Imagine a computer chip that remembers information without needing constant power, like a light switch that stays in position. This technology allows AI to process data right on a device instead of sending it to a distant cloud. It works like a super-efficient brain that uses tiny amounts of electricity to recognize voices or images.
Current AI hardware for edge devices consumes too much power and suffers from high latency, limiting the ability to run complex neural networks on battery-powered devices.
A non-volatile memory cell using hafnium zirconate and FeFET-2 circuits integrated into a memory-augmented neural network accelerator.
If you are a wearable device manufacturer dealing with short battery life in voice-activated gadgets — this project developed a ferroelectric accelerator that provides a 2500x gain in energy-efficiency. This allows for complex AI processing on the wrist without draining the battery.
If you are an ADAS sensor provider dealing with high latency in image recognition — this project developed a BEoL integrated memory-augmented neural network. This enables faster, local processing of visual data with ultra-low power consumption.
If you are a smart factory sensor developer dealing with the high cost of powering thousands of edge nodes — this project developed FeFET-2 circuits. These circuits break the POPS/W barrier to enable scalable, low-power intelligence at the machine level.
Based on available project data, the project focuses on using low-cost, high-density BEoL integrated technology, though specific price points are not listed.
Yes, the project specifically targets a scalable edge accelerator using BEoL (Back End of Line) integration, which is designed for scalable systems integration.
Based on available project data, licensing terms are not specified, but the consortium includes 6 industrial partners and 2 SMEs who are developing the technology.
The technology uses BEoL integration and standard CMOS field effect transistors, allowing it to be integrated into existing semiconductor manufacturing flows.
The project is active from 2024-01-01 to 2027-12-31.
The consortium is heavily industry-weighted with 46% industrial participation (6 companies, including 2 SMEs), suggesting a strong push toward commercial viability. With 13 partners across 5 European countries and a collaboration with South Korea, the project covers the full value chain from material science to system simulation.
Contact the Commissariat à l'énergie atomique et aux énergies alternatives (CEA) in France.
Contact us to connect with the Ferro4EdgeAI consortium for early adoption of ferroelectric AI hardware.
