Diminishing Returns on Knowledge in Operations Management

Transcription

1 PICMET Diminishing Returns on Knowledge in Operations Management Charles Weber and Asser Fayed October 12-15, 2008 Washington, DC, USA ETM Slide # 1

2 Abstract A empirically grounded model of the operating curve a semiconductor fabrication facility, which is sufficiently accurate to make capitalization decisions, has been developed. The model is used to simulate the performance of a hypothetical semiconductor fabrication facility that operates under very realistic conditions. Results of the simulation show that the value of additional technological knowledge can be negative. Learning more of a good thing is not always a good idea! Slide # 2

3 Outline About technological knowledge Optimizing the operating curve Prior approaches Output maximization Value-driven approach. Insights pertaining to the value of knowledge Slide # 3

4 The Traditional Learning Curve (adapted from Bohn, 1994, p. 66) Production Something happens Improved Cost and Quality Statistical linkage between input and output See, for example, Wright (1936); Searle and Gody (1945); Alchian (1963); Rapping (1965); Hayes and Clark (1985); Dutton and Thomas (1984); Argote and Epple (1990). Slide # 4

5 Inside the Learning Curve (adapted from Bohn, 1994, p. 66) Explicit recognition of knowledge. See, for example, Adler and Clark (1991); Jaikumar and Bohn (1992). Definition of Technological Knowledge (Bohn, 1994): understanding the effects of input variables (x) on the output (Y). Y = f(x 1,.., x n ); uncharacterized function! PICMET Slide # 5

6 Volume Learning and Production Knowledge Mishina (1999) Studied Boeing s B-17 heavy bomber production during World War II, which underwent a 60-fold increase in output over a period of less than four years. Found that investments pertaining to a scale-up triggered learning activity. Increases in production volume drives learning Factory encounters new experiences, new phenomena and new challenges, which Mishina concludes drive learning. Slide # 6

7 The Value of Technological Knowledge Has not been characterized The value of the same set of skills May vary under different circumstances Such as different production volume Or different positions on the operating curve Research Question: Is learning more of a good thing always a good idea? Slide # 7

8 The Operating Curve A manifestation of Little s Law Tradeoff between throughput and cycle time (Aurand and Miller, 1997). Decreasing cycle time requires reduction of lot starts; insufficient product output? Loading a factory to capacity Increases cycle time and WIP excess inventory, Greater vulnerability to production problems Decreased flexibility. Slide # 8

9 Profitability and the Operating Curve in Semiconductor Manufacturing Profitability in semiconductor manufacturing, to a large degree, depends upon how a fab (fabrication facility) manages its operating curve. Every fab responds to a particular economic environment and is subject to a particular set of constraints (Fayed and Dunnigan, 2007). Every fab has to optimize it own operating curve. Slide # 9

10 Prior Approaches to Optimizing the Operating Curve Learning by experience (Potti & Whitaker, 2003) Simulation of demand (Cakanyildrim and Roundy, 2002) Simulation of capacity (Nazzal, et al., 2006) Queuing theory (Leachman, 2004) To date, no general solution to the operating curve problem has been found. Slide # 10

11 Characterizing the Operating Curve (Fayed & Weber, 2008) Accurate measurement of setup time, qualification time and process time for each recipe, availability, uptime and throughput of every piece of equipment. All these variables are integrated into a queuing model (Hopp & Spearman, 2000) that represents all processes, flows, operations, equipment, recipes and qualifications in the fab. The output of this model is the expected operating curve of the fab. Slide # 11

12 Validation of Model (Fayed & Weber, 2008) Simulate current conditions (OC-1) Validate OC-1 at two actual operating points of the fab (OC-1 actual). The predicted and actual values agreed within 2%. Cycle time rises modestly until the fab is at 90% capacity and rises dramatically as the fab load approaches capacity. Slide # 12

