Nordic Semiconductor, a leading global provider of low-power wireless connectivity solutions, is bringing AI intelligence and functionality to the smallest battery-powered IoT devices. With its industry-leading ultra-low-power edge AI solutions, Nordic is accelerating the arrival of a new generation of devices integrating edge AI intelligence, perfectly combining energy efficiency with unparalleled ease of use for developers.

· 

nRF54LM20B – A brand new ultra-low power, large memory wireless SoC, equipped with an Axon NPU

· 

· 

Custom Neuton Model – Industry-Leading Ultra-Small Edge AI Model

· 

· 

Nordic Edge AI Lab – Development tools to simplify and accelerate edge AI development

· 

"AI factories train intelligence, while Nordic deploys it—on devices, at the edge, wherever the world operates," said Vegard Wollan, CEO of Nordic Semiconductor. "Edge AI is no longer an option, but the only way to achieve security, privacy, and sustainability at scale. Nordic's edge AI solutions enable millisecond-level decision-making, eliminate latency to and from the cloud, ensure compliance through local processing, and deliver significantly improved battery life for billions of connected devices. This is the new standard for ultra-low-power edge AI, and Nordic is defining it." In 2023, Nordic acquired Atlazo and its Axon technology. The nRF54LM20B SoC is the first large-memory product in the nRF54L series, integrating the Axon Neural Processing Unit (NPU), a highly efficient AI hardware accelerator that significantly enhances performance for edge AI workloads. Compared to similar solutions, Axon offers up to 7x performance improvement and up to 8x energy efficiency in tasks such as sound classification, keyword recognition, and image detection.

The nRF54LM20B SoC combines an Axon NPU with 2 MB NVM, 512 KB RAM, a 128 MHz Arm® Cortex-M33 plus RISC-V coprocessor, high-speed USB, up to 66 GPIOs, and Nordic's fourth-generation ultra-low-power 2.4 GHz radio, supporting features such as Bluetooth® LE, Bluetooth Channel Sounding, and Matter over Thread.

Neuton models are ultra-small, CPU-driven edge AI models, typically less than 5 KB, 10 times smaller than other CPU-driven models, and faster and more efficient. Nordic Edge AI Lab helps developers generate custom Neuton models for applications such as anomaly detection, activity and gesture recognition, and biometric monitoring—providing privacy-preserving real-time intelligence with minimal battery and limited memory, without relying on cloud services.

Leveraging Nordic Edge AI Lab and the Neuton model, intelligent technologies can easily move from concept to reality. A recent deployment case exemplifies this: a global supply chain solutions company easily upgraded its intelligent tracking devices using AI models developed by Nordic Edge AI Lab, enabling direct detection of real-world events such as shocks, swaying, and transportation on the nRF54L series SoCs. Thanks to Nordic's nRF Cloud Lifecycle Services, these AI-driven insights were seamlessly deployed across the entire fleet without any operational disruption.

Nordic Edge AI Lab

"With Nordic Edge AI Lab, the Neuton model, and the Axon NPU, Nordic makes advanced on-device AI easily accessible to every embedded developer," said Oyvind Strom, Executive Vice President of Short Range Business Unit at Nordic Semiconductor. "Developers can quickly and easily develop and achieve the disruptive performance needed for a wide range of applications, from wearables to industrial sensing—all thanks to Nordic's trusted, ultra-low-power hardware solutions." As intelligent technologies move to the edge, the demand for OTA management and deep observability is growing. Meanwhile, cloud-based lifecycle services remain crucial for device management, embedded observability, and location services. Manufacturers increasingly need continuous insights into device performance—not only for real-time product improvements but also to meet growing regulatory and customer demands. Leveraging data from deployed devices can enhance functionality and optimize performance without impacting user experience. This ensures connected products can evolve securely and efficiently throughout their lifecycle.

Nordic Edge AI Lab and custom Neuton models are now available for Nordic Wireless nRF54 series SoCs and cellular IoT SiP modules. The nRF54LM20B SoC with an Axon NPU is currently being sampled to select customers and is expected to be fully available for development by early Q2 2026.

Meanwhile, MIPS has also released a RISC-V-based NPU.

The MIPS S8200 provides a software-first RISC-V NPU

GlobalFoundries' MIPS recently released details of its MIPS S8200 processor IP, designed to support next-generation AI workloads on embedded platforms. As the latest member of the MIPS Atlas RISC-V processor IP family, the MIPS S8200 RISC-V NPU supports edge Transformer and Agentic language AI models, significantly improving efficiency and performance. This innovation marks a significant step forward in providing standards-based, scalable physical AI solutions for autonomous edge platforms across multiple markets.

MIPS CEO Sameer Wasson stated, "MIPS is bringing physical AI to autonomous edge platforms that require modern AI capabilities in a scalable, efficient, open, and standards-based manner. The MIPS S8200 RISC-V NPU significantly enhances the performance and efficiency of customers' autonomous platforms. The MIPS virtual platform Atlas Explorer enables customers to accelerate progress and begin model optimization before chip development. MIPS is empowering autonomous edge platforms, bringing AI from the data center into physical commercial and industrial products." ForwardEdge ASIC, a wholly owned commercial subsidiary of Lockheed Martin, is a leader in trusted microelectronics solutions for mission-critical applications in the aerospace, defense, and other high-reliability industries. The company has selected the MIPS S8200 for its upcoming high-performance application-specific ASIC, which will be used in autonomous platforms.

Frank Ferrante, CEO of ForwardEdge ASIC, stated, "We are committed to building high-performance solutions for autonomous platforms, which requires best-in-class performance and area density when running the latest AI models. The MIPS S8200 provides the superior performance and efficiency our advanced mission-critical platforms need." The MIPS S8200 delivers true multimodal intelligence, meeting the demands of performing physical AI in autonomous edge platforms, such as neural networks for multiple functions including vision, radar, and speech, as well as complex context awareness. MIPS' software-first approach allows customers to begin model optimization on a virtual platform using the MIPS Atlas Explorer. These core models enable developers to collaboratively design hardware and software, improving efficiency and optimizing applications. Combined with the open, modular RISC-V standard, MIPS reduces platform development costs.

In 2027, MIPS expects to launch the first silicon reference platform samples featuring MIPS S8200 neural processing units to accelerate the application of physical AI in autonomous edge devices. These evaluation platforms will showcase the performance and efficiency of the MIPS AI platform and drive the development of the toolchain ecosystem, thereby supporting the adoption of RISC-V (including vector and matrix extensions) in autonomous edge applications.

Source: Compiled from Nordic

View more at EASELINK