Recently, the SOI Consortium held its annual Silicon Valley Symposium. I was only able to attend in the afternoon since I spend the morning at the ESD Alliance workshop on digital marketing (for more on that, see my post Digital Marketing—With No Hands on the Wheel, but there will also be more posts to come).
Also, between the SOI Symposium and today, Samsung held their own Foundry Forum. They reported on the roadmaps for their FD-SOI process. You can read about it in my blog post from the forum.
At one of the first sessions I attended, Dan Hutcheson of VLSI Research updated the work that he presented at the same event a couple of years ago. In 2016 he ran a survey across the industry on attitudes to FD-SOI. As he described it: I talked to a wide range; FinFET bigots, SOI bigots, shades in-between.
He ran the same survey again this year to see what had changed. He met with decision makers and influencers representing 51% of the IC market and 59% of the IP market, across a wide swath of product categories. About half of them were the same individuals as in 2016. See the pie chart at the top left of this post for a more detailed breakdown.
This time, the industry was less divided, with far fewer bigots. It's no longer a stark choice of FD-SOI or FinFET, more that many people are using both or would consider both. The market now sees FD-SOI and FinFET as complementary technologies. A few people say "it's over and fins won" but most people "see fins and FD as complementary technologies."
In 2016 the biggest issues that people had with FD-SOI was:
About 3/4 of people would consider running dual roadmaps, and only 1/4 would not. The overwhelming reason for this was the choice depends on the application and the market size. Some people won't even consider FinFET since they don't need the performance or have to move their IP.
Dan said that choosing between FD and fins is like choosing where to go out for dinner. French or Italian? Everyone likes Italian, but French is for special occasions. FinFETs are used for the highest volumes that need performance, integration, and density. FD-SOI is better for complex high-mixed-signal SoCs. Customers views have matured and they see both as useful, depending on the application, especially the fact that more and more RF integration is happening and this can be done on FD but not really on FinFET. As a result, FD-SOI is seen as uniquely positioned for 5G/mmwave. Fin parasitics are the killer here: "you can't do on-chip 5G with FinFETs, the parasitics are too high."
As a result, FD-SOI's advantages in RF has tipped the scales in its favor. Power is the other advantage, but Dan reckons that body-biasing was oversold and caused a negative perception of complexity. The chart above shows the reasons for picking FD-SOI, with RF and RF/mixed-signal integration being the two biggest reasons.
Many people perceive FD-SOI as the "IoT Technology" but IoT doesn't really have a market definition. However, it does have certain characteristics: high mix of analog/digital/RF, tight power budgets, high thermal range (especially automotive), and some thermally sensitive applications (such as image sensor processors, where the image sensor sits on top and is affected by heat).
FD-SOI equivalent nodes run about 2 years behind FinFET. That allows FD-SOI to level fin learning to achieve lower cost, However, there are things you can do with FD-SOI that FinFET cannot do (primarily RF and high precision analog) meaning that there is a fragmented market with a complex decision matrix. The above chart shows the roadmaps for FinFET and FD-SOI by technology and year. The fact that GF has a roadmap to 12nm and presumably beyond gives certainty for the future.
Dan had some advice, based on the interviews, for what needs to happen going forward:
Sign up for Sunday Brunch, the weekly Breakfast Bytes email.