DAC 2020: The State of the Industry

Monday was the first day of DAC, the Design Automation Conference. Of course, like everything else it was fully virtual. Having been in EDA for nearly 40 years, my usual experience at DAC was not being able to walk more than about ten feet without running into someone I already know. It's not quite the same experience online. It was even less of my normal experience since I came down ill for a few days so I didn't attend many events live. However, I've been working my way through some of them. In that sense, it is better than if it was live since I'd probably just have taken myself off home to recover.

As I said in my DAC preview post, I've seen earlier versions of many of these talks, and I've already covered them in Breakfast Bytes over the last year or so. For example, this is the third time I've seen Cerebras present their amazing wafer-scale "chip".

A couple of presentations that I did attend were on the industry as a whole and how well it has held up in the current WFH era. One was by Wally Rhines, who is now CEO of Cornami, but was CEO of Mentor for many years, and before that ran Texas Instruments' semiconductor division.

The chart above shows how the semiconductor market had negative growth in 2018. But this was entirely due to the softening of memory prices. Logic actually grew.

Volumes are going up, too, with 10 new 300mm (12") fabs scheduled to open this year. So revenue will go up, too. Note that the chart above shows the change year-to-year and so you need to add all the bards up to get the capacity. One of the most amazing constants in the semiconductor industry is that for 60 years semiconductor revenue per unit area has been close to $32.98 per square inch, despite new process generations, new equipment, and orders of magnitude increase in the size of the market, the fabless transition, and so on. When I started in the industry in the early 1980s, it was 3-micron HMOS with two layers of metal. That was before we had standard cells so I'm not quite sure of what the transistors per square inch were, but around 20,000, I'm guessing. Now we are in hundreds of millions. And it costs the same: $33.

Other things Wally covered:

Automotive has been the fastest-growing segment of the semiconductor market and is now up to nearly 10%. But there are over 500 companies who have announced they will build electric cars or light trucks, and 277 autonomous drive programs. Unit volumes have actually been declining before we got to where we are today, but that is outweighed by semiconductor content going up. It's actually the same story in mobile, with unit volumes declining but semiconductor content up due to 5G.

EDA remains strong since semiconductor companies have not been laying engineers off. They are just too difficult to hire. But EDA is actually growing driven by machine learning. And not just end-markets, we are leveraging it inside our flows. For example, Cadence's Digital Full Flow with Machine Learning that Wally showed.

Another big trend is system companies designing their own silicon. This is the fastest-growing segment of foundry customers, growing from essentially nothing in 2011 to 19% today, a CAGR of 35% (compared to CAGR of 3.1% for fabless and CAGR of 2.3% for IDM).

Wally wrapped up with a short discussion on Fully Homomorphic Encryption (or FHE). This is Wally's new project as he is CEO of Cornami designing chips (with quite a lot of Cadence tools) to implement FHE. For more on that, see my post Cornami and Trusted Data.

DATE 2021

Let's quickly change conferences, since I'm the publicity chair for DATE 2021 next February. So let me point out that the call for papers for DATE 2021 is now out. Submissions have to be in by September 14 (mostly). Full details are on the DATE website. It still has not been finally decided exactly what form the conference will take. You can see my preview post at Make a DATE for the Alps Next Ski Season.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.