Machine Learning based Test Quality Improvement
Machine Learning based Test Escape Detection
In test of large scale integration (LSI) circuit, test escape is always regarded as a critical issue since significant cost is imposed to manufacturing cost. We propose a novel outlier screening method for test escape. The proposed method exploits variational autoencoder (VAE) that is widely used to design complex generative model in artificial neural network field. While a typical autoencoder (AE) simply extracts features of training data, the VAE does it as probability distribution, and thus it can avoid potential risk of overfitting by using the probability distributions as a regularizer. Moreover, the proposed method effectively detects test escapes by utilizing the probability distributions as likelihood function of good chips. Through experiments using an industrial production test data, we demonstrate that the proposed method detects test escapes more than approximately 8.5 times as compared to that using a conventional AE-based method.