Presentation Overview Modeling and Optimization in Semiconductor Manufacturing • Speaker introduction • Motivation • Directions in fab modeling and optimization James R. Morrison Associate Professor Industrial & Systems Engineering – Photolithography equipment models – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design • Concluding remarks ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 1 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 2 Presentation Overview Speaker Introduction • Speaker introduction IBM • Motivation • Directions in fab modeling and optimization – Photolithography equipment models – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design • Concluding remarks ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 3 • • • • • • IBM, PhD in Microelectronics Electrical and Computer Division Engineering, UIUC Fab “Queueing Operations network Engineering analysis of semiconductor manufacturing” 5Advisor: years PR Kumar Associate Professor Corresponding co-chair• Semiconductor equipment • Queueing models Manufacturing Automation Industrial IEEE Technical & Systems Committee Engineering on Semiconductor Since 2009 2008 • Research on practical models and optimization ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 4 Presentation Overview Motivation (1) • Speaker introduction • Semiconductor manufacturing – Global revenue in 2013: NT$ 9,540 billion (US$ 318 billion) • Motivation [1] • Directions in fab modeling and optimization • Construction costs – 300 mm wafer fab: – 450 mm wafer fab: – Photolithography equipment models NT$150 billion NT$300-450 billion (US$ 5 billion [2]) (US$10-15 billion) – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design – 1996-1999: Fab production control method earned Samsung NT$ 15 billion (US$ 1 billion [3]) additional revenue – 2005: IBM’s 30 independent supply chains merged into a single global system and saved NT$ 180 billion (US$ 6 billion [4]) – … • Concluding remarks ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 5 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 6 Motivation (2) Presentation Overview • Clustered photolithography tools (CPT) – – – – • Significant value for improvements • Speaker introduction Purchase cost of NT$ 0.6-3 billion (US$ 20-100 M [5]) The most expensive tool in a fabricator Typically the bottleneck of the fabricator Key yield and cycle time contributor • Motivation • Directions in fab modeling and optimization – Photolithography equipment models – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design • Concluding remarks [5] ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 7 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 8 System Description: CPT (1) System Description: CPT (2) [6] Conceptual diagram of a CPT (slightly simplified) Scanner Pre-scan processes Clustered Photolithography Tool P2 P1 Wafers Enter P2 P1 P3 Scanner P6 P5 P4 P2 Wafer handling robots P11 Wafers Exit Multi-cluster tool, robot in each cluster, IF buffers, STK buffer Scanner is often the CPT bottleneck Largely deterministic process times Process time can vary by product Setups between lots (reticle changes, pre-scan setup, …) Wafer handling robot decision policy & deadlock prevention P11 P8 P9 P10 P8 P9 P11 P7 P8 Post-scan processes Conceptual diagram of a CPT (robots “removed”) Wafers Enter Pre-scan processes buffer buffer P2 buffer buffer buffer buffer Post-scan processes … • • • • • • Buffer P4 P1 P2 P1 … P2 Scanner P6 P11 … P11 P11 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 9 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 10 System Description: Performance Metrics Models for CPTs Notation – Al: arrival time of lot l to the tool queue – Sl : start time of lot l on a tool – Cl : completion time of lot l on a tool – Wl : wafers in lot l • Performance measures Computation time – Cycle time of lot l: TlCT := Cl - Al – Process time of lot l: TlPT := Cl - Sl – Throughput time of lot l: TlTT := min{ TlPT, Cl – Cl-1 } • T1TT • Wafers Exit Models with various levels of detail Detailed Model Linear Model “Everything” A(k1) A(k1), B Affine Models A(k1), B(k1) A(k1), B(k1, k2) Flow Line Models Parametric flow lines Empirical flow lines Collect Tool Log Data Train a set of parameters T2TT T3TT Lot 1 With complete tool log data Lot 2 Lot 3 Time ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 11 Exit Recursion Models With wafer in/out log data Simulate models With lot in/out log data ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 12 Computational Comparison Linear Model Affine Model ER Model Accuracy Assessment Model Computation Time Linear Model 0.44 Affine Model-B 2.01 Linear Model CT LRT TT CT LRT TT CT LRT TT Affine Model-B(k) 2.35 Affine Models CT LRT TT CT LRT TT CT LRT TT Affine Model-B(k1,k2) 1.57 ER Models CT LRT TT CT LRT TT CT LRT TT ER Model - Tool Log 0.88 Flow Line Models CT LRT TT CT LRT TT CT LRT TT ER Model - Wafer Log 0.73 ER Model - Lot Log 0.36 Flow Line Model 134.38 Empirical FL Model 64.66 Detailed Simulation 11404.98 FL Model DS Same Sample, Same Parameter Different Sample, Same Parameter Different Sample, Different Parameter • Errors relative to detailed model – Error of 20%+ – Error 5-20% – Error 0-5% ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 13 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 14 Application