the ongoing challenge - tutorial the illusion of capacity

Fordyce, Fournier, Milne, SinghIllusion of FAB Capacity in Central Planning – hunt for CAPAVAIL

1

The Ongoing Challenge - Tutorial

The Illusion Of Capacity

Dr. Ken Fordyce & John Fournier, IBMProf. John Milne, Clarkson University & Dr. Harpal Singh, CEO Arkieva

** Dr. Horst Zisgen, IBM, Rich Burda, Gary Sullivan (IBM, retired), Peter Lyon (IBM retired), Prof Chi-Tai Wang NCU (Taiwan)

How do we determine what a FAB can produce in what time frame under what conditions?

What Methods are available and used with what Frequency?

How do they compare?

part 5 of 4


2

Where we just visited & where we are going

• Just Visited - How can we do a better job of representing FAB capacity within traditional central planning models. This topic is broken into two parts:– Part 1 - describing the core challenge, simply describing the

current process and the basic nature of the question. – Part 2 – indentifying options to actually improve the

representation of FAB capacity within central planning • Example link EPOS with PROFIT• “clearing function” that modifies capacity available or cycle time• Heuristic estimate of resource entity

• Going to Visit - If we narrow our focus to just FAB planning and open to methods other than the traditional balance equations, what methods are available and how do they compare?


3

The Never Ending Loop

CapacityPlanningSystem

Client Demand

Capacity Limits

Wafer StartProfile

CentralPlanningEngine

EquipmentLoad Reports

Feedback Loop for Constrained Toolsets

CT Commit


4

Typical FAB Questions

• Which lots are going to exit the FAB next week?• What is the near-term and long-term capability of a

toolset (or fab)?• What is the expected incoming WIP at a toolset?• Will toolset operational outs rise or fall in the near future?• What will be the effect of adding/removing a certain tool

from production?• What is the impact of expedites on cycletime?• Is the toolset’s current deployment adequate for

incoming WIP? • Where should resources be deployed today?• Is there a window of opportunity for equipment work?


5

Typical FAB Questions

• What tool sets are “broken” (insufficient capacity for projected load), what are the options to fix broken tool sets, and what is the impact of these actions?

• Are the wafer start limits accurate?

• Is the cycle time commit accurate?

• When are the lots in the line exiting?

• When would a projected set of wafer starts exit?


6

Emerging FAB Question

• What is the impact of complex process time windows– Capacity required– Cycle time impact– Near term lot completion estimates

• Different than “best practices” for schedule dispatch (not more important, just different)


9

Short List of “methods” to answer questionstypically either start driven (forward) or demand driven (backward)

• Algebraic methods to allocate steady state workload to tool sets • Extending the algebraic method to handle deployment or cascade (essentially CAPS)• “race” track with / without limits (as wafers)• WIP Projector type logic working sectors, cycle times, and aggregate capacity• Daily output planning• SLIM• Rule Based simulation• Time Slide (TALC)• Queuing models• Queuing models and static capacity planning• queuing networks / fluid models (EPOS)• discrete event simulation (WIPSIM)• Clearing Functions• Column generation optimization type simulation• Combination of simulation and optimization• Clearing “function” incorporated into optimization• Transport equations via partial differential equations

ResearchCompare, contrastClassify best use

Examples WIPSIM vs EPOS

EPOS vs Transport Equations

New textProduction Planning and Control for

Semiconductor Wafer Fabrication FacilitiesMonch, Fowler, et al Springer


10

Clearing Functions within LP

• Armbruster & Uzsoy 2012, “Continuous Dynamic Models, Clearing Functions, and Discrete-Event Simulation in Aggregate Production Planning”, Production Planning Tutorials in Operations Research”, isbn 978-0-9843378-3-5, dx.doi.org/10.1287/educ.1120.0102

– ACF: allocated Clearing Function pp. 10-13• Asmundsson et al, “Production planning models with resources

subject to congestion,” Naval Res. Logistics, vol. 56, no. 2, pp. 142–157, Mar. 2009.

