GPT-4 generative AI is rapidly expanding its application boundaries, chatting, writing articles, programming, and drawing, and now generative AI can already make chips.
Researchers at New York University's Tandon School of Engineering communicated with GPT-4 to realize the design of microprocessor chips by AI, and then the research team sent the design to the manufacturer for manufacturing and successfully taped it out.
The development of any type of hardware starts with describing what functions the hardware should develop in normal language. Specially trained engineers will write into a hardware language (HDL), such as Verilog, to create the actual circuit elements that allow the hardware to perform its tasks.
GPT-4 was able to generate viable Verilog through back-and-forth dialogue, then send the benchmark and processor to the Skywater 130nm shuttle for tapeout.
This research is very important and has the advantage of reducing human error in the HDL conversion process, thereby increasing productivity and shortening design and time-to-market, and allowing more creative designs, semiconductor chip design threshold will be lowered, even without this People with technical skills in the field can also participate, and the innovation and forward-looking of this research have new breakthroughs.
This research resulted in the first fully AI-generated HDL for making physical chips. This research shows that AI can also benefit hardware manufacturing, especially when it is used for conversations, where you can refine hardware designs through back and forth conversations, an important direction for the future of chip manufacturing.
The research paper has been published on the Arxiv platform , the paper is entitled Chip-Chat: Challenges and Opportunities in Conversational Hardware Design