Microbolometer development and production at Indigo Systems Bill Terre, Bob Cannata, Pat Franklin, Alfredo Gonzalez, Eric Kurth, Hiep Ly, Bill Parrish, Kevin Peters, Tommie Romeo, Bob VanYsseldyk Indigo Systems, 50 Castilian Dr. Goleta CA. 93117

ABSTRACT While microbolometers have been in production for several years, the number of companies producing them is quite small. Indigo Systems has entered into the development and production of VOx based microbolometers, at its Goleta facility. Through the investment of significant capital, Indigo has established a high volume production facility based on the silicon industry model. The 6-inch, cassette-to-cassette, highly automated facility is capable of yielding hundreds of thousands of die per year. Discussed in the paper will be the design and layout of the facility, performance of the devices, as well as yield, trend and throughput data. Keywords: Microbolometer, uncooled infrared detectors, FPA, vanadium oxide

Introduction For the past 18 months, Indigo Systems has been developing a silicon microbolometer fabrication capability. This capability is unique in many ways. Conceived and facilitated from the beginning to accommodate high volume manufacturing, the Indigo bolometer factory is patterned after today’s modern, commercial, silicon foundries. With a throughput capacity of 20, 6-inch wafers per day, well over 100,000 yielded, mid format devices can be produced annually. Maximum processing efficiency is achieved by employment of all cassette-to-cassette wafer handling. This attribute allows the continuous flow of wafers through the vacuum deposition equipment while preserving the vacuum conditions required for the individual processing steps (i.e. not venting to atmosphere between depositions). Average cycle time is approximately 3 weeks from wafer start to the completion of released die. Minimization of cycle time was a key design tenet in an effort to reduce the amount of material in process at any given time. Particular attention was paid to the integration of metrology into the processing line as a means to embed continuous process feedback. At a cost of well over $25M, the equipment acquired to facilitate Indigo’s bolometer foundry, represents the state-of-the-art in silicon microbolometer processing.

Cleanroom & Equipment Description

The cleanroom layout was developed entirely from scratch with minimal constraints other than the intended goal of creating a very high volume, high throughput facility. Based on previous experience in high volume semiconductor fabrication facilities (fabs), a modest size footprint was chosen encompassing only 1,000 square feet. The rationale for this size was based on the desire to create a very efficient process flow. The enabling element for this efficiency is based on cassette-to-cassette wafer processing. The second motivation

for efficient space utilization was scalability. Scalability meaning that as the commercial demand for bolometers exceeded the intended 105 capability, another replicate fab could be created somewhat simply. Designed to operate as a class 1,000 facility, real-time monitoring equipment typically records particle count levels less than the class 100 standards. This level of cleanliness is vital to minimizing defect densities which ultimately control yields. The photolithographic area is located within the confines of the class 1,000 space. But as a further step towards yield control, it is isolated from the other processing equipments by glass partitions and an automatically activated sliding door. This design feature ensures the best possible guards against contamination in the most sensitive processing region.

Photolithography One of the crucial elements to any high quality silicon foundry is the inherent photolithographic capability. Use of sub-micron design rules is an essential element in the design of high performance bolometers. The challenge is to realize these small features, with regularity, in a high volume flow. A Saturn Ultratech 1X stepper, with a 0.6µ minimum feature size was specifically chosen to meet the demanding requirements. Capable of stepping 80, mid format devices on a 6” wafer in less than 60 seconds, this stepper is well suited for keeping up with the high throughput demands required of the photolithographic equipment. Most microbolometer designs require 10 to 15 masking steps. This means each wafer must pass through the photo area at least that many times prior to completion. Often, photolithography can be a throughput constraint. The selection of a highly capable machine is the mitigation for this constraint. Complementing the stepper is a Karl Suss projection aligner. The projection aligner, while not compatible with the demands for most bolometer photolithographic operations, is ideally suited for the die row and column ID process. This operation cannot be performed on the stepper and the accuracy requirements are not very demanding. This makes projection alignment a good choice. Cassette-compatible track coaters have been employed for the application of photo resist and polyimide. Metrology and visual inspections are performed on a Nanometrics AFT 4100 and a Nikon Optiphot respectively. Each of these equipments is additionally compatible with wafer cassettes to maintain the overall throughput goals as well as minimize handling. Vacuum Deposition The heart of the bolometer process is the vanadium oxide deposition chamber system. Originally designed to perform thin film deposition of more traditional materials, this machine has been significantly altered for optimization of the vanadium oxide (VOx) deposition process. Of particular note is the incorporation of an in-situ residual gas analysis (RGA) device. The output from the RGA is tied to the regulation of the inputs to the constituents of the plasma. In other words, there is real time closed loop control over the VOx deposition. This feature, in conjunction with the wafer cassette load locks, dramatically increases the wafer-to-wafer repeatability as well as the intra-wafer uniformity. The metal depositions are all performed on an RF diode sputter deposition system. This is a 5-chamber cluster tool and is capable of depositing all of the required materials. Nickel, Nickel-Chromium, Titanium, Aluminum and Zirconium are the typical metals deposited by this machine. Each of the chambers has a sputter pre-clean capability to ensure optimum surface preparation prior to any deposition. The requirement

