Factory Operations Research Center (FORCe II)

Transcription

1 Factory Operations Research Center (FORCe II) SRC/ISMT 2004-OJ-1214 Configuration, monitoring and control of semiconductor supply chains Jan. 7, 2005 Shi-Chung Chang (task 1) Argon Chen (task 2) Yon Chou (task 3) National Taiwan University

2 Multiple Threads of Manufacturing Services Design Optional Captive Fab Optional Captive C/P control owner control point Fab C/P Packaging Final Test Packaging Final Test C/P Packaging Final Test Physical Layer Fab C/P Packaging Final Test

3 Challenges of Manufacturing Services Effective collaboration between engineering and manufacturing Reliable delivery Supply and service monitoring and control customer business eng monitoring & control supply chain reliable delivery

4 Supply Chain Configuration, Monitoring and Control Objectives: to enhance Predictability Scalability Service differentiability Control Robust configuration Monitoring Dynamic control Task 2 Task 3 variety Supply Chains variations Performance variation Task 1: behavior modeling

5 ask 1: Empirical Behavior Modeling PI: Shi-Chung Chang Co-PIs: Da-Yin Liao, Argon Chen To develop methodology: 1. Definition of quality of service (QoS) metrics Scalability Controllability Service Differentiability 2. Modeling and simulation Performance Variability (engineering and business) Capacity allocation & control configuration &control uncertainties Empirical Behavior Modeling environment settings performance metrics

6 id-year Progress Task 1 Definition of Performance Metrics Have identified reduced set of QoS metrics from SCOR Have defined the QoS translation problem among nodes of supply chain Is developing a queueing network-based QoS translation method Fab Behavior Modeling Have developed a baseline model open queueing nework with priority mean and variability Have defined a response surface fitting problem Is developing response surface modeling method Simulation Have developed a baseline Fab simulator based on QN models

7 Performance Metrics Definition SCOR-based six categories quality, cost, cycle time, delivery, speed, and service Service Differentiation Demand SSC Levels Supply FAB Process CP Process ASSY Process FT Process Design Houses IDM Supply Supply Supply Supply Chain & nodal Requirements

8 Q Co CT D Yield Key Level 1 Metrics for SSC Yield Variability % of Downtime due to Non-availability of WIP Capacity Utilization Finished Goods Inventory Carrying Costs In-process Failure Rates Machine Wait Time Actual-to-theoretical Cycle Time Make Cycle Time Ratio of Actual to Theoretical Cycle Time % of Orders Scheduled to Customer Request Average Days per Schedule Change Delivery Performance to Customer Request Date Forecast Accuracy Sp Sv % Schedules Changed within Suppliers Lead Time % Schedules Generated within Suppliers Lead Time Intra-Manufacturing Re-Plan Cycle Schedule Achievement Schedule Interval Re-Plan Cycle Time Responsiveness Lead Time % Orders/Lines Received On-Time to Demand Requirement %Orders/Lines Received with Correct Shipping Documents Delivery Performance to Customer Request Date Forecast Cycle

9 Characterization of Variability Sources Process varieties Engineering changes Operation excursions Demand plan Arrival Parameters Service Parameters Characterization by Mean & Variance Nodal Performance Measures Hybrid models Response surface Priority queueing Simulation (λ h, (λ l, 2 C h ) 2 C l ) queue T h, T l machine 2 (τ, C s ) d h, d l

10 Behavior Modeling Methodology Priority open queueing network (OQN) Nodal and system characterization by mean and variance Response surface matching with empirical data Simulation for performance prediction/model adaptation Uncertainties Environment settings Configuration & control Inputs Response surface PQN model matching Demand Model adaptation Simulator Simulated performance Monitored actual or empirical performance Supply FAB Process CP Process ASSY Process Supply FT Process Design Houses IDM Monitoring Supply Supply Supply

11 Priority OQN Model: Fab Example External Arrivals b 1 b 2 b M b M+1 Machine Group b M+2 Machine Group b 2M Ma c h in e Group 1 2 M b J-M+1 b J-M+2 b J Departures flow of type A parts flow of type B parts Node : Group of identical failure prone machines Queue : Infinite buffer for each step Job Class : Part type Arrival : General independent processes Service : General time distribution (single/batch, failure) Routing : Deterministic with feedback Discipline : Priority

12 Decomposition Approximation Two Notions Each Network Node as an Independent GI/G/m Queue Two Parameters, Mean & SCV, to Characterize Arrival & Service Processes Arrival Arrival Parameters Service Parameters Nodal Nodal Performance Measures

13 Three Flow Operations Merging i i q ij j j λ = λ C j 2 aj 0j = a + n i= 1 λ q ij n 2 j + Caib ij i= 1 i Splitting i i q ij j j λ = λ q C ij 2 ij i ij = q C ij di q ij Flow Through a Queue 2 ( τ, C ) Input s Output i 2 i ( λ,c ) 2 a ( d, C d ) Departure Rate d = λ Inter-departure Time SCV C 2 d = ( ρ ) C + ρ C 1 a s

