Eta Compute’s Low-Power Architecture Offers Long-Lasting AI Vision Workloads on a Single Battery

By Saumitra Jagdale

Freelance Technology Writer

August 12, 2021


Eta Compute’s Low-Power Architecture Offers Long-Lasting AI Vision Workloads on a Single Battery

Designing low-power architectures with intensive computational capacities decides the feasibility of machine learning applications. It comes with the challenge of accommodating high processing power and a more extended battery without compromising the compact form factor. So, Eta Compute’s design technology supports developers with limitless application options while consuming minimal power along with the complete assistance of the Edge Impulse machine learning development platform.

Image Credit: Product Page

Designing and prototyping for Machine Learning on embedded devices have always been complex. To address these challenges, developers need to step up and meet the following demands when designing for embedded devices:

  • High processing power
  • Low power devices
  • Time to market
  • Lower the overall cost and risk of the system
  • Longer battery life

Today, AI technology is gaining momentum in embedded vision and image processing applications as developers increasingly employ deep learning and neural networks to improve object detection and classification. However, running computationally intensive tasks like vision algorithms on embedded devices is a challenge in itself. Additionally, the power consumption required to achieve the processing performance has rendered these capabilities impracticable for many potential use cases. 

This drive to integrate visual intelligence into the end devices gave rise to the field of embedded vision. Thus, neural network and embedded software designers are looking for practical ways to develop machine learning for edge applications that are less tedious and time-consuming.  

Fortunately, there exists a solution to these issues. The Eta Compute’s ECM3532 AI vision board, together with Edge Impulse’s embedded machine learning platform, makes vision feasible in many new power-sensitive applications. The AI vision board tackles every aspect of designing and building an ML application for IoT and low-power edge devices. Therefore, embedded software designers can simplify successful integrations into embedded edge devices, saving months of development time.

The ECM3532 AI Vision Board is excellent for prototyping, field testing, and deployment of AI embedded vision applications due to its compact form factor, embedded battery, low-power IoT, and Bluetooth low-energy connection. Its ultra-low-power operation overcomes the limitations of tethered systems with short battery life and high power consumption.

Deeper Look into AI Vision

The ECM3532 TENSAI SoC targets AI workloads and feature a hybrid multicore Cortex-M3 microcontroller with a clock speed of 100 Mhz and an ultra-low-power CoolFlux 16-bit DSP dedicated to speed up ML operations. The compact (1.5”x1.5”) AI Vision board performs efficiently with neural networks due to the embedded ECM3532 Neural Sensor Processor built with unique self-timed continuous voltage and frequency scaling at the run time using Eta Compute’s CVFS technology. This produces optimized energy for inference of ultra-low-power machine learning algorithms. 


The main highlight of the onboard sensors is a Himax HM0360 camera enabling a range of vision-based applications such as image classification, person and object detection, and counting. There is also a built-in microphone for sound classification and keyword spotting, as well as a Texas Instruments OPT3001 Ambient Light Sensor and a 6-axis accelerometer/gyroscope for gesture and defect detection.

With 64 Mbit of serial flash, there is ample space for your tinyML models and data logging. Despite its compact size and low power consumption, the board enables vision applications that last for months on a single CR2032 battery. The board also has an extension that simplifies the addition of other RF interfaces through a Micro SD card slot. 

Eta Compute also comes with a SparkFun FTDI Basic Breakout -3.3V board, enabling micro-USB programming. For debugging, the board also has serial wire debug (SWD) CoreSight connectors for both the ECM3532 and the A31R118. For this, a separate JLink or Ulink probe is recommended.

Edge Impulse’s Software Support to Deploy Vision Workloads

The ECM3532 AI vision board is versatile and comes with the support of Edge impulse’s machine learning development platform to deploy ultra-low-power computer vision applications at the edge. This technology is used for rapid neural network development to make energy-efficient vision endpoints seamless. Thus, it allows data sampling, building models, and deploying trained ML models. 

The firm also partnered with Eta Compute to incorporate their TENSAI Flow software. This enables the developers to create the next generation of intelligent device solutions with embedded machine learning.  

The TENSAI Flow software minimizes development risk by verifying feasibility and proof of concept. Hence, it enables the seamless development of machine learning applications for IoT and low-power edge devices. It comes with a neural network compiler, a neural network zoo, middleware that includes FreeRTOS, HAL, sensor frameworks, and IoT/cloud capabilities. The TENSAI Platform allows AI performance in the 1mW range, which is exceptionally low compared to existing systems, particularly for image processing.

Image Credit: Embedded

TENSAI Flow allows developers to securely gather and store training data via its interface with Edge Impulse, allowing clients to train once and keep real-world models for future development. For Eta Compute's TENSAI SoC, the software automatically optimizes TensorFlow Lite AI models, resulting in high optimization and power efficiency.

Using the TENSAI Flow, the TENSAI processor can load AI models that easily incorporate sensor interfaces. TENSAI Flow lays the groundwork for automatically provisioning and connecting devices to the cloud. As well as updates firmware over the air when new models or data become available.


With the development of extremely powerful, low-cost, and energy-efficient processors, embedded vision systems are gaining momentum. The ECM3532 AI vision board is an ultra-low-power AI platform with sensors that can run multiple algorithms, focusing on embedded vision. This Neural Sensor Platform provides more significant processing at the industry's lowest power profiles. As a result, battery capacity is no longer a barrier to IoT consumer and industrial applications. 

Hence, Eta compute has enabled its ECM3532 AI Vision Board to reduce risk and costs while accelerating development time for AI vision solutions.

For more detailed information, you can head over to the official page of Eta Compute.

Saumitra Jagdale is a Backend Developer, Freelance Technical Author, Global AI Ambassador (SwissCognitive), Open-source Contributor in Python projects, Leader of Tensorflow Community India, and Passionate AI/ML Enthusiast.

More from Saumitra