Kernel Based Functional Test Analysis Framework for Test Content Optimization

The dramatic increase in design complexity of modern circuits challenges our ability to verify their functional correctness. Functional verification is critical, because an undetected bug in a design may result in significant financial loss for a company. Among the techniques and methodologies available for functional verification, simulation-based verification is prevalent in industry because of its linear and predictable complexity and its flexibility to be applied to any design. Success of simulation-based verification depends on generating effective verification tests that can achieve the verification objectives in short amount of time. Therefore, test content preparation is one of the most important tasks while generating thousand of tests to verify the design.;When preparing test content for the pre-silicon verification, its objective is to generate effective verification tests that can achieve high verification coverage quickly. When preparing test content for post-silicon validation, its objective is to maximize the frequency of hitting a few targets, such as worst-case power. Currently, verification and validation methodologies rely on direct and/or constrained random test generation. Due to the difficulty in modeling and time/resource constraints in simulation, the evaluation of test coverage with respect to the effects of interest may not be accurate. As a result, test content optimization is limited by the available information on test coverage. To compensate this limited information, additional knowledge on test coverage has to be learned by domain experts who prepare the test content. This learning process is usually not automatic and can be quite ineffective.;In this dissertation, we propose a kernel based functional test analysis framework for test content optimization, which is applicable in pre-silicon verification and post-silicon validation. The framework relies on the data learning methodologies that automate most of the learning process for acquiring the additional missing knowledge during the test generation. Then, the learned knowledge is applied in a novel test selection approach for test content optimization.;We have successfully applied this framework in both pre-silicon and post-silicon applications. Experimental results on MIPS and OpenSparc T2 processors have demonstrated the effectiveness of this approach.