Skip to main content

Predicting Bugs, Validating Bugs

Conference: DVCLUB Europe | AI/ML in Verification
Speaker: Daniel Hansson, Cadence
Speaker Title: Predicting Bugs, Validating Bugs

Using ML-trained models it is possible to predict which commits are most likely to contain bugs. This can be done without running any simulations. By focusing on the high-risk commits the debug process becomes faster. However, it is important to validate the conclusions to separate good predictions from bad predictions. This presentation will show how that is done. With this solid foundation you can build upon this to create a larger tool chain for debugging.

Key Points:

  • ML-trained models can predict high-risk commits without simulation.
  • However, the predictions must be validated to be useful
  • Validated results is a solid foundation for a larger tool chain for debugging
Speaker Biography:

Daniel Hansson has 25 years of experience as ASIC designer, project manager and tool developer at Ericsson, Texas Instruments, ARC (now part of Synopsys), Verifyter and Cadence. Daniel has spent the last 10+ years working on automatic debug and ML-based bug prediction with 2 patents granted in the field.


DVCLUB Europe is made possible through the generosity of our sponsors.

Close Menu