design of complex stochastic systems: semiconductor wafer …xs3d.kaist.ac.kr/paperdata/invited...
TRANSCRIPT
James R. Morrison
KAIST, Department of Industrial and Systems Engineering
xS3D Lab Students:
Seunghwan Jung, Jonghoe Kim, Minsung Kim and Kyungsu Park
Design of Complex Stochastic Systems: Semiconductor Wafer Fabrication
and Port Services
2nd KI International Symposium, September 4 - 9, 2008
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 2
Presentation Overview
• Part I: Design of stochastic systems
• System description: Photolithography clusters in computer chip fabrication
• Why does a system-level and stochastic perspective matter?
• Design for throughput
• Part II: Design of service systems
• Port system decoupling leads to service concept
• Concluding remarks
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 3
System Description: Photolithography (1)
• Semiconductor wafer/LCD fabrication are key industries• 2007 worldwide revenue: W270x109 (US$270 billion)
• 2007 Samsung Electronics revenue: W21x109 (US$21 billion)
• 2007 Korean GDP: ~ W1,000x109 (US$ 1 trillion)
• State-of-the-art fabricator construction: ~ US$3 billion
• Photolithography cluster tools• Key toolset in semiconductor wafer fabrication
• Typically a fabricator bottleneck (even with dozens of tools)
• Cost: ~ W20x109 (US$20 million)
[1] Gartner Research, http://www.gartner.com, [2] CIA World Fact Book[3] EE Times Asia, http://www.eetasia.com, 2007/07/20, [4] Scanner image courtesy of ASML
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 4
System Description: Photolithography (2)
• Operation of a of photolithography cluster tool• Process modules coat each wafer with photosensitive films
• Photolithography scanner exposes the film to a light pattern
• Process modules develop the image on the film
Wafer path for coating
Wafer path for develop
Scan
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 5
System Description: Photolithography (3)
• Several modules may be devoted to a single process
• Scanner will be the system bottleneck (least throughput potential)
• Process time for wafer j in module mi is a constant tji
• Wafers of different kinds may require different process times
• May have a buffer before the bottleneck (scanner)
m1
m5
mA2 mB
2 mA3 m4
mB8 mA
8 mC7 mB
7 mA7 mB
6 mA6
mB3 mC
3
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 6
System Description: Photolithography (4)
• Design question: When should the robots advance the wafers?
• Design of steady-state operation• Axiomatic design has shown that periodic operation is a good design1
• Cyclic (periodic) robot schedule is throughput optimal2
• Design of transient operation?• Transients are generally ignored, and if not…
• Typical objective is to reestablish steady state operation
[1] Hilario L. Oh and Tae-Sik Lee, “A synchronous algorithm to reduce complexity in wafer flow,” Proceedings of the 1st
International Conference on Axiomatic Design (ICAD), pp. 87-92, June 21-23, 2000.[2] M. Dawande, N. Geismar and S. Sethi, “Dominance of cyclic solutions and some open problems in scheduling bufferlessrobotic cells,” SIAM Review, vol. 47, pp. 207-721, 2005.
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 7
System-Level Perspective (1)
• Good steady-state design ensures maximum throughput potential (this is what the tool supplier quotes as speed)
• In practice:• Modules may require a setup between different types of wafers
• Modules conduct self-cleaning operations
• Photolithography scanner pauses production when settings change
• Tool must be emptied before maintenance and filled after
• Production may wait while monitor (test) wafers are run
• These events may be considered to occur randomly
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 8
System-Level Perspective (2)
Throughput Potential and Non Steady-State Phenomenon
80 wph
Waf
ers
Per
Ho
ur
Ideal steady state
Return from maintenance
69 wphModule clean and monitor Wafer to
wafer interactions
61 wph
55 wph
0 wph
Non steady-state events cause
dramatic loss!
