Arm officially launched the Ethos-U85 NPU. As an advanced product for edge AI, the Ethos-U85 is specifically designed for fields such as industrial automation and video surveillance, delivering performance four times higher than its predecessor.
The Ethos-U85 achieves 20% greater energy efficiency compared to previous models and supports common neural networks with 85% utilization. Its architecture adapts to systems based on Arm Cortex-M/A processor cores and demonstrates strong resilience to memory latency.
The Ethos-U85 NPU supports flexible configurations ranging from 128 to 2048 MAC units, delivering up to 4 TOPs of computational power at a maximum frequency of 1GHz.
This NPU product utilizes the same toolchain as the Ethos-U55/65, enabling developers to get started quickly, while supporting mainstream AI frameworks such as TensorFlow Lite and PyTorch.
Companies pioneering the adoption of the Ethos-U85 NPU include Infineon and Alif Semiconductor (IT Home Note: The latter was founded in 2019 and was one of the first companies to adopt the Ethos-U55).
Additionally, Arm simultaneously launched the Corstone-320 IoT reference design platform, integrating the Cortex-M85 CPU, Mali-C55 IPS, and Ethos-U85 NPU to deliver powerful performance for edge AI applications like audio and vision.
The Corstone-320 reference design platform incorporates Arm virtual hardware, enabling developers to begin software development before physical chips are finalized, significantly reducing time-to-market for edge AI devices.
Paul Williamson, Senior Vice President and General Manager of Arm's IoT Business Unit, stated: As edge AI deployments continue to scale, IC designers face increasingly complex systems and software, rapidly growing AI performance demands, and urgent time-to-market pressures. Simultaneously, software developers seek a more unified and simplified development experience to integrate more easily with new AI frameworks and libraries. Arm's new technologies address these needs, accelerating the adoption of edge AI deployments."