Arm Cortex-M MCUs Can Get an Instant Machine Learning and Inference Facelift with NanoEdge AI

February 26, 2020 by Gary Elinoff

Cartesiam's NanoEdge AI Studio simplifies the task of integrating machine learning directly into everyday edge devices that are based on Arm Cortex-M MCUs.

Cartesiam, a startup focused on AI at the edge, has announced NanoEdge AI Studio, the first integrated development environment (IDE) that enables easy machine learning (ML) and inference directly on Arm Cortex-M MCUs. 


What NanoEdge AI Does

NanoEdge AI Studio makes it easy to build a machine learning static library to embed within programs running on Arm Cortex-M MCUs. MCUs embedded within edge devices can be empowered to locally learn, infer, and predict from directly inside the microcontroller, interfacing with the physical environment in the most direct manner possible.

NanoEdge AI Studio runs entirely on the user's Windows or Linux PC. The studio offers heightened security, simply because no data is transmitted to the cloud where it might be intercepted.


NanoEdge AI Studio MCU

NanoEdge AI Studio works with Arm Cortex-M MCUs. Image used courtesy of Cartesiam

Once the designer has described the eventual environment, NanoEdge AI Studio will automatically calculate, optimize, and test the best algorithmic solution. The chosen algorithm can be embedded as a C library into the MCU. Libraries as small as 4K are possible.


How NanoEdge AI Does It

There are five basic steps need to impart AI capabilities to edge devices using NanoEdge AI Studio.

  1. Choose an Arm Cortex-M MCU—M0 to M7—and specify the amount of RAM and the type of sensor to be employed.
  2. Provide contextual sensor data in the form of regular and abnormal signals.
  3. Automatically get a useful library combination out of 500 million possibilities.
  4. Before compilation, test the produced NanoEdge AI library within the emulator. No need for a physical MCU.
  5. Download the final static library to the target MCU. It’s ready to learn and infer.

This solution is designed to slash both the cost and development times necessary to bring unsupervised learning, inference, and prediction to edge devices. These advantages are geared to make AI capabilities practical for devices where low costs and low power consumption is essential.

As Marc Dupaquier, general manager and co-founder of Cartesiam, puts it, “Cartesiam’s NanoEdge AI Studio offers a completely different approach, with a cost- and time-efficient and self-learning AI.”


Diagram of how NanoEdge AI Studio works

Diagram of how NanoEdge AI Studio works. Image used courtesy of Cartesiam

He goes on to state that “It allows any embedded designer to develop application-specific machine learning libraries quickly and run the program inside the microcontroller right where the signal becomes data. It’s the only solution that can run both machine learning and inference on the microcontroller.”


No AI Scientist Needed

Until now, implementing AI in embedded devices such as Arm MCUs has been an expensive, time-consuming process, requiring the expertise of very specialized data scientists. But now, as Cartesiam explains, the NanoEdge AI Studio allows “your embedded developer [to] become your AI developer.” 


Benchmark NanoEdge AI

Benchmark within the NanoEdge AI software. Screenshot used courtesy of Cartesiam


A specialized data scientist is no longer required. 


The Security Benefits of AI at the Edge

Speed is an obvious benefit of having AI capacity at the edge where electronic devices interface directly with the physical world. But there is the vitally important security factor to consider. 

As described by Marc Dupaquier, “As far as security and privacy are concerned, learning an initial state locally reduces data exchanges over the network and prevents risk of falsification or intrusion.”


Proven Field Applications

Cartesiam has revealed a few proven field applications for NanoEdge AI Studio. Some of these applications have looked like: 

  • An air conditioner using AI to inform when air filters need replacement
  • Lithium-ion batteries monitoring at the cell level and using AI to warn of potential fire risk
  • Home appliances becoming smart home appliances and improving service
  • Bob Assistant learning engine pattern and alerting users of anomalies
  • Lacroix Electronics perfecting the maintenance of its reflow ovens


A demonstration of how NanoEdge AI and STM32Cube.AI can work hand in hand

A demonstration of how NanoEdge AI and STM32Cube.AI can work hand in hand. Screenshot used courtesy of STMicroelectronics

Cartesiam allows embedded designers to try NanoEdge AI for free, enabling them to create machine learning libraries for autonomous learning and inferring at the edge.