Edge refers to the edge of a network, relative to data centers or cloud computing, meaning the edge of a network (e.g., the Internet). Compared to large models (e.g., ChatGPT), Edge AI can run on edge devices, and even on small sensors.
The development of Edge AI benefits from the continuous enhancement of computing power in microcontrollers (MCUs) or microprocessors (MPUs), providing computational support for data processing and analysis software. Additionally, the development of the Internet of Things (IoT) has also promoted the growth of Edge AI, allowing for the deployment of Edge AI using existing IoT infrastructure, enabling more efficient data processing and transmission.
Edge AI involves knowledge from multiple disciplines, such as data analysis, machine learning algorithms, and compilers. It is challenging for an individual or a company to possess comprehensive knowledge to develop Edge AI products, creating innovative development opportunities for the toolchain in Edge AI product development. The toolchain helps accelerate the implementation of Edge AI projects. Moreover, several solution companies have emerged focusing on Edge AI.
“To do a good job, one must sharpen their tools.” This article reviews several Edge AI development platforms.
Edge Impulse
Edge Impulse is a machine learning development tool provider that offers a suite of tools for enterprise teams to build and deploy machine learning models on embedded devices. Edge Impulse provides powerful automation and low-code features, making it easier to create valuable datasets and develop advanced artificial intelligence for edge devices. It has been used by wearable health device manufacturers such as Oura, Know Labs, and NOWATCH, as well as industrial organizations like NASA, top chip suppliers, and over 100,000 developers across more than 250,000 ML projects, establishing Edge Impulse as a trusted platform for enterprises and developers.
Main Applications: Predictive maintenance, asset tracking & monitoring, human & animal perception, etc.
SensiML
SensiML simplifies the software tools for TinyML code development. The SensiML Analytics Toolkit provides an end-to-end development platform covering data collection, labeling, algorithm, and firmware auto-generation and testing, addressing the most common failure issues in AI projects through unique data collection and labeling methods. SensiML offers comprehensive functionalities, from simple point-and-click model creation to complete pipeline control for model building and testing workflows.
The SensiML Toolkit supports various Arm® Cortex®-M and above microcontroller cores, Intel® x86-based CPUs, and heterogeneous core SoCs like Quicklogic’s S3 platform, and is optimized for FPGAs.
Main Applications: Predictive maintenance, anomaly detection, behavior recognition, gesture recognition, motion analysis, smart lighting control, smart cities and urban roads, wearable devices, biometrics, acoustic feature detection, multi-device sensor fusion, process control and detection, etc.
Stream Analyze
Stream Analyze provides an end-to-end platform for developing Edge AI solutions, capable of efficiently developing, training, deploying, and coordinating analysis, computation, and AI models. This platform is designed for resource-constrained edge devices, including MCUs based on Arm Cortex-M and Cortex-A. It combines real-time stream analysis with device edge AI capabilities. The real-time data stream architecture used is specifically designed for edge devices and microcontrollers, enabling streaming analysis on very resource-limited devices. The platform is easy to learn and use, suitable for data scientists, engineers, domain experts, and others without deep coding skills. The main target customers of Stream Analyze are large industrial and automotive companies, enabling a wide range of use cases in areas such as product intelligence for new services, product usage analysis in product development, operational improvements, predictive maintenance, etc.
Stream Analyze also offers a free community version tool for real-time data stream management and analysis on edge devices, with an interactive environment that significantly shortens the model development process, achieving rapid and efficient progress from concept to deployment.
Main Applications: Mining loaders, autonomous vehicles, lawnmowers
221e
221e is a global leader in intelligent device precision sensing solutions. The AI sensor fusion algorithms provide exceptional accuracy and consistent results, with significant cost advantages. Since 2012, it has been providing innovative sensing solutions powering various products, including wearable devices, sporting equipment, and industrial applications.
Main Applications: Fitness and sports science, personal protective equipment, entertainment and VR, health and biomedicine, automotive and mobility, automation and robotics, environmental quality monitoring, construction and civil engineering, equipment and machinery, etc.
Octonion
Octonion provides industrial machine intelligence solutions for continuous diagnostics and predictive machine lifespan at the network edge. Having invested in embedded AI R&D activities five years ago, Octonion focuses on developing sensor data analysis software optimized for low-power microcontrollers, accumulating deep expertise in unsupervised machine learning, AI model personalization, and industrial equipment behavior prediction.
Main Applications: Industrial predictive maintenance, etc.
Qeexo AutoML
Qeexo is a company that provides machine learning automation solutions for embedded edge devices (Cortex M0-M4). The one-click fully automated Qeexo AutoML platform allows customers to quickly create machine learning solutions using sensor data in highly constrained environments, applicable in industries, IoT, wearable devices, automotive, mobile devices, and more. Over 300 million devices worldwide are equipped with AI based on Qeexo AutoML.
