M05 Dependability | From Data to Actions: Applications of Data Analytics in Semiconductor Manufacturing & Test

This part motivates the need, the challenges, and the benefits of using data analytics and discusses its utility on actual industrial problems.

This part of the tutorial will cover the application of data analytics toward developing low-cost test and built-in test approaches. In general, test cost reduction can be achieved by replacing or eliminating a subset or all of the standard tests that are used for measuring the performances specified in the data sheet while maintaining test accuracy. We will discuss methods for leveraging the power of machine learning to replace standard specification tests by predicting their outcome from lower-cost alternative test measurements and to reduce the number of standard specification tests that are finally carried out by identifying intricate correlations amongst them.

This part focuses on the application of data analytics for developing tests that are adjusted dynamically on a device-by-device, wafer-by-wafer, or lot-by-lot basis, rather than chosen statically for all devices. This dynamic adjustment aims at choosing the most appropriate test limits, test content, and/or test flow based on historical and real-time test data. Adaptive test can further reduce test cost and improve test quality and is indispensable in dealing with process variation and/or test equipment drift.

This part discusses an orthogonal direction for leveraging statistical learning by exploiting wafer-level and lot-level correlations rather than die-level correlations. Recent research has shown great promise in reducing test cost by sampling a measurement at a sparse subset of die locations on each wafer/lot, and then training statistical spatial/spatiotemporal models to predict these measurements at unobserved die/wafer locations.

This part introduces a process variation decomposition method which offers insight regarding the correlation between manufacturing and test outcomes. Specifically, process monitoring through e-tests facilitates various applications, including yield loss attribution, yield estimation, outlier detection, yield forecasting in production migration and fab attestation.