Autonomous Process Control via Cloud-Based Reinforcement Learning Agents

About the Session

Semiconductor manufacturing demands precise control of complex, high-variability processes. Traditional static control systems often fall short in adapting to dynamic shifts in tool behavior and product requirements. This presentation introduces a cloud-native framework that leverages reinforcement learning (RL) to enable autonomous process control across semiconductor fabrication environments. RL agents are trained in the cloud using a combination of historical process data and live sensor inputs to discover optimal control strategies. These strategies are then deployed to edge controllers, allowing real-time, low-latency adjustments to process parameters directly at the tool level. This cloud-to-edge orchestration supports continuous learning, recipe optimization, and fault response with minimal human intervention. The session will cover system architecture, data flow design, and deployment considerations, along with performance benchmarks and potential applications. The approach aims to enhance yield, reduce process variability, and move toward intelligent, self-optimizing fab operations.

Speaker

Tarun Parmar

Principal Software Engineer at Skyworks Solutions