Development of a microcomputer equipped with AI functions - Industry first* - Learns and infers without the need for a network, predicts equipment anomalies -

*March 18, 2025, based on ROHM research on microcontroller products

ROHM Co., Ltd. has developed the ML63Q253x-NNNxx / ML63Q255x-NNNxx (hereafter referred to as AI microcontrollers), a microcontroller equipped with AI (artificial intelligence) functions that enables failure sign detection and deterioration prediction using sensing data in all types of equipment, including industrial equipment such as motors. This product is the industry's first microcontroller that can perform learning and inference independently without the need for a network.

The new product is a unique on-device AI solution calledSolist-AI Soloist AITo achieve this, a simple three-layer neural network algorithm is used. Learning and inference are possible on a standalone microcontroller, without relying on the cloud or a network.

Typically, endpoint AI trains in the cloud and performs inference on the device, so it requires a network connection, and these processing models perform inference in software, requiring GPUs and high-performance CPUs.
On the other hand, ROHM's AI microcontrollers are endpoint AI, but because on-device learning allows the microcontroller to perform both learning and inference independently, they can flexibly respond to variations in the installation environment and even within the same model. The proprietary AI accelerator "AxlCORE-ODL" speeds up AI processing by approximately 1,000 times compared to ROHM's microcontrollers that use conventional software methods (conditions: theoretical value at 12MHz), and detects abnormalities that are "different from usual" in real time and outputs them as numerical values. In addition, because high-speed learning (on-site learning) is possible at the installation site of the equipment, it can also be retrofitted to existing equipment.

The AI microcontroller is equipped with a 32-bit Arm® Cortex®-M0+ Core, a CAN FD controller, a 3-phase motor control PWM, and a 2-unit A/D converter, achieving power saving performance with a power consumption of approximately 40mW. It is ideal for failure prediction detection in industrial equipment, residential equipment, and home appliances.

Please see here for details.

・ ROHM website

・ Product introduction materials 

Inquiry

If you have any questions about our products or would like to request a quote, please contact us here.