ROHM Semiconductor has announced new microcontrollers featuring built-in Artificial Intelligence capabilities, designed to execute both learning and inference tasks independently, without the need for cloud or network connectivity. These devices, part numbers ML63Q253x-NNNxx and ML63Q255x-NNNxx, represent a significant first: they enable real-time fault prediction and degradation forecasting directly on devices such as industrial motors and equipment, thereby addressing long-standing automation challenges around predictive maintenance and operational downtime.
The microcontrollers (MCUs) utilize a dedicated 3-layer neural network algorithm, implemented via ROHM’s proprietary Solist-AI solution and powered by the AxlCORE-ODL accelerator. This architecture achieves up to 1,000x higher Artificial Intelligence processing speed compared to ROHM’s software-based MCUs at the same clock frequency. The MCUs leverage a 32-bit Arm Cortex-M0+ core running at 48MHz, with integrated CAN FD controller, motor control PWM, and dual A/D converters. Their power consumption is kept around 40mW, making them notably energy efficient for a broad spectrum of embedded and industrial uses. The design specifically supports scenarios where traditional network-dependent and cloud-heavy Artificial Intelligence models fall short, such as factory floors with variable or restricted connectivity, or legacy equipment in need of retrofitting.
ROHM´s Artificial Intelligence MCUs will be available in a portfolio of 16 models, covering diverse memory sizes, packages, and pin counts. Eight models in TQFP package have entered mass production, with evaluation boards and select devices now purchasable via online distributors. The ecosystem is supported with accessible development tools, including the Solist-AI Sim simulator for pre-deployment evaluation and the Solist-AI Scope real-time viewer for effectiveness assessment. Potential application areas span from condition monitoring of motors and batteries to advanced anomaly detection in residential appliances, industrial robotics, and factory automation sensors, all benefitting from localized learning and adaptability to installation-specific variations.
