A Machine Learning Approach to Predict Yield with Thermal Variation on Silicon Based on CTS
Michael Chang, Simon Kao, Patrick Xue, Bryant Hsu, Andrew A. Chien, Kevin Chung, Robby Ho · 2020
In this paper we propose an accurate machine learning technique of thermal impact on silicon within a given system to improve compliance test standard (CTS) result to predict the yield analysis based on artificial neural network and the regression based polynomial regression. The disadvantage of conventional CTS is that the temperature effect is not considered. Study shows that silicon is very sensitive to thermal variation especially the eye height is greatly reduced as temperature increases. This variation is due to the poly-resistor temperature dependence of the transmitter and the frequency response temperature of the continuous time linear equalization of the receiver. Not addressing the thermal impacts may result in system malfunction with large temperature span. The goal during is to find the issue of insufficient profit at early design stage, build up high reliability method and achieve on system-level success.