Ultra-Low-Power Wireless SoC Integrates an NPU

March 26, 2026

Blog

The Nordic Semiconductor nRF54LM20B is a wireless SoC in the company’s nRF54L series that combines an MCU, low-power wireless connectivity, and on-chip AI acceleration. The device integrates up to 2 Mbytes of nonvolatile memory and 512 kbytes of RAM, the largest memory configuration in the series.

It also includes an expanded peripheral set, high-speed USB, and support for as many as 66 GPIOs, allowing it to serve as a central controller in embedded designs that require multiple interfaces.

An integrated Axon NPU is included to accelerate machine-learning inference at the edge. Offloading inference workloads from the CPU to the NPU can improve execution efficiency and increase inference performance by roughly an order of magnitude compared with running the same workloads on the CPU alone. The architecture is intended to allow AI tasks to be handled locally without relying on cloud resources.

Wireless connectivity is handled through an integrated 2.4-GHz radio supporting Bluetooth LE as well as protocols used in IoT networking such as Matter, Thread, Zigbee, and proprietary 2.4-GHz links with data rates up to 4 Mbits/s. Additional capabilities include Bluetooth Mesh and Channel Sounding. When paired with companion devices from the Nordic Semiconductor nRF70 Series, the platform can also support Wi-Fi connectivity.

Radio power consumption is specified at roughly 3.4 mA in receive mode and 4.8 mA when transmitting at 0 dBm. Sleep-mode current ranges from about 0.7 to 4.0 µA at 3 V, depending on the operating state. Maximum transmit power reaches +8 dBm in the CSP package and +7 dBm in the QFN version. Receiver sensitivity is specified at -96 dBm for 1-Mb/s Bluetooth LE operation and -101 dBm for 802.15.4.

The device is intended to simplify embedded designs that require wireless connectivity and local AI processing by integrating compute, radio, and acceleration resources on a single chip. This approach can reduce external components while supporting battery-powered operation through relatively low active and standby power consumption. The architecture also incorporates hardware security features intended to help designers meet common IoT security and regulatory requirements.