ET-DRL+AF-PID: Co-Suppression of Multi-Physics Disturbances for Sub-Nanometer Motion Control at 3 nm Node

Ultra-precision motion control in the field of semiconductor lithography and optical manufacturing faces the core challenge of coordinated suppression of multiple physical disturbances: thermal gradients lead to 12-nanometer positioning drift, mechanical resonance causes 20-nanometer vibration amplitude, and electromagnetic interference causes ±5% thrust fluctuation. Traditional proportional-integral-differential control is difficult to meet the 5-nanometer trajectory tracking accuracy requirement due to bandwidth limitations, and model predictive control is limited by more than 500 microseconds of computing delay and cannot adapt to high-speed scanning scenarios. Pure deep reinforcement learning lacks engineering feasibility due to the demand for millions of training samples. This study proposed a fusion architecture of event-triggered deep reinforcement learning and adaptive fuzzy proportional integral differential, breaking through technical bottlenecks through three innovations.