Microelectronics research Optimized Chips for Local AI

CEA-Leti hat seinen Sitz auf dem Minatec-Campus in Grenoble.
CEA-Leti is located at the Minatec campus in Grenoble, France.

»Edge AI« – artificial intelligence in local devices – is the field where Europe can best bundle its AI competencies. France's CEA-Leti research institute wants to develop chips that are optimized for this very purpose.

 Artificial intelligence is already penetrating our everyday lives: with language assistants such as Alexa and Siri, face recognition in photos, the "Smart Compose" e-mail assistant from Google and much more. But this is just the beginning: AI will soon penetrate more and more areas that are security-critical or strongly linked to the protection of personal data. Some of these applications require a strict real-time capability - e.g. autonomous driving.

In such a context, AI in the cloud is not sufficient; rather, intelligence must be available at the local level. This means that the algorithms must be processed locally, independent of the cloud. The systems must react completely autonomously and make decisions. Working independently of the cloud can also strengthen cyber security, because then an attack via the network is not possible.

To put this into practice, however, a wide range of hardware and software improvements are needed. The CEA-Leti (Laboratoire d'electronique des technologies de l'information), a research institute under the French Atomic Energy Agency (CEA) and the largest research institute for electronics and information technology in France, has set itself the task of working on these improvements. In 1972, a spin-off from Leti created the company STMicroelectronics and Leti is still today ST's "home and farm research institute".

Unlike America and China

Emmanuel Sabonnadiere, CEO of CEA-Leti (photo), sees artificial intelligence in the area of conflict between the USA and China: "There is the American model in which users get all services free of charge, but pay with their data. And the Chinese model, in which we see fundamental ethical problems. In both cases, no value is placed on the protection of personal data. Europe must find its own way that takes personal rights into account."

Both the American and the Chinese models rely on the cloud to bundle and process data. And this is precisely where Sabonnadiere sees an opportunity for Europe to differentiate itself in the race for artificial intelligence. The key word is "Edge AI", artificial intelligence at the edge of the network, implemented on local devices, independent of the cloud.

Dr. Handel Jones, CEO of Intelligence Business Strategies, who was invited as a guest speaker to the CEA-Leti Innovation Day, also makes the same point: "With IMEC, CEA-Leti and Fraunhofer, Europe is leading in the innovation of sensors that will occupy a key position as data suppliers for AI. As the market for sensors will grow strongly, he sees Europe as well positioned, but there is no reason to rest on the previous successes. "Industrialisation has taken place over a period of 200 years. AI will bring even more extensive upheavals, but in a period of 20 years. Many jobs will be lost and we must now create new industries so that we do not have massive unemployment soon".

According to CEA-Leti, there are three application levels of artificial intelligence (Fig.): 

  • The cloud, in which the initial training of neural networks will also take place in the future, because this process requires enormous amounts of data and equally large computing power.
  • Medium-size systems with powerful edge computers such as autonomous vehicles or robots
  • Small and mobile systems, e.g. wearables for medical electronics, which must operate independently of the power supply network and possibly also the data network.