Complex manufacturing processes are the heart of semiconductor manufacturing. A semiconductor chip is a highly miniaturized, integrated circuit (IC) consisting of thousands of components. Semiconductor manufacturing starts with thin discs, called wafers, (typically) made of silicon. A large number of usually identical chips can be produced on each wafer by fabricating the ICs layer by layer in a wafer fabrication facility (wafer fab). The corresponding step is referred to as the Fab step. Next, electrical tests that identify the individual dies that are likely to fail when packaged are performed in the Probe facility. An electronic map of the condition of each die is made so that only the good ones will be used. The probed wafers are then sent to an Assembly facility where the good dies are put into an appropriate package. The assembled dies are sent to a test facility where they are tested to ensure that only good products are sent to customers. The tested devices are then sent to regional warehouses or directly to customers. Wafer fabrication and probe are often called the front-end and assembly and test are called the back-end.

Supply chain management (SCM) problems have become more and more important in the last decade. This has been caused by the fact that front-end operations are often performed in highly industrialized nations, while back-end operations are typically carried out in countries where labor rates are cheaper. Moreover, there are centers of competencies (e.g. bumping) that may consist of only a few process steps that may be done in a different company owned facility or remotely by a subcontractor. These centers of competencies speed up innovations and reduce costs, but increase the complexity of SCM.

The semiconductor industry is capital intensive with the cost of an entire wafer fab up to nearly $10 billion US caused primarily by extremely expensive machines, some up to $100 million US each. The manufacturing process is very complex due to the reentrant flows in combination with very long cycle times and the multiple sources of uncertainty involved. Capacity expansions are very expensive and time-consuming. This kind of decision is based on demand forecasts for the next years. Because of the rapidly changing environment, the demand is highly volatile. Consequently, the forecast is rarely accurate. The semiconductor industry is an extreme field for SCM solutions from an algorithmic as well as from a software and information systems point of view. The huge size of the supply chains involved, the pervasive presence of different kinds of uncertainties and the rapid pace of change leads to an environment that places approaches developed in other industries under major stress. Modeling and analysis approaches that are successful in this industry are likely to find applications in other areas, and to significantly advance the state of the art in their fields (cf. [1]).

The purpose of this seminar was to bring together researchers from different disciplines including information systems, computer science, industrial engineering, operations research, and supply chain management whose central interest is in modeling, analyzing, and designing complex and large-scale supply chains as in the semiconductor industry. Moreover, practitioners from the semiconductor industry who have frequently articulated their perception that academic research does not always address the real problems faced by the industry brought in their domain knowledge to make sure that progress towards applicability and feasibility would be made during this seminar. The seminar had 26 attendees from ten different countries (see participant list at the end of the report). We had participants from leading semiconductor companies Infineon Technologies and Intel Corp. as well as researchers who work closely with ST Microelectronics, Globalfoundries, and Taiwan Semiconductor Manufacturing Company (TSMC).

A primary purpose of the workshop was to extend the scope of the academic research community from single wafer fabs to the entire semiconductor supply chain. We show the principle architecture of the planning and control system of a semiconductor supply chain in Figure 1.

Seminar Objectives

The first objective of the seminar consisted of developing a research agenda for semiconductor supply chain modeling and analysis topics. This includes innovative modeling approaches for supply chain network planning, demand planning, master planning, and detailed production planning and scheduling in semiconductor supply chains. But it also includes ideas on how to design the related future information systems.

The research agenda was developed around the following two main topics:

• Topic 1: Novel planning and scheduling approaches that can deal with the complexity and stochasticity of the semiconductor supply chain:
  • Many planning approaches on the SC-level are based on (distributed) hierarchical and generally deterministic approaches to deal with the sheer complexity of the semiconductor supply chain. The role of anticipation of lower level behavior in upper level decision-making is still not well understood and has to be studied in more detail. Because a semiconductor supply chain contains many different, often autonomous decision-making entities including humans, negotiation approaches are typical in such distributed hierarchical systems for planning and control. It should be researched how such negotiation approaches can be automated and which decisions should be made by humans.
  • The overall cycle times in a typical semiconductor supply chain are on the order of 10 to 15 weeks. Therefore lead times have to be modeled in planning formulations. Using lead times as exogenous parameters in planning formulations leads to a well-known circularity because the cycle time depends in a nonlinear manner on the resource utilization which is a result of the release decisions made by the planning approach. Different types of clearing functions have to be researched in the semiconductor supply chain context.
  • Approaches to demand planning that take the product life cycle into account have to be studied. The interaction of demand planning and supply chain planning has to be investigated.
  • Different ways to anticipate stochasticity including robust optimization, approximate dynamic programming, and stochastic programming have to be researched in the semiconductor supply chain context.
  • Different ways to appropriately deal with stochasticity including rolling planning techniques and inventory holding strategies have to be studied.
  • Generation of scenarios and other distribution parameters for planning problems in supply chains using data mining techniques have to be researched.
  • Because of the complexity of supply chains, long computing times still hinder the usage of analytic solution approaches especially for what-if analysis. The role of state-of-the-art computing techniques including parallel computing on Graphics Processing Units (GPU) machines or Cloud computing techniques in decision-making for semiconductor supply chains has to be investigated.
• Topic 2: Future information systems and supply chain management in the semiconductor industry:
  • Understanding the limitations of today’s packaged software for supply chain management in the semiconductor industry.
  • Proposing alternative software solutions including software agents and service-oriented computing for planning and scheduling applications in the supply chain context.
  • Integration concepts for state-of-the-art computing techniques to get models that are computationally tractable and address the different uncertainties encountered in this industry.
  • Approaches to embed real time simulation techniques in current and future information systems to support decision-making in semiconductor supply chains.
  • Understanding the interaction of human agents with information systems.

The implementation of ERP, APS, and MES systems in semiconductor supply chains provides both an opportunity and the need for development of supply-chain wide integrated production planning and scheduling solutions. Therefore, we think that the second topic is important and should be also addressed in the research agenda. Research related only to the first main topic is not sufficient.

The second objective of the seminar consisted of identifying the core elements of a conceptual reference model for planning and control of a supply chain in the semiconductor industry that can be used for analysis and performance assessment purposes and to foster a common understanding in the research community both in academia and industry. This included specifying reference planning and control activities, the major information flows, and their interaction with a reference system of a physical supply chain. Due to the inherent complexity of semiconductor supply chains it requires simulation of the physical supply chain to understand the interactions between the planning and control components and the physical supply chain, to find solution approaches to problems and to verify them in the risk-free simulation environment before implementing them. There are widely accepted reference (simulation) models for single wafer fabs, mainly developed in the Measurement and Improvement of Manufacturing Capacity (MIMAC) project (led by one of the organizers of this Dagstuhl seminar) 20 years ago that are still used by many academic researchers working with the semiconductor industry.

Existing reference models on the planning and control level like the Supply Chain Operations (SCOR) reference model and the supply chain planning (SCP) matrix are too generic to be useful for detailed analysis and have to be refined considerably to cover the important domain-specific aspects of semiconductor supply chains.

The Process

In the opening session, the organizers welcomed the participants and acknowledged Infineon Technologies as a sponsor of the seminar. Next, the participants each introduced themselves. This was followed by an overview of the goals and objectives of the seminar and a detailed review of the seminar program including the ground rules for interactions.

The remainder of the day on Monday consisted of four industry overview talks (by Hans Ehm, Kenneth Fordyce, Chen-Fu Chien, and Irfan Ovacik) and a review of the literature related to modeling an analysis of semiconductor supply chains (by Lars M"onch and Reha Uzsoy). Tuesday and half a day on Wednesday were devoted to presentations and discussions about the various elements of the semiconductor supply chain planning and control systems shown in Figure 1 above. See Table 1 for a list of topics and presenters and Section 3 for abstracts of the presentations.

Wednesday afternoon was the excursion that was enjoyed by the participants. Thursday was devoted to 3 breakout sessions with report outs on the topics in Table 2. Section 4 has the breakout report outs.

The first set of breakout sessions had four groups focus on the individual elements in Figure 1 and one group focus on a semiconductor supply chain reference model. The second set of breakouts had three groups consider the interaction between various elements in Figure 1, one group talked about the incorporation of humans in the supply chain, and one discussed how to go from the reference model to a specific semiconductor supply chain model instance.