14 Traffic Equations: F( ). Traffic Rate λ a n = λ δ e n = λ δ e n + + M λ mn m= 1 M λ a m m= 1 Traffic Variability C 2 an = a n + M m= 1 C 2 am q b mn mn ( λ an τ n, 2 C an 2 C sn ) Performance/ QoS Measures : WIP, Cycle Time,... Interaction among Nodes

15 QoS Translation Given Higher Level/Coarse QoS spec. Y Service Node Parameters = ( τ, Flow Routing Information Q Priority OQN model F(α, θ, Q) FCFS Discipline for Each Priority Derive by solving External control specs. Y = F( α, θ,q) 2 θ ) n C sn α ( λ 2 =, ) Nodal Level QoS reponsibility y = G( ω, θ) e C e

16 Response Surface Modeling Given Empirical I/O Characterization (I, O) Service Node Capacity m n Flow Routing Information Q Priority OQN model F(α, θ, Q) FCFS Discipline for Each Priority Fit F(α, θ, Q) to (I, O) and derive 2 Node characteristic parameters θ = ( τn, C sn )

17 Modeling Capacity Allocation Deduct Capacity Allocated to Higher Priority P.C.A: roportional apacity llocation No PULL+PCA* PUSH+PCA FLOW_IN Estimation CONVERGE? Yes MAX_FLOW_IN Next Priority Targets (Capac. Alloc.) & Cycle Time Estimates for the priority

18 Deliverables task 1 July 2005 Selection and definition of key QoS metrics Translation algorithm of QoS from chain to nodes Fab behavioral model Priority, capacity allocation, source of variation Fab behavioral simulator July 2006 Methodology generalization to the service thread from design house, fab to circuit probe Methodology and tool integration with control (task 3) and optimization (task 2)

19 ask 2: Robust Allocation and Monitoring PI: Argon Chen Co-PI s: David Chiang, Andy Guo Will develop: 1. A baseline supply chain allocation strategy Robustness on performance Robustness on performance variability Quadratic approximation 2. Supply chain sensitivity and monitoring 2 nd moment performance of priority queueing network Decomposition of supply chain performance Ranges of optimality and feasibility Trigger of supply chain control actions

20 id-year Progress Task 2 Supply chain simulation model Have defined environment variables and variability sources Have defined control policies for various supply chain threads Have built a preliminary simulation model using ARENA Supply chain allocation programming Have defined allocation decision variables Have formulated constraints Have started development of implementation strategies Supply chain allocation optimization Have studied quadratic programming methodologies Have studied Wolfe-dual based algorithm Have studied piecewise linear programming methodologies

21 Semiconductor Supply Chain Fab CP Assm FT SC Control Point SC Route SC Route SC Control Point

22 Supply Chain Routes and Threads Thread (i) Route (r) i=1 r=1 r=2 r=3 i=2 i=3

23 Supply Chain Allocation X rikq (%) Proportion of production for product type k at service-level q allocated to supply chain thread i of route r Route 1 X 11kq % Service level q X 12kq % Product k Route 2 X 21kq % X 22kq %

24 Supply Chain Behavior Model Uncertain Factors v rikq varieties eng changes exceptions.. demands Allocation % x rikq.. Empirical Model E(y jkq )=f jkq (x rikq v rikq, e rikq ) SD(y jkq )=g jkq (x rikq v rikq, e rikq ).. Performance Metrics y jkq capacity business thread mode scheduling policies.. Environment Settings e tikq y jkq : the jth performance index for product k at service level q

25 Supply Chain Constraints (I) Product Mix Constraints The proportion of product type k to total production Priority Mix Constraints r i q X rikq = ρ k The proportion of service-level q production to total production k r i k X rikq φ q q

26 Supply Chain Constraints (II) Demands Fulfillment Constraint The total production is equal to or less than the demand r i k q X rikq 1 Example: X rikq = 0.95 r i k q meaning 95% of demand will be fulfilled

27 Supply Chain Constraints (III) Route Mix Constraints The proportion of production allocated to route r can not exceed a predetermined limit i k q Thread Mix Constraints X rikq α The proportion of production allocated to thread i can not exceed a predetermined limit r r r k q X rikq β i i

28 Supply Chain Constraints (IV) Resource (Capacity) Constraints The proportion of capacity consumed by route r cannot exceed a given proportion of route r capacity to the total capacity k i q Xrikq * mrk Cr r C 1 r r Where m rk =Uω ki : the percent use of route r by one percent of production for product type k allocated to route r C r : the proportion of route r available capacity to total capacity C r : the proportion of available capacity to total capacity U(%): capacity utilization (production to total capacity ratio) ω ki : the capacity of route r consumed by one unit of product type k

