../

Minutes

Impressions of RV

  • Software running somewhere that we want to enforce certain behavior at runtime
    • So runtime in the general sense rather than pertaining to hardware
  • RV on software behavior that prevent “bad” actions
    • This is in contrast to what we want usually do in RV where the assertions are meant to be the stuff we want the system to possess.
  • Augmenting hardware to act as a shield
  • Monitoring AI agents was the hot topic

CVA6 Updates

  • Multiple templates in the config
  • How are we selecting templates
    • Based on bug reports on CVA6 and ask an LLM what the template should be
    • These templates try to generalize over the existing bugs
    • Can we go the other way around? Go from the architecture down?
    • Arithmetic operators in Texada?
      • Can we do a post processing on the traces to calculate the interim values
    • The templates are then negative which is the opposite of what we were mining before
    • How does this relate to Periscope

BOOM Updates

- Still the same
- Mining still takes a long time
- G(A -> B)
- G(A)? Haven't tried this yet

Mine

  • If LLMs are bad at Verilog, should we be augmenting them with skills?
  • Does running the faq_extraction under the agentic setting change things?