Dynamic Simulation: Guiding Manufacturing from Process Mechanisms to Factory Operations G. W. Rubloff Director, Institute for Systems Research Professor, Materials Science & Engineering, and Electrical & Computer Engineering University of Maryland Dynamics plays a critical role in the behavior and performance of semiconductor manufacturing from the unit process level to full factory operations, yet major gaps exist in our ability to simulate the consequences of this dynamics. At the process level, process models can provide a reasonable description of steady-state process behavior, but the realities of semiconductor equipment dictate that both total process times and thermal histories depend on the dynamics of the equipment and control systems, as well as on the raw process itself. We have developed physically-based dynamic simulation strategies which accurately reflect time-dependent behavior of equipment, process, sensor, and control systems, and we have used them to understand and optimize equipment systems and process recipes. Another dimension of dynamics appears in the behavior of cluster tools, where the tool architecture, process module populations, and scheduling algorithms add further dynamics to tool behavior. We have integrated reduced-order process models, reflecting dynamic unit process simulations, with discrete event simulations of cluster tool performance to enable co-optimization of process recipes, cluster tool configurations, and their scheduling algorithms. Finally, we have incorporated these integrated models into factory-level operational models to facilitate the evaluation of factory-level performance as a function of process, equipment, and logistics choices. These simulation strategies seem attractive in terms of their ability to represent dynamics, from continuous parameter dynamic recipes at the unit process level, to discrete-event dynamics associated with scheduling and throughput at the factory level. AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
1
Dynamic Simulation: Guiding Manufacturing from Process Mechanisms to Factory Operations G. W. Rubloff
OUTLINE
• •
Director, Institute for Systems Research Professor, Materials Science & Engineering, and Electrical & Computer Engineering University of Maryland
Motivation Dynamic simulation: – Unit process & factory infrastructure systems level (continuous parameter systems) – Cluster tool & factory operations levels (discrete event systems)
•
Heterogeneous simulation environments for optimization and control
•
Opportunities
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
2
Acknowledgements •
Dynamic equipment and process simulation – – –
•
Integrating continous and discrete dynamic systems – – – –
•
J. W. Herrmann L. Henn-Lecordier S. Marcus M. Fu
Other – – –
•
G. Lu, B. Levy, A. Rose N. Gupta, M. Bora F. Shadman (U. Arizona)
R. Adomaitis K. Jensen (MIT) M. Kushner (U. Illinois)
Support – – –
National Science Foundation Semiconductor Research Corporation Sematech
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
3
Motivation
•
Modeling & simulation are basis for efficient manufacturing enterprises => virtual manufacturing
•
Dominant modeling and simulation strategies – Process technology – molecular-scale chemistry/physics for rates, microfeature profile evolution – Equipment technology – steady-state computational fluid dynamics, plasma behavior – Factory operations – discrete event simulation, optimization/control algorithms (including cluster tool behavior)
•
Limitations – Dynamics in simulation largely limited to logistics at factory and cluster tool levels – Dynamics of equipment & process rarely treated, though crucial for unit process metrics (throughput, consumables use, ESH, …) – Equipment & process dynamics should carry forward into consequences at factory level
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
4
Process & Equipment Modeling Multiscale Multiscale modeling modeling Multiple Multiple length length scales scales Chemistry Chemistry & & physics physics have have consequences consequences at at multiple multiple length scales length scales Process Process models models available available in in numerous numerous cases cases Fundamental Fundamental Empirical Empirical
Dynamics Dynamics treated treated at at atomic atomic and and microfeature microfeature scales scales
product
Atomic Scale (1 nm)
reactant
Material & interface properties Surface & gas phase chemistry Energy-enhanced reaction
Microfeature Scale (1 um) 3-D profile shape Process chemistry & physics Reactant flux distribution
