Uploaded March 2026 | Updated September 2026, 2 weeks ago
At Embedded World 2026, Electromaker visited the Nordic Semiconductor booth to see two approaches to edge AI on Nordic hardware. The demo compares Neuton models running on a CPU with the Axon neural processing unit found in the nRF54LM20B.
Neuton models run directly on the CPU of devices such as the nRF54L15. These models process sensor data such as accelerometer input to recognize gestures. The system identifies motion patterns including circles, swipes, and taps. The device sends only the recognized result over Bluetooth. Raw data stays on the device, which protects privacy and reduces energy use.
Neuton models are generated through the Nordic AI Lab platform. Developers upload labeled data sets and the platform produces optimized models automatically. These models reach similar accuracy to larger TensorFlow Lite models while remaining up to ten times smaller and faster.
The Axon NPU supports more demanding AI workloads. A wake word demo shows voice command recognition running either on the CPU or on the Axon NPU. CPU execution takes about 73 milliseconds per inference and consumes around 185 microcoulombs. The Axon NPU performs the same task in about 6.5 milliseconds and uses under 20 microcoulombs. Developers access both technologies through the nRF Connect SDK and supporting tools.
Learn more about Nordic AI tools and development platforms:
nordicsemi.com
Timestamps:
Introduction to Nordic Edge AI Demos – 0:00
Overview of Neuton Models and Axon NPU – 0:23
Gesture Recognition with nRF54L15 – 0:54
Edge AI Motion Detection Demo – 1:07
How Neuton Models Are Generated – 1:25
Running AI Inference on the CPU – 2:10
Why Edge AI Keeps Data Local – 2:35
Introduction to the Axon Neural Processing Unit – 3:26
Voice Command Demo with Wake Word – 4:30
Comparing CPU and NPU Performance – 5:27
Inference Speed and Energy Measurements – 5:55
AI Performance Gains with Axon NPU – 6:38
Developer Access Through Nordic AI Lab – 7:28
Tools and SDK Support for Axon – 8:05
▬ Support Us! ▬▬▬▬▬▬▬▬▬▬
For all the latest products, projects and articles, visit our website at electromaker.io
💡 **Stay Connected with Electromaker!**
youtube.com/channel/UCiMO2NHYWNiVTzyGsPYn4DA?sub_confirmation=1
facebook.com/electromaker.io
twitter.com/ElectromakerIO
linkedin.com/company/electromaker
At Embedded World 2026, Electromaker visited the Nordic Semiconductor booth to see two approaches to edge AI on Nordic hardware. The demo compares Neuton models running on a CPU with the Axon neural processing unit found in the nRF54LM20B.
Neuton models run directly on the CPU of devices such as the nRF54L15. These models process sensor data such as accelerometer input to recognize gestures. The system identifies motion patterns including circles, swipes, and taps. The device sends only the recognized result over Bluetooth. Raw data stays on the device, which protects privacy and reduces energy use.
Neuton models are generated through the Nordic AI Lab platform. Developers upload labeled data sets and the platform produces optimized models automatically. These models reach similar accuracy to larger TensorFlow Lite models while remaining up to ten times smaller and faster.
The Axon NPU supports more demanding AI workloads. A wake word demo shows voice command recognition running either on the CPU or on the Axon NPU. CPU execution takes about 73 milliseconds per inference and consumes around 185 microcoulombs. The Axon NPU performs the same task in about 6.5 milliseconds and uses under 20 microcoulombs. Developers access both technologies through the nRF Connect SDK and supporting tools.
Learn more about Nordic AI tools and development platforms:
nordicsemi.com
Timestamps:
Introduction to Nordic Edge AI Demos – 0:00
Overview of Neuton Models and Axon NPU – 0:23
Gesture Recognition with nRF54L15 – 0:54
Edge AI Motion Detection Demo – 1:07
How Neuton Models Are Generated – 1:25
Running AI Inference on the CPU – 2:10
Why Edge AI Keeps Data Local – 2:35
Introduction to the Axon Neural Processing Unit – 3:26
Voice Command Demo with Wake Word – 4:30
Comparing CPU and NPU Performance – 5:27
Inference Speed and Energy Measurements – 5:55
AI Performance Gains with Axon NPU – 6:38
Developer Access Through Nordic AI Lab – 7:28
Tools and SDK Support for Axon – 8:05
▬ Support Us! ▬▬▬▬▬▬▬▬▬▬
For all the latest products, projects and articles, visit our website at electromaker.io
💡 **Stay Connected with Electromaker!**
youtube.com/channel/UCiMO2NHYWNiVTzyGsPYn4DA?sub_confirmation=1
facebook.com/electromaker.io
twitter.com/ElectromakerIO
linkedin.com/company/electromaker
