Cloud Hosting

Microchip has announced the launch of a new software package designed to put machine learning workloads onto eight-, 16-, and 32-bit microcontrollers and processors: the MPLAB Machine Learning Development Suite.

"Machine Learning is the new normal for embedded controllers, and utilizing it at the edge allows a product to be efficient, more secure and use less power than systems that rely on cloud communication for processing," claims Microchip's Rodger Richey of the core benefits behind on-device machine learning with resource-constrained hardware, known as "tinyML." Microchip's unique, integrated solution is designed for embedded engineers and is the first to support not just 32-bit MCUs and MPUs [Microcontroller Units and Microprocessor Units], but also 8- and 16-bit devices to enable efficient product development."

Designed for use alongside the MPLAB X Integrated Development Environment (IDE), the machine learning toolkit allows the developer to build machine learning models suitable for flashing to Microchip's various microcontroller and processor parts taking into account their limited resources compared to desktop computers or cloud servers. Driven by AutoML and with the option to use cloud computing resources to find the best algorithm for a given task, the package aims to cover feature extraction, training, validation, and testing in one, with an application programming interface (API) convertible to Python.

While Microchip had already supported the use of existing deep neural network (DNN) models from TensorFlow Lite on its microcontrollers, the launch of the MPLAB Machine Learning Development Suite demonstrates a desire to provide everything a developer needs to build something from the ground up and joins MPLAB Harmony V3 and the VectorBlox accelerator Software Development Kit (SDK), the latter designed for use with Microchip's various field-programmable gate array (FPGA) parts, in the company's on-device machine learning line-up.

The software is free for trial use on up to 1GB of data and with 2,500 labels plus five hours a month of AutoML CPU time, but no rights to deploy models for purposes other than evaluation; a standard license offers 10GB of data, unlimited labels, and 10 hours a month of CPU time, plus a license to deploy models in production for $89 a month; a "pro" license increases the CPU time to 250 hours a year (20.8 hours a month) and offers the option to output source code, rather than a pre-compiled library.

More information on the MPLAB Machine Learning Development Suite is available on the Microchip website, along with a getting-started guide which walks the reader through creating models for fan state monitoring and gesture recognition and running them on SAM D21 and AVR devices.

See the article here:
Microchip Launches the MPLAB Machine Learning Development Suite for 8-, 16-, 32-Bit MCUs and MPUs - Hackster.io