BUT… BUT… Dynamics Dynamics rarely rarely treated treated at at equipment equipment level level RTP RTP the the exception exception
Control Control system system dynamics dynamics dependent dependent on on equipment equipment transient transient behavior behavior
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
Tool Scale (1 m) Uniformity Loading & microloading Pattern factor Equipment design & architecture
AVS00inv.dynsim.ppt
11/20/00
5
Dynamics Through the Process Cycle RTCVD polySi Deposition from SiH4 Real-time mass spectrometry
RTCVD process cycle Establish reactant gas pressures Heat wafer rapidly (lamps) Deposit film at elevated pressure/temperature Cool wafer and pump out gases
Gas On
Lamps On
Gas Off Lamps Off
10-3
Partial Pressure (arb. units)
10-4
SiH2+ -5
10
H2+
650oC
-6
10
10-7
SiH4 reactant depletion
10-8
10-9
750oC
10-10
H2 reaction product 10-11 0
25
50
75
100
125
150
Time (sec)
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
6
Dynamics Through the Process Cycle RTCVD polySi Deposition from SiH4 Real-time mass spectrometry
RTCVD process cycle Establish reactant gas pressures Heat wafer rapidly (lamps) Deposit film at elevated pressure/temperature Cool wafer and pump out gases
Process cycle time Raw Process time
Overhead time
Steady-state process behavior occurs only during raw process time
Gas On
Lamps On
Gas Off Lamps Off
10-3
10-4
Partial Pressure (arb. units)
Significant overhead time accompanies process cycle:
SiH2+ -5
10
overhead time = process cycle time – raw process time
H2+
650oC
-6
10
10-7
Dynamics of process and equipment through the process cycle determines total process cycle time
SiH4 reactant depletion
10-8
10-9
Total process cycle time is the critical throughput metric for factory level logistics performance and optimization
750oC
10-10
H2 reaction product 10-11 0
25
50
75
100
125
150
Time (sec)
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
7
Continuous parameter systems for dynamic process & equipment modeling AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
8
Dynamic Simulator for RTCVD PolySi Visual Solutions, Inc.
Time-dependent behavior of equipment, process, sensor, and control system through process cycle
Film FilmThickness Thickness(A) (A)
QMS QMS Partial PartialPressures Pressures Ar, Ar,SiH SiH4,,HH2 4
2
wafer waferTT(oC) (oC) growth growthrate rate (A/sx10) (A/sx10)
Multi-level Multi-level Structure Structure Second Second Level Level Compound Compound Block Block
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
9
Dynamic Simulator for RTCVD PolySi •
Dynamic simulation can realistically represent complex systems, including – – – –
•
•
Valves, MFC’s vs. time, status Lamp power vs. time Overall process timing, conditions
• • • • • • •
Equipment Simulator
• • • • • •
temperatures
CVD Reaction • • •
Process Simulator Sensors • Total and partial pressures • Temperatures • Valve and MFC status Controls • PID controlers for temperature and pressure • Lamp power output control • Throttle valve positions
Manufacturing FoM Simulator
Manufacturing Process Efficiency • • •
G. W. Rubloff ã2000
Wafer absorptivity, emissivity Wafer thermal mass Wafer radiation, conduction Wafer temperature Temperature control system Process-dependent absorptivity, emissivity Convective heat loss in fluid flow
Sensors and Sensors and Control System Control System
Platforms commercially available (Windows) Exploit rapidly growing software base
AVS Natl Symp, Oct 2000
•
Vacuum chambers Mass flow controllers Pumps, valves Conductances, volumes Partial and total presssures Pressure control system Viscous/fluid flow
partial pressures
timing/dynamics subtle systematics systems analysis optimization sensor-in-tool models control system design training ==> learning
Heat Flow
Gas Flow
Numerous applications – – – – –
•
Process Recipe
Results validated against experiment – –
•
equipment process sensors control
• • •
Cycle time Consumables volume Energy consumption
AVS00inv.dynsim.ppt
Gas phase transport Boundary layer transport Surface-condition-dependent reaction rates - surface kinetics
rate s Wafer State • • • •
Deposition rate Film thickness Thickness control system Product properties uniformity, conformality, material quality, topography, reliability
Environmental Assessment • • • •
Gaseous emissions Reactant utilization Power consumption Solid waste
11/20/00
10
Flow Rate Dependence of Mass Spec Sensor Signal
– –
minor if fixed process recipe tractable for varying recipes with simulator available
System dynamics introduces complexity in sensor response
Mass spec H2 signal during polySi RTCVD at 750 oC, 5.0 torr SiH4/Ar for 40 sec H2 QMS Signal (amp.)
