Nordic Semi nRF54LM20B wireless SoC integrates 128 MHz Axon NPU for Edge AI workloads

Abstract

Nordic Semiconductor has introduced the nRF54LM20B wireless SoC, the first nRF54L microcontroller to integrate a dedicated 128 MHz Axon Neural Processing Unit (NPU) for high-efficiency edge AI tasks. This NPU delivers substantial performance benefits, claiming up to 7 times faster speeds and 8 times higher energy efficiency compared to competing wireless solutions. The SoC maintains a heterogeneous architecture, combining a 128 MHz Arm Cortex-M33 CPU with a 128 MHz RISC-V coprocessor, while offering extensive 2.4 GHz radio connectivity.

Report

Key Highlights

• AI Integration: The nRF54LM20B is the first nRF54L series chip to embed the 128 MHz Axon NPU specifically for offloading Edge AI workloads like sound classification and image detection.
• Performance Gains: The Axon NPU offers exceptional efficiency, reported to be up to 7x faster and 8x more energy-efficient than unnamed competing wireless solutions, and 15x faster than running the same tasks purely on the Cortex-M33 CPU.
• Heterogeneous Architecture: The SoC includes a 128 MHz Arm Cortex-M33 main processor and a 128 MHz RISC-V coprocessor (FLPR).
• Connectivity: Features an ultra-low-power 2.4 GHz radio supporting modern protocols including Bluetooth LE, Bluetooth Channel Sounding, Matter over Thread, Zigbee, Aliro, and Amazon Sidewalk.
• Software Ecosystem: Supported by the nRF Connect SDK and the "Nordic Edge AI Lab," which facilitates the generation of ultra-tiny Neuton models (typically under 5 KB).
• Availability: The nRF54LM20B is currently sampling, with general availability scheduled for Q2 2026.

Technical Details

Implications

• Mainstreaming Edge AI: The inclusion of a powerful, dedicated NPU in a standard ultra-low-power wireless MCU confirms that complex AI/ML inference (like keyword spotting and biometric monitoring) is becoming a prerequisite for competitive IoT devices, moving away from CPU-only implementations.
• Validation of RISC-V in Heterogeneous Design: The nRF54LM20B continues Nordic's strategy of utilizing a high-performance RISC-V coprocessor (FLPR) alongside the primary Arm core and the new Axon NPU. This reinforces the RISC-V architecture's strength as a specialized, low-power auxiliary processor for real-time tasks within complex commercial embedded systems.
• Increased Competition in Low-Power Silicon: The significant performance and energy efficiency claims set a new benchmark for wireless SoCs, pressuring competitors to accelerate the integration of specialized AI hardware to remain competitive in the IoT and consumer electronics markets.
• Developer Tooling Focus: The introduction of the Nordic Edge AI Lab and ultra-small Neuton models demonstrates Nordic's commitment to providing a complete, accessible toolchain necessary for developers to quickly deploy efficient, cloud-independent AI solutions on their hardware.