.png)
Signal processing and machine learning touch almost every aspect of daily life but are largely unseen and often overlooked. This includes the phone calls we make (signal processing) to the recommendation engines (machine learning) that power Spotify or Netflix. It pervades all aspects of daily life and is only growing. And yet, much of the chatter in tech today (hi Twitter) is focused on size: what is the use case for bigger GPUs? Can we use more compute power to train larger neural nets? But today – right now – there are billions of devices with small microcontrollers. Think smartwatches, wearables, kitchen appliances, and all IoT devices. As chips get smaller the opportunity gets larger. With the right software, these devices have the compute power necessary for efficient but powerful use cases. These tiny low-power devices at the “edge” are the biggest frontier of machine learning and signal processing.
(To nerd out for a second) As an electrical engineer, I have a deep appreciation for the decades of research and advancements in embedded devices and the capabilities it has unlocked. When I was in graduate school, Intel paved the way with their 14 nanometer FinFET transistor, which was a groundbreaking announcement that resulted in a days-long debate on how much further Moore’s law could go. This month, less than a decade later, IBM announced a 2 nanometer chip, an unfathomably small size just a few years ago. For reference, this is about the same size as the width of DNA! Coupled with the well-documented advancements in AI/ML that have taken place in the last decade, we are at the beginning of an explosion of use cases for embedded machine learning.
When I met with Zach and Jan at Edge Impulse, it was clear that they deeply understood the expansionary opportunity of embedded machine learning. They built a platform that empowers developers to go from idea to creation and deployment within minutes. At Canaan, we believe in their vision of 1) empowering the developer, 2) embracing the complex hardware ecosystem, and 3) end-to-end application development. The combination of these together gives the company an unparalleled edge (pun intended) towards building the category defining company in embedded machine learning.
Empowering the developer: Today, thousands of developers (and enterprises) are using Edge Impulse for everything from personalized sleep monitoring to predictive analytics within industrial environments, where sparse connectivity and/or power constraints made it previously impossible to fully leverage device compute capabilities. Empowering developers is the key to unlocking expansive market potential - we saw this with PCs, mobile, and IoT. Now with Edge Impulse, developers can leverage the community of forums and educational courses to build the next generation of applications at the edge (22k+ and counting).
Embracing the complex hardware ecosystem: From the various Arduino and Raspberry Pi boards for consumers, to more custom chips by Nvidia and Arm used for enterprise and industrial use cases, the “where to start” question has often been the biggest bottleneck in application development at the edge. Zach, Jan and the team have taken the approach to partner with the complex hardware ecosystem and provide a unified developer platform that makes it easy to not only train and test your machine learning model, but optimize your application by helping you select the right hardware endpoint. Edge Impulse is enabling a rare win-win-win for the software developer, the hardware vendor, and the novel use case of embedded machine learning.
End-to-end application development: A big part of the “magic” of Edge Impulse is the end-to-end capabilities for their platform. Many of us hobby developers have had the experience of wanting to build a quick application around the house (e.g., automatic garage door opener triggered by your car’s movement in your security camera) but end up spending most of the time trying to integrate multiple software platforms together and ultimately give up. The problem in enterprises is multiplied with added complexities around custom sensors, security requirements, and power constraints. Edge Impulse’s focus on an end-to-end solution, where a developer can connect sensors, collect data, train models, and deploy them at the edge all on their platform truly is a “magical” experience.
Today, we have super computers in our home, in our pocket, on our wrist, and even on our finger. As we push the boundaries of next generation chips and processors as well as AI/ML, the biggest opportunity will be unlocking the power of machine learning on the tiniest of devices. By doing so, Edge Impulse has the power to truly democratize machine learning for everyone. That is a mission we at Canaan are excited to back with our friends at Acrew Capital, Fika Ventures, Momenta Ventures, and Knollwood Investment Advisory. Thank you Zach and Jan for letting us be a part of the Edge Impulse journey.