Mass spec sensitive to reactor flow rate at constant pressure Dynamic simulator captures flow rate dependence Sensor is influenced by process dynamics Consequences:
-10 2.5x10
__o__o__o___ Experimental ___________ Simulation
-10 2.0x10
200 sccm
-10 1.5x10
500 sccm -10 1.0x10
1000 sccm
-11 5.0x10
0.0 0
25
50
75
100
125
150
TIME (sec.)
Sensor-in-tool model not just a sensor model
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
11
Unit Process Optimization for Manufacturing & Environment Constant flow rate 300 sccm
Constant temperature o 650 C
50
50
750°C
Environment Manufacturing
30
40
SiH4 Utilization (%)
SiH4 Utilization
SiH4 Utilization (%)
40
700°C
20
650°C 10
0 0
2
4
30
100 sccm 20
200 sccm 300 sccm 450 sccm 600 sccm 1000 sccm
10
0
6
0
2
4
6
8
100
Manufacturing
150
650°C
Process Cycle Time (sec)
Process Cycle Time
Process Cycle Time (sec)
100 sccm 80
60
700°C 40
750°C 20
0
A
B
0
2
4
6
Pressure (torr) at which heating begins
Dynamic, continuous parameter simulation è Optimization for desired utility functions Innovation in process recipes AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
200 sccm 100
300 sccm 450 sccm 600 sccm 1000 sccm
50
A
B
0 0
2
4
6
8
Pressure (torr) at which heating begins
A: start gas flow and heating simultaneously B: start heating after gas flow established AVS00inv.dynsim.ppt
11/20/00
12
Hardware-Software CoDesign Central wafer handler
W CVD Reactor Load lock
W CVD Reactor Process and wafer state: mass spec chemical sensing
Ulvac ERA-1000 W CVD cluster tool mass spec
xj (t)
Dynamic tool simulator
Controller operates real tool
PC VisSim
Ulvac controller
Pump system PC LabView
Equipment state: valve status, pressures, flows, temperatures, ...
Controller operates virtual tool (simulator)
Integrate Integrate equipment equipment state state signals signals with with chemical chemical process process sensor sensor signals signals Validate Validate dynamic dynamic simulator simulator against against experiment experiment
New Brooks controller AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
Implement Implement new new tool tool controller controller Use Use dynamic dynamic simulator simulator to to debug debug and and enhance enhance controller controller software software
AVS00inv.dynsim.ppt
11/20/00
13
Impurity Concentrations in Liquid Source Delivery (w/ Motorola)
Impurity Concentration (ppm) in Delivered Gas
700
600
total pressure (psi) NBout = ò FxBgdt
500 Gas Phase
400 Liquid Phase
Impurity Concentration in Delivered Gas (ppm)
300
200
Experimental Data for a 20000 lb. Tank (DuPont)
100 liquid-dry point
0 100
80
60
40
20
0
% of Initial Content Remaining AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
14
Impurity Concentration Delivery Profile vs. Source Temperature 1000 o
Impurity Level in Gas (ppm)
900 800 700
-20 C
Changes in the Gas Phase Impurity Level (ppm) as the Cylinder is emptied at Various Cylinder Temperatures. The Initial Total Impurity Level in the Liquid is 100 ppm.