[1] James R. Morrison, Beverly Bortnick and Donald P. Martin, “Performance evaluation of serial photolithography clusters: Queueing models, throughput and workload sequencing,” Proceedings of the 2006 IEEE/SEMI Advanced Semiconductor Manufacturing Conference, Boston, MA, pp. 44-49, May 2006. *2+ James R. Morrison and Donald P. Martin, “Performance evaluation of photolithography cluster tools: Queueing and throughput models,” OR Spectrum (Springer), Vol. 29, No. 3, pp. 375-389, July 2007.
Actual throughput
Actual throughput x 1.45 = Ideal throughput
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 9
Axiomatic Design at the System Level (1)
• Goal:• Design photolithography cluster tool
• Take a system-level approach (explicitly address non steady-state)
• Constraints: • Module process times are known and fixed (tj
i) – scanner is bottleneck
• Buffer modules may be placed only just before the bottleneck (scanner)
• Setups may be required between wafers of different types
• Wafer may have a maximum allowed residency time in each module (time window: [tj
i, tji + rj
i] ) – process is a success if this is obeyed
• Objective(s):• Maximize steady-state throughput
• Minimize the effect of disturbances on throughput (“non steady-state”)
• Minimize variation in module residence times for like wafers
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 10
Axiomatic Design at the System Level (2)
Functional Requirements:
• FR1. Conduct wafer processes to exceed bottleneck rate (lB)
• FR2. Transport wafers (physically + orchestration)
Design Parameters:
• DP1. Process modules
• DP2. Robots and algorithms
2
10
2
1
DP
DP
XX
X
FR
FR
Decoupled design X = Relationship between DP and FR0 = Negligible relation between DP and FR
[1] Axiomatic Design: Advances and Applications, Nam Pyo Suh, Oxford University Press, 2001
So long as our robots/algorithms obey the process time windows, this will remain 0!
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 11
Axiomatic Design at the System Level (3)
Functional Requirements:• FR1.1. Position, enter/exit wafers
• FR1.2. Conduct process• FR1.2.1. Prepare for process
• FR1.2.2. Maintain process quality
• FR1.2.3. Conduct process
• FR1.2.4. Ensure process rate (<lB)
• FR2.1. Physically move wafers
• FR2.2. Orchestrate steady state (SS) operation• FR2.2.1. Decouple SS process times
• FR2.2.2. Minimize SS wait to transport
• FR2.3. Orchestrate transient operation
Design Parameters:• DP1.1. Method for positioning
• DP1.2. System for processing• DP1.2.1. Recipe setup system
• FR1.2.2. Module cleaning system
• FR1.2.3. Process modules
• FR1.2.4. Sufficient number of modules
• DP2.1. Robots
• DP2.2. Algorithms for steady state (SS) operation• DP2.2.1. Cyclic schedule
• DP2.2.2. Algorithm to minimize waiting
• DP2.3. Algorithms/structure for transient operation
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 12
Axiomatic Design at the System Level (4)
Functional Requirements:
• FR 2.3. Orchestrate transient operation
• FR2.3.1. Reestablish SS
• FR2.3.2. Protect from disturbance
• FR2.3.3. Recover time lost due to disturbance
• FR2.3.4. Maintain decoupling for wafers not delayed by disturbance
• FR2.3.5. Decouple and minimize transient wafer residency times
• FR2.3.6. Replenish protection from delay
Design Parameters:
• DP 2.3. Algorithms/structure for transient operation
• DP2.3.1. Algorithm to return to SS
• DP2.3.2. Buffer before the bottleneck
• DP2.3.3. Algorithm to minimize distance between normal/disturbed wafers
• DP2.3.4. Same SS cyclic schedule for wafers not disturbed
• DP2.3.5. Algorithm to decouple and minimize delayed wafers
• DP2.3.6. Wafer input rate to tool (= process rate of prescan bottleneck)
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 13
Axiomatic Design at the System Level (5)
• Concepts of the design (design parameters)
Scan
DP2.3.4. Same SS cyclic schedule for wafers downstream of(after) disturbance
DP2.3.2. BufferDP2.3.3. Minimum distance
between delayed/normal wafers
DP2.3.5. Decouple transient transport
DP2.3.6. Input rate
DP2.3.1. Algorithm to return all wafers to SS cyclic schedule