• Asmundsson et al, “Tractable nonlinear production planning models for semiconductor wafer fabrication facilities,” IEEE Trans. Semicond. Manuf., vol. 19, no. 1, pp. 95–111, 2006


11


12

Sources

• Tibbits, B, 1993, “Flexible simulation of a complex semiconductor manufacturing line using a rule-based”, IBM R&D Journal

• Sullivan, G, 1992, “Daily Output Planning”, Expert Systems with Applications

• Sullivan, G, 1995, “Dynamically Generated Rapid Response Capacity Planning Model”

• Bermon, S., and S. Hood. 1999. Capacity optimization planning system (CAPS). Interfaces 29 (5): 31–50.

• Bagchi, S et al 2008. A Full-factory Simulator As A Daily Decision-support Tool For 300mm Wafer Fabrication Productivity, MASM 2008

• Zisgen, H. et al 2008. A Queueing Network Based System To Model Capacity And Cycle Time For Semiconductor Fabrication, MASM 2008

• Schelasin, R 2011. Using Static Capacity Modeling And Queuing Theory Equations To Predict Factory Cycle Time Performance In Semiconductor Manufacturing, MASM 2011

• Levy, J. et al 2010,”Method For Determining Amount of Product Released Into a Time Sensitive Operation”, MASM 2010


13

Sources• Armbruster & Uzsoy 2012, “Continuous Dynamic Models, Clearing Functions, and

Discrete-Event Simulation in Aggregate Production Planning”, Production Planning Tutorials in Operations Research”, isbn 978-0-9843378-3-5, dx.doi.org/10.1287/educ.1120.0102

• Kacar, Irdeem, & Uzsoy, 2011, “An Experimental Comparison of Production Planning Using Clearing Functions and Iterative Linear Programming-Simulation Algorithms”, IEEE Transactions on Semiconductor MFG,

• Irdeem, Kacar, & Uzsoy, 2010, An Exploratory Analysis of Two Iterative Linear Programming—Simulation Approaches for Production Planning, Ieee Transactions On Semiconductor Manufacturing, Vol. 23, No. 3, August 2010

• Asmundsson et al, “Production planning models with resources subject to congestion,” Naval Res. Logistics, vol. 56, no. 2, pp. 142–157, Mar. 2009.


• H. Missbauer. Models of the transient behaviour of production units to optimize the aggregate material flow. International Journal of Production Economics 118(2):387{397, 2002.

• H. Missbauer and R. Uzsoy. Optimization models for production planning. K. Kempf,• P. Keskinocak, and R. Uzsoy, eds. Planning Production and Inventories in the

Extended Enter-• prise: A State of the Art Handbook. Springer Verlag, New York, 437{508,A47 2011.• [ Modigliani and F. E. Hohn. Production planning over time and the nature of the

expectation


14

Presentations at MASM 2012

• Using Iterative Simulation to Incorporate Load-Dependent Lead Times in Master Planning Heuristics; Lars Moench and Thomas Ponsignon (Infineon). Abstract: In this paper, we consider heuristics for master planning in semiconductor manufacturing. While lead times are typically assumed as fixed in production planning, we use iterative simulation to take load-dependent lead times into account. An AutoSched AP simulation model of a semiconductor supply chain is used for implementing the scheme. Simulation results show that the iterative scheme converges fast and leads to less variable, more profitable production plans compared to planes obtained by the fixed lead time approach.

• Product Mix Optimization for a Semiconductor Fab: Modeling Approaches and Decomposition Techniques; Andreas Klemmt, Martin Romauch, Walter Laure (Infineon Technologies). Abstract: For optimizing a semiconductor fab we are aiming to match the production capabilities, capacities and the demand in the most profitable way. In this paper we address a linear model of the product mix problem considering product dependent demand limits (obligations and demand forecast) and profits while respecting the the capacity bounds of the production facility. Since the capacity consumption is highly depended on choosing from different production alternatives we are implicitly solving a static capacity planning problem for each product mix. This kind of planning approach is supported by the fluid flow concept of complete resource pooling in high traffic. We propose a general model that considers a wide range of objectives and we introduce a heuristic based on a decomposition of the static capacity planning problem. The computational study of the approaches is based on real world data and on randomly generated instances.


