Two-Stage Sequence Model for Maximum Throughput in Cluster Tools

Abstract

Cluster tool is a core manufacturing system in semiconductor industry. Optimizing the schedule of operations of a cluster tool is important because it is directly connected with its productivity. The scheduling becomes more complicated as the number of operating steps increases. There have been extensive studies to model the cluster tool operations and predict its throughput for a given configuration. However, the theoretical models cannot reflect realtime issues and the state-of-the-art throughput models are hard to be applied to predict scheduling parameters. In this work, we characterize the unique behavioral pattern of a key scheduling parameter towards the cluster tool throughput, and propose a novel deep-learning model that effectively identifies the best scheduling parameters. A two-stage model is designed that consists of an one-dimensional convolution neural network and a semantic segmentation network. Our experimental results show that the proposed model shows a superial accuracy than the state-of-the-art DNN solution for the best scheduling parameter detection.