13 Capital Purchase Decisions -- Expansions to the fab were planned by predicting operating curves under particular sets of conditions (OC-2 and OC-3). OC-2 was validated at two data points after expansion (OC-2 actual). Predicted values and actual values for cycle time differed by less than 3%. Further expansion has been simulated (OC-3). Capacity Planning Slide # 13

14 Optimizing the Operating Curve We define the operational effectiveness ratio (OER) as throughput divided by cycle time. OER is the general performance metric for the fab. We run our model in a manner that shows OER as a function of throughput and cycle time. We validate the model by comparing predicted conditions to actual conditions observed in the fab. If necessary, we adjust the model to predict actual data more accurately. Slide # 14

15 Key Finding-1 (Fayed & Weber, 2008) OER is nearly optimal for a wide range of throughput regardless of the level of capital investment. This suggests stable fab operations are possible if the fab s operating curve is well characterized. Slide # 15

16 Key Finding-2 (Fayed & Weber, 2008) OER is near optimal levels for a very narrow range of cycle time regardless of the size of the fab s equipment set. Not surprisingly, optimal cycle time decreases as the equipment base increases. Slide # 16

17 Implications A relatively wide range of throughput will yield a particular, nearly optimal cycle time. However, if a fab s throughput exceeds its near-optimal range, then the operational effectiveness and fab productivity fall off the cliff. The value of volume learning becomes highly negative! Slide # 17

18 An Optimal Level of Knowledge Optimal cycle time for every fab needs to be targeted. A model such as the one presented in this paper will be required to determine the optimal cycle and the throughput range that will yield a nearly optimal cycle time. The optimal level of volume production knowledge has to be discovered by trial and error or be simulated in advance. Slide # 18

19 Limitations of Model Useful heuristic that assists with capacity planning and output maximization, but... Suggests no approach towards maximizing profit. A value-driven approach is required. Incorporates revenue, cost, profit and time. Slide # 19

20 Operating a Modern Fab New products get introduced into the fab on a regular basis, Product unit price can decay over time. Planning for profitability is likely to include attempts to optimize the product mix that the fab realizes. The fab s management would like to run products with a high unit price. However, the fab capacity is limited, and overloading the fab will increase cycle time. The commitment to realizing older, less profitable products may have to be abandoned in favor of running newer, more profitable ones. How do you manage in this environment? Slide # 20

21 Simulation of Realistic Modern Fab (Fayed & Weber, 2008) Capacity = 1000 wafer starts per day 190 surviving dice per 200-mm wafer Highly nonlinear operating curve Decaying unit price -- P(t) = $100*10-t/(365 days) (order of magnitude per year) At t=0 the fab starts 100 wafers per day of a new product. Output metrics OER, Cumulative Revenue and Profit Slide # 21

22 The Value-Driven Operating Curve Success is a function of the operating curve. Maximum OER at 810 wafers per day. Maximum cumulative revenue at 900 wafers per day. (Life of product=365 days.) All variables drop off the cliff near maximum capacity. Slide # 22

23 Cumulative Profit as a Function of Time Fab respectively runs at 72%, 81%, 90%, 96% and 99% of its maximum capacity of 1000 wafers per day. The point of maximum cumulative profit is the ideal time to stop processing the new product (exit time). It is a function of fab throughput. A fab that is nearly fully loaded can never make a profit. 81% 90% 72% 96% 99% Slide # 23

24 Optimizing the Operating Curve Optimal exit time varies near-linearly with fab throughput. OER is at a maximum at 810 wafers per day. Cumulative profit is at a maximum at 900 wafers per day. Using OER as metric results in $3M of unrealized gain. Slide # 24

25 Implications for Operations Only a minority of fabs are capable of making accurate data-driven decisions that maximize profit. (Weber, 2006) Most fabs are still driven by maximizing output. The direct connection between the operating curve and profitability has not yet been made. Millions of US $ in gain are not realized as a consequence. Slide # 25

