Arm has announced a new neural processing unit (NPU) and a new IoT Reference Design Platform. The company believes these new advancements will help support edge AI applications. 

Arm Ethos-U85 improves performance by 4x compared to the previous version and also has 20% higher power efficiency. It also has 2048 MAC units, compared to the previous version’s 128. 

It supports Transformer Networks and Convolutional Neural Networks (CNNs) for AI inference, which opens up new use cases with AI, especially when it comes to vision and generative AI. This allows it to support tasks such as understanding videos, filling in the missing pieces of images, or analyzing data from multiple cameras in order to classify images and identify objects.

The company also announced a new IoT Reference Design Platform called Corstone-320. It includes the software, tools, and support needed for developers to build applications for the edge. 

Arm believes that Corstone-320 will enable benefits such as improved performance, improved memory bandwidth, greater energy efficiency, lower design costs, and more. 

Corstone-320 also utilizes Arm Ethos-U85, as well as Arm Cortex-M85, Arm Mali-C55 ISP, and Arm CoreLink DMA-350. 

“As AI adoption grows, everyone from startup innovators to the world’s biggest microcontroller players are converging on Arm as the platform of choice to deliver their AI solutions from cloud to edge,” Paul Williamson, SVP and GM of the IoT LoB at Arm, wrote in a press release.