Remaining Useful Life Estimation of Lenses for an Ion Beam Etching Tool in Semiconductor Manufacturing Using Deep Convolutional Neural Networks

Abstract

Maintenance plays a significant role in semiconductor manufacturing as plant yield, factory downtime and operation cost are all closely related to maintenance efficiency. Accordingly, maintenance strategies in semiconductor manufacturing industries are increasingly shifting from traditional preventive maintenance (PM) to more efficient predictive maintenance (PdM). PdM uses manufacturing process data to develop predictive models for remaining useful life (RUL) estimation of key equipment components. Traditional approaches to building predictive models for RUL estimation involve manual selection of features from manufacturing process data. This paper proposes to use deep convolutional neural networks (CNN) for the task of estimating RUL of lenses for an ion beam etch tool in semiconductor manufacturing. The proposed approach has the advantage of automatic feature extraction through the use of convolution and pool filters along the temporal dimension of the optical emission spectroscopy (OES) data from the endpoint detection system. Simulation studies demonstrate the feasibility and the effectiveness of the proposed approach.