26 Summary and Conclusion This paper has presented a model of a fab operating curve that is sufficiently accurate to make capitalization decisions. The model has been used to simulate the performance of a hypothetical fab that operates under very realistic conditions. Simulation shows that significant additional revenue can be generated by shifting from today s output-driven approach to the operating curve to one that is value-driven. Slide # 26

27 Implications Pertaining to Technological Knowledge The value of volume learning is not always positive. Near maximum capacity revenue and profit drop dramatically. Simulation determines the optimal level of volume production knowledge. Slide # 27

28 List of References (1) Adler, P. S. and Clark, K. B., Behind the learning curve: A sketch of the learning process. Management Science 37(3), Alchian, A Reliability of progress curves in airframe production. Econometrica 31, Argote, L. and Epple, D Learning curves in manufacturing. Science 247, Aurand, S. S. and Miller, Peter J The Operating Curve: A Method to Measure and Benchmark Manufacturing Line Productivity. IEEE/SEMI- ASMC, Boston, MA, Sept , 1997, Bohn, R. E Measuring and managing technological knowledge. Sloan Management Review 36(1), Çakanyildirim, M. and Roundy, R. O SeDFAM: semiconductor demand forecast accuracy model. IIE Transactions 34(5), Dutton, J. M. and Thomas, A Treating progress functions as a managerial opportunity. Academy of Management Review 9(2), Slide # 28

29 List of References (2) Fayed, A. and Dunnigan, B Characterizing the Operating Curve How Can Semiconductor Fabs Grade Themselves? ISSM, Silicon Valley, October 2007, Pages Fayed, A. and Weber C. M Optimizing the Operating Curve: How can every fab maximize its performance? IEEE/SEMI-ASMC, Boston, MA, May 6, Jaikumar, R. and Bohn, R A dynamic approach to operations management: An alternative approach to static optimization. International Journal of Production Economics 27, Hayes, R. H. and Clark, K. B Exploring the source of productivity: Differences at the factory level, in: Clark, K. B., Hayes, R. H. and Lorentz, C. (eds.), The Uneasy Alliance: Managing the Productivity-Technology Dilemma. Boston, MA, Harvard Business School Press, Hopp, W. and Spearman, M Factory Physics, McGraw-Hill/Irwin; 2nd edition. Slide # 29

30 List of References (3) Leachman, R. C The economics of speed. presented at Portland State University s ETM Seminar, Portland, OR, Nov. 6, Leachman, R. C. and Ding, S Integration of speed economics into decision-making for manufacturing management. International Journal of Production Economics, Mishina, K Learning by new experiences: Revisiting the Flying Fortress learning curve. In Lamoreaux et al. (eds.) Learning by doing in markets, firms and countries. A National Bureau of Economic Research Conference Report. The University of Chicago Press, Chicago, IL. Nazzal, D., Mollaghasemi, M. and Anderson, D A simulationbased evaluation of the cost of cycle time reduction in Agere Systems wafer fabrication facility -- a case study. International Journal of Production Economics, Potti, K. and Whitaker, M Cycle Time Reduction at a Major Texas Instruments Wafer Fab. IEEE/SEMI-ASMC, 3/31 4/1, 2003, Slide # 30

31 List of References (4) Rapping, L Learning and the World War II production functions. Review of Economics and Statistics 48, Reinhardt, U. E Break-Even Analysis for Lockheed s Tri Star: An Application of Financial Theory, The Journal of Finance 28(4), Searle, A. D. and Gody, C. S Productivity increases in selected wartime shipbuilding processes. Monthly Labor Review 60, Weber, C Yield Learning and the Sources of Profitability in Semiconductor Manufacturing and Process Development. IEEE Transactions on Semiconductor Manufacturing 17(4), Weber, C. M. Do learning organizations have strokes of genius? Proceedings of PICMET 06, Istanbul, Turkey, July 8-13, Wright, T. P Factors affecting the cost of airplanes. Journal of Aeronautical Science 3, Slide # 31