Wally Rhines, CEO Emeritus of Mentor, a Siemens business, delivered a bullish prognosis for the semiconductor and EDA sectors in a talk at the beginning of the Design Automation Conference (DAC) in Las Vegas this week (June 3, 2019), insisting that the strong growth of the past two years is not a flash in the pan for an otherwise maturing industry.
“We’ve recently had a dramatic increase in growth of the semiconductor industry,” Rhines said. “It had been coming off a period of less than 3 percent compound annual growth. And in 2015 and 2016 it was less than one half of a percent. It seemed the semiconductor industry was no longer a growth industry, period. Then, all of a sudden, it changed.”
Rhines pointed to an increase in semiconductor sales of 22 percent in 2017 and 15.5 percent in 2018. “We were off to the races. And yet when you look at the forecasts coming out now they are down as low as minus 6.5 percent. Was that growth just a two-year phenomenon? Or are there some underlying differences?” he asked rhetorically.
Memory price movements
The strong rise in sales over the past two years was driven by memory, he said, and mostly by higher average selling prices. “It wasn’t because people were suddenly using more memory. It was a price phenomenon.”
It was a growth spurt that also came with a cost.
“We know from history we have to obey the learning curve. Whenever you spend time above the line you will have to pay for it later. That’s what’s really driving these forecasts of a negative year in 2019 and probably 2020 as well,” Rhines claimed.
Despite these ups and downs, Rhines argued the industry is in rude health. One reason is the steady long-term trend to increase the semiconductor content of end systems. “We are definitely moving above the 16 percent range in terms of semiconductor content in systems,” Rhines said. Automotive systems represent the most dramatic example of that, due to the twin development paths of electric drivetrains and advanced driver assistance system (ADAS) units. But across the board, systems companies are getting into chip design.
“Chips are being designed by appliance companies like LG, as well as Tesla and Bosch. EDA is now in a tremendous boom because the customer base has moved to new people,” Rhines said.
He added that if you look at semiconductor wafer shipments, the proportion attributable to systems companies has risen over the past few years to 17 percent while the share of production for fabless companies has fallen from a peak of 80 per cent in 2012 to two-thirds this year. IDMs have also registered a drop in share. The current 16 percent share is a little smaller than it was when the fabless companies peaked.
Hidden systems value
The trouble for forecasters, Rhines explained, is that systems companies’ sales are not readily reflected in market numbers because all analysts can see is the raw wafer output and not the imputed market value. “They tend to hide how fast the industry is growing,” he said.
A second positive note for the semiconductor industry and EDA suppliers lies in a jump in venture funding for startups working in chip design, many of them beneficiaries of Chinese government and private equity. The main driver for this lies in the push by many companies to enter the market for self-driving vehicle technology, and the wider AI and machine-learning markets.
“Domain-specific accelerators are driving a new wave of growth. Traditional von Neumann architectures have dominated [computer] architecture for decades. If you want to do pattern recognition, von Neumann is not very effective. We have new types of design coming in and the investment in semiconductor companies has done a major reversal in trend. All those VCs who only wanted to invest in social networking a few years ago now want to invest in semiconductors,” Rhines said. He did sound a note of caution, though, about the longevity of the Chinese funds and whether they would follow through in future years, having concentrated to date on early funding rounds for a large number of startups.
Given the increase in engineering charges for leading-edge chip design, Rhines asked: “How can startups afford the design of new chips?” Part of the answer appears to lie in a change in the level of abstraction of design for these datapath-focused chips. “They don’t have a lot of legacy. They are doing data paths using high-level synthesis. They are starting with C++ and automatically synthesizing RTL. And they can easily move to different nodes. keeping the same source code.”