Open-Source AI Loop Aims to Optimize RISC-V CPU Design

An independent developer launched an open-source research project in April aimed at exploring how artificial intelligence models can improve RISC-V CPU architecture. The project, called auto-arch-tournament, was posted to GitHub under the handle FeSens and has been tracking performance results on the HWEBench evaluation platform. The developer noted curiosity about whether AI models can meaningfully influence CPU design decisions, and the open-source loop provides a way for others to run similar experiments.

According to the post, the question driving the project remains active: how effectively can different AI models enhance a RISC-V CPU? The developer has been documenting outcomes on the HWEBench v2 benchmark, creating a public record of what works and what does not. The code and methodology are available for anyone interested in running the research loop from the repository.

Around the same time, OpenAI published a blog post about advancing chip design with their models. The independent developer acknowledged the coincidence, noting it is “fun to see OpenAI exploring a similar question in today’s chip design post.” The timing highlights how quickly AI-assisted hardware design is moving into the mainstream, and the open-source effort provides an earlier point of comparison.

The auto-arch-tournament project sets up a competitive framework where multiple AI models can be pitted against each other in CPU optimization tasks. Each run generates data that gets recorded on the HWEBench platform, allowing trends to emerge over time. The developer described the process as a way to gather reproducible results rather than relying on isolated anecdotes.

Hardware architecture and AI capability intersect in ways that are still being mapped. By keeping the loop open-source, the creator hopes to lower the barrier for other researchers to participate. The repository includes the necessary scripts to get started, and the HWEBench site hosts the accumulated data from previous runs. Anyone with access can pull the code, select which models to test, and contribute new results to the growing dataset.

The post also touches on the broader shift toward using large models for engineering tasks. Chip design has long required deep domain expertise, but the availability of capable AI systems opens the possibility of automating or augmenting certain steps. Whether the auto-arch-tournament approach will prove practical for production-grade designs remains to be seen, but the experiment contributes a data point in a space where public benchmarks are still rare.

Developers and researchers looking to explore the boundary between AI assistance and hardware design now have a concrete starting point. The open-source nature means the barrier to entry is low, and the public tracking on HWEBench means results are not lost to closed systems. The developer’s April head start over the more widely noted OpenAI post shows that independent work in this area predates the recent surge of attention.