29 Supply Chain Allocation Optimization Goal Programming Objectives: Max or Min E(y jkq )=f jkq (x rikq v rikq, e rikq ) Min SD(y jkq )=g jkq (x rikq ri ikq, e rikq ) Business Scenarios: e rikq Stochastic Elements: v rikq Constraints: Resources Constraints Demands Fulfillment Constraints Product-mix Constraints Priority-mix Constraints Route-mix Constraints, etc Service Priority (q = 1) Min f jk1 (SD(y jk1 )) Max or min f jk1 (E(y jk1 )) x rik1 Goal Constraints Service Priority (q = 2) Min f jk2 (SD(y jk2 )) Max or min f jk2 (E(y jk2 )) x rik2 Goal Constraints x rik(q-1) Goal Constraints Service Priority (q = Q) Min f jkq (SD(y jkq )) Max or min f jkq (E(y jkq ))

30 Solution Methodology Quadratic Stochastic Goal-Programming Transform the model to a piecewise quadratic programming model Construct Wolfe-dual based algorithm for the piecewise quadratic programming model Develop a preemptive goal programming approach for differentiable service priorities Perform sensitivity analysis through parametric quadratic programming

31 Implementation Case 1: Order fulfilled based on X rikq Order Route r=1 r=2 Thread i=1 i=2 i=1 i=2 X 11A1 X 12A1 X 21A1 X 22A1 X rikq buckets max

32 Implementation Case 2: Order fulfilled by a lower priority Order Route r=1 r=2 r=3 r=4 Thread i=1 i=2 i=1 i=2 i=1 i=1 X rikq uckets X 11A1 X 12A1 X 21A1 X 22A1 over over over over X 31A2 max X 41A2

33 Deliverables Task 2 Supply chain quadratic goal programming model and solution (Model, Methodology, Report) (July- 05) Supply chain simulation model (March-05) Supply chain planning goals (April-05) Supply chain goal programs (May-05) Baseline supply chain allocation model and solution (July-05) Supply chain sensitivity analysis and monitoring methodology (Model, Methodology, Report) (July- 06)

34 Supports Needed Task II Supply chain network data Number of supply chain levels Number of facilities at each level Capacity and capability of each facility Locations of facilities, etc. Supply chain operations data Facility reliability data Cycle time Dispatching policies Control policies Order fulfillment policies, etc. Supply chain allocation practice Supply chain performance data

35 Task 3: Dynamic Control Will develop: 1. A control model for demand support Salient scope: Sales channel PI: Yon Chou Co-PI: Shi-Chung Chang advanced info. of inventory and business plan Fab 1 2 C/P 2. A control method to enhance delivery, speed and service Microchip Company orders Fab dynamic events in eng. & mfg n C/P

36 id-year Progress Task 3 1. A control model for demand support Have defined the problem scope Have outlined the model 2. A control method to enhance delivery, speed and service Have developed a workload flow model Have developed an integer program (with preliminary implementation)

37 3.1 Salient feature Demand (technology, product, etc.) has a life cycle true demand growth mature obsolete time Demand forecasts and channel inventory are signals. The total demand is a more reliable estimate. forecast Mean-reverting model

38 3.1 A control model for demand support Objectives To monitor demand-capacity mismatch in medium-long term To support the demand-capacity synchronization by capacity decisions (expansion, reservation, prioritization) Model scope Relationship between capacity, cycle time, WIP and throughput Integrating channel inventory and demand dynamics with supply capability Channel inventory Demand dynamics Fab 1 2 C/P capacity capacity

39 3.1 Elements of the model Channel inventory: an input, based on market intelligence data Demand dynamics Demand lifecycle Demand learning effect Supply capability of the nodes Cycle time, WIP, throughput Objective functional I(t) X (t): cumulative demand by time t X ( t) = k( X ( t)) WIP = Capacity allocation (to control shortage points) ( TH )( CT )

40 3.2 Delivery control Objectives: To assess the impact of dynamic events on the performance under high-mix environment To identify feasible revision, shortfall points and delay information in order to enhance delivery, speed and service differentiated services Schedule, Events Delivery Control Service quality Feasible revision Delay information

41 3.2 Workload variation propagation Elements Events: uncertain job arrivals, urgent orders, disrupting events, and material availability Modeling of capacity loss due to variety, variation and dynamic events Cumulative workload Cumulative workload Shop 1 Shop K time

42 3.2 Behavior modeling of re-allocation Entities of allocation schedule T time periods (weeks) K shops (nodes) J orders Variables X j k, Dynamic events Hold Hold-release Order insertion, t {0,1} Is coding an integer program for studying the behavior

43 3.2 Variety-efficient relationship There are many parallel machine systems in semiconductor manufacturing. How to measure variety? How to characterize the relationship between variety and efficiency? Efficiency hypothesis Variety

44 Deliverables task 3 A control model for demand support (Model, Report) (July-05) A delivery control method (Methodology, Report) (July-06)