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 14
Axiomatic Design at the System Level (6)
• Design matrix for the transient operation
6.3.2
5.3.2
4.3.2
3.3.2
2.3.2
1.3.2
0000
00000
00000
0000
00000
00000
6.3.2
5.3.2
4.3.2
3.3.2
2.3.2
1.3.2
DP
DP
DP
DP
DP
DP
XX
X
X
XX
X
X
FR
FR
FR
FR
FR
FR
Decoupled design
X = Relationship between DP and FR0 = Negligible relation between DP and FRNote: The entire design matrix is decoupled
Return to SS
Protect
Recover
Preserve SS
Decouple transient
Replenish
Algorithm
Buffer
Algorithm
Same cycle
Algorithm
Input rate
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 15
Axiomatic Design at the System Level (7)
• Relationship between design parameters our objectives
Design Parameters:• DP2.3.1. Algorithm to return to SS
• DP2.3.2. Buffer before the bottleneck
• DP2.3.3. Algorithm to minimize distance between normal/disturbed wafers
• DP2.3.4. Same SS cyclic schedule for wafers not disturbed
• DP2.3.5. Algorithm to decouple and minimize residency times
• DP2.3.6. Wafer input rate to tool (= process rate of prescanbottleneck)
Objective Functions:• Maximize steady-state
throughput
• Minimize the effect of disturbances on throughput
• Minimize variation in module residence times for like wafers
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 16
Axiomatic Design at the System Level (8)
• The flavor of the algorithms: Subset of DP2.3.3. • Algorithm to minimize disturbance due to different wafer classes
• Let x(i,j) = start time in module i for wafer j
• Let c(i,j) = completion time from module i of wafer j
• Minimize the time between wafer exits from the tool
• Guarantee similar wafers experience same module residency times
• Equivalent formulation as a mathematical program:
1
1
11
,...,21,1,1,1max,1
m
k
j
m
j
k
j
kMm
tttjxjcjx
Mmtttjxjx
jx
m
k
j
m
j
k
j
k ,...,1,1,1,1
subject to ,1min
1
1
11
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 17
Presentation Overview
• Part I: Design of stochastic systems
• System description: Photolithography clusters in computer chip fabrication
• Why does a system-level and stochastic perspective matter?
• Design for throughput
• Part II: Design of service systems
• Port system decoupling leads to service concept
• Concluding remarks
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 18
Port Service Systems and Decoupling (1)
• Goal: Design port service systems
• Method: Axiomatic design
• Functional Requirements (subset): Port service• Unload goods from input carrier
• Load goods to output carrier
• Transfer incoming goods from sea to land
• Transfer outgoing goods from land to sea
• To maintain the independence of these requirements:• Mobile floating port (MFP) or “mobile harbor”
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 19
Port Service Systems and Decoupling (2)
• Typical interpretation of “Transfer goods from sea to land”• Porting service should be directly tied to a specific land-based port
• This interpretation is not solution neutral!
• Independence of MFP from its nominal land port• Design MFP to be relatively small and agile
• Decoupling from the land based port entirely allows MFP to:• Serve congested ports world-wide
• Provide a genuine port service!
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 20
Presentation Overview
• Part I: Design of stochastic systems
• System description: Photolithography clusters in computer chip fabrication
• Why does a system-level and stochastic perspective matter?
• Design for throughput
• Part II: Design of service systems
• Port system decoupling leads to service concept
• Concluding remarks
2nd KI International Symposium – Daejeon, South Korea – September 4, 2008 – 21
Concluding Remarks
• Part I: Design of stochastic systems
• System perspective on the design of photolithography clusters
• Such disturbances can/often play a dominant role in performance!
• Design can allow numerous objective functions
• Next steps: Conclude algorithm design and quantify performance
• Part II: Design of service systems
• Port system decoupling leads to service concept
• Next Steps: Continued service perspective within the Axiomatic Design framework