Novel test analysis to improve structural coverage — A commercial experiment

Novel tests are important in simulation-based functional verification because they provide coverage of difficult-to-verify corners. In this work, we did an experimental study on how to learn from the novel tests to help improve structural coverage in functional verification. A feature-based learning methodology is proposed to diagnose the reasons why a novel test contributes to coverage of a target block. The extracted rules can be used as constraints to improve test generation. Our experiments were conducted based on a simulation environment for verifying a commercial dual-thread low-power processor core. In one case, we improved the toggle coverage of a block in load store unit to 100%, which was otherwise difficult without learning.