ML for Verification: Does it Work or Doesn’t It?
TimeTuesday, July 12th1:30pm - 3pm PDT
Location2010, Level 2
DescriptionThe ML contribution to verification today is limited to point solutions for internal engines. The goal of the panel is to discuss why there are no major breakthrough in the use of ML in functional verification and what can be done to change this picture.
AI in general and ML specifically are hot research topics in the EDA research community. The past few years saw many tools, technologies, and methodologies that utilize the power of AI and ML influence EDA in many areas, such as place and route, timing closure, and synthesis.
One area that is almost missing in this picture is functional verification. On the face of it, functional verification is a natural candidate for massive use of ML technologies. The verification process produces large amount of data that can be used to train ML-based solutions. Moreover, the highly automated verification process contains several manual bottlenecks that can benefit from ML. Examples of such bottlenecks are coverage closure and debug/triage.