. of Scientific Instruments, vol. 40, no. 5, May 1969 pp. 681-683. [116] G. W. Neudeck, The PN Junction Diode, Reading, Massachusetts: Addison-Wesley Publishing Company, 1989. [117] D. Cartasegna, A. Cito, F. Conso, A. Donida, M. Grassi, L. Malvasi, G. Rescio, and P. Malcovati, “Smart RFID label for monitoring the preservation conditions of food,” Proc. ISCAS 2009, pp. 1161-1164. [118] W. H. Steinecker, M. P. Rowe and E. T. Zellers, “Model of vapor-induced resistivity changes in gold-thiolate monolayer-protected nanoparticle sensor films,” Anal. Chem., vol. 79, pp. 4977-4986, 2007. [119] W. Yao, X. Chen, and J. Zhang, “A capacitive humidity sensor based on gold-PVA core-shell nanocomposites,” Sensors and Actuators B 145 (2010) pp. 327-333. 174

[120] C. Wang, C. Wu, I. Chen and Y. Huang, “Humidity sensors based on silica nanoparticle aerogel thin films,” Sensors and Actuators B, 107 (2005) pp. 402-410. [121] M. Zhu, C. Aikens, F. Hollander, G. Schatz, R. Jin, “Correlating the crystal structure of a thiol-protected Au25 cluster and optical properties,” J. Am. Chem. Soc. vol. 130, pp. 58835885, 2008. [122] H. Qian, W. T. Eckenhoff, Y. Zhu, T. Pintauer, and R. Jin, “Total structure determination of thiolate-protected Au38 Nanoparticles,” J. Am. Chem. Soc. vol. 132, pp. 8280-8281, 2010. [123] H. Qian and R. Jin, “Controlling nanoparticles with atomic precision: the case of Au144(SCH2CH2Ph)60”, Nano Letters, vol. 9, no. 12, pp. 4083-4087, 2009. [124] D. W. Green and R. H. Perry, Perry’s Chemical Engineers’ Handbook, 8th ed. New York: McGraw-Hill, 2008. [125] Kaye and Laby Tables of Physical and Chemical Constants, Nat. Phys. Lab., Teddington, U.K., 2009. [Online]. Available: http://www.kayelaby.npl.co.uk [126] C. Cornila, A. Hierlemann, R. Lenggenhager, P. Malcovati, H. Baltes, G. Noetzel, U. Weimar, and W. Gopel, “Capacitive sensors in CMOS technology with polymer coating,” Sensors and Actuators B vol. 24/25, no. 1-3, pp. 357-361. [127] Solvent Boiling Point Chart, BrandTech Scientific, Inc., Essex, CT., 2012. [Online]. Available: http://www.brandtech.com/solbps.asp [128] H. Qian and R. Jin, “Ambient synthesis of Au144(SR)60 nanoclusters in methanol,” Chem. Mater., vol. 23, pp. 2209-2217, 2011. [129] R. P. Manginell, G. C. Frye-Mason, R. J. Kottenstette, P. R. Lewis, and C. Channy Wong, “Microfabricated planar preconcentrator,” Tech. Digest 2000 Sol.-State Sensor and Actuator Workshop, 2000, pp. 179-182. [130] F. Bender, N. Barie, G. Romoudis, A. Voigt, and M. Rapp, “Development of a preconcentration unit for a SAW sensor micro array and its use for indoor air quality monitoring,” Sensors and Actuators B vol. 93, 2003, pp. 135-141. [131] S. M. Cho, Y. J. Kim, G. S. Heo, and S.-M. Shin, “Two-step preconcentration for analysis of exhaled gas of human breath with electronic nose,” Sensors and Actuators B vol. 117, 2006, pp. 50-57. [132] C.-J. Lu, and E. T. Zellers, “A dual-adsorbent preconcentrator for a portable indoor-VOC microsensor system,” Anal. Chem. vol. 73, 2001, pp. 3449-3457. [133] E. T. Zellers, M. Morishita, and Q.-Y. Cai, “Evaluating porous-layer open-tubular capillaries as vapor preconcentrators in a microanalytical system,” Sensors and Actuators B, vol. 67, 2000, pp. 244-253. [134] I. Voiculescu, R. A. McGill, M. E. Zaghloul, D. Mott, J. Stepnowski, S. Stepnowski, H. Summers, V. Nguyen, S. Ross, K. Walsh, and M. Martin, “Micropreconcentrator for enhanced trace detection of explosives and chemical agents,” IEEE Sensors J., vol. 6, No. 5, 2006 pp. 1094-1104. [135] B. Alfeeli, D. Cho, M. Ashraf-Khorassani, L. T. Taylor, and M. Agah, “MEMS-based multiinlet/outlet preconcentrator coated by inkjet printing of polymer adsorbents,” Sensors and Actuators B, vol. 133 No. 1, 2008, pp. 24-32. [136] B. Bae, J. Yeom, A. D. Radadia, R. I. Masel, and M. A. Shannon, “A fully-integrated MEMS preconcentrator for rapid gas sampling,” Proc. Transducers 2007, pp. 1497-1500. 175

[137] P. R. Lewis, R. P. Manginell, D. R. Adkins, R. J. Kottenstette, D. R. Wheeler, S. S. Sokolowski, D. E. Trudell, J. E. Byrnes, M. Okandan, J. M. Bauer, R. G. Manley and G. C. FryeMason, “Recent advancements in the gas-phase microchemlab,” IEEE Sensors J., vol. 6, no. 3, 2006, pp. 784-795. [138] W.-C. Tian, S. W. Pang, C.-J. Lu, and E. T. Zellers, “Microfabricated preconcentratorfocuser for a microscale gas chromatograph,” J. of Micromech. Syst., vol. 12, no. 3, 2003, pp. 264-272. [139] W.-C. Tian, H. K. L. Chan, C.-J. Lu, S. W. Pang and E. T. Zellers, “Multiple-stage microfabricated preconcentrator-focuser for a micro gas chromatography system,” J. of Micromech. Syst., vol. 14, no. 3, 2005, pp. 498-507. [140] M. Zuurbier, G. Hoek, P. Van den Hazel, and B. Brunekreef, “Minute ventilation of cyclists, car and bus passengers: an experimental study,” Environmental Health 2009, 8:48 [141] M. Graf, D. Barrettino, K.-U. Kirstein and A. Hierlemann, “CMOS microhotplate sensor system for operating temperatures up to 500°C,” Sensors and Actuators B vol. 117, 2006, pp. 346-352. [142] P. J. Gilgunn, J. F. Alfaro, and G. K. Fedder, “Aspect ratio dependent etch modulation for CMOS-MEMS applications,” Proc. Transducers 2009, pp. 216-219. [143] “Tenax TA adsorbent resin physical properties,”, Scientific Instrument Services, 2010 [Online]. Available: http://www.sisweb.com/index/referenc/tenaxtam.htm [144] F. P. Incropera and D. P. DeWitt, Fundamentals of Heat and Mass Transfer, New York, NY: John Wiley and Sons, Inc, 1996. [145] D. Helmig, “Air analysis by gas chromatography,” J. of Chromatography A, vol. 843, 1999, pp. 129-146.

176