Wafer yield prediction using derived spatial variables

Abstract

Unreliable chips tend to form spatial clusters on semiconductor wafers. The spatial patterns of these defects are largely reflected in functional testing results. However, the spatial cluster information of unreliable chips has not been fully used to predict the performance in field use in the literature. This paper proposes a novel wafer yield prediction model that incorporates the spatial clustering information in functional testing. Fused LASSO is first adopted to derive variables based on the spatial distribution of defect clusters. Then, a logistic regression model is used to predict the final yield (ratio of chips that remain functional until expected lifetime) with derived spatial covariates and functional testing values. The proposed model is evaluated both on real production wafers and in an extensive simulation study. The results show that by explicitly considering the characteristics of defect clusters, our proposed model provides improved performance compared to existing methods. Moreover, the cross‐validation experiments prove that our approach is capable of using historical data to predict yield on newly produced wafers.

A Wafer Bin Map “Relaxed” Clustering Algorithm for Improving Semiconductor Production Yield

“…This approach integrates data transformation with k−means and particle swarm optimization (PSO) clustering algorithms, assessing results through adaptive response theory (ART) neural networks. In [23] Hang Dong et al analyse the spatial patterns of defective chips on Wafer Bin Maps, proposing a wafer yield prediction model that incorporates the spatial clustering information in functional test-ing. Liukkonen and Hiltunen [24] combine Self-Organizing Map (SOM) network and k−means clustering algorithm for analysing systematic defect patterns on spatially oriented wafer maps.…”

“…These "identified problems" make up an important knowledge base (also known as rules or defect patterns [23]) useful for dealing with manufacturing problems. Then, in the production line, a wafer's defects can be compared with the set of known defect patterns [29] to verify the possible presence of already known "rules": if the defective wafer does not belong to any known feature (at least with an acceptable degree of confidence), a new defect pattern is submitted to the attention of the analyst who, if that pattern is internally consistent, formalises the recognition of a new yield issue (rule) and inserts it in the database of known issues.…”

A Review and Analysis of Automatic Optical Inspection and Quality Monitoring Methods in Electronics Industry

“…They showed that their proposed method is rotation invariant and can work correctly in noisy images with the variations of defect locations and angle. However, this method directly generated the probability of failure according to physical locations and did not explicitly separate clustered defects and random defects [369]. Li and Huang in [75] proposed a hybrid approach combining the supervised SVM classifier with the unsupervised SOM clustering for binary bin defect pattern classification.…”

Rotation-Invariant Wafer Map Pattern Classification With Convolutional Neural Networks

“…Early studies have focused on extracting useful features from wafer maps based on feature engineering. Notable and widely-used feature extraction methods in the literature include projection [8]- [10], [12], geometry [8], [9], [12], clustering [11] and density-based methods [13]. Any prediction model, such as logistic regression, tree ensembles, support vector machines and neural networks, can be built on top of these hand-engineered features.…”