Arm Expands Chip Lineup for Demanding Edge AI
Arm has expanded its Flexible Access portfolio with new designs, including the Ethos-U85 and Cortex-M52. These chips are specifically built for demanding edge AI and digital signal processing workloads like real-time video processing. The expansion is aimed at helping developers optimize on-device AI performance before committing to silicon.
The new Ethos-U85 NPU delivers a 4x performance increase and is 20% more energy efficient compared to previous generations like the Ethos-U55 and U65. It can be scaled up to 4 tera-operations per second (TOPS) at 1 GHz and is designed to handle transformer-based models, which are fundamental to the latest language and vision AI. The Cortex-M52, featuring Arm's Helium technology, is engineered for power-sensitive and cost-constrained devices. It provides a significant performance boost for Digital Signal Processing (DSP) and Machine Learning (ML) tasks—up to a 5.6x improvement for ML and 2.7x for DSP over the Cortex-M33—without needing a dedicated compute unit. This hardware expansion is part of Arm's broader strategy to reduce risk and cost for developers through its Flexible Access program. The program allows startups, now defined as those with up to $50 million in funding, to experiment and prototype with Arm's IP for free, only paying license fees when they are ready for manufacturing. Over 100 companies have already launched more than 400 chips through this initiative. The move reinforces Arm's focus on the growing edge AI market, where processing happens locally on a device rather than in the cloud. Arm CEO Rene Haas has emphasized that shifting AI inference from power-hungry data centers to edge devices is crucial for energy sustainability as AI workloads scale. The global edge AI processor market was valued at $2.5 billion in 2021 and is projected to reach $9.6 billion by 2030.
