Adaptive Post-silicon Server Validation using Machine Learning

This paper mainly focuses on providing solutions for efficient feature validation. Modern day server processors and computer systems are developed with billions of transistors. Validation of such a complex systems is playing a crucial role in current research. Pre-silicon validation is not enough to get a full system functional coverage. Post-silicon validation is a necessary step to validate these complex systems and to determine the escaped functional silicon bugs during presilicon validation. During post-silicon validation in order to get a full system functional coverage there are more number of features for testing. Applying all the features manually and going through the each test results is difficult to maintain. In order to reduce resource requirements for determining test failure signature, and to reduce the time to debug the failure, introduced the machine learning in current validation environment. The proposed validation algorithm in this paper, which is very useful in feature validation of server processer’s and is adaptive to the previous validation learning’s. Validation is mainly carried out for power management features provided by Advanced Configuration and Power Interface specification. The functional coverage implemented for important power management features namely processor power states, processor performance states and thermal states. This feature coverage analysis is provided through graphical plots. General Terms Algorithm, Functional coverage, Machine learning, Processor Power and Thermal management, Validation