NXP Announces Major Partnership for Arm Ethos-U55 Neural Processing Unit in Machine Learning Applications

Shanghai, China – February 28, 2020 – NXP Semiconductors NV (NASDAQ: NXPI) today announced a major partnership for the Arm Ethos-U55 microNPU (Neural Processing Unit). This is a machine learning (ML) processor for resource-constrained industrial and Internet of Things (IoT) edge devices. As a leading innovator in the microcontroller (MCU) industry, NXP plans to implement the Ethos-U55 in its real-time subsystems for Arm Cortex-M-based microcontrollers (MCUs), crossover MCUs and applications processors. The expansion builds on the company’s growing portfolio of machine learning offerings, including the recently released i.MX 8M Plus applications processor with a dedicated NPU.

The highly configurable Ethos-U55 works in tandem with the Cortex-M core to achieve a smaller form factor, and also provides more than 30x improvement in inference performance compared to the Cortex-M alone (even with high-performance MCUs). The Ethos-U55 is tailored for accelerating machine learning inference in space-constrained embedded and IoT devices. Its advanced compression technology saves energy and significantly reduces the size of machine learning models, enabling the execution of neural networks previously only possible on large systems. In addition, with a unified toolchain with Cortex-M, developers can develop machine learning applications in the familiar Cortex-M development environment in a simplified and seamless way. Ethos-U55 end-to-end support from training to runtime inference deployment will be accessible through NXP’s eIQ machine learning development environment.

NXP supports its comprehensive portfolio of machine learning computing elements (CPU, GPU, DSP and NPU) through its eIQ machine learning development environment. The environment provides a selection of commonly used open source inference engines to meet the performance needs of specific computing elements. Using NXP edge processors and eIQ tools, customers can easily build many machine learning applications, including object detection, facial and gesture recognition, natural language processing, and predictive maintenance.