Opportunities: Capacity Optimization Application Opportunities: Toolset Agility (1) • Fundamental process/robot bottleneck analysis & mitigation buffer buffer Pre-scan processes Post-scan processes … Wafers Enter P1 P1 P2 P2 … P2 buffer buffer buffer buffer Scanner P6 P11 … • More complicated analysis – Buffer size implications – Manufacturing environment & mitigation – Penultimate dominating process ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 15 Wafers Exit • Wafers are commonly admitted to a CPT as soon as possible – Deployment opportunity of the lot is reduced – High priority hot lots experience additional queueing – Lot/wafer residency time and buffer level greater than required P11 P11 • Question: When should wafers be admitted to the CPT? – Maintain throughput capacity – Minimize residency time and thereby increase agility ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 16 Application Opportunities: Toolset Agility (2) Application Opportunities: Toolset Agility (3) • Results: Detailed CPT model JIT Trade-off between throughput and wafer residency time Lexicographic Multi-Objective Linear Program (LMOLP) ([14]) ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 17 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 18 Application Opportunities: Fab Simulation/Optimization Presentation Overview • Equipment and fabricator simulations are used to – Predict value of changes to fabricator capacity – Predict value of changes to fabricator production control policies – Predict capacity of fabricators • Want expressive, accurate and computationally tractable models to help make decisions on US$ billions – Future manufacturing facilities will cost US$15 billion – High quality models enable improved decisions • Speaker introduction • Motivation • Directions in fab modeling and optimization – Photolithography equipment models – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design • Concluding remarks • Can also be used for model based optimization – Production control ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 19 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 20 Brief Overview: Preventive Maintenance PM Decision Tiers • “The care and servicing of equipment… including tests, measurements, adjustments and parts replacement, performed specifically to prevent faults from occurring” – Wikipedia.org • Planned PM activities – Increase planned downtime – Overall equipment availability – Equipment reliability – Unplanned tool failures • FOCUS: Tier 1 with a roughly yearly perspective • QUESTION: How frequently to conduct PMs? “PM planning” PMs are essential for manufacturing performance… need careful consideration ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 21 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 22 Fundamental Tradeoffs: Frequency of Setups (1) Fundamental Tradeoffs: Frequency of Setups (2) • Consider a major PM in a cluster tool – Service activities • 30 primary components to service • 2 hours/component on average • Each must be conducted once every month One month cycle: 30 separate PMs (1 component each) Setup: 6 hrs Tool uptime: 16 hrs 1 component PM: 2 hrs One month cycle: 1 PM (All 30 components at once) – Setup activities • Cool down, vent, pump, conditioning, qualification • Duration about 6 hours total each time • How many components should we service in each PM? – How many setups do we have per month? – What is the resulting tool availability? ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 23 Tool availability = 66.7% … Setup: 6 hrs Tool uptime: 654 hrs Tool availability = 90.8% 30 component PM: 60 hrs • Tradeoffs – Fewer setups Increased availability – Fewer setups Large WIP bubbles • Goal: Determine how often we plan to take tool down for PM ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 24 System Description: G/G/m-Queue PM Plan Optimization (1) Servers Waiting room Customers arrive Served customers exit ⋮ • Nonlinear programming problem • There are n types of PM. Min !" , … , !$ <//-queue with tool failure> • • • • • • • The number of servers , customer interarrival times are generally distributed with rate The service time of a customer has general distribution with rate Generally distributed failure intervals with mean Time-based preemptive events Exponentially distributed available intervals of mean The mean availability of the server A The coefficient of variation of the downtimes ,the coefficient of variation of the service times The effective coefficient of variation of the service time , , ≔ ∑*+, - &'( . )( ∑*+, /0( 1( (system loading constraint) 2+ 3 !( 3 4+ 5 6 1,2, … , 7 , 2 1 1 1 Subject to (System loading constraint means 1 for stability) 1 1 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 25 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 26 PM Plan Optimization (2) Numerical Results ∏*+, !( • Parameters in the NLP8) 6 • :+ : The number of PM occurrence of the type 58<in =0, 8) ?. • @* : When there are n types of 8<A,the total number of PM occurrence in =0, 8) ?. • B+ :The PM occurrence probability of the type 58< in =0, 8) ?. • C( : The mean down time of the type 58<. ( • Examples inspired by fab data ( ( Parameters ( Total duration (hrs) DE) ( DE) , @* 6 ∑*+, ( C( 6 • :+ 6 • !" , … , !$ 6 ∑*+, B+ C( 6 JK LM )( , B+ 6 F( , G$ )( DE) ∑$ (O"F( N( ( • !" … , !