We often get asked about TinyML – what it is, how it works, how we at SensiML fit into the picture, and for some real-world examples of how it can it useful. Let’s start with the definition. For this we turn to the authoritative body, the tinyML Foundation (tinyml.org).
Here’s what they have to say:
“Tiny machine learning is broadly defined as a fast growing field of machine learning technologies and applications including hardware, algorithms and software capable of performing on-device sensor data analytics at extremely low power, typically in the mW range and below, and hence enabling a variety of always-on use-cases and targeting battery operated devices.”
That’s a pretty clear and succinct description, so hopefully so far so good. How does SensiML fit into the picture? It turns out (and not by accident), that our tools are exceptionally good at implementing sophisticated AI/ML algorithms on compute-limited platforms. These include nearly all 32-bit, 16-bit and even 8-bit microcontrollers – exactly the kind of devices used in most edge sensing applications. Our tools have also been designed to be easy to use, which means that developers building or upgrading data sensing IoT endpoints can quickly and efficiently add local machine learning capabilities to their designs. The net result is intelligent endpoints which can be easily implemented and consume little power running on low-cost standard off-the-shelf microcontrollers from a wide range of vendors.
These include stand-alone microcontrollers from rapidly growing list of companies including Broadcom, Infineon, Microchip Technology, Nordic Semiconductor, NXP, onsemi, QuickLogic, Silicon Labs, and ST Microelectronics as well as SoC-embedded processor IP such as ARM, ARC and RISC-V.
As for example applications, we’ve seen a wide range of those as well. They include smart IoT devices using audio “wake words” or other acoustic event triggers that invoke an action or notification when a keyword is spoken or specific sound is detected, or motion recognition systems that perform an action when a specific movement or gesture is detected. Another set of applications is industrial machine predictive maintenance in which sensors monitor equipment for known fault conditions or other anomalous behavior. Still other common uses are in smart building applications where low-cost passive IR sensors can provide contextual insight into building occupancy and usage for controlling HVAC, lighting, and building services without resorting to the invasive presence of cameras.
One thing these all have in common is the need for high quality predictive models that transform raw time-series sensor data streams into meaningful application insight on low power (often battery-powered) low-cost devices, and that’s exactly what we at SensiML can help enable.
Join SensiML at the tinyML Summit 2022
March 29-30
Hyatt Regency
San Francisco Airport
SensiML, a gold sponsor of TinyML Foundation, will be demonstrating the SensiML Analytics Toolkit and presenting:
“Suitability of TinyML for addressing predictive maintenance in high tech manufacturing”.
