U.S.-China relations, chip outlook, circling back to EDA’s early days.
Trade restrictions against China were a regular topic of discussion at both DAC and Semicon West this week.
Five Chinese startups exhibited prominently at this week’s Design Automation Conference, in the wake of increasingly restrictive trade regulations that limit the sale of U.S. and European EDA tools used to develop advanced semiconductors.
Most attendees interviewed said privately they had expected this development because China cannot continue developing complex chips without EDA tools. The five startups — EasyLogic, Phlexing, Rigoron Software, Semitronics, and X-Epic — collectively offer a range of tools, including functional and formal verification, emulation, debug, ECO management, sign-off, and test. How competitive they become remains to be seen, but the presence of these companies at the biggest conference for EDA and IP certainly was noticed.
On the Semicon West side, analysts were mostly bullish on the potential for language models like ChatGPT in particular, and AI in general, to help support growth in the wafer processing equipment market, though the initial impact will be minimal. Stacy Rasgon, managing director and senior analyst at Bernstein, Mark Lipacis, managing director at Jefferies, Lori Keith, director of research at Parnassus Investments, and Ulrike Hoffman-Buchardi, partner at Tudor Investments discussed progress in data privacy, the impact of the China-U.S. trade war, and sustainability policy at Semicon West today. Jane Lee, chip reporter for Reuters moderated the discussion.
Looking at present chip and wafer processing equipment markets, Rasgon observed that the industry is experiencing the greatest memory downturn since the financial crisis and the burst of the tech bubble. Nonetheless, panelists expect the market to rebound as it has in the past, largely driven by the proliferation of semiconductors to multiple electronics markets that will sustain long-term growth. Lipacis noted that another driver that makes this cycle different is the industry growth in trailing-edge manufacturing equipment (around 90nm), which is happening for the first time.
Adobe and Microsoft are making strides in data privacy policy according to Parnessus Investments’ Keith. She noted that sustainability models have become important to investors that associate risk with poor climate policy. Actual reductions in carbon emissions are expected by investors as well. Productivity advances will be one key result of implementation of AI models, largely a result of parallel processing capability according to Hoffmann-Burchardi.
The U.S. ban on capital equipment that enables 14nm and beyond seems to be having little impact on NVIDIA’s profitability, according to Rasgon. He estimated the company took only a 6% hit on sales due to the switch to lower performing GPU A800 and H800. Hoffmann-Burchardi commented that investors are weighing the risk associated with exposure to the China market. Unfortunately, investors should not assume there is no black market for wafer processing equipment according to Berstein’s Rasgon. In the short term, Chinese chipmakers have been stockpiling wafer processing equipment. The industry has yet to realize the overall impact of the ban.
Fig. 1: AI’s influence on growth, China-U.S. trade war, and the importance of climate policy were discussed among (right to left) Ulrike Hoffman-Buchardi of Tudor Investments, Lori Keith of Parnassus Investments, Mark Lipacis of Jefferies, Stacy Rasgon of Berstein, and moderator Jane Lee of Reuters. Source: Semiconductor Engineering/Laura Peters
Also at DAC, EDA luminaries Joe Costello and Wally Rhines recounted what they saw as the biggest breakthrough in EDA technology. “EDA is made up of breakthroughs in so many different places, and I finally decided the biggest breakthrough in technology was Joe Costello,” said Rhines. “Until Joe came along, you started an EDA company, you grew it, and you competed. But Joe changed the game. He came in and started acquiring companies, sometimes for exorbitant multiples, because the rest of us didn’t know they had already built a foundation and tools that just hadn’t taken off. What he did was foster what became a sort of farm club, almost like the pharmaceutical industry where big pharma companies don’t develop anything. They just acquire it from the innovative entrepreneurs who develop it. And that’s what he did for EDA, so that we went through the next 25 years, just picking up the innovations that were created in the industry. He doesn’t know that was his contribution, but that’s what I think was the biggest breakthrough.”
For Costello, the biggest breakthrough was the ASIC design methodology. “When I started, people were doing schematic diagrams, and there was hardly any simulation. You were doing hand layout. People were doing Rubylith on top of light tables. That was really the way people were doing everything. And to move to an HDL-based thing where you were essentially programming the hardware, and then you had automated tools from synthesis to place and route, and simulation like Verilog — that was what triggered the gigantic breakthrough.”
Left to right, ARM’s Brian Fuller (moderator), Joe Costello (Metrics, Kwikbit, Arrikto, Acromove), and Wally Rhines (Cornami). Source: Semiconductor Engineering/Ann Mutschler
Interestingly, Costello actually recruited Rhines to work at Cadence before he left TI to lead Mentor.
Rhines and Costello also reflected on what they consider their greatest individual contributions.
“Mentor, in its early days, had this great relationship with a company called ECAD,” Rhines said. “They had developed a physical verification tool called Dracula, so Mentor OEM’d it and created most of the business they had. It was really a good contributor to the business. Then Joe came along and acquired the company, and we used to joke that this was a marriage made in heaven. SDA was a company that had no products but great marketing, and ECAD had a great product and no marketing. And so the marriage worked [and created Cadence]. That left Mentor out in the cold. So Mentor came to TI, and we had pretty good physical verification. We made a deal that they could OEM our physical verification to replace Dracula, and they called it Checkmate. A guy named Lawrence Broad was in charge, and he collected hundreds of designs that he ran through Checkmate. It was only about 20% market share, Dracula remained about 80%. So when I got [to Mentor], the contract with TI was expiring. They said, ‘We want to develop the next generation.’ I had no idea how talented Lawrence was. I said, ‘OK, I guess we have to do this,’ and I had actually signed the license to OEM it. I went back to TI, and they said, ‘Go take a hike, we’re not interested.’ So we bought out their rights, and funded Lawrence to go off and develop Calibre just to fill in a hole, but it turned out to be enormously valuable over time.”
Costello, meanwhile, took a whole different approach at Cadence. “We essentially got mergers and acquisitions to be a real thing. People now don’t think about it, but people didn’t do mergers and acquisitions at that time anywhere in technology. Cisco did it later, after we did it in the EDA industry. And so doing the mergers and acquisitions changed the face of the industry in a very positive way, because it allowed you to build a complete suite of tools that would be virtually impossible in a reasonable, relevant timeframe. But going for the full software approach at the time, when I took over Cadence, all the leading companies were doing turnkey systems. So Mentor was OEMing its hardware, other people were building their hardware. The nastiest fight in the history of Cadence was when I said, ‘We’re not going to do hardware. That’s a stupid idea.’ And the fight that we had over that was vicious. Now you look back and it seems like such a stupid thing. How we could even argue that point? But that’s the way the religion at the time was. You’re going to OEM and sell the hardware to double or triple your revenue, or whatever it was, and I said, ‘It doesn’t make any sense. You don’t add any value in the hardware side. Don’t do things where you don’t add value.’ Another thing that changed was when we changed the licensing scheme from a perpetual license to an annual license. Those are the business things that I think made a difference.”
Related
DAC/SEMICON West Addresses Top Issues, Trends For Chips
The chip industry returns in full force after a four-year gap.
We did the same thing in photomasks that Joe Costello did in EDA, although it was less about innovation and more about scale. In about ten years, the industry went from 80/20 captive/merchant to 20/80. Later, as mask technology became critical to extending optical lithography, it began returning to captive.