Adaptive Plasma Parameter Control via Reinforcement Learning for Mitigation of Aspect Ratio Dependent Etching (ARDE) in Deep TSV Fabrication

**Abstract:** Aspect Ratio Dependent Etching (ARDE) poses a significant challenge in the fabrication of Deep Through-Silicon Vias (TSVs), leading to non-uniform etching profiles and compromised device performance. This paper presents a novel approach to dynamically control plasma parameters during Reactive Ion Etching (RIE) using a Reinforcement Learning (RL) agent. The RL agent learns an optimal control policy to minimize ARDE by adaptively adjusting gas flow rates, RF power, and chamber pressure in real-time based on continuously monitored etching endpoint signals. This approach promises a significant improvement in TSV etch uniformity and overall fabrication yield, a direct pathway to commercialization within 2-3 years.