Research On Prediction Of IC Wafer Yield Using Multivariate Piecewise Model Basing On PCA

With the China government launched a "National IC industry development outline for promotion", IC manufacturing flourishes in our country these years. The main semiconductor manufacturing enterprises in our country are OEM. So, the yield improvement and prediction are very important topic for semiconductor industry.There are a lot of studies and achievements of the yield prediction and control in researchers from the last century 60’s. The early research works are mainly focus on the relationship between yield and line defects. With the development of semiconductor technology, the process design becomes more and more complex. It makes higher requirement of yield prediction.In this paper, we propose a new mode combining with two reasons of yield losing. The first reason is inline defects. And the second reason is design weakness. We use a multivariate piecewise function model in order to reduce the prediction error. The principal component analysis (PCA) is used to reduce the dimensions and redundant information. And the decision tree method is used to find the suitable segment point. The logistic regression model is also used to improve the accuracy. In finally, the prediction error is less than previous researches. And that, the advantage of the multiple piecewise function is that it can explain the two kinds of yield losing reason, and provide engineering improvement suggestion to engineers.Based on this statistical method, a set of standard procedure can be built. The computer program is available for the standardization process, which is automation and convenience to production.