Product Name: nRF54LM20B

Manufacturer: Nordic Semiconductor

Product Category: Computing Hardware, Software and Systems

Supporting Documentation (if available)

What is the nRF54LM20B?

The nRF54LM20B is an ultra-low-power, large-memory wireless SoC delivering AI intelligence to the smallest battery-powered devices. It marks the next generation of on-device AI performance for its power class, enabling real-time audio, image, or high-rate time series data processing, while dramatically accelerating and simplifying AI development at all levels.

This is achieved through an advanced SoC architecture, including the first integration of Nordic’s Axon Neural Processing Unit (NPU) hardware accelerator, high-memory availability (2MB NVM and 512KB RAM), market-leading energy efficiency (193 CoreMark/mA at 3V), and multiprotocol connectivity. SoC design is combined with Nordic’s proprietary ultra-tiny custom Neuton AI models and access to Nordic’s Edge AI Lab Development tool, an automated no-code platform for building these compact AI models, as well as complete SDK and cloud lifecycle support.

Introducing a new level of AI support and performance

Until now, adding AI capabilities to wireless low-power IoT devices meant choosing between acceptable battery life and performance. While discrete NPUs are available, they add cost and complexity. Running TensorFlow Lite models on the CPU is too slow and consumes high amounts of memory, and both approaches ultimately spend too much of the power budget.

The nRF54LM20B addresses these limitations by integrating an NPU directly into an ultra-low-power SoC. The Axon NPU offloads inference workloads that would otherwise be run on the CPU, delivering 15x faster inference performance while significantly improving energy efficiency. Continuous or near-real-time AI processing can now run on battery-powered devices while maintaining long operational lifetimes. By processing data locally rather than transmitting raw sensor data continuously to the cloud, the device also reduces bandwidth usage and connectivity dependency, lowers latency, improves responsiveness, and minimizes exposure of sensitive data.

The SoC takes full advantage of Nordic’s proprietary ultra-tiny Neuton models. Unlike conventional models requiring large memory footprints and significant optimisation, Neuton models are typically under 5 KB and up to 10x smaller, faster, and more efficient than other CPU-run models.

Simplifying things further still, Nordic’s Edge AI Lab provides a no-code workflow for automatically generating AI models – requiring little-to-no ML expertise. The platform automates signal processing, feature extraction, and model optimization, saving significant time and resources. Teams can deploy lightweight Neuton models directly on the CPU for simpler AI tasks, or generate accelerated models optimized for execution on the Axon NPU.

These capabilities are all further supported by Nordic’s comprehensive development ecosystem, including nRF Connect SDK, cloud services, and device management infrastructure. Together, these capabilities enable Nordic to offer an end-to-end wireless development platform spanning hardware, AI acceleration, embedded software, connectivity, security, cloud integration, and lifecycle device management. Scalable edge AI development has never been more accessible.

nRF54LM20B performance highlights & specs

The proprietary Axon Neural Processing Unit (NPU) specializes in accelerating edge AI models, including TensorFlow Lite. It delivers up to 15x faster inference performance than CPU-only execution, with up to 7x higher performance and 8x better efficiency than the closest competing wireless NPU.

This, in combination with its high-memory availability of 2MB NVM and 512KB RAM, and 128 MHz Arm® Cortex-M33 processor and RISC-V coprocessor, delivers an SoC that sets a new standard for ultra-low-power edge AI hardware design.

The SoC also integrates Nordic’s fourth-generation ultra-low-power 2.4 GHz radio, supporting Bluetooth® LE, Bluetooth Channel Sounding, Matter over Thread, Zigbee, as well as various connectivity ecosystems. It has a broad range of peripherals, including a high-speed USB and up to 66 GPIOs, particularly useful for next-gen human interface devices, gaming peripherals, or smart home devices (such as Matter implementations), all of which will increasingly rely on edge-AI capabilities and sensing.

It also features TrustZone® isolation, tamper detection, and cryptographic engine with side-channel leakage protection.

Performance highlights

Processing efficiency and performance: 193 CoreMark/mA @ 3 V, 503 CoreMark®

Enhanced edge AI capabilities thanks to integrated Axon NPU

Radio power consumption: 3.4 mA for RX and 4.8 mA for TX @ 0 dBm (@ 3 V)

Sleep modes power consumption: from 0.7 μA to 4.0 μA (@ 3 V)

Maximum TX power: +8 dBm (CSP) / +7 dBm (QFN) 

Sensitivity: -96 dBm (1M Bluetooth LE), -101 dBm (802.15.4)

Optimized for the lowest total average current



Key benefits

Increased speed and efficiency of local AI inference tasks, unlocking new application possibilities

Reduced design complexity, size, and cost

Reliable and robust wireless connectivity

Extended battery life enabled by lower power AI processing and wireless operation

Secure on-device processing and storage of sensitive data

Simplified embedded AI development through Neuton models and Edge AI Lab

Complete development support through Nordic’s chip-to-cloud ecosystem