The final Friday set of breakouts included three groups that discussed process models of multiple elements from Figure 1 and the flow of information needed between the elements to provide core elements of a reference model. Another group discussed the role of agents in a semiconductor company’s supply chain. The final breakout group discussed the level of detail needed in a top down reference model. Friday consisted of a discussion on the required core elements of a reference model for semiconductor supply chains and a wrap-up session.

Key Take Aways

here were a number of key findings and areas for future research that were identified in the seminar. We will first summarize some of the key findings and will follow this with some areas for future research.

One of the first findings was that the participants generally agreed that the different elements in Figure 1 are reasonably well understood by both the industrial and academic communities, but the interactions between the elements are less well understood. Having said this, a number of the software solutions for the elements are not geared toward the complexities of the semiconductor industry (e.g. ATP/APS systems are generally focused on profit maximization and ignore many of the system complexities). Second, it appears that there are still limitations in solution approaches in practice such as: capacity generally is expressed without regard to mix; fixed lead times are generally still assumed despite research done on clearing functions for planning; and ignoring all but production lots when developing plans. Third, as indicated above both the industrial and academic participants generally agree that the integration of the decisions made by the different elements is often fairly ad hoc and could/should be improved. Finally, the participants generally agreed that there does not currently exist an adequate reference model for the semiconductor supply chain. In fact, there is not even a reasonable set of data sets that describe instances of the semiconductor supply chain such as the MIMAC datasets at the factory level. There is some indication that a reference model and incorporating human behavior of the various decision makers on the supply chain level will help to better understand supply chains producing and containing semiconductors.

In addition to the findings mentioned above, several areas for future research were identified. An overarching idea was that the future research should focus more on formulation of appropriate models because this is fundamentally more important than the actual solution techniques chosen. Some of the future research areas are included below:

• Using event-driven process chains (EPCs) to model/visualize planning processes.
• Developing better integration of various decisions made in the elements of Figure 1.
• Combining rolling horizon strategies with demand forecast evolution models.
• Incorporating sustainability aspects into supply chain models.
• Developing stochastic model versions of current deterministic models.
• Incorporating the behavior of human decision makers (this will be useful, but challenging).
• Exploring the use of different simulation paradigms (systems dynamics, agent-based, hybrid models, reduced simulation models) to model and analyze semiconductor supply chains.

Next Steps

As a way to further the discussion of and collaboration on the topics of the seminar, Prof. Lars Möonch, Prof. Chen-Fu Chien, Prof. Stéphane Dauzère-Pérès, Hans Ehm, and Prof. John Fowler are guest editing a special issue of the International Journal of Production Research (IJPR) entitled Modeling and Analysis of Semiconductor Supply Chains. The deadline for submission is September 1, 2016. This date was selected to allow time for ideas created by the participants of the seminar to be incorporated into papers for the special issue. The Call for Papers can be found at the following address: http://explore.tandfonline.com/cfp/est/semiconductor-supply-chains-call

Aknowledgements

The seminar organizers would like to thank Infineon Technologies AG for their support of the seminar. The seminar also would not have been nearly as productive without the active contribution of every attendee, and for that the organizers are extremely grateful.

References

1. Chien, C.-F., Dauzère-Pérès, S., Ehm, H., Fowler, J. W., Jiang, Z., Krishnaswamy, S., Mönch, L., Uzsoy, R. (2011): Modeling and Analysis of Semiconductor Manufacturing in a Shrinking World: Challenges and Successes. European Journal of Industrial Engineering, 5(3), 254–271, 2011. http://dx.doi.org/10.1504/EJIE.2011.041616
2. Mönch, L., Fowler, J. W., Mason, S. J. (2013): Production Planning and Control for Semiconductor Wafer Fabrication Facilities: Modeling, Analysis, and Systems. Springer Operations Research/Computer Science Interfaces, New York, Vol. 52, 2013. ISBN 978-1-4899-9901-6.

Creative Commons BY 3.0 Unported license

Chen-Fu Chien, Hans Ehm, John Fowler, and Lars Mönch