-10oC
600
0oC
500
10oC
400
20oC 30oC
300 o
40 C
200 100 0 100
80
60
40
20
0
% of Initial Content Remaining AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
15
WaterSim, v/3.0 Education Module • Collaboration with F. Shadman, Director, NSF ERC for Environmentally Benign Semiconductor Manufacturing, U. Arizona • Physically based simulator in learning environment provides practical experience • Explore design parameters, analyze/visualize data, experimental history-keeping B. Levy et al
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
16
Water Recycling Simulator for Engineering & Control System Design • Dynamic system visualization and analysis for ESH metrics & upset response • Complexity approaching a full chemical plant B. Levy et al
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
17
Dynamic Simulator Complexity
•
30,000 Vissim elements; reverse osmosis example
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
18
Dynamic Network Reconfiguration • • • •
Prototype for simulator network reconfiguration from graphic user interface Generalized approach to network development for dynamic simulation Choose UPW treatment processes and connect with data “wires” Automatically generates underlying network model in simulation engine B. Levy et al
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
19
Manufacturing Training Equipment and process training based on dynamic simulation EquiPSim: Equipment and Process Simulation
www.isr.umd.edu/CELS/ AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
20
EquiPSim Learning Modules (Equipment and Process Simulation) Physics-based dynamic simulation Active learning through exploration, anytime, anywhere Powerful learning aides Tightly coupled guidance Learning histories Distance collaboration & consultation
Extendible authoring architecture
www.isr.umd.edu/CELS/ AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
21
Discrete event systems for operations modeling AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
22
Single Wafer W CVD Cluster Tool Example Tool Configuration W CVD W CVD Lot process time (makespan) Lot Makespan vs. Pressure different temperatures) vs. pressure(at(at different temperatures) 1350
Orient & Degassing
Load Lock
Lot Makespan (seconds)
1300 1250 1200
T = 448
1150
T = 464 T = 484
1100
T = 492
1050 1000 950 900 64
68
72
76
80
84
88
92
96
Pressure (torr)
Results using: • Cluster tool simulator with scheduling capability (L. Shruben) • W CVD Process RSM linked to cluster tool simulator
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
23
Cluster Tool Simulator 3-stage 3-stage cluster clustertool tool
Input times Input times
Simulation Simulation results results
Lot Lotprocess processtime time &&utilization utilization
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
24
Optimizing Cluster Tool Scheduling Search algorithms in our independent cluster tool simulator can find optimal and near-optimal operations sequences. These algorithms can give substantial reduction (to 40%) in lot processing time compared to conventional dispatching rules, which are common in industrial practice (Shin and Lee, SMOMS, 1999).
Percent PercentReduction Reductionin inTotal TotalLot LotProcessing ProcessingTime Time for Optimized Sequence over Dispatching Rule for Optimized Sequence over Dispatching Rule
40 35 30 25 20 15 10 5 0
Category Short Medium Long
Move & Raw process exchange time time 1-10 secs. 20-40 secs. 10-20 secs. 10-20 secs. 20-40 secs. 1-10 secs.
J.W. Herrmann and M.-Q. T. Nguyen, “Improving cluster tool performance by finding the optimal sequence of wafer handler moves,” in preparation for J. Scheduling. ISR Technical Report 2000-3.
CT1-1
CT1-2
CT2-1
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
CT2-2
CT1-1-1 CT1-2-2 CT2-2-1 CT2-2-2 AVS00inv.dynsim.ppt
11/20/00
25
Factory Operations •
Discrete event simulation for entire factory – e.g., Factory Explorer
•
Inputs – Process sequence, # tools per step, time per step, re-entrant flow, scheduling algorithms, …
•
Outputs – Throughput, cycle time, WIP, …
•
Huge impact on manufacturing cost – Essential everywhere
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
26
Continuous parameter and discrete event systems see see L. L. Henn-Lecordier’s Henn-Lecordier’s talk talk MS-ThA5, MS-ThA5, Thursday Thursday 3:20pm, 3:20pm, Room Room 304 304
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
27
Integrating Process & Operations Models Enterprise management NSF/SRC program M. Fu S. Marcus J. W. Herrmann G. W. Rubloff
Full factory operations
Operations models => manufacturing performance
Module (subfactory) integration
Sector operations
Relationship not systematically analyzed
