New project looks at low-power chips for AI applications

By eeNews Europe

The programme will last three years. It will focus on developing process technology and hardware platforms that use emerging memory technologies for neuromorphic computing. The project will look to enable applications that currently need cloud-based server racks and execute them in battery-powered mobile devices.

Sending data to the cloud costs energy, adds latency, and creates conflicts with privacy legislation. As such, edge AI applications will require intelligent energy-efficient local processing.

TEMPO will use process technology platforms that are being developed by European research technology organisations and cooperating foundries in the project. These technology platforms will be combined with application and hardware knowledge from additional partners. TEMPO will look at current solutions at device, architecture and application level, and build and expand the technology roadmap for European AI hardware platforms. The memories being investigated include MRAM (imec), FeRAM (Fraunhofer) and RRAM (CEA-Leti). These will be used to implement both spiking neural network (SNN) and deep neural network (DNN) accelerators for use cases ranging from consumer to automotive and medical applications.

More information

Related news

imec and CEA-Leti work together on AI and quantum computing

ASML and IMEC collaborate to take EUV lithography to high-NA

Imec demonstrates spin-orbit torque MRAM manufacturability on 300mm Silicon

Imec shows optimized process flows for Ge-based devices