15


• An Evaluation of an Option Contract in Semiconductor Supply Chains Konstanze Knoblich (Infineon Technologies) and Cathal Heavey and Peter Williams (University of Limerick), Abstract: The purpose of this paper is to evaluate an option contract within a semiconductor supply chain consisting of one semiconductor manufacturer and one customer. In an option contract the customer pays an upfront fee (option price) for an option to purchase product. A simulation model is used to compare the performance of an option contract against a standard supply contract used in a semiconductor supply chain in terms of delivery performance and costs for the supply chain partners

18

Time Target(s)

Zone of Control for Process Time Window ManagementBasic Flow

MA 1 MA 2 MA i MA nStart of ZOC End ZOC

Each Lot in the Process Window Zone of Control mustMake it through N manufacturing actions (MA) under the time target

This gets complicated QuicklyLet’s take a look

Dangerstarts

DangerEnds

19

Time Target(s)

Zone of Control for Process Time Window Management - manufacturing activities and tool sets

MA 1 MA 2 MA i MA nStart of ZOC

Tool Set A

Tool Set B

Tool Set C

Tool Set D

End ZOC

Each MA is handled by a Different Tool GroupThe number and types of tools in each tool group vary

In traditional capacity analysis there is no cost in waitingExcept for the accumulation of cycle time

Now need to look at capabilities of the entire suite or zone of toolsForces additional idle without WIP

20

Time Target(s)


ZOC lots ZOC lots ZOC lots ZOC lots

Tool Set A

Tool Set B

Tool Set C

Tool Set D

End ZOC

Process Time Window Lots (called ZOC lots) are “spread”across the zone of control

Zone of Control for Process Time Window Management - manufacturing activities, tool sets, and ZOC lots

21

Time Target(s)


ZOC lots ZOC lots ZOC lots ZOC lots

Tool Set A

non ZOC lots

non ZOC lots

non ZOC lots

non ZOC lots

Tool Set B

Tool Set C

Tool Set D

End ZOC

Zone of Control for Process Time Window Management - manufacturing activities, tool sets, ZOC lots, non ZOC lotsNow the problem GETS Interesting. The tool sets must also process non

time window lots. Called non-ZOC lots

Can This fillIdle w/o WIP


28

Core of Algebraic Method

• Algebra methods – the FAB plans to start three products (A, B, and C) at the rate of 10, 30, and 20 wafer starts per day. Product A requires 3 passes at photo, Product B required 4 passes, and C requires 2 passes. With some simple arithmetic I can determine the work load on the photo tools in terms of passes.


29

Basics of FAB Planning


30

Basics of FAB Planning

• Focus on matching assets with demand• Three major classes

– Aggregate FAB planning– Deployment or near term tool planning– WIP Projection

• Forward flow of starts dominate method as opposed to pulling to meet demand

• Wide variation in methods• Wide variation in how much FAB complexity of

deployment and operating curve is handled


31

Basics of Algebraic Approach to Aggregate FAB Planning


32

Basics of Aggregate Factory Planning

• focused on assessing the ability of the factory to meet certain demand looking to identify “broken” toolsets and creating the capacity inputs required by central planning.

• Demand is stated as a starts profile and a lead (cycle) time commit for each part.

• Various levels of sophistication in handling operating curve, deployment, mix variability, etc

• The key challenges for the factory planner are:– Determine if the workload can be allocated across the tools in

such a way that all of the workload can be allocate without violating capacity constraints

– If insufficient capacity exits• find the optimal mix of workload that can be met without violating

capacity constraints• find the optimal allocation that either minimizes additional capacity

needed incorporating some type of fair share of pain


33

Factory load or starts

Statement

Representative Factory Routes often

aggregating parts into a family

Aggregate Factory Capacity Plan

Tool Set Character

istics

Reports on required utilization levels and capacity loss points

High level statement of capacity based specified

cycle time to limit demand on the factory

How additional capacity enables

the factory to handle more starts

Allocation of tools to

families

Factory Planning

Model has key

relationships


34

Tool A Tool B Tool Cno tools covering

oper

oper001 1 1 0 2

oper002 1 1 0 2

oper003 0 1 1 2

oper004 0 0 1 1

oper005 0 1 1 2

oper006 1 0 0 1

oper007 1 0 0 1

number opers tool covers 4 4 3

Table 5.1: Deployment Information for PSO Group

Ingredient # 1Which Tools Can Handle Which Operations

1 – oper/tool link active0 – not allowed


35

Tool A Tool B Tool Coper001 4 20 5oper002 15 20 6oper003 10 15 8oper004 10 9 20oper005 5 5 5oper006 8 10 10oper007 10 10 10

