Optimization on system design and operation for automated material handling system in semiconductor FABs

Abstract

This dissertation addresses the design and operation problems of the overhead hoist transportation system under practical issues. We focus on the practical issues of the specific environment called ``mega'' semiconductor FABs that a massive number of transportation requests are processed with a large number of OHT vehicles in unified OHT track layout. With the ever increasing size and complexity of modern FABs, practical issues that has been neglected in small FABs plays as important factors by affecting overall performance of the OHT system. First, the vehicles interfere with each other in frequent and unpredictable ways. Second, the resource limitation of wireless power supply (WPS) system, that supply electricity to the vehicles wirelessly, imposes blocks and deadlocks on the vehicle movement. These issues cause transportation delay, so OHT system possibly becomes the ``bottleneck'' of entire manufacturing process. However, the current semiconductor FABs rely on ad hoc approaches and manual interventions to handle these issues. Moreover, most related research has neglected these issues.

In this dissertation, we discuss two types of the design and optimization problem considering practical issues: (1) optimal design and allocation of WPS system, (2) Dynamic route guidance of OHT vehicles considering congestion in the track. WPS system design involves the determination of two important parameters: WPS capacity and area coverage. For the first topic related to WPS system, we derived the methodology to optimally determine the WPS capacity for minimizing the investment cost while maintaining reliability of the operation. For the second topic, we are investigating methodology to determine track partitioning of the WPS systems for minimizing transportation delay. Dynamic route guidance is to develop the adaptive routing method for avoiding congestion with travelling data of vehicles being generated in the FABs. Q learning, one of the reinforcement learning techniques, are used to adaptively control the routes of vehicles. Multi-step adaptation framework is proposed to utilize traffic data efficiently. Heuristic adaptation method is derived from the framework. Developed algorithm prevents and avoids congestion in OHT track with low calculation burden which is favorable for the actual FABs.

This dissertation makes following contributions: (1) It is the first study to investigate the impact of WPS system on the overall performance of OHT system and derive the optimization problem on the determination of WPS capacity with solution procedure that exploits properties of the problem. (2) It is the first study to define track partitioning problem of WPS units and propose the solution procedure based on the mathematical approach. (3) We propose reinforcement learning based adaptive dynamic OHT vehicle routing method that can avoid congestion with low calculation burden especially for the highly complicated modern FABs

it outperforms the best-known benchmark algorithm especially in the highly complicated modern FABs