for the deposition of sequential metal films can be easily accommodated by the robotic movement of wafers, from one process to another, within the vacuum space. The deposition of oxides and nitrides is performed in a single machine, namely an Applied Material P-5000. Largely intended for use commercial foundries, this machine, as originally offered was optimized for very high volume depositions. Typically, these high volumes are achieved through processes that are high temperature and high deposition rate. Unfortunately, these processing conditions were not compatible with all of our other bolometer processes. A significant amount of process development was needed to modify the previously optimized deposition parameters to those compatible with our bolometer design. Micro Machining Much of the micro machining processes are performed by reactive ion etching (RIE). In addition to the two deposition chambers in the P-5000, there are also two RIE chambers. These chambers are where the majority of the micro machining operations take place. The luxury of having two RIE chambers within the same machine is of particular benefit in avoiding the contamination issues associated with etching multiple materials in the same confines. For those materials that do not lend themselves to rapid material removal by RIE (e.g. metals) or where rate and uniformity control are required, an ion mill is employed. With a variable position wafer chuck, uniform material removal can be accommodated even in very densely featured areas on the wafer. Helium-backed wafer cooling of the platen assures wafer temperatures never exceed the pre-determined design thresholds even under the most aggressive milling conditions. The removal of the sacrificial layer, while not classically a micro machining operation, is performed in an ozone asher. This system can be used to clear the polyamide at either the die, or wafer level. This attribute was of particular benefit during the development phase when die level release was important. Under high volume conditions, wafer level release will likely be employed. Testing Final device performance is performed on Electroglass wafer probers. These probers, like all other equipment, are compatible with cassette loading. Custom hardware and software was developed internally to interface with the probers. Data collected at the wafer level is stored in data files located on servers and is accessible to all computers within the company. This ease of access to the data greatly enhanced the ability to digest the results and take the appropriate actions. Metrology The key to any high volume, commercially viable, silicon processing capability is the achievement of high process yields. Implicit in attaining high yields is good process control. Good process control can not be realized without data on the performance of the process. The only way to get data on the process is through the use of metrology. This piece of philosophy and the degree of success one has in implementing it has everything to do with the success of the processing capability. While the investment in processing equipment in the Indigo fab was substantial, the investment in metrology made up almost half of that investment. While the importance of visual inspection cannot be ignored, it is necessary but not sufficient. Visual inspection within a high volume fab is impractical. Embedded in the Indigo fab is a critical dimensioning

