feature article
Subscribe Now

Qeexo Takes Misery Out of EdgeML

Startup Takes a Dose of its Own Medicine

“It’s what you learn after you know it all that counts.” — John Wooden

Invention, meet your mother, Necessity. It’s an oft-told tale. A programmer hacks together a tool to solve a particular problem, then realizes the tool has broader applicability than he thought. He refines it a bit so it can be used over and over. Sometimes, the tool is even more valuable than the work product. And sometimes, that insight leads to a whole new company. 

Just like it did with Qeexo

Sang Won Lee and his colleagues from Carnegie Mellon University had been working as programmers-for-hire for several years, specializing in clever ways to detect and characterize fingertip presses on smartphone screens, using just little microcontrollers. Their technique relied on some clever ML models and inference code, which ran on ARM Cortex-M0 and M4 MCUs. They productized that as FingerSense, and life was rosy. 

Problem was, the team had to redo everything for each new project. Every screen is different, every vendor wants something different, every sensor suite is different. That meant lots of travel, lots of on-site tweaking with customer hardware, and lots of sleepless nights. It was a traveling minstrel show called ML at the Edge. 

Time to automate the process. They polished up their in-house tools and the next year cranked out 56 new variants of FingerSense without ever leaving the office. The light dawned. “Hey, we need to productize this thing.” Thus was born a new product, AutoML

The idea is that you feed AutoML the data from your sensors (accelerometers, gyroscopes, thermometers, microphones, etc.) and let it build a model. A few sliders and radio buttons let you tweak sampling rate, weighting, your favorite algorithm, target MCU, permissible code footprint, and some other variables. Press the blue GO button and you’re done. AutoML hands back executable Cortex-M code, ready to download. It’s zero-coding for MCU-based ML. 

Sang says even if you are an experienced ML coder, AutoML is faster, and therefore, more profitable for programmers and their employers. There’s little to be gained from hand-crafting models and massaging input data. Let the tool do it based on observable criteria like code size and latency. And, if you’re not an experienced ML programmer, so much the better. AutoML can make you look like one to your boss. 

Like a lot of online development tools, there are three pricing tiers. Bronze level is free (for now) and comes with 2GB of online storage for sensor data and resulting models. Silver and Gold levels permit more simultaneous users, include more storage, more training, and more hardware support. Subscription pricing for the latter two tiers is negotiable. 

Machine learning is terra incognita for most of us, like DSPs and GPUs of past years, or like VR now. There’s demand for the talent but no supply. That makes automated tools like AutoML vitally important. Experts may sniff that it’s like putting training wheels on a Ducati. If you don’t know how to operate the machine, stay off it. But product deadlines won’t wait for us to come up to speed. Nobody complains about using a C compiler instead of an assembler, or Verilog instead of a protractor and mechanical pencil. Tools like AutoML raise the level of abstraction and increase productivity by broadening the developer base. It’s a gateway to a whole new world of ML at the edge. 

Leave a Reply

featured blogs
Nov 25, 2020
It constantly amazes me how there are always multiple ways of doing things. The problem is that sometimes it'€™s hard to decide which option is best....
Nov 25, 2020
[From the last episode: We looked at what it takes to generate data that can be used to train machine-learning .] We take a break from learning how IoT technology works for one of our occasional posts on how IoT technology is used. In this case, we look at trucking fleet mana...
Nov 25, 2020
It might seem simple, but database units and accuracy directly relate to the artwork generated, and it is possible to misunderstand the artwork format as it relates to the board setup. Thirty years... [[ Click on the title to access the full blog on the Cadence Community sit...
Nov 23, 2020
Readers of the Samtec blog know we are always talking about next-gen speed. Current channels rates are running at 56 Gbps PAM4. However, system designers are starting to look at 112 Gbps PAM4 data rates. Intuition would say that bleeding edge data rates like 112 Gbps PAM4 onl...

featured video

Product Update: Broad Portfolio of DesignWare IP for Mobile SoCs

Sponsored by Synopsys

Get the latest update on DesignWare IP® for mobile SoCs, including MIPI C-PHY/D-PHY, USB 3.1, and UFS, which provide the necessary throughput, bandwidth, and efficiency for today’s advanced mobile SoCs.

Click here for more information about DesignWare IP for 5G Mobile

featured paper

Streamlining functional safety certification in automotive and industrial

Sponsored by Texas Instruments

Functional safety design takes rigor, documentation and time to get it right. Whether you’re designing for the factory floor or cars on the highway, this white paper explains how TI is making it easier for you to find and use its integrated circuits (ICs) in your functional safety designs.

Click here to download the whitepaper

Featured Chalk Talk

Mom, I Have a Digital Twin? Now You Tell Me?

Sponsored by Cadence Design Systems

Today, one engineer’s “system” is another engineer’s “component.” The complexity of system-level design has skyrocketed with the new wave of intelligent systems. In this world, optimizing electronic system designs requires digital twins, shifting left, virtual platforms, and emulation to sort everything out. In this episode of Chalk Talk, Amelia Dalton chats with Frank Schirrmeister of Cadence Design Systems about system-level optimization.

Click here for more information