Process RSM’s
Experimental results
Process models => technology performance AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
Dynamic simulations Reactor- & featurescale models 11/20/00
28
Novellus Multistep/Multiwafer Process Module for W CVD 3000
Growth Growth
P = 68
2500
Sputter Clean / Liner Dep
Growth
P = 92
Total time (sec)
Nucleation
P = 80
2000
1500
Growth
1000
Transfer point Entrance Load Lock
Exit Load Lock
500
Lot process time (makespan) vs. growth temperature at different pressures
0 448 452 456 460 464 468 472 476 480 484 488492 Temperature (C)
Results using: • Deterministic cluster tool simulator (Excel) • W CVD Process RSM linked to cluster tool simulator AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
29
HSE Modeling Architecture User Interface TiN PVD Thickness Pressure Power Pumpdown time TiN PVD cluster tool # process chambers Scheduling algorithm OD time Robot move time
30 80 600 2.5
nm torr W min
2 push 15 sec 6 sec
Via2 W CVD Thickness Pressure Temperature Pumpdown time
270 80 470 2
nm torr C min
W CVD cluster tool # process chambers Scheduling algorithm OD time Robot move time
3 pull 12 5
sec sec
Oxide 3 CVD Thickness Pressure Power Pumpdown time OD time Robot move time
30 80 500 1.9 9.6 4.5
nm torr W min sec sec
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
TiN PVD Process Model
Factory Model (discrete event simulator)
TiN PVD Cluster Tool Model
Step Time (min) … TiN PVD liner 1.5 Via 1 W CVD fill 2.4 CMP 1.9 Clean 0.8 Oxide 2 CVD 2.3 P/R apply 0.7 Litho 3.5 P/R remove 0.7 TiN PVD liner 1.5 Via 2 W CVD fill 3.2 CMP 2.1 Clean 0.8 Oxide 3 CVD 2.9 P/R apply 0.7 Litho 3.2 …
Alternate: replace by compact model W CVD Process Model
Oxide CVD Process/Equipment Model/Simulator
W CVD Cluster Tool Model
# tools 2 3 2 2 3 1 3 1 2 3 2 2 3 1 2
Models and Simulators Continuous & discrete-event Static and dynamic Empirical and physics-based AVS00inv.dynsim.ppt
11/20/00
30
Heterogeneous Simulation Environment (HSE) HSE HSEintegrates: integrates:
•• •• ••
USER GROUP: process, equipment, & operations engineers
Process Processmodels models Cluster tool Cluster toolmodels models Factory Factoryoperation operationmodels models
User Interface Operations engineer
Inputs: Process input parameters Cluster tool configuration Equipment overhead time Factory resources Changes in technology & product mix
Outputs:
Process Parameters
Throughput, Cycle Time
Simulation Supervisor
Factory Simulator
Raw Process Times
Lot Process Times
Equipment engineer
Process Simulators
Process engineer
Cluster Tool Simulators
Process module and cluster tool efficiency Factory throughput, cycle time Sensitivity of input parameters Future: optimization, cost-of-ownership, yield, risk, reduced-order model generation
J. W. Herrmann G. W. Rubloff
see see L. L. Henn-Lecordier’s Henn-Lecordier’s talk talk MS-ThA5, MS-ThA5, Thursday Thursday 3:20pm, 3:20pm, Room Room 304 304 AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
31
Heterogeneous Simulation Environment
Cluster Clustertool tool scheduling scheduling
Process Process recipe recipe
Sensitivity Sensitivity analysis analysis
Cluster Clustertool tool configuration configuration
Factory Factory simulation simulation
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
32
Continuous parameter systems
Discrete event systems
Factory metrics Factory infrastructure systems
Factory operations
Factory sectors/subsystems
Integrated/cluster tools Unit process & equipment
•
Dynamic simulation systems can be – – – –
Vertically integrated from unit process to factory metrics Hybrid, including both continuous parameter and discrete systems character Physically accurate with regard to transient and control systems response Usable and valuable to a broad community
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
33
•
Exploiting dynamic simulation – – – – –
•
Constructed from mechanistic or empirical models Emphasizes key manufacturing metrics Both continuous and discrete parameter systems Build on commercially available software platforms Integrate with effective user environments
Building hierarchical modeling structures – Enables system-level evaluation – Reveals vertical interactions in system behavior o E.g., impact of unit process changes on factory operations
– Model reduction methods o Capture and integrate knowledge at multiple levels o Keep system-level models manageable
– Integrating software supervisors o Organize and execute different simulation levels which are linked
– Effective user environments o Graphical interfaces and experimentation tools to facilitate engineering design and analysis
AVS Natl Symp, Oct 2000 G. W. Rubloff ã2000
AVS00inv.dynsim.ppt
11/20/00
34
