Reinforcement Learning for Efficient Scheduling in Complex Semiconductor Equipment

Abstract

Semiconductor cluster tools add an integral component to the modern semiconductor manufacturing process. These complex tools provide a flexible deployment option to group multiple processing steps into a single piece of equipment, allowing for more efficient processing. They also contribute to a reduction in the number of times a wafer must go through the atmospheric-vacuum-atmospheric cycle. These highly automated tools present a complex scheduling challenge where process-specific requirements are balanced against a need to achieve maximum wafer throughput in a fault tolerant manner. Previous work demonstrated that a reinforcement learning algorithm would be suitable for automated generation of efficient planners for simple tools. This investigation looked at how these same techniques could be extended to operate on more complex equipment.