Weebit Nano Demos Live Edge AI on ReRAM Silicon
Weebit Nano is demonstrating live edge AI inference on silicon using its Resistive RAM (ReRAM) technology at the embedded world conference. The company's ReRAM is being promoted as an enabling technology for low-power AI inference on edge devices. The live demos aim to showcase the practical application of the technology in real-world scenarios.
- Weebit's ReRAM is positioned as a successor to flash memory, particularly at advanced process nodes where embedded flash is difficult to scale; the demonstration chip is built on a 22nm process, a node where flash is not viable. - The technology offers a significant performance advantage with 10 to 100 times greater endurance than flash, supporting up to 1 million write cycles compared to the typical 10,000 for flash. - In a specific joint demonstration with the company EMASS, the ReRAM stores neural network weights for a gesture recognition system, showcasing its intended use for AI model storage in power-constrained edge devices. - Weebit has entered into licensing agreements to integrate its ReRAM technology into the products of major semiconductor manufacturers, including Texas Instruments and onsemi. - The demonstration partner, EMASS, specifically pivoted from MRAM technology to Weebit's ReRAM, citing ReRAM as being better suited for next-generation systems in IoT, automotive, and consumer electronics. - As a Back-End-of-Line (BEOL) technology, ReRAM is simpler and more cost-effective to integrate into standard manufacturing flows than flash, which is a Front-End-of-Line (FEOL) process. - Long-term, ReRAM is considered a foundational technology for neuromorphic computing, as its physical ability to change resistance is analogous to synapses in a biological brain, enabling future in-memory computing paradigms.
