You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$18,000 left to reach the funding goal.
Our open kit showed that Codex passed 15/15 runs on 5 small Verilog designs. Next, test harder designs, add formal verification, and measure whether agents cheat. Everything will be public.
Test harder designs, add formal verification, and measure whether agents cheat. The work builds on the open benchmark github.com/missionfission/veriflow-rtl-benchmark, where Codex passed 15/15 runs on 5 small designs, checked against hidden reference tests, lint and synthesis. All code, results and datasets will be public.
Minimum $5,000 (compute only). Goal $18,000: compute and API $6k, 4 months of stipend $10k, tools $2k.
Khushal Sethi founded DragonX Systems, which builds agentic, GPU-native EDA (dragonx-systems.com). He was a Lead Design Engineer on Tensilica at Cadence and cofounded Kiwi AI, where he was Head of Engineering. He holds an MS from Stanford and a B.Tech from IIT Delhi (AIR 79). His publications include IEDM 2021, A-SSCC 2020, ISCAS 2019, NOCS 2019, BioCAS 2018 and DRAGON (arXiv 2022). He built the open benchmark github.com/missionfission/veriflow-rtl-benchmark. Google Scholar: scholar.google.com/citations?user=T6DyDNcAAAAJ
Agents may already saturate the easy designs, or formal checks may not scale. Either way, the negative results and the dataset still get published.
$50,000 USD from Indian government grants.