Experts at the table, part 1: How many process technologies do we need, and why new ones are being developed; FD-SOI as a platform for IoT, RF and MEMS/NEMS.
Semiconductor Engineering sat down to discuss finFETs, 22nm FD-SOI and how the how the market will segment over the next few years with Marie Semeria, CEO of Leti; Patrick Soheili, vice president of product management and corporate development at eSilicon; Paul Boudre, CEO of Soitec; and Subramani Kengeri, vice president of global design solutions at GlobalFoundries. What follows are excerpts of that conversation.
SE: Where is the pain being felt in manufacturing and design?
Soheili: The problem we see is the readiness of the ecosystem’s off-the-shelf IP. In the ASIC world, customers can’t afford to wait for creation and validation of IP for an end product they need. This all has to happen for something that is ready in order to get ASIC customers into the fray. This is different from big fabless companies, which may develop their own IP, but in the case of ASICs they rely on the ecosystem. And that ecosystem relies on the creation and validation of IP.
Kengeri: The biggest challenge is whether we can continue to provide the power and performance advantages that customers have gotten used to over the past 30 years. How can we provide that? That is really the industry’s challenge. Fundamentally, that became a bigger issue when bulk CMOS began slowing down from a device point of view, from slowing to leakage to things not really scaling. And then there’s the optical side. With single patterning, it was not capable of moving to the next level so we moved to double patterning. All of that came in around the 20nm node, where the device was problematic. The back end of line was a problem. We looked at multiple options, and for two years we did all the pathfinding and looked at what’s next. From a front-end point of view, both the FD-SOI and finFETs are fantastic candidates. They came very close when we had 20 different criteria, including ecosystem enablement and customer adoption concerns to the technical aspects from reliability to performance to scalability. We scored it. FD-SOI came pretty close to finFET, but the reason it didn’t beat finFET was because the first generation didn’t have performance comparable to finFET. So the question now is how we continue to offer the value to our customers. Otherwise there is no incentive to move and there won’t be any growth or innovation.
Semeria: Fully depleted SOI provides an opportunity to open a new era for the Internet of Things and multiple applications. It’s a good time because the technology is ready, customers are there, and there is a window into the market. For this technology, it is really a turning point where we have gone from research to product. We are fully supportive of the entire ecosystem, and we are fully committed to supporting the scale-up of 22nm FD-SOI.
Boudre: Clearly, the biggest challenge I see is to provide technology for new waves of growth that are coming. We are talking about electric cars, IoT, mobile applications and connectivity. At the end of the day the biggest challenges are performance compared to what it was five years ago and power consumption. We clearly need to bring a low-power technology to the market. But if you think about the applications we’re bringing to market, these are consumer applications. With the consumer market, a big concern is cost. It took a few years for the industry to recognize that finFET is not going to enable IoT, and it will not enable low-cost and medium-cost mobile applications.
SE: But all of this depends on how you define IoT, right? IoT is everything from the edge device all the way to the cloud, and everything connected in between.
Boudre: Agreed. And from a system perspective, finFET applications are really going to be driven by big data.
SE: On the edge node, we keep hearing that mainstream is 55nm or even higher. How did we suddenly get down to 22nm FD-SOI? And if everything else is going to be big data on finFETs, what’s in the middle?
Kengeri: It’s not one technology that fits the whole chain. On the higher end there will be a need for more processing, and that market can support the premium for finFETs. It will continue to be performance-driven at any cost, and that’s okay. As you come down toward the edge of the device, which is what we call the sensor side, cost and power are going to become critical. Performance is second. Customers will choose the optimum node, depending on where they are in that IoT system. 40nm is where customers will go for IoT, and there are many reasons for that—RF capability, non-volatile availability on that planar node, and it’s reasonably cost-effective. But if there was another node that could do better than 40nm with all these features and lower power, they’ll go for that. In the last few months, we’ve been talking to our customers and they’re very excited about 22nm FD-SOI.
Semeria: I agree, because fully depleted SOI at 28nm and 22nm can be considered a platform. So it’s compatible with RF function, photonics, MEMS, NEMs (nanoelectromechanical devices), so we are able to put new functions on this platform. It’s a way to utilize the whole system and to address many applications with low cost.
Kengeri: And the industry trend is moving from cost per transistor to cost per function. The more you can integrate on a single node, the better it’s going to be.
Soheili: The market is far more segmented than that. There is a significant place for 55nm and 40nm. There is NVM, BCD (bipolar CMOS DMOS), RF that haven’t improved on certain technologies, and there are applications that from a cost or size or power point of view will drive one versus the other. Will 22nm or 28nm FD-SOI be successful? Absolutely, because they have specific advantages in their segment. How big that will be depends on how ready the manufacturability will be at the right time. It has to hit the market at the right time, and it depends on what other options the market has at the time of deployment. Time will tell.
SE: And those options could be everything from the availability of other flavors of 28nm, right?
Soheili: Yes. There are five flavors of 28nm that are trying to compete today, either by distracting or overwhelming the market, and they’ve got huge manufacturability experience behind it. And there’s a strong ecosystem that follows it. So the options are plenty, and the advantages on paper for 22nm FD-SOI are fantastic. The question is who can get to market fastest while realizing the advantages.
SE: But several years ago when 28nm FD-SOI came out, there was a lot of talk about fabulous power numbers and reasonable performance. What’s changed that wasn’t in place then?
Boudre: What changes is the industry has spent another two years trying to bring up finFETs with the right speed and the right cost. Today, everyone agrees this is difficult and costly. It really positions finFETs for specific applications, which we were not prepared to admit two years ago. Two years ago we were thinking it would be the next node after 28nm, because 20nm didn’t really work very well. So the industry was thinking that was the natural next step. It’s not just EUV. It’s the complexity of the ecosystem around finFETs. We need finFETs. But this is the first time that we see a new platform being implemented on two nodes already that will give an opportunity for new products over the next 10 years for applications that haven’t yet reached the market. These are new applications. You can talk to a lot of CEOs in Silicon Valley. If these applications can get into the supply chain, they will go for it because they need low power and they need to stay in a stable mode in terms of IP, their designs, and their ecosystem. It brings what was not here two years ago, when we all had the convictions that finFETs would be faster, cheaper, and easier to get done.
Kengeri: The first generation of FD-SOI did not have the performance. That was one of the issues. Now we have the second generation, which is able to compete with finFETs. That is really the key change.
SE: Are you really going head to head with finFETs, though? This is another slice of the market.
Kengeri: FinFETs will have their place.
Boudre: Yes, and what’s changed is that this was not ready two years ago.
Kengeri: We are targeting low-end mobiles, wearables, and everything else. FinFET will command a premium for CPUs and high-end smartphones. Everything else will be FD-SOI. We also believe a pure-play foundry endorsing this and bringing this to market is really important. Previous ones were not necessarily pure-play foundries. There will be a few others. We also are running into physics. FinFETs have higher capacitance. There is a real technical reason why FD-SOI is getting attention. Most of the industry is at 28nm. The early adopters have moved to 14nm. The rest are still trying to figure out what is the best node to move to. There was no choice other than finFETs until now. So all of those things are coming together. That’s really the inflection point.
Soheili: We’ve been involved with FD-SOI since 2013, and we’ve been designing technology for FD-SOI for a while. The No. 1 concern has been capacity and manufacturing. Crolles going to Dresden and South Korea and expanding its base of manufacturing is a huge change.
SE: How about the need for a second source?
Soheili: It’s desirable but not necessary. From our perspective, customers we took into FD-SOI were concerned that this was being made in an R&D fab, so what do they do for volume?