R&D powerhouse CEA-Leti plays a major role in the SOI ecosystem. We had a chance recently to catch up with CEO Emmanuel Sabonnadière, who shares his outlook.
Advanced Substrate News (ASN): Can you tell us about your vision for SOI at CEA-Leti?
Emmanuel Sabonnadière (ES): CEA-Leti wants to be a beacon for the More-than-Moore segment of the business. In the semicon world, most of the production is in Asia, most of the IP is in the US, but in fact most of the big R&D centers are in Europe. For Europe, it makes a lot of sense for the three big RTOs – CEA-Leti, imec and Fraunhofer – to collaborate. We have MOUs with Fraunhofer and imec, and they have signed collaboration agreements with each other, too. So imec is more FinFET, and we are more SOI, and on the memory side, they are more standalone, we are more embedded. Fraunhofer is especially good at packaging and integration. Together, we can tackle serious challenges. In particular, there are two we want to target: quantum computing on silicon and edge AI – which are SOI. CEA-Leti and imec signed a specific MOU on these subjects. The three of us are well supported by the European Commission because they highly value what our work represents for Europe.
All we do is for the benefit of the big players in the semiconductor ecosystem but also the start-ups. And, of course, we are helping to highly promote FD-SOI in China.
CEA-Leti really launched it all with the invention of Smart Cut™ wafer production technology, and we continue to support Soitec in advanced development.
In FD-SOI, you have 5G. But the next wave is edge AI, and that wave is probably even bigger.
ASN: What is CEA-Leti’s role in driving FD-SOI in edge AI?In FD-SOI, the route to 12nm and below 10nm is very serious for us. -- Leti CEO Emmanuel Sabonnadière Click To Tweet
ES: In FD-SOI, the route to 12nm and below 10nm is very serious for us. When we look at AI, and look at the model, we see two schools of thought: in the American & Chinese model personal data goes to the cloud; but in Europe, there are strong laws about data privacy. So how do you move your AI system down? That’s the big story. The mobile system has to learn by itself. The intelligence needs to be embedded locally and run on the local, personal data set. So we need some powerful computing, but additionally we have to analyze the personal data set of each mobile owner locally. One school believes in CMOS doing everything and interacting with an extremely powerful neural network. The second school says let’s rebalance what you use in CMOS with what’s needed for neural networks, and ask how you can add sensors on top of that. That’s what we’re dealing with at CEA-Leti. The sensors are very low power – microwatts – and the neural network is powerful but also very low power. So the CMOS part doesn’t have to be so powerful – I can get 10 teraOPS for 1 watt – it’s probably good enough. 3D architectures will form a fully dedicated MPU for edge AI. That’s why we’re working with imec to make this heterogeneous CMOS using our CoolCube™ technology with both FinFETs and FD-SOI in a monolithic architecture. The 3D work by imec is complementary to ours. This is an important edge AI program.
However, 3D production is a completely different animal than what you have today. There are not many fabs that are able to do so many layers. But CoolCube – we know how to do that.
ASN: What other areas of SOI is CEA-Leti involved in?
ES: Photonics, digital RF [on FD-SOI] and quantum computing.
Quantum computing will be based on 28nm FD-SOI. SOI has the right physical properties. [Note: Leti has found that the electron spin in a SOI transistor can be manipulated by pure electrical signals, which enable fast and scalable spin qubits, the elementary bricks of future quantum processors.]
On photonics – silicon photonics – it’s been around for so long but nobody wanted to jump in. One of our spin-offs, Scintil Photonics is a startup in silicon photonics with both the III-V laser and fiber integrated on silicon. They have a pretty interesting solution – let’s see how it will develop. [Note: CEA-Leti presented 21 papers, including five invited, at Photonics West 2020. Read about that here.]
ASN: What do you see as the drivers?
We have decided strategically on four verticals. One is quantum. One is edge AI. A third one is imagers and LiDARs to control an autonomous car. We want to have something really integrated here with multiple sensors. The fourth one is 5G digital. CEA [CEA-Leti’s parent organization at the national level] is working on three transformations: environmental, digital and health. So they asked us to work on these three transformations.
ASN: And you’ve got your next Leti Innovation Days coming up.
ES: Yes! Leti Innovation Days in Grenoble will gather more than 1,000 attendees – from companies, institutes and universities in Asia, Japan, the US and Europe. They will include more than 200 C-level decision makers to explore the most pressing issues and key opportunities in the areas mentioned previously, as well as other fields. [Note: the date has been moved from June to Oct. 12-16, 2020.]
At this year’s Grenoble event we’ll be showcasing our innovations in semiconductors involving sensors, haptics, imaging, computing, radio and smart power. We also host annual LIDs in Japan, Taiwan, and for the first time, in Korea.
This year at LID in Grenoble we also will host an FD-SOI design-and-technology event co-organized with the SOI Consortium on June 23rd. [Details on that will be available shortly.]
ASN: Can you tell us a little more about how CEA-Leti functions?
ES: Only 15% of our €330 million annual budget comes from CEA. All the rest comes directly or indirectly from our industrial contracts. We have bilateral contracts with customers, which have very strong confidentiality agreements, which typically means we can’t tell you what we’re doing or for whom. While that can be frustrating, it’s also something that helps make the institute so powerful.
In addition to industrial contracts, we also invest in what we see as key growth opportunities – as was the case for FD-SOI. Because of the business we’re doing with industry, we can reinvest money we’ve made in advanced technologies.
We also help capitalize our start-ups, and some of them have been very successful. We have created 65 start-ups over the last ten years. While in software hundreds of start-ups can be created every day, in deep tech, six per year, with a couple of unicorns like Soitec, make for good returns. We then re-invest those returns. It’s deep tech, so we’re extremely careful that when we launch we are prepared. The proof of concept has to be pre-financed, and the technological roadmap has to meet the market trends. We love the enthusiasm of our engineers, and are careful to train them thoroughly for business launches. We have a program called Challenge First Steps, in which there is training for engineers in how to use finance, how to manage technology, how to do a robust market study, and defend your plan before a panel of business experts. Then there’s a year of incubation – in some cases, we are also an incubator for the first 18 months, and we offer a strong safety net.
ASN: What about other synergies?
ES: We have relationships with world-leading academic and research institutes, like Stanford, Caltech, MIT, EPFL in Switzerland, Jurich in Germany, and Zhangjiang Lab in China for a collaboration organized by Dr. Xi Wang. [Editor’s note: Dr. Xi Wang has been the leading proponent of SOI in China for over a decade as head of the Shanghai Academy of Sciences. He’s now the country’s Vice-Minister of Science & Technology.] In the Zhangjiang Lab, we are also working on technology with SITRI [an innovation center spin-off of the Shanghai Academy of Sciences established to accelerate More-than-Moore solutions for IoT]. We have created the SOI Academy which mirrors the work of the SOI Consortium but aimed more at training designers on FD-SOI PDKs. We do two week-long sessions per year, with 200 designers per session.