Arm has introduced the Ethos-U85, a new addition to its Ethos line-up of neural processing units (NPUs), aimed at enhancing artificial intelligence (AI) performance on edge devices. The company has also released a Reference Design Platform for seamless integration of the new NPU. Devices equipped with the Ethos-U85 are expected to support advanced generative AI models by next year, marking a step forward in edge computing capabilities.
Revolutionizing Edge AI with Enhanced Performance and Efficiency
The Ethos-U85, part of Arm's Ethos-U series of microNPUs, is designed to work alongside Cortex-M processors, offering a fourfold increase in performance and a 20% improvement in power efficiency compared to its predecessors. This boost is attributed to the NPU's configurable design, which allows for between 128 to 2,048 multiply-accumulate units, enabling up to 4 TOPs (trillion operations per second) at 1 GHz. Paul Williamson, SVP & GM of Arm's IoT Line of Business, highlighted the evolving demands on embedded systems, from handling extensive data volumes to executing complex AI and machine learning (ML) tasks. The Ethos-U85's ability to support both Transformer Networks and Convolutional Neural Networks (CNNs) is poised to unlock new possibilities in vision and generative AI applications, such as image classification and object detection.
Empowering Developers with Tools and Compatibility
To complement the Ethos-U85, Arm has introduced the Corstone-320 IoT Reference Design Platform, enabling hardware partners to expedite chip design and development. This platform combines the Ethos-U85 with Cortex-M85, Arm's highest-performing microcontroller design, and the Mali-C55 image signal processor. Furthermore, the Ethos-U85 is compatible with high-end Armv9 Cortex-A CPUs, broadening its applicability to a wider range of devices requiring efficient edge inference. Arm's commitment to developer support is evident in the NPU's compatibility with existing toolchains and popular AI frameworks like TensorFlow Lite and PyTorch, ensuring a smooth transition for developers already working with Ethos products.
Arm envisions the Ethos-U85 playing a crucial role in the deployment of generative AI models in edge devices across various sectors, including smart home, retail, and industrial settings. The announcement comes as the AI chip market is becoming very competitive. Nvidia has taken a leading role with its H100 processor, which has helped the company boost its revenues. Microsoft has gone proprietary with its own AI chips called Azure Maia and Azure Cobalt, while OpenAI is also developing its own processing system.
