Foreseeing the Invisible Data Mistake Using Machine Learning for Devices That Are Prone

Abstract

One kind of test escape that corrupts data without anyone noticing is known as a Silent Data Error (SDE). At the scales used by cloud service providers, they are noticeable even at very low DPPM levels. The costly system level test (SLT) is used to screen for certain faults in high-volume production that appear as system dependent errors (SDEs). Because these faults are so subtle, semiconductor devices that are susceptible to SDEs don't show any obvious patterns or abnormalities in the test data distributions. Therefore, it is difficult to screen these defective devices using ATE utilizing statistical death limits. We suggest using a Supervised Machine Learning (ML) strategy to discover inherent patterns in a real-world test dataset in order to speed up the process of identifying such defective devices before system testing. Results from experiments reveal that when compared to both conventional supervised and unsupervised approaches, the adopted supervised learning framework using an ensemble of feature selection algorithms performs far better. Index Terms—machine learning, silent data mistakes