scanning electron microscope (CD SEM). This piece of metrology performs automated measurements on predetermined critical features and reports the data into the process data base. This machine is integral to the fab and is a key feature in attaining high throughput with high yields. Film stress measurements are made on a laser-based stress gauge and the indices of refraction are monitored by conventional elipsomerty. Height measurements typically associated with photoresist thicknesses are performed on the Nanometrics AFT 4100. Several visual inspection stations are present thought the facility and all are outfitted with Nikon Optiphots. All VOx sheet resistance is measurements are made on a fourpoint probe machine. The development of the final device design and process characterization was supported by two key equipments. A convention imaging SEM was used routinely to visually assess device characteristics. Of even more value, was the focused ion beam (FIB) SEM. This apparatus is capable of making a vertical slice through the plane of the bolometer and imaging the cross section with its integral SEM. Additionally, precise dimensional data can be acquired on the areas of interest. This capability proved to be invaluable in the rapid development of Indigo’s bolometers. Development Time Line Figure1 illustrates the development time for the realization of the bolometer fab. From the acquisition of the bare building, to the qualification of the Omega camera using an Indigo bolometer, the total elapsed time was 32 months. Outfitting of the bare building began in June of 2000 and concluded 4 months later in September. Several months were spent acquiring the processing equipment and in November of 2000 installation began. 7 months later, in June of 2001, all equipment had been installed and qualified to the desired processing capability. The development of the actual bolometer process began in August of 2002 with the processing of the first devices on fanouts. In this case, fanouts are 6” wafers with the top layer of metal in the actual readout integrated circuit (ROIC) replicated (these fanouts were made in the bolometer fab). The fanouts were primarily used to get a preliminary look at the bolometer design, the processing adequacy and most importantly, the process interactions. A significant number of fanout were processed, lending valuable insights into the processing capabilities without having to bear the expense of actual ROICs. An additional feature of the fanout architecture is the connection between 11 individual pixel elements and the chip I/O pads on every die on the wafer. This feature allowed direct electrical measurements to be made at the fanout level as well.

1st ROIC Lot Started

1st Fanout Lot Started

Feb 2003 Jan 2002

Aug 2001

03 20

02 20

Compliant Devices

Cleanroom Complete

Aug 2002 01 20

Sept 2000 00 20

Equipment Installation Begins

Nov 2000

All Equipment Qualified

June 2001

June 2000

Figure 1. Bolometer development timeline

With the successful completion of the fanout processing, processing on actual ROICs began in January of 2002. By August of 2002, devices that were fully radiometericaly compliant were completed. Responsivity, TCR, noise, response uniformity were all measured and deemed to be compliant to the design goals. Several devices were packaged in Dewars compatible with the Omega camera. A number of anomalies previously not observed at the device level, were unfortunately observed at the camera level. This forced further investigative work as well as additional device engineering. Ultimately, all of the problems observed in the camera were related to the device flatness. The original design goals underestimated the degree by which the bridges of the bolometer must be planar to the ROIC surface. Significant efforts were spent on the understanding and engineering of the deposition parameters that effect thin film stresses. By February of 2003, all of the issues were resolved and Omega cameras using Indigo bolometers were fully qualified and shipments began. Process Capabilities The ultimate measurement of process capability is the device yields. Figure 2 shows the device yields over an 11 wafer sample size. The first statistic is the yield of the silicon multiplexer prior to any processing.

SPIE USE, V. 3 5074-54 (p.5 of 9) / Color: No / Format: Letter/ AF: Letter / Date: 2003-04-01 09:11:46

These wafers are manufactured at a commercial foundry and are demonstrating an average yield of 82 % as measured at incoming inspection. As can be additionally seen, the average corresponding bolometer die yield for these multiplexers is running at about 57 %. This puts the end-to-end yield at an aggregate of 47%.

9811 Bolometer Yield 140

Yield (ROIC Die or FPAs)

120 100 80

ROICs FPAs

60 40 20 0 62

45

37

42

104

105

110

111

109

106

103

Wafer Number

Figure 2. Die yields

The average sensitivity of these devices is approximately 45mK as measured at f1.6, 30Hz frame rate and 23ºC ambient. This equates to ~ 18mK when normalized to f1.0. This average value was calculated over the most recent 500, 160 x 120 FPAs produced. Other process capabilities of interest are the VOx sheet rho uniformity which runs less than 2.5%, 1-sigma over the 150mm wafer (excluding the 12mm edge band). The average uniformity for uncorrected FPA output voltage is 20% (sigma/mean) and the average uniformity for uncorrected FPA response is less than 1%, 1sigma. The average TCR is measured at -2.4%/degree K at room temperature. A plot of the variation in TCR with temperature can be seen in figure 3. Of particular interest is the linearity of this slope over a wide temperature range. As can been seen, the performance is very linear out to +85º C. This property makes the VOx produced at Indigo very compatible with high temperature applications such as firefighting and automotive.

