Strategic Capacity Planning for the Semiconductor Industry:Current 

Industrial Practice and New Directions

Robin Roundy, MetinCakanyildirimWoonghee Timothy Huh, Feng Zhang

School of Operations Research and Industrial Engineering

CornellUniversity

We summarize a multi-year research effort designed to provide useful tools for capacity planning decisions in the semiconductor industry.The decisions are crucial and challenging.The business environment is volatile, but equipment has long procurement lead times and is extremely expensive. We will review and evaluate current business practices.We present methods for quantifying the errors in demand forecasts. We present a novel approach for multi-dimensional demand modeling, and discuss practical and algorithmic implications of different stockout cost models.We present efficient algorithms for provably solvable versions of the capacity planning problem.

Workforce Agility: Models for gaining insight into when and how to use cross-training

Wallace J. Hopp

BreedUniversity Professor

Northwestern University

We examine issues that affect the nature of the “best” strategy for cross-training workers in serial production environments.Using a combination of optimization and simulation models, we examine strategies based on chaining, which enable labor capacity to be shared very flexibly.Our results suggest that chaining strategies work robustly well given a suitable task allocation policy and that they offer promise as practical workforce management tools.

Using Transaction Data for the Design of Sequential, Multi-unit, Online Auctions

Abraham Seidmann

SimonGraduateSchool of Business Administration

University of Rochester

Internet auctions for consumers' goods are an increasingly popular selling venue.The Internet's computational ability makes possible the sale of multiple units of the same good in a single auction.We found that many sellers, instead of offering the entire inventory at a single auction, split it into sequential auctions of smaller lots, to reduce the negative market impact of large lots.In the short lecture we plan to show how the available inventory should be split into multiple lots and how many sequential auctions should be run.We also investigate how managers can leverage information technology to improve the design of future auctions. 

Assuming a truth-revealing ascending auction model, we quantify the effect of auction lot size on the closing price.We then develop a stochastic optimization model for allocating inventory across multiple auctions.Solving the dynamic programming formulation, we prove that the lot size drops from period to period.The intensity of the decline increases in the holding costs and the website's traffic intensity, while decreasing in the dispersion of consumers' valuations of the good.Finally, we extend this model to dynamically incorporate the results of previous auctions as feedback into the design of consecutive auctions, updating the lot size and number of auctions.We demonstrate how information signals from previous auctions should be used to update the auctioneer's belief s about the customers' valuation distribution, thereby significantly increasing the sellers' profit potential.We conclude the seminar with several practical examples that show the economic benefits of using detailed transaction data for the design of sequential, multi-unit, online auctions and how these benefits are influenced by the inventory holding costs, bid traffic, and the dispersion of consumers' valuations.(Joint research with Professors Pinker and Vakrat.)

RECENT RESULTS ON TWO-MOMENT APPROXIMATIONS, SAFETY LEAD TIMES, AND HYBRID PUSH-PULL SYSTEMS

RajanSuri

Center for Quick Response Manufacturing

University of Wisconsin-Madison

Dr. Buzacott's pioneering works in the areas of modeling job shops, analyzing safety stock and safety lead times, and hybrid material control systems, have inspired many researchers over the years. In this talk we will show some of the latest results in these three areas: (i) simple two-moment approximations for fork-join stations which can be used to model kanban and other material control systems in job shops; (ii) numerical results showing that safety lead times are better than safety stock for multi-product systems over a wide range of parameters; (iii) case studies on implementations of a hybrid push-pull system, called POLCA, at several factories, explaining why the system was chosen and the performance results seen thus far. This talk represents joint work with Ananth Krishnamurthy and Mary Vernon at the University of Wisconsin-Madison.

I-CLIPS : One way to materialize a stochastic model of a manufacturing system

Nico Vandaele

University of AntwerpBelgium

Building stochastic models that support operational and managerial decision making in manufacturing systems is an intriguing academic research activity. However, transferring these models to industry is even more challenging and it includes not only issues like software development, a dedicated IT-approach and the effective implementation but also a well-thought educational effort. We discuss the software I-CLIPS, which is the materialisation of a multi-product queuing network with embedded lot sizing optimisation. I-CLIPS serves three purposes: better planning, better tuning and better levers for improvement.

