Edge Impulse Flattening the Learning Curve for TinyML Deployment

Subscribe To Download This Insight

By Lian Jye Su | 4Q 2021 | IN-6317

This insight discusses the recent announcements from Edge Impulse Imagine 2021 and the current state of edge Machine Learning (ML) software and service industry.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


Key Announcements During Imagine 2021


Edge Impulse, a San Jose-based Machine Learning Operation (MLOps) platform vendor, held its first industrial event, Edge Impulse Imagine, in September 2021. Known for its focus on Tiny Machine Learning (TinyML), the young startup, founded in 2019, continues to bring new solution updates, hardware partners, and software upgrades that accelerate TinyML adoption and deployment:

  • Security certification: Edge Impulse will be American Institute of Certified Public Accountants (AICPA) SOC 2 certified by the end of 2022, allowing it to work with enterprises.
  • White-label software: Edge Impulse partners can now white-label all Edge Impulse solutions (Edge Impulse Studio, EON Compiler, EON Tuner) and make them a part of their offering.
  • New hardware partners: Edge Impulse continues to add more TinyML processor vendors as its hardware partners. Syntiant, a key TinyML processor vendor, is the latest addition to the list. In addition, the company also partners with ALIF Semiconductor to launch the first commercial Arm Cortex-M55 and Ethos-U55 processor. Arm Cortex-M55 is the first Arm Cortex-M series processor that features Arm Helium vector processing technology for high ML performance, while Ethos-U55 is an ultra-low powered ML accelerator.
  • Brand new models: Edge Impulse has developed a new object detection model for TinyML devices and a few-word multilingual keyword spotting model.

Edge Impulse's End-to-End Platforms for MLOps


At its core, Edge Impulse develops a low-to-no code visual MLOps platform, enabling the design and development of ML algorithms embedded in edge devices. Designed for all developers, Edge Impulse aims to help users unfamiliar with ML train, test, deploy, and maintain embedded ML models. By creating embedded production-grade ML models, developers and product manufacturers can bring intelligence to these devices via a firmware upgradse or over-the-air (OTA) updates.

In September 2020, Edge Impulse introduced Edge Optimized Neural (EON) Compiler, a proprietary compiler that optimizes large ML models for resource constraint devices. In addition, edge Impulse offers a Software Development Kit (SDK) and libraries that help quantize the models, reducing the memory and storage requirements while maintaining the same level of accuracy. Finally, for developers who look for more assistance during the development process, EON Tuner is Edge Impulse’s answer to AutoML; EON Tuner helps developers identify the right embedded ML models based on the target device, perform end-to-end optimization, and reduce the total time from data collection to commercial deployment.

All these announcements aim to accelerate the market adoption of TinyML. The TinyML industry has shown a lot of promise, as models are now applied to analyze various types of data sources, including temperature, vibration, audio, and vision, on resource-constrained hardware. Currently, there is significant momentum in consumer technologies, including smartphones, wearables, and True Wireless (TWS) earbuds. However, the next five years of TinyML development will be driven by wide-scale deployment of in industrial, healthcare, and smart city applications. Key use cases include low-powered machine vision, asset tracking, predictive maintenance, hearing aids, patient health monitoring and alert, and ambient scene recognition.

A Growing Market


Edge Impulse has positioned itself to capitalize on this growth. Edge Impulse software suite is handy for developers who are still exploring the concept of TinyML. As a result, the company is reporting strong growth in the number of developers on its platform. The promise of TinyML has also led to the emergence of several innovative startups targeting AutoML-based model optimization techniques and MLOps platforms for embedded ML applications. For example, Imagimob, Latent AI, and SensiML focus on providing MLOps platforms, while Neural Magic, Nota, Pumerai, and Qeexo offer specialized model optimization techniques for embedded ML models in end devices.

More importantly, all these platforms offer zero code or low code solutions that developers can use without extensive computing expertise. The future of TinyML lies in a higher level of automation through low or zero code design. This lowers the barrier to entry for end-users who do not possess data science or embedded ML expertise and enables them to perform MLOps in a seamless manner. Thus, through the right partnerships and collaboration, ABI Research believes that the global edge AI Software-as-a-Service (SaaS) and turnkey service market will grow at a Cumulative Average Growth Rate (CAGR) of 46% between 2020 and 2025. The market size is expected to reach US$7.2 billion by 2025, approximately 25% of the US$28 billion global edge AI market, which is also a market in which Edge Impulse is expected to be a big player and key influencer.