Nordic Semiconductor, a provider of low-power wireless connectivity solutions, has released its nRF54L series of SoCs. These integrate an NPU and the Nordic Edge AI Lab. Consequently, this enables developers to more easily run artificial intelligence directly on devices and provides greater power efficiency.
The nRF54LM20B is the company’s latest system-on-a-chip designed for IoT and embedded devices. It is an ultra-low-power, high-memory wireless SoC that integrates the Axon NPU. Furthermore, Nordic Semiconductor has also integrated the Axon neural network processing unit, which helps to efficiently run AI workloads directly on the device. Its Nordic Edge AI Lab is a new development tool that simplifies and accelerates edge AI development.
Nordic Semiconductor CEO Vegard Wollan commented on the importance of edge AI: “Edge AI is…the only way to achieve security, privacy, and sustainability at scale. Nordic’s edge AI solutions enable millisecond-level decision-making without round trips to the cloud, ensure compliance through local processing, and significantly extend battery life for billions of connected devices.”
Axon AI technology enables scalable, energy-efficient AI at the edge.
With the release of the nRF54LM20B SoC, Nordic has introduced the first high-memory chip in its NRF54L series. This also integrates Axon technology, such as the Axon neural network processing unit — a result of Nordic’s acquisition of Atlazo in 2023. According to the company, this unit acts as an “ultra-efficient AI hardware accelerator.” As a result, it runs demanding AI tasks and workloads more efficiently than a standard CPU.
Axon is reportedly seven times faster than competing solutions and eight times more efficient in tasks including keyword recognition and classification.
The nRF54LM20B is compatible with the Axon NPU. It features 2 MB NVM, 512 KB RAM, high-speed USB, a 128 MHz Arm Cortex-M33 plus RISC-V coprocessor, and 66 GPIOs. It also includes Nordic Semiconductor’s fourth-generation ultra-low-power 2.4 GHz radio. This radio supports modern wireless standards such as Bluetooth Low Energy (Bluetooth LE) and Bluetooth Channel Sounding.
Powerful Intelligence with Tiny Models: Neuton Edge AI
Neuton’s CPU-based edge AI models are incredibly compact, typically less than 5 KB. Yet, they are ten times faster and more efficient than comparable CPU-based models. Through the Nordic Edge AI Lab, developers can create new Neuton models for anomaly detection, biometric monitoring, and activity recognition. This not only protects user privacy but also offers extremely low power consumption and memory footprint. Thus, it is highly practical even for small, battery-powered devices.
Oyvind Strom, Executive Vice President of Short-Range Wireless at Nordic Semiconductor, stated that Nordic’s latest offerings make “advanced on-device AI truly accessible to every embedded developer.”
As more devices are deployed at the edge, manufacturers need to find new methods for OTA management. While intelligence is shifting to the edge, cloud-based services remain crucial for device management, monitoring, and location services.
By collecting and analyzing real-time data from deployed devices, manufacturers can enhance features and improve performance without disrupting the user experience. This enables connected products to evolve in a secure and efficient manner. Moreover, Nordic’s hardware, model, and software ecosystem will lead this trend, empowering developers to bring intelligent, low-power edge AI to billions of IoT devices.