A research team from New York State University (NYU) has managed to develop a semiconductor chip without use hardware description language (HDL). Using plain English and the terms and examples it contains, the team was able to identify and describe a part of a semiconductor processor – typically using a specialized HDL language such as Verilog.
The chip developed by the research team using ChatGPT was not a full-fledged processor – it had nothing to do with an Intel or AMD processor. The semiconductor created is a processor element: the logic responsible for creating a new 8-bit microprocessor architecture batteries – Registers (memory) in which the results of intermediate calculations are stored until the main calculation is completed.
Teams typically work in stages to bring the chip to design and production. One such step is to translate the English language describing the chip and its capabilities into a selected HDL that represents the actual geometry, density and general arrangement of various elements within the chip required for the actual implementation etching Processor.
In this case, ChatGPT allowed engineers to skip the HDL phase. According to the researchers, they anticipate there will be less human error in the HDL translation process, which will improve productivity, reduce chip development time and time-to-market, and allow for more creative designs.
Researchers used commercial and open-source Large Language Models (LLMs) to work on eight hardware design cases, working from plain English text to the Verilog equivalent (HDL) in real-world interaction between engineers and the LLM. “The result of this study was, in our opinion, the first HDL, generated entirely by AI to make a physical chipsaid Dr. Hammond Pierce, associate professor at NYU Tandon and a member of the research team.
“Some AI models, such as OpenAI’s ChatGPT and Google’s Bard, can generate code in different programming languages, but their application to hardware design has not yet been extensively explored. This study shows that AI can also be useful in device manufacturing, especially when used in conversation to improve designs.Pierce added.
Of even greater concern is the desire to eliminate the need for free HDL ownership for chip designers. Because it is an extremely specialized and complex field, this is a relatively rare skill that is very difficult to master. Of course, automating part of this process would be a boon as it would speed up the work of existing professionals while onboarding and training new professionals. However, there is a risk that this capability will become entirely dependent on software that relies on power and, in the case of ChatGPT, a connection to a server.