Investigators
Michael Fu, Steven Marcus (Univ. of Maryland)
Emmanuel Fernandez (Univ. of Cincinnati)
Industrial Liaisons
Gurshaman S. Baweja, Texas Instruments Incorporated
Wayne F. Carriker, Intel Corporation
Robin L. Hoskinson, Intel Corporation
Ben-Rachel Igal, Intel Corporation
Mani Janakiram, Intel Corporation
Shekar Krishnaswamy, Motorola, Incorporated
Ying Tat Leung, IBM Corporation
Matilda O'Connor, Advanced Micro Devices, Inc.
Madhav Rangaswami, Intel Corporation
Ramesh Rao, National Semiconductor Corporation
Anticipated Primary Result
* Models, algorithms, and a suite of software tools to automate and/or guide in optimally scheduling and coordinating PM tasks for bottleneck tools in a semiconductor fab. Software will be designed for seamless integration with existing commercial discrete-event simulation software packages such as AutoSched AP and WorkStream. 2002 Annual Report
Background
A critical need for the proposed research was identified through interactions with industrial partners during the final year of the SRC Task 491.001, "Markov Decision Processes for Integrating Life Cycle Dynamics into Fab-Level Decision Making." Preventive maintenance of bottleneck tool sets was identified as an area of critical need for improvement, leading to the development of some initial models. A summer internship at AMD by a Ph.D. student from Maryland led to specific algorithmic formulation for a single cluster tool (CVD TiN), in the form of a mixed integer programming model. Preliminary simulation studies were conducted, the results of which indicated very promising potential for the approach. However, the model did not take into consideration a number of critical factors in the fab, including the consideration of tool failure and other unplanned stochastic events such as process drift, and did not incorporate production planning and process control information into the algorithms. Furthermore, there was a clear need for further research on other tool sets such as photolithography.
Description
We propose to develop models, algorithms and a suite of software tools that can be easily employed to obtain close-to-optimal PM schedules for semiconductor manufacturing. The models and algorithms will be general enough so that they can satisfy the needs of any specific fab, and the software tools will cover all of the usual bottleneck tool sets in the fabs. Traditionally, maintenance and production planning have been addressed in isolation. This can lead to a substantial under-utilization of equipment in semiconductor manufacturing fabs, which are characterized by re-entrant flows and high correlation between different tools. In addition to considering stochastic unplanned events such as tool failure or process drift, our approach will explicitly take into consideration production control information (e.g., WIP), and future planned production schedules. Another feature of our proposed approach is the model integration of the interdependence of different PMs in the fab, e.g., consolidation of PM tasks for increasing tool availability. Models and algorithms will be developed to cover the major bottleneck tools in a fab. Comprehensive validation through simulation, and use of real data (reliability, WIP levels, etc.) will be performed. These efforts will be further guided by conducting a comprehensive survey of PM practices in industry during the first year. A deliverable from this phase of the project will an extensive simulation engine, based on commercial software (e.g., ASAP, IBM OSL) that can be used in industry to benchmark current practices and gauge potential gains. Efforts during the second and third year will increasingly concentrate on software tool development and transfer to industry. A clear and focused effort will be made to develop software that is fully compatible with commercial software used in industry, in order to facilitate integration in the form of technology transfer to 3rd party commercial vendors.