NXP Semiconductor announced a lead partnership Monday with Arm over a machine learning (ML) accelerator that is focused on industrial and IoT edge devices.
NXP said it intends to use Arm’s Ethos U-55 micro Neural Processing Unit (NPU) accelerator in Cortex-M microcontrollers as well as crossover MCU’s and real-time subsystems in processors for industrial and edge uses.
NXP described the Ethos-U55 accelerator as highly configurable and able to work with the Cortex-M core to provide a small footprint with a 30x increase in inference performance when matched against other microcontrollers. An Arm whitepaper (PDF) also described the Ethos-U55 with Cortex-M as offering low-power consumption and a 480x ML performance improvement over previous Cortex-M generations. The microNPU is a new class of ML processor and comes in four different configurations with maximum performance of up to 0.5 trillion operations each second in a 16nm process.
The Ethos-U55 can work with the Cortex-M55, Cortex-M7, Cortex M-33 and Cortex-M4. Using a single Cortex-M toolchain can simplify and reduce AI application development time, Arm said in its whitepaper.
“The greatest potential for the next computing revolutions lies in scaling AI to the billions of smaller, power-constrained endpoint devices,” Arm said.
NXP boasted its comprehensive portfolio of ML compute elements include CPUs, GPUs, DSPs and NPUs that work with its elQ machine learning development tools. Customers can use the tools and NXP edge processor building applications including object detection, face and gesture recognition, natural language processing and predictive maintenance.
NXP recently announced i.MX 8M Plus processors to add to its ML product line. When combined with the quad Arm Cortex-A53 processor, the NPU can operate at up to 2.3 TOPS, NXP said.
Also on Monday, NXP announced the availability of its i.MX RT600 crossover MCU family for low-power, secure edge applications. It is available for suggested resale at $4.50 per MCU in 10,000-unit quantities. An evaluation kit is priced at $129.
RELATED: Fastest growing tech job roles: engineer/scientist in data or ML