News

Arm Releases the Cortex-M52 Aimed At Low-Power Embedded AI

November 22, 2023 by Aaron Carman

Rounding out its portfolio, the latest Arm processor—announced today—sports AI performance in low-power silicon.

In a bid to bring AI capabilities to more applications, today Arm has announced the Arm Cortex-M52 processor. The new Cortex CPU core comes equipped with several familiar additions such as DSP support and ML acceleration and rounds out the Cortex-M lineup of Arm processors by providing designers with a low-cost and high-efficiency option for ML processing.

 

The Arm Cortex-M52 includes many features for AI/ML development while maintaining low power and high efficiency, making it a candidate for embedded applications.

The Arm Cortex-M52 includes many features for AI/ML development while maintaining low power and high efficiency, making it a candidate for embedded applications.

 

As designers begin to incorporate AI into more applications, a frequent pain-point is balancing the size, power, and cost of the device with the performance necessary to support even the simplest ML models. As such, the Cortex-M52 targets applications where cost and efficiency are key metrics in addition to pure performance.

In order to understand what the Cortex-M52 offers designers as well as how it could add more AI interaction to everyday life, we sat down with Paul Williamson, senior vice president and general manager of the IoT Line of Business at Arm to learn what inspired the Cortex-M52 and what applications it could impact.

 

Extending AI’s Applications

Although large AI models such as GPT-4 are often what comes to mind after hearing “AI”, the presence and impacts of AI/ML models extend far beyond powerful generative AI. In fact, AI has already found itself to be quite impactful in mid-range applications. When discussing the extent to which AI is currently utilized, Williamson said, “Many people just don’t realize how much AI is already being deployed into embedded devices, and it’s making a difference in various applications across our homes, cities, and industries.”

 

“AI is critical in harnessing that intelligence from the vast amount of data we’re producing and collecting with these digital devices.”

 

Despite the breadth of uses for AI models, however, they inherently require a considerable amount of computation and signal processing, and are usually augmented by extra silicon like NPUs. As such, in applications that prioritize cost or size, bulky AI processors may not be feasible.

 

In an example predictive maintenance workload, the Cortex-M52 is able to address all signal conditioning needs, allowing designers to simplify the embedded AI/ML development process.

In an example predictive maintenance workload, the Cortex-M52 is able to address all signal conditioning needs, allowing designers to simplify the embedded AI/ML development process.

 

These kinds of devices are exactly the ones that Arm hopes to address with its latest Cortex processor. The Cortex-M52 strikes a careful balance between AI/ML performance and size, cost, and area, allowing designers to use the processor in a slew of new applications that were previously not possible with current generation processors.

 

Boosting Performance

The Cortex-M52 contains several features that distinguish it from previous generations and current members of the Cortex-M family. Namely, its reduction in size, power, and cost allow it to target high-efficiency applications that do not require as much computational power.

Compared to previous Cortex generations, the Cortex-M52 sports a 5.6× performance increase in ML and a 2.7× increase in DSP, giving designers access to considerable upgrades when developing ML-enabled applications. In addition, the Cortex-M52 includes Arm’s Helium, PACBTI, and TrustZone technologies, all of which can also be found in the top-of-the-line Cortex-M85.

The Cortex-M52 is complemented by a comprehensive software suite, allowing designers to forego multiple environments for a single development toolchain for embedded code, DSP, and AI/ML programming. This not only improves the development experience, but also creates an overall shorter time to market thanks to unified development tools.

 

Compared to previous generations of Arm processors, the Cortex-M52 supports greater DSP and ML performance, while sporting a smaller die area compared to the Cortex-M55.

Compared to previous generations of Arm processors, the Cortex-M52 supports greater DSP and ML performance, while sporting a smaller die area compared to the Cortex-M55.

 

The improved ML performance combined with the 23% reduction in die area could greatly increase the ability for designers to incorporate AI into more devices. Williamson discussed some of the secondary benefits of the Cortex-M52.

“We believe Cortex M 52 will be critical in removing barriers to ML adoption and enabling deployment on the smallest of devices, bringing AI into the reach of everyone, enabling innovation at scale, and ensuring developers continue to draw on the broad and proven software tooling ecosystem that Arm provides,” he said.

 

Well Suited for Embedded AI

All-in-all, the new Cortex-M52 could be a big step in bringing AI capabilities to low-power or embedded applications. This ultimately allows for greater performance, and a unique paradigm shift for low-power, high efficiency applications. In Arm’s case, they believe that applications such as predictive maintenance, industrial control, and sensor fusion will all benefit from the data processing performance of AI/ML models.

 

The Cortex-M52 addresses high-efficiency applications, where designers can leverage simple AI models to improve performance while maintaining stringent requirements.

The Cortex-M52 addresses high-efficiency applications, where designers can leverage simple AI models to improve performance while maintaining stringent requirements.

 

When wrapping up how the new Arm chip could benefit designers, Williamson said, “We are bringing AI to even the smallest devices, and we believe this will enable the industry to scale IoT device deployments even further as we continue to build an AI enabled future on Arm.” So, while the Cortex-M52 may be the first step to making AI completely pervasive, it is a good sign that we may soon be incorporating intelligence in all new applications.

 

All images used courtesy of Arm