An Automatic Approach to Evaluate Assertions' Quality Based on Data-Mining Metrics

Abstract

The effectiveness of Assertion-Based Verification (ABV) depends on the quality of assertions. Assertions can be manually or automatically generated. In both cases assertion generation is error prone and needs high expertise. Moreover, the number of generated assertions is generally too large. Thus, assertion qualification is necessary to evaluate the quality of generated assertions to assist verification engineers to select only the highest quality assertions for systems' verification. Most of the current works for assertion qualification are based on fault injection analysis, which requires long simulation time. To fill in the gap, this work proposes a new automatic data mining-based approach for assertions already defined for a design, which in contrast to the state-of-the-art can evaluate assertions' quality precisely within a very short simulation time. Experimental results support the benefit of the proposed methodology.