5HVLVWDQFHN

100

α = -2.41%/K

10 0.0034

0.0033

0.0032

0.0031

0.003

0.0029

0.0028

0.0027

1/Temperature

Figure 3. TCR variation with temperature

FPA configurations All of the focus in the initial push to the production of bolometers was on the small format 160 x 120, 51µ devices. This was done in an effort to support the production of the Omega camera. The next area of focus will be the mid format configuration, namely a 320 x 256 array with smaller format pixels. Several multiplexer designs have been realized previously in anticipation of the bolometer fabrication capability. A pictorial representation can be seen in figure 4. This family of devices all share the unique features of the small format device, namely: on-chip pixel level offset correction, temperature invariant operation (TEC-less) and in some cases, on-chip A/D conversion.

160x120 51um Pixels

320x256 25um Pixels

320x240 38um Pixels 320x120 38um Pixels

Figure 4. Uncooled ROIC family

Efforts beyond the mid format devices would logically extend to the larger formats (640 x 512 or 1280 x 960). This work will likely come as the commercial and or Military market coalesces to the point that it makes economic sense. To date, all investments into the bolometer fabrication capability have been internally funded. Decisions to develop focal planes in various configurations have to be made based on the return on investment. Lessons Learned The majority of the processes required in making bolometers are fairly standard in the silicon industry today. Each of these processes, if taken individually, is quite easy to master given adequate processing equipment. In addition, the micro-machining techniques are actually trivial compared to the current state-of-the-art in micro-machining. Given all this, it would seem that making bolometers should not be too difficult. Surprisingly, that was not our experience. Judging by the time others in the industry have required to develop the capability, this experience does not seem isolated. What makes bolometers a challenge are the unique process interactions. It was initially thought that the control of the VOx process would be the most challenging element in the development. In fact, due the upfront emphasis placed on the acquisition of a for-purpose VOx deposition machine, this aspect has turned into one of the least challenging. Far more challenging were the interactive effects of the sequential processes on one another. For example, a processing phenomenon at masking step 7 can have deleterious effects on the results of masking step 3, even though the results at masking step 3 looked perfect. The net effect is that changes in masking step 3 must be made in anticipation of the interaction of subsequent operations. From our experience, it would seem that these interactions can only be observed empirically. Efforts to predict these effects were only partially successful. Another challenge is that every piece of processing equipment is unique. There may be volumes of literature describing how to deposit silicon dioxide but, to achieve the desire film properties specific to a given set of design parameters, the individual piece of deposition equipment must be taken into account. Even two identical machines from the same vendor can behave quite differently when trying to create consistent and uniform results. The use of top quality equipment is necessary but not totally sufficient. Acquiring good machines does not guarantee good results. Contrarily, the use of bad machines almost certainly guarantees bad results. Size does matter. The use of 6-inch processing equipment (or larger) is critical in achieving the per unit cost goals. The cost to process a lot or maintain a machine is almost independent of the wafer size. With the fabrication costs fixed, the more die that yield from a process, the cheaper the individual die cost. The liberal and efficient use of metrology cannot be under emphasized. Good yields can be randomly achieved with minimal effort. Achieving good yields consistently can only come from the diligent application of statistical process control techniques. Without the embedded capabilities to collect the data that define the process control limits, the process is “flying blind”. This open loop condition will assuredly result in undesirable processing results and poor overall yields.

.

Summary Indigo Systems has succeeded in developing a state-of-the-art micro bolometer fabrication facility in a very modest span of time. The features of the fab are unmatched with any capabilities that have been reported to exist in the industry today. With a capacity for very high volumes and a demonstrated ability to produce devices with a very high yield, this facility represents a rather unique capability. Over the most recent 500 FPAs delivered to camera assembly, an average sensitivity of 45 mK (measured @ f 1.6, 30 Hz, 23ºC) was achieved. This equates to ~ 18 mK when normalized to f1.0. The end-to-end processing yield has been shown to be ~50% with an average cycle time less than 3 weeks. The fab has been sized to a capacity of 20, 6-inch wafers per day. This facility uniquely embodies high volume, high performance and high yields simultaneously, all of the elements required to meet the commercial demands of the uncooled micro bolometer marketplace.

Acknowledgements This paper is dedicated to the memory of Hiep Ly. His unselfish contributions to the realization of this world class bolometer capability will be remembered by all of those who had the pleasure of working with him. He will be sorely missed.