On January 4, 2023, TDK Corporation acquired Qeexo. Qeexo will become a wholly-owned subsidiary of TDK.
Main Applications: Industrial & manufacturing, transportation & automotive, smart buildings, energy & power generation
Archetype AI
Archetype AI is a cutting-edge AI company focused on developing Physical AI foundational models. Physical AI is a new type of AI that can perceive, understand, and reason about the surrounding world. The “Large Behavior Model” (LBM) developed by Archetype AI is a multimodal AI foundational model that integrates real-time sensor data and natural language, revealing hidden behavioral patterns in unstructured sensor data.
Main Applications: TVs, smart speakers, and smart appliances, etc.
NanoEdge AI Studio
NXP Semiconductor’s NanoEdge™ AI Studio is a new machine learning (ML) technology that allows end-users to easily enjoy true innovation outcomes. In just a few steps, developers can create the best ML library for their projects based on minimal data.
One of the major advantages of NanoEdge™ AI Studio is that it requires no specialized data science skills to use. Any software developer using the Studio can create optimal ML libraries without AI skills through its user-friendly environment.
Main Applications: Predictive maintenance, environmental awareness, human activity, asset tracking, human-machine interfaces, object detection, biometrics, image classification, etc.
Imagimob
Imagimob is a fast-growing startup dedicated to driving innovation in Edge AI and tinyML, supporting future smart products. The company is headquartered in Stockholm, Sweden, and has been serving global customers in the automotive, manufacturing, healthcare, and lifestyle industries since 2013. In 2020, Imagimob launched IMAGIMOB Studio for rapidly and easily developing end-to-end Edge AI applications for resource-constrained devices. IMAGIMOB Studio provides guidance and support throughout the development process, increasing productivity and accelerating time to market.
In May 2023, Infineon Technologies acquired 100% of Imagimob. Imagimob is part of the Connected Secure Systems (CSS) division.
Main Applications: Audio classification, predictive maintenance, gesture recognition, signal classification, fall detection, material detection, etc.
Reality AI
Reality AI offers extensive embedded AI and TinyML solutions for advanced non-visual sensing in automotive, industrial, and consumer products, perfectly compatible with Renesas embedded processing and IoT products. They provide advanced mathematical methods for signal processing to enable fast and efficient machine learning inference, applicable to small MCUs and powerful MPUs. Using Reality AI Tools®, a software environment supporting the complete product development lifecycle, users can automatically explore sensor data and generate optimized models. Reality AI Tools include analytical features that can find the best sensors or sensor combinations, sensor placement locations, and automatically generate component specifications, fully explaining model functions in time and frequency domains.
In July 2022, Renesas Electronics acquired the excellent supplier of embedded AI solutions, Reality Analytics, Inc. (Reality AI), making it an indirect wholly-owned subsidiary of Renesas Electronics.
Main Applications: Motor control (predictive maintenance, anomaly detection, and intelligent control feedback), HVAC, automotive safety alerts (horns, distance detection, positioning), etc.
eIQ® ML
eIQ is NXP’s machine learning tool. The NXP® eIQ® machine learning (ML) software development environment supports the use of ML algorithms on NXP EdgeVerse™ microcontrollers and microprocessors (including i.MX RT crossover MCUs and i.MX series application processors). The eIQ ML software includes ML workflow tools called eIQ Toolkit, as well as inference engines, neural network compilers, and optimization libraries.
The eIQ Toolkit supports machine learning development using an intuitive GUI (eIQ Portal), development workflow tools, and command-line host tool options (as part of the eIQ ML software development environment). The eIQ Portal is an intuitive graphical user interface (GUI) developed in exclusive collaboration with Au-Zone Technologies, simplifying ML development.
The output software provided by the eIQ Portal can seamlessly output to DeepViewRT™, TensorFlow™ Lite, TensorFlow Lite Micro, Glow, Arm NN, and ONNX runtime inference engines.
Main Applications: Object detection and recognition, voice command and keyword recognition, anomaly detection, image and video processing, others (smart wearables, smart factories and buildings, healthcare and diagnostics, augmented reality, logistics, public safety).
Conclusion
The toolchain for Edge AI mainly focuses on the development of AI software algorithms, including data analysis processing, model training, model conversion, and model deployment. Due to the specialization and complexity of Edge AI development, software platforms oriented towards AI development have emerged, simplifying the open process and making it easier to develop Edge AI-related products. With the demands for model training, iteration, and upgrades in Edge AI, SaaS platforms oriented towards Edge AI have emerged, providing software as a service. With the already deployed IoT infrastructure, Edge AI will empower more devices with autonomous intelligence.