||Verification Futures 2022 (click here to see full programme)
||Using Neural Networks to Select Test Cases for Coverage Closure
||We compare two recently published methods for using neural networks to select test cases in order to close coverage more quickly. Both methods rely on the identification of tests that contrast with, or are novel with respect to, the tests that have so far been simulated. Results of the use of the two methods on the same industrial-strength data set are reported and show that both approaches can lead to considerable savings in time to close coverage. This indicates that the novelty of a test can be used as a proxy for the likelihood that a test hits new coverage. Since determining novelty using machine learning techniques uses only open source licenses and is typically much quicker than simulation this has potential to reduce cost and schedules for projects for which coverage closure is time-consuming.
||Tim is a Senior Principal Verification Engineer at Infineon Technologies, based in Bristol. His research focus is improving the efficiency of the verification of complex IPs, subsystems and SoCs including by the use of machine learning. He is also the Global Verification Manager for Infineon’s current family of Aurix3G microcontrollers and is a Visiting Industrial fellow at the University of Bristol.
- Comparison of two machine learning techniques for novel test selection.
- Results of using the two methods to speed coverage closure for a significant data set indicate novelty is a reasonable proxy for hitting new coverage.
- Potential for real-life savings in cost and schedule for projects for which coverage closure is time-consuming.