Towards Autonomous Accelerator Design: FPGA Accelerator Generation with SECDA

arXiv·medium signal

Sharma, Fu, and Haris extend the SECDA methodology toward autonomous generation of FPGA-based accelerators for modern AI workloads, navigating a large and complex hardware design space. It is an early step toward agents that help design the compute substrate they run on, compressing a traditionally expert-heavy hardware design loop. Of interest to anyone watching agentic design move from software into silicon.