$ 6 • !" , … , !$ 6 • A !" , … , !$ 6 G$ PQR=? S T Y Y S 6 6 T ST S T DE) G$ 6 H( )( 1( DI( 66 14.8349 3 3 , !" , … , !$ 6 ST J K VM T S 6 1 VM 6 W+ Original Plan Type 2 Type 1 Type 2 Total duration (hrs) 240 720 52.5677 403.4836 PM duration (hrs) 66 30 14.4561 16.8118 Setup time (hrs) 3 4 3 4 0.6694 System loading 0.7632 0.8123 Cycle time (hrs) 40.4364 22.9174 0.6653 0.6164 System loading 0.7046 0.7605 Cycle time (hrs) 35.7362 21.1243 Z [ ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 27 Optimal Plan Type 1 53.9452 Setup time (hrs) 0.7125 Parameters Availability T T ∑$ (O" U( =C( ?S 240 Optimal Plan PM duration (hrs) Availability ( Multiple PM Types Single PM Type Original Plan ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 28 Presentation Overview • Speaker introduction • Motivation • Directions in fab modeling and optimization Motivation for Analytic Engineering Chain Management • Supply Chain Management (SCM) – 2005: IBM’s 30 independent supply chains merged into a single global system and saved NT$ 180 billion (US$ 6 billion [4]) – Competitive advantage for Infineon – Required for success in many industries – Photolithography equipment models – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design • Concluding remarks • Engineering Chain Management – Engineering chain consists of all activities from design of the product to design of the manufacturing system • Opportunities – Largely addressed by business philosophy or project management – Each component shares limited information with other components ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 29 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 30 Concept for AECM Presentation Overview • Speaker introduction • Motivation • Directions in fab modeling and optimization – Photolithography equipment models – PM planning optimization – Convergence of fab perspectives for analytic engineering chain design • Concluding remarks ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 31 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 32 Concluding Remarks • Directions in fab modeling and optimization – Photolithography equipment models – PM planning optimization Modeling and Optimization in Semiconductor Manufacturing – Convergence of fab perspectives for analytic engineering chain design • Application opportunities – CPT capacity optimization, CPT toolset agility Questions and Discussion – Fab simulation/optimization – Combined perspectives for next generation optimization James R. Morrison Associate Professor • Future directions Industrial & Systems Engineering – Industry engagement – Further model development ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 33 References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. HIS iSuppli April 2011 Elpida Memory, Inc., available at http://www.eplida.com, Leachman, Robert C., Jeenyoung Kang, and Vincent Lin. "SLIM: Short cycle time and low inventory in manufacturing at samsung electronics." Interfaces32.1 (2002): 61-77 http://www.forbes.com/forbes/2003/0811/076.html Roger H. French and V. Hoang, “Immersion Lithography: Photomask and Wafer-Level Materials,” Tran. Annual Review of Materials Research, Vol. 39, 93-126 Hyun Joong Yoon and Doo Yong Lee, “Deadlock-free scheduling of photolithography equipment in semiconductor fabrication,” IEEE Trans. Semi. Mfg., vol. 17, no. 1, pp. 42-54, 2004 Avi-Itzhak, B. "A sequence of service stations with arbitrary input and regular service times." Management Science 11.5 (1965): 565-571 Friedman, Henry D. "Reduction methods for tandem queuing systems." Operations Research 13.1 (1965): 121-131 Park, Kyungsu, and James R. Morrison. "Performance evaluation of deterministic flow lines: Redundant modules and application to semiconductor manufacturing equipment." Automation Science and Engineering (CASE), 2010 IEEE Conference on. IEEE, 2010 Morrison, James R. "Deterministic flow lines with applications." Automation Science and Engineering, IEEE Transactions on 7.2 (2010): 228-239 Morrison, James R. "Multiclass flow line models of semiconductor manufacturing equipment for fab-level simulation." Automation Science and Engineering, IEEE Transactions on 8.1 (2011): 81-94 Kim, Woo-sung, and James R. Morrison, “On the steady state behavior of deterministic flow lines with random arrivals.” Accepted June 14, 2014 for IEEE Transactions on Automation Science and Engineering (IEEE) Kim, Woo-sung and James R. Morrison, “The throughput rate of serial production lines with regular process times and random setups: Markovian models and applications to semiconductor manufacturing,” Computers & Operations Research (Elsevier), Online at http://dx.doi.org/10.1016/j.cor.2014.03.022, April 4, 2014. Park, Kyungsu and James R. Morrison, “Controlled wafer release in clustered photolithography tools: Flexible flow line job release scheduling and an LMOLP heuristic,” IEEE Transactions on Automation Science and Engineering (IEEE), Online at http://dx.doi.org/10.1109/TASE.2014.2311997, April 7, 2014. Longest waiting pair: [7] Geismar, H.N.; Sriskandarajah, C.; Ramanan, N., "Increasing throughput for robotic cells with parallel Machines and multiple robots," IEEE Trans. Auto. Sci. and Eng., vol.1, no.1, pp.84,89, Jul 2004 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 35 ©2014 –James R. Morrison – TSMC Visit – August 15, 2014 – 34