NVIDIA on Monday released new research on the use of AI-based chatbots in the semiconductor design process.
Because modern chips are made up of circuits made up of tens of billions of transistors, one of the technology industry's most difficult tasks is understanding how to arrange them on a piece of silicon, which can take thousands of engineers up to two years.
NVIDIA chipsets are among the most complex in the industry and are essential to powering technologies such as popular chatbots (ChatGPT).
Nvidia on Monday unveiled a study that used so-called large language models, the technology behind chatbots, and augmented it with 30 years of data from its chip design archives. One of the first uses of this step is to leverage the company's long history to answer questions.
“It turns out that a lot of our senior designers spend a lot of time answering questions from junior designers,” Bill Daly, chief scientist at Nvidia, told Reuters. “So, it has become a norm for inexperienced designers to ask a chatbot question. This can save a lot of time for experienced designers.
The study concluded that by adding large amounts of data based on a company's experience, relatively simple chatbots can become more accurate than advanced chatbots, which Nvidia says can help control system costs.
Another advantage of the company is the use of artificial intelligence to create code. Engineers spend a lot of their time finding parts of the chip that don't work and using testing tools to find out why, Daily said.
To perform this test, the AI system can quickly write code; This is called the "script" that runs the tool.
“The goal is not to automate processes or replace people, but to give people superpowers to make them more productive,” Dailey said.