With the addition of the Cortex-M line-up to its controller core designs, Arm is set to integrate AI into connected devices and other low-power hardware.
The British chip giant has introduced the Cortex-M52, touted as its smallest and most cost-effective processor. It features Helium vector processing extensions aimed at accelerating machine learning (ML) and digital signal processing (DSP) tasks.
To help users leverage these capabilities, Arm is offering a streamlined application development program.
Despite the buzz around advanced AI technologies like conceptual AI, Arm envisions a more straightforward approach to AI implementation.
Paul Williamson, Arm’s Senior Vice President and General Manager for its IoT division, emphasized the importance of enabling machine learning-optimized processing for even the smallest and most power-efficient devices to fully harness AI’s potential in the Internet of Things landscape.
AI plays a crucial role in extracting valuable insights from the vast amounts of data generated by electronic devices. This transformative capability is poised to empower devices to become smarter and more efficient over time.
Arm’s Cortex-M series predominantly consists of single-core 32-bit designs tailored for compact size, low power consumption, and cost-effectiveness.
The Cortex-M52 is expected to support a range of applications, including vibration or anomaly detection, sensor fusion for enhanced data accuracy, and keyword detection through natural language processing (NLP).
Helium, also known as M-Profile Vector Extensions (MVE), introduces over 150 additional scalar and vector instructions, operating on 128-bit data. The Cortex-M52, as the initial implementation of this technology, aligns with these advancements.
The integration of Helium results in significant performance boosts, with DSP operations running 2.7 times faster and ML tasks achieving up to 5.6 times higher performance compared to earlier iterations like the Cortex-M33, all without requiring a dedicated neural processor (NPU).
Transitioning to the Cortex-M52 not only enhances efficiency but also improves energy efficiency and reduces footprint by 23%, according to Williamson.
The Cortex-M52 is positioned to facilitate the adoption of ML, enabling its deployment on even the smallest devices and democratizing access to AI technology.
To simplify app development for the Cortex-M52, Arm is incorporating AI support into a unified application framework.
Developers can leverage Helium technology to optimize DSP and ML functions within their applications efficiently, without delving into the intricacies of the underlying hardware.
Arm’s ecosystem partners are developing tools to ensure the seamless portability of ML models across different systems, facilitating testing and deployment in diverse environments.
The Cortex-M52 will be available on Shoulder Virtual Hardware, a cloud-based platform enabling software development before the physical silicon is ready.
Chip manufacturers are authorized to integrate the Cortex-M52 into end products such as system-on-chip (SoC) solutions. The first products based on the Cortex-M52 are expected to hit the market in 2024.
While these chips are primarily targeted at the \(1 to \)2 price range, there is potential for them to be featured in more sophisticated and higher-priced devices, as highlighted by Williamson.