​To fully run artificial intelligence on the edge, the institute is working on an unsupervised learning neural network using advanced FD-SOI and a mix of other technologies. These include embedded non-volatile memory (NVM), 3D integration, and new design tools.

Sabonnadière said this new approach is expected to exceed the performance levels of current digital deep learning with neural networks that are capable of handling time-domain signals, sound and speech—and may produce a first "killer app" for advanced SOI.

AI will require compact and power-efficient circuits for the inference phase, when neural networks infer things based on new data they receive, close to the end user. The combination of FD-SOI, 3D integration, and NVM opens a path towards dedicated circuits with major performance improvement within the limited power budgets of distributed electronics.

'Very Aggressive Technology Nodes'

CEA-Leti, which created FD-SOI technology and launched Soitec in 1992, has continuously worked to improve and scale the technology, especially in very aggressive technological nodes that will be used in Edge AI. Currently, FD-SOI technology is integrated into millions of consoles and smartphones and it's now becoming a standard across nomadic electronics, automotive engineering, and the Internet of Things (IoT) sectors. The applications all require high-performance processing at very low energy consumption. In the future, FD-SOI probably will become standard in the connected-car market, which requires highly energy efficient processing, Sabonnadière said.

Looking to the future, Sabonnadière noted that CEA-Leti has demonstrated the advanced FD-SOI roadmap for 12nm and beyond. It is in discussions with top-tier semiconductor companies to pursue new mobile and automotive applications using the technology. CEA-Leti also expects to explore development of an FD-SOI platform in China.

CEA-Leti and SITRI

These two organizations initiated an ongoing collaboration in 2016. SITRI, the Shanghai Industrial μTechnology Research Institute, is an international innovation center focused on globally accelerating the innovation and commercialization of More than Moore technologies to power the Internet of Things.