../
Meeting with Tommy
- Certified designs from open core.
- Assertion bench has too much
- They have tb too. I was to generate traces.
Coverage guided reduction in the number of assertions “good” assertions
Meeting Notes
- Quantifying the assertion quality is the most important thing
- This translates across projects. So would be a great contribution
- Move to even smaller designs. Things like a 1-bit adder, flip-flop etc… So that you can reason effectively
- Objectively prove that coverage is the most important thing. #(assertions) is meaningless without assertions
- Rerun HARM and Goldmine
- Measure how long lean takes to run stuff. See if you can make a comparison on the time front
- Coverage guided assertion
- Can you construct a minset of assertions?
- Given a large set of assertions, measure the coverage first. Then remove an assertion and measure coverage. If the coverage doesn’t reduce then throw out the assertion
- Can you detect tautological or trivially true assertions?
- How to deal with semantically duplicate assertions ?
- Develop small tools that claude can run to improve mining
- Example would be a hierarchy tool
- A tool that can return the wires of the parent
- Take a look at claude code’s source and see if there are any inspiration that you can take from that