CES 2026: Nordic nRF54LM20B Wireless SoC Adds Axon NPU for Edge AI

At CES2026, Nordic Semiconductor has expanded its nRF54L series with the release of the nRF54LM20B, a new low-power wireless SoC designed for Artificial Intelligence (AI) and Machine Learning (ML) applications on the edge. Building on the high-memory architecture of the previously announced nRF54LM20A, the new ‘B’ variant integrates a dedicated Neural Processing Unit (NPU) to address the increasing demand for on-device intelligence in battery-powered IoT devices.

Nordic Semiconductor’s nRF54LM20B ultra-low power wireless SoC features an integrated Axon NPU and the largest memory of the nRF54L series. Video used courtesy of Nordic Semiconductor

The nRF54LM20B and Axon NPU

The most notable addition to the nRF54LM20B is the integrated Axon Neural Processing Unit (NPU). While the SoC still features the same 128 MHz Arm Cortex-M33 application processor and RISC-V coprocessor found in the standard “A” variant, the addition of the Axon NPU changes how the chip handles AI workloads.

Nordic mentions that the Axon NPU can deliver up to a 15× improvement in both speed and energy efficiency for on-device AI inferencing compared to running the same workloads on the CPU alone. This makes it possible to run demanding edge-AI tasks such as sound classification, keyword spotting, and anomaly detection while keeping power consumption very low, which is critical for battery-powered devices.

Like all other Nordic MCU this one is also compatible with Nordic’s nRF Connect SDK. On top of that, they have also released two ultra-tiny Neuton edge AI models that are under 5 KB and up to 10 times smaller, faster, and more efficient than other models. Nordic has also introduced the “Nordic Edge AI Lab” and support for custom Neuton models. These ultra-tiny AI models are optimized to run efficiently on the nRF54L series, enabling developers to implement sophisticated features like gesture recognition and biometric monitoring without relying on cloud connectivity.

High-Memory and Processing Power

Like its predecessor, the nRF54LM20B is also designed for applicatons that need a relatively large memory footprint. It comes with 2 MB of non-volatile memory (NVM) and 512 KB of RAM, which is more than enough to support Matter-based smart home platforms as well as advanced edge AI workloads.

The nRF54LM20B features a multicore architecture (Arm Cortex-M33 and RISC-V) with an integrated Axon NPU, 2MB NVM, and a 2.4 GHz radio, designed for advanced Edge AI and IoT applications. Image used courtesy of Nordic Semiconductor

The SoC also uses a dual-core design, combining an Arm Cortex-M33 application processor with a RISC-V coprocessor. This setup allows tasks to be split efficiently, with the main core running application logic while the coprocessor and NPU take care of real-time data processing and AI inference.

Connectivity, Security, and Peripherals

The nRF54LM20B retains the robust connectivity features of the nRF54L series. Its fourth-generation ultra-low-power 2.4 GHz radio supports:

• Bluetooth LE (including Bluetooth 6.0 Channel Sounding)
• Bluetooth Mesh
• Matter, Thread, and Zigbee
• Proprietary 2.4 GHz protocols (up to 4 Mbps)

The SoC is also compatible with the nRF70 Series companion ICs for Wi-Fi connectivity.

In terms of peripherals, the chip includes high-speed USB for fast data transfer and firmware updates, up to 66 GPIOs, and a 14-bit ADC. Security remains a priority, with TrustZone isolation, tamper detection, and cryptographic accelerators included to meet modern IoT security standards.

Nordic Semiconductor’s nRF54LM20B is available in a CSP package (left) and QFN package (right). Images used courtesy of Nordic Semiconductor

A New Standard for Edge AI

The nRF54LM20B targets next-generation devices where privacy, low latency, and long battery life are essential. By running AI workloads locally on the Axon NPU, devices can respond in milliseconds without relying on cloud connectivity, while keeping sensitive user data on the device itself.

Nordic mentions that the nRF54LM20B low power AI SoC is currently sampling to selected customers, with broad availability expected in early Q2 2026. More information is available on the Products page and the Press Release page.