The A3 Workshop on Agentic Approaches to Architecture
The A3 workshop explores the paradigm shift toward reasoning-capable AI agents in computer architecture. We invite submissions that advance the frontier of agentic methodologies throughout the hardware/software co-design ecosystem.
The A3 Workshop invites researchers to submit short papers, position pieces, and "blue-sky" ideas that use AI agents to solve hardware problems. If you are applying agentic techniques to computer architecture, VLSI, HW/SW co-design, or anywhere else across the hardware/software stack, we want to hear from you.
A3 is a forum for research that puts agents to work on hardware/software co-design problems. How can agents produce better architectures, chips, and systems? Submissions may explore topics such as:
Results from deploying agents on co-design problems: architecture and systems design exploration, RTL generation and PPA optimization, physical design, verification, and HW/SW co-design.
Building and analyzing techniques that make agents better at co-design tasks: integrating agents into design flows and EDA tools, domain-specific benchmarks and evaluation, and simulation environments for agentic exploration.
HW/SW co-design and verification tools today are built for humans. Agents doing hardware design may benefit greatly from changes to these tools, such as more streamlined interfaces and high-speed evaluators.
Submissions are encouraged on agentic themes, including but not limited to the following areas.
Agents that discover novel solutions for CPUs, GPUs, and accelerator architectures; autonomous RTL design and PPA optimization; physical design and verification.
Agents that cut across the system stack, co-designing hardware together with compilers, operating systems, and applications.
Agents that tune and reconfigure running systems online using runtime feedback.
The scaffolding that lets agents do hardware/software co-design well: benchmarks and metrics for agentic hardware-design tasks; simulation techniques for facilitating agentic design exploration; debugging agent trajectories in design flows.
We also welcome submissions on the broader set of agentic themes across the hardware/software stack, including autonomous system design, OS/hardware co-design, autonomous EDA & RTL generation, dynamic adaptation, evaluation & benchmarking, simulation environments, reproducibility & open-source, and verification.
Paper format guidelines are below:
| Length | Format |
|---|---|
| 2–4 pages (excluding references) | Double-column ACM/IEEE style |
Other requirements will broadly follow those of conferences like ASPLOS, ISCA, and MICRO. Use of AI-assistance in paper writing is welcomed, but must be acknowledged at the end of the paper. Human authors are responsible for the entire contents, tone, and quality of their paper.
Submission link will be available soon!
Have questions? Reach the organizers at a3-workshop@eecs.berkeley.edu.
Every paper is reviewed by human experts on the program committee, who use their own evaluation of the paper and their judgment to recommend feedback and a decision. To support them, the workshop track embraces an AI-Native Review Process, a collaboration between human experts and AI reviewers that provides grounded, thorough, and thoughtful feedback. Each paper is also evaluated by AI agents with different personas, which do not pass judgement on acceptance or rejection. These AI-based reviews are supplied to the program committee to inform the experts' own evaluation.
Reach the organizers at a3-workshop@eecs.berkeley.edu.