Why it matters: While AI algorithms are seemingly everywhere, processing on the most popular platforms require powerful server GPUs to provide customers with their generative services. Arm is introducing a new dedicated chip design, set to provide AI acceleration even in the most affordable IoT devices starting next year.
The Arm Cortex-M52 is the smallest and most cost-efficient processor designed for AI acceleration applications, according to the company. This latest design from the UK-based fabless firm promises to deliver "enhanced" AI capabilities to Internet of Things (IoT) devices, as Arm states, without the need for a separate computing unit.
Paul Williamson, Arm's SVP and general manager for the company's IoT business, emphasized the need to bring machine learning optimized processing to "even the smallest and lowest-power" endpoint devices to fully realize the potential of AI in IoT. Despite AI's ubiquity, Williamson noted, harnessing the "intelligence" from the vast amounts of data flowing through digital devices requires IoT appliances that are smarter and more capable.
The Cortex-M52 chip design incorporates Arm's Helium technology, adding 150 new scalar and vector instructions to the Armv8.1-M Cortex-M lineup (including Cortex-M55, Cortex-M85). Compared to the previous Cortex-M generation, Helium instructions can provide up to 5.6 times more performance in machine learning algorithms and up to 2.7 times more performance for digital signal processing (DSP) workloads.
Security remains a critical aspect, as Arm explains. Cortex-M52 implements the latest security extensions for Armv8.1-M (PACBTI, Arm TrustZone). The new chip design also facilitates a "modern development flow," according to Arm, giving developers access to a unified toolchain for the Cortex-M platform with full support for AI workflows.
Before Cortex-M52, developers had to use a combination of CPU, DSP, and NPU units with three different software toolkits to achieve the ML and DSP performance now natively provided by the new design.
Arm states that a single toolchain is sufficient now. Cortex-M52 is fully compatible with software written for Cortex-M55 and Cortex-M85, and the new chip will also be available through the Arm Virtual Hardware cloud platform for pre-silicon software development.
Cortex-M52 technology can be licensed for integration into very low-cost IoT products, with chips costing $1-2 likely representing the majority of production, according to Williamson. He added that the chip can also be integrated into "slightly richer," more capable IoT devices.
https://www.techspot.com/news/100942-arm-cortex-m52-chip-brings-ai-acceleration-low.html
