It is “the world’s first implementation of the Ethos-U85 in an endpoint AI MCU”, according to Arm general manager of IoT Paul Williamson.
“With Alif’s high bandwidth memory and power management, [the new parts] support NPU operators required by transformer networks to execute generative AI workloads,” according to Alif. “Transformer networks enrich endpoint use cases by focusing on the most relevant parts of older portions of an acquired input sequence while processing in parallel the current acquired sequence to enable deeper contextualisation. For localised use this results in a more satisfying experience when using multimodal AI in products focused on vision, voice, text and sensor fusion, without relying on the cloud.”
The company has, so far, only revealed that there is more support for embedded vision and audio applications than in the first generation, and a wider internal memory bus will improve bandwidth for the NPU, dual MIPI-CSI camera interfaces and 200frame/s image signal processor. A pair of external memory interfaces will transfer 800Mbyte/s.
More wake options are also promised, from a broader range of sram data retention options.
E4, E6 and E8 are the new base part numbers – Electronics Weekly has asked Alif which parts have what resources – so wath this space if this might be important to you.
