Virtual metrology for semiconductor manufacturing: Focus on transfer learning

Abstract

Nowadays, virtual metrology models for semiconductor manufacturing aim to be scalable. A Virtual Metrology (VM) system is intended to cover a wide spectrum of production contexts. However, due to the large numbers of possible combinations of recipes, tools and chambers, it becomes intractable to model each context separately. This work presents a VM modeling approach based on the paradigm of transfer learning in a fragmented production context. The approach exploits a 2-Dimensional Convolutional Neural Network (2D-CNN) architecture, namely Spatial Pyramid Pooling Network (SPP-net), to perform the transfer learning from source to target domains with input of different sizes. We implemented several transfer learning strategies on a benchmark dataset provided by the Prognostics and Health Management competition in 2016. The main key points of the proposed approach are discussed based on the findings gathered from the numerical analysis.