Arm Ethos-U85 NCU in MCUs for local transformer networks

According to Arm IoT general manager Paul Williamson, this is “the world’s first implementation of the Ethos-U85 in an end-to-end AI microcontroller.”

“With Alif’s high-bandwidth memory and power management, [the new parts] support NPU operators needed by transformer networks to run generative AI workloads,” Alif said. “Transformer networks enrich endpoint use cases by focusing on the most important parts of the older parts of the received input sequence, while simultaneously processing the current received sequence to provide deeper contextualization. When used locally, this leads to a more enjoyable experience using multimodal AI in products focused on combining image, voice, text and sensors, without the need for the cloud.”

For now, the company has only said that there is more support for embedded video and audio applications than in the first generation, as well as a wider internal memory bus for improved NPU throughput, dual MIPI-CSI camera interfaces, and 200fps. image signal processor. Speeds of 800 MB/s will be available through a pair of external memory interfaces.

More wake-up options are also promised, including more SRAM storage options.

E4, E6 and E8 are the new base part numbers. Electronics Weekly asked Alif which parts have which resources – so keep an eye on this section if that might be important to you.