Practical machine learning for tools
These slides were presented at SEMICON Europa, 2019.
Semiconductor equipment makers (OEMs) are under pressure to produce tools that offer the maximum possible substrate throughput while maintaining high process quality levels. Although actual processing time is often constrained by physical limits based on the underlying requirements of the process itself, overall throughput in a tool is also subject to non-productive time delays such as transfer time, equipment speed limitations, and vacuum pump down. Therefore, minimizing non-productive tool time is a key element in maximizing overall throughput. This case study investigates efforts made to use machine learning to improve throughput in a complex cluster tool. Traditional methods to find planner optimizations were successful, but required extensive amounts of non-recurring engineering expense. This led to an exploration of how modern machine learning algorithms could be used to streamline the planner optimization process and find solutions beyond what a human engineer could discover.
Want the presentation? Tell us about yourself.