Recently I decided to experiment a bit with AI assisted FPGA design.
I wanted to use the HyperRam device that is present on AMD's recently released SCU35P Evaluation Board but I needed some IP to talk it. After spending some time to read the datasheet for a HyperRam device, I felt this would be a good medium level complexity project to try AI on.
This is part of my overall interest in experimenting with some ideas for a Spartan UltraScale+ based drone flight controller.
The results are quite amazing. Using ChatGPT-Codex, in a few days, I had running simulations with meaningful test coverage.
I spent the time afterwards focusing on changes needed to actually have a working design after Vivado implementation.
This design is verified working on a SCU35P Evaluation Board running for about 5 days continuously. It should also work for other UltraScale+ devices. I also added 7-Series FPGA support. Passing in simulation, but not verified in hardware yet.