The 6th annual Silicon Catalyst Semiconductor Industry Forum occurred on Thursday, November 9th,2023, in Menlo Park, Calif. A panel of semiconductor industry veterans addressed the subject, “How will AI affect the semiconductor manufacturing industry?” The group discussed how and when AI would change the way chips are made and how odd the upcoming “AI wonderland” will turn out to be.
AI Wonderland – from Silicon Catalyst
Ivo Bolsens, AMD senior vice president, said, “We are entering an era of electronic design creation,” He projected that artificial intelligence will soon be able to fill out most of a chip’s architecture from high-level specs. However, he believes AI will only be able to cross the last mile soon.
Bolsens gave an example from a recent trip to Austin. “I fly to Austin, take a car to the office’s parking lot, then I walk into the building,” he explained. “AI is like flying; it gets you very close to where you must be very quickly. From there, you must employ more traditional methods of operation. That is the opportunity AI provides chip designers. It just won’t reach the parking lot to the office.”
Deirdre Hanford, chief security officer at Synopsys’, stated that her organization is already “wrapping an AI harness around our [design] tools.” “People are currently figuring out where AI can be deployed in the chip design process,” she said.
Moshe Gavrielov, former Xilinx CEO and current member of TSMC’s board of directors, predicted that AI would eventually be used to construct standard cell libraries. Building such libraries, he says, is “very complex,” with “a lot of corner cases. Computers can create these libraries with fewer people, higher quality, and density.”
AI and Analog Circuit Design was another topic of the day. Gavrielov also emphasized AI’s ability to work with analog circuits. “AI can take analog libraries and move them from generation to generation [of technology] automatically; this used to be incredibly time-consuming, error-prone, and difficult.”
But how soon will this change occur? According to Gavrielov, “Benefits to the end customer will be obvious in a short time frame; there is a threshold, and once we cross that threshold, the dam will burst, and it will be amazing to see the revolution in [chip] design that will happen.”
Gavrielov recalled the last major revolution in chip design, the transfer to electronic design automation (EDA). That transformation, he claims, took more than 30 years. “I think the transition that AI will bring will happen in a third to a fifth of the time and will have a much bigger impact,” he said. “In five years, in less than 10 years, design will be done differently than today.”
But, as Hanford assures, chip designers should not be concerned. She said, “We will keep automating as an industry, and it will accelerate, but people, I don’t think our industry is facing the same threat as paralegals; we will just move up in abstraction.”
Moderator David French, Silicon Catalyst board member and SigmaSense CEO asked participants to consider AI’s environmental impact beyond its effects on design. The massive processing needed to build AI models worried him about energy and carbon emissions. We have exciting answers for that. AMD vice president Bolsens said, “We have taken current architectures and extended from them to respond to the emerging needs of AI,” Thus, “they are being used inefficiently, typically only exploiting 10 or 20 percent of the compute capabilities of the hardware [and] wasting a lot of power.”
“These are early, sloppy innings,” Gavrielov retorted. Bolsens believes much is possible. “The good thing about AI is that it is a narrow class of problems, in terms of characteristics of the compute it requires,” he said further. “So new compute architectures will arise that take advantage of that to be more efficient.”
“And AI is about data, not just compute,” Bolsens said. “Most of the power used today is in moving data to compute. To reduce data transit power, you’ll examine technologies and memory architectures that connect computation to data.”
The research on AI was also discussed on this occasion. The amount of processing power necessary to do AI research has created a divide between large enterprises, universities, and tiny startups. “A startup doesn’t have enough compute hours” to perform AI research, Hanford said, adding that “university computer resources are frequently insufficient as well.”
She said, “If you want to do research in this space, you have to go to Meta or Microsoft or have a very well-funded startup,” adding, “We should make sure that crazy research continues to be done at universities,” she said, “which will require the creation of something like a national AI resource.”
However, according to Bolsens, the tendency toward open-source projects will be helpful. “That allows people to leverage work from others in the field.”
In the end, Hanford warned the gathering of chip-company executives, entrepreneurs, and investors, “Every single group has to think about how AI will disrupt their mission or make them more productive; AI should impact every business operation, large or small. You’re in troauble if you don’t think of it as a paradigm break or something that can put you out of business.”
