../

Meeting with Kevin

Minutes

  • Yasas has started work on the BOOM core
  • Tommy’s email about security bugs
    • HACK@DAC has datasets with “golden” designs and then with bugs
    • Can we use this for our benefit in some way?
    • Periscope only tells us whether an assertion is consistent or not
    • It doesn’t tell us anything about how good it is
    • 20260825T131153-hack_at_dac_and_periscope
  • Elong’s work on security leaks
    • Is ours related to that?
    • Are there parallels that we can draw?
    • Detecting leakage bugs may be too ambitious
  • To use CVA6 to hunt for bugs requires us to have some sort of spec? I think bug hunting implicitly implies the existence of a spec.

LLM work

  • How does asking an LLM to include the module header impact the results?
    • I need to concretely prove that this is a thing
  • Start working on using specifications
  • Not using verilog ?
  • Just generate properties and then ask the LLM itself to generate the module declarations
    • Is this different than including it in the prompt
    • Make sure across designs/iterations

LLM + CVA6

  • CVA6 commit history and attempting to extract information from that

Thesis

  • Clarify exactly what your problem statement is
  • “Methodology to optimize LLMs through mostly prompts for any arbitrary task” may be too broad
    • It might also be very model specific
    • Can we pick something smaller?