Spare Parts Inventory Management at the Royal Netherlands Navy

dr. W.H.M. Zijm

University of Twente, the Netherlands

In this talk we present models and techniques that have been developed to support the design of a new spare parts management system at the Royal Netherlands Navy. Since we focus on high system availability at ships, failed components or subassemblies are initially replaced by spare items, while the failed items are later repaired and then stored for future re-use. Since military installations on ships often have a complex product structure, and since spare parts are kept in stock both at the ships and at a central facility, we deal with multi-echelon, multi-indenture inventory systems. Our goal is to establishoptimal base stock levels of all components at all sites in order to maximize system availability given limited investment budgets. We extend the analysis to repair shops with limited capacities, each modeled as a BCMP queuing network. In addition, we present results for complex real life systems, indicating that substantial improvements in system availability can be obtained, while at the same time reducing inventory investments with more that 50 %, leading to estimated savings of more than 25 million Euro in inventory investment.

Kanban, CONWIP, PAC, Information Flow, and Multiple Loop Network Decomposition

Stanley B. Gershwin

Department of Mechanical Engineering

Massachusetts Institute of Technology

In many material flow systems (including kanban, CONWIP and PAC), the flow of material is controlled by the presence or absence of tokens. 

In such systems, a token is generated by the arrival of an order, the arrival of an item of raw material, the departure of a finished item, or by the completion of an operation.The presence of this token is required before some other operation may occur.To generalize, there may be many tokens taking different (but fixed) paths in the system, and the space allowed for the storage of tokens at specific points in the system may be limited.

We describe a class of flow network models which can help the analysis and design of such control mechanisms, and of the manufacturing systems controlled by such mechanisms.Material flows in one part of such a network and tokens flow in the rest of it, but no distinction is made between material and token flow in the analysis.The issuance of a token at the completion of an operation may be viewed as a disassembly, and the requirement that a token be present for an operation to take place may be viewed as an assembly.

These networks are a generalization of more restricted systems that have been analyzed by decomposition methods.In such systems, buffers (links) are finite, machines (nodes) are unreliable, machines may perform assembly and disassembly operations, but the network is limited to acyclic or tree structures.In the larger class, there is no such structural limitation.In particular, there may be multiple loops.

Loops create a new phenomenon that does not exist in acyclic systems: a strong correlation among the quantities in buffers at any time. This is because there is an invariant associated with each loop.(For example, in closed loops where the material visits all the machines and buffers in order, the population is constant.)

We describe a new decomposition method for the performance analysis of these systems.

On the Value of Advance Demand Information in Production/Inventory Systems

FikriKaraesmen

Laboratoire Genie Industriel

EcoleCentrale Paris.

Inspired by the earlier work of Buzacott and Shanthikumar, we investigate the value of advance demand information in production/inventory systems. For a single-stage make-to-stock queue, we assess the value of using advance demand information under a variety of assumptions on the cost of obtaining advance demand lead time, and on the delivery timing requirements. This analysis enables us to identify conditions under which advance demand information may bring significant benefits in capacitated systems. (joint work with G. Liberopoulos and Y. Dallery)

What is Missing to Enable Optimization of Inventory Deployment and Supply Planning?

Sridhar Tayur

GSIA 

CarnegieMellonUniversity

Tremendous opportunity exists if we can bridge Top Academic Research and the commercial marketplace. The timing is appropriate because investments in ERP and APS have failed to provide the return, and the practitioners are getting frustrated with existing vendors. Frankly, data is also available at an adequate level. I would like to celebrate John's workshop by showcasing some significant practical contributions that have been made when our community's research has been appropriately applied. This may also lead to a discussion on how to make our research more practically useful in a widespread manner.

The Evolution of Manufacturing System Models: A Personal View

John Buzacott

SchulichSchool of Business

YorkUniversity

This talk outlines the way in which models of manufacturing systems have evolved over the last 50-60 years. It is presented from a personal perspective: how awareness of issues arose and how models developed. The role of direct and indirect industry interactions combined with student originality and commitment will be stressed.