Table 5.3: RPT information

Ingredient # 2Raw Process Time (RPT) per widget per time unit

for Tool / Operation Pairing


36

Tool A Tool B Tool Caverage daily

workloadoper001 4 20 999999 40oper002 15 20 999999 40oper003 999999 15 8 40oper004 999999 999999 20 40oper005 999999 5 5 50oper006 8 999999 999999 50oper007 10 999999 999999 50

capac avl 720 1152 1296

Table 5.4: Average Daily Demand and Capacity Available

Ingredient # 3Average Daily Work load for each operationBased on Starts Profile for some time unit


37



capac avl 720 1152 1296


Ingredient # 4Average Capacity Available for each tool Based on tool analysis for some time unit


38

Ingredient # 5 “cycle time tax” to reduce capacity available

to effective capacity availableTo account for required idle without WIP


39



capac avl 720 1152 1296


Ingredients - operation to tool link - raw process time - capacity available - workload or demand - cycle time tax (not show in this example)

Assign “portions” of tools to operationsTo cover workload

And not exceed available capacity


40

Tool A Tool B Tool C demand met constraint

average daily workload delta

oper001 23.0 17.0 0.0 40.0 ≤ 40 0.0oper002 0.0 30.0 0.0 30.0 ≤ 40 10.0oper003 0.0 10.0 30.0 40.0 ≤ 40 0.0oper004 0.0 0.0 40.0 40.0 ≤ 40 0.0oper005 0.0 10.0 30.0 40.0 ≤ 50 10.0oper006 40.0 0.0 0.0 40.0 ≤ 50 10.0oper007 30.0 0.0 0.0 30.0 ≤ 50 20.0

cap used 712.0 1140.0 1191.0 total unmet dmd 50.0constraint ≤ ≤ ≤cap avl 720.0 1152.0 1296.0delta 8.0 12.0 105.0

Table 5.5: Allocation Decision and Results where capacity limits are not violated

assign lots to tools to minimize unmet demand without exceeding capac avail or workload

Allocation Decision # 1

Demand on Oper001 (20) met as followsTool A takes 23 units and Tool B gets 17

Total Workload on Tool A is(23x4) + (40x8 ) +(30x10 ) =

92 + 320 + 300 = 712



capac avl 720 1152 1296


Workload on Tool A is 092 = 23 x 4Workload on Tool B is 340 = 17 x 20


41

Tool A Tool B Tool C demand met constraint

average daily workload delta


cap used 712.0 1140.0 1191.0 total unmet dmd 50.0constraint ≤ ≤ ≤cap avl 720.0 1152.0 1296.0delta 8.0 12.0 105.0




All capacity constraints metSome workload

Not met


42

Tool A Tool B Tool C demand met constraintaverage daily workload delta


cap used 991.8 1340.0 1241.0 total unmet dmd 0.0constraint ≤ ≤ ≤cap avl 720.0 1152.0 1296.0delta -271.8 -188.0 55.0




capacity constraints NOT met All workload met


43

Clearing Functions within LP

• Armbruster & Uzsoy 2012, “Continuous Dynamic Models, Clearing Functions, and Discrete-Event Simulation in Aggregate Production Planning”, Production Planning Tutorials in Operations Research”, isbn 978-0-9843378-3-5, dx.doi.org/10.1287/educ.1120.0102

– ACF: allocated Clearing Function pp. 10-13• Asmundsson et al, “Production planning models with resources

subject to congestion,” Naval Res. Logistics, vol. 56, no. 2, pp. 142–157, Mar. 2009.



44

Sources

• Tibbits, B, 1993, “Flexible simulation of a complex semiconductor manufacturing line using a rule-based”, IBM R&D Journal

• Sullivan, G, 1992, “Daily Output Planning”, Expert Systems with Applications

• Sullivan, G, 1995, “Dynamically Generated Rapid Response Capacity Planning Model”

• Bermon, S., and S. Hood. 1999. Capacity optimization planning system (CAPS). Interfaces 29 (5): 31–50.

