A Machine Learning-Based Method for Assisted Analysis and Decision Making of Wafer Yield

With the rapid development of the semiconductor industry, the process technology has become more complex, and it is increasingly important to maximize the control of defects in production and improve wafer yield. Defects shown in Wafer Bin Maps (WBMS) are often strongly correlated with local system failures in production. In view of this, this paper proposes the convolutional neural networks (CNNs) architecture based on machine learning, and realizes the classification of hybrid defects by designing a separate classification model of minimum defects and using the probability analysis of recognition results. The purpose is to classify and statistics the faults of the wafer diagram fault diagram, perform yield analysis and calculation, identify defect types according to the wafer (fault pattern recognition), and take corresponding decisions to adjust the process in view of the systematic fault chip. The effectiveness of the method is proved by experiments, and the identification accuracy of single defects is higher, and the detection performance of mixed defects is also improved, and the overall performance is better.