Chip design firm Arm Holdings Plc said today it’s aiming to bring advanced artificial intelligence workloads to the smallest of internet of things devices with the launch of its new Cortex-M52 processor that’s capable of more powerful computation with limited resources.

The company said its partners have been asking for access to on-device intelligence that can be deployed in smaller, battery-powered devices at the network edge. Their goal is to achieve greater privacy and reliability at lower cost, which can be obtained by reducing the reliance on costly, cloud-based servers.

The Cortex-M52 processor is designed to power what Arm calls AIoT operations that require a significant boost in digital signal processing and machine learning performance, without the cost overhead of a dedicated accelerator. Customers will be able to embed intelligence into applications and devices at the edge at much lower cost than is possible today.

The idea, said Paul Williamson, senior vice president and general manager of Arm’s IoT line of business, is to “enable AI everywhere, on any device.”

Previously, to achieve the level of performance delivered by the Cortex-M52, it would have been necessary to combine a central processing unit with a dedicated signal processor and neural processing unit. This complexity meant that developers had to write, debug and tune their code for the chips using three separate tool chains, three compilers, three debuggers and so on. In other words, three times the work.

However, the Cortex-M52 integrates the DSP and NPU capabilities directly with the CPU, enabling developers to use a single toolchain and architecture for building AIoT devices.

According to Arm, the performance boost delivered by the Cortex-M52 is derived from its Arm Helium technology, which is designed to enhance the DSP speed of smaller, low-powered devices. The technology has previously been integrated into products at the network edge, and by embedding it within the Cortex-M52, Arm says it can now bring this increased performance to more power-constrained devices.

Companies that are already running AIoT operations will benefit from a simplified migration path, with the ability to upgrade from Arm’s earlier-generation Cortex-M33 and Cortex-M4 chips. When they do so, the Cortex-M52 will enable richer AI-powered voice and vision experiences for applications including automotive and industrial control, predictive maintenance and wearable sensor fusion, the company said.

Cortex-M52 is the latest addition to the Armv8.1-M Cortex-M family of low-powered processors, which also includes the more powerful Cortex-M55 and Cortex-M85 chips. It provides optimal performance for AI at a new efficiency point, enabling new capabilities to be integrated with the microcontrollers that control things such as drones and sensors. Arm said chip manufacturers will now be able to offer their customers a better tradeoff between performance and cost.

Arm said the Cortex-M52 is fully software-compatible with the Cortex-M55 and Cortex-M85, meaning developers will be able to access the growing ecosystem of software and tools built around its Helium technology. The company is also providing free software libraries and an extensive knowledge base from its partner ecosystem. To aid developers further, the Cortex-M52 will also be made available on the Arm Virtual Hardware service, so they can get started in developing ahead of its general availability.

The company expects its partners to start shipping ships sometime next year.

With reporting from Robert Hof

Images: Arm