• Bagchi, S et al 2008. A Full-factory Simulator As A Daily Decision-support Tool For 300mm Wafer Fabrication Productivity, MASM 2008

• Zisgen, H. et al 2008. A Queueing Network Based System To Model Capacity And Cycle Time For Semiconductor Fabrication, MASM 2008

• Schelasin, R 2011. Using Static Capacity Modeling And Queuing Theory Equations To Predict Factory Cycle Time Performance In Semiconductor Manufacturing, MASM 2011

• Levy, J. et al 2010,”Method For Determining Amount of Product Released Into a Time Sensitive Operation”, MASM 2010


45

Sources• Armbruster & Uzsoy 2012, “Continuous Dynamic Models, Clearing Functions, and Discrete-Event

Simulation in Aggregate Production Planning”, Production Planning Tutorials in Operations Research”, isbn 978-0-9843378-3-5, dx.doi.org/10.1287/educ.1120.0102

• Kacar, Irdeem, & Uzsoy, 2011, “An Experimental Comparison of Production Planning Using Clearing Functions and Iterative Linear Programming-Simulation Algorithms”, IEEE Transactions on Semiconductor MFG,

• Irdeem, Kacar, & Uzsoy, 2010, An Exploratory Analysis of Two Iterative Linear Programming—Simulation Approaches for Production Planning, Ieee Transactions On Semiconductor Manufacturing, Vol. 23, No. 3, August 2010

• Asmundsson et al, “Production planning models with resources subject to congestion,” Naval Res. Logistics, vol. 56, no. 2, pp. 142–157, Mar. 2009.


• H. Missbauer. Models of the transient behaviour of production units to optimize the aggregate material flow. International Journal of Production Economics 118(2):387{397, 2002.

• H. Missbauer and R. Uzsoy. Optimization models for production planning. K. Kempf,• P. Keskinocak, and R. Uzsoy, eds. Planning Production and Inventories in the Extended Enter-• prise: A State of the Art Handbook. Springer Verlag, New York, 437{508,A47 2011.• [60] F. Modigliani and F. E. Hohn. Production planning over time and the nature of the

expectation


46


• Using Iterative Simulation to Incorporate Load-Dependent Lead Times in Master Planning Heuristics; Lars Moench and Thomas Ponsignon (Infineon). Abstract: In this paper, we consider heuristics for master planning in semiconductor manufacturing. While lead times are typically assumed as fixed in production planning, we use iterative simulation to take load-dependent lead times into account. An AutoSched AP simulation model of a semiconductor supply chain is used for implementing the scheme. Simulation results show that the iterative scheme converges fast and leads to less variable, more profitable production plans compared to planes obtained by the fixed lead time approach.

• Product Mix Optimization for a Semiconductor Fab: Modeling Approaches and Decomposition Techniques; Andreas Klemmt, Martin Romauch, Walter Laure (Infineon Technologies). Abstract: For optimizing a semiconductor fab we are aiming to match the production capabilities, capacities and the demand in the most profitable way. In this paper we address a linear model of the product mix problem considering product dependent demand limits (obligations and demand forecast) and profits while respecting the the capacity bounds of the production facility. Since the capacity consumption is highly depended on choosing from different production alternatives we are implicitly solving a static capacity planning problem for each product mix. This kind of planning approach is supported by the fluid flow concept of complete resource pooling in high traffic. We propose a general model that considers a wide range of objectives and we introduce a heuristic based on a decomposition of the static capacity planning problem. The computational study of the approaches is based on real world data and on randomly generated instances.


47


• An Evaluation of an Option Contract in Semiconductor Supply Chains Konstanze Knoblich (Infineon Technologies) and Cathal Heavey and Peter Williams (University of Limerick), Abstract: The purpose of this paper is to evaluate an option contract within a semiconductor supply chain consisting of one semiconductor manufacturer and one customer. In an option contract the customer pays an upfront fee (option price) for an option to purchase product. A simulation model is used to compare the performance of an option contract against a standard supply contract used in a semiconductor supply chain in terms of delivery performance and costs for the supply chain partners


48

the ongoing challenge - tutorial the illusion of capacity

Documents

fab produce

fab planninghow

representation of fab

emerging fab questionwhat

capacity available

certain tool

broken tool sets

insufficient capacity