The MCUs are the industry’s first to independently execute both learning and inference without relying on a network connection.
The development of these MCUs (ML63Q253x-NNNxx / ML63Q255x-NNNxx) is a response to the need for more efficient operation of equipment and machinery and early failure detection and enhanced maintenance efficiency. Equipment manufacturers require solutions that allow real-time monitoring of operational status while avoiding the drawbacks of network latency and security risks.
While standard AI processing models typically depend on network connectivity and high-performance CPUs, which can be costly and difficult to install these new AI MCUs enable standalone AI learning and inference directly on the device.
Network-independent, these solutions support early anomaly detection before equipment failure – contributing to a more stable, efficient system operation by reducing maintenance costs and the risk of line stoppages.
These products adopt a simple 3-layer neural network algorithm to implement ROHM’s proprietary on-device AI solution “Solist-AI”, which enables the MCUs to perform learning and inference independently.
AI processing models are generally classified into three types: cloud-based, edge, and endpoint AI.
Cloud-based AI performs both training and inference in the cloud, while edge AI utilises a combination of cloud and on-site systems connected via a network. Typical endpoint AI conducts training in the cloud and performs inference on local devices, so network connection is still required, and inference is usually performed via software, necessitating the use of GPUs or high-performance CPUs.
ROHM’s AI MCUs, by contrast, although categorised as endpoint AI, can independently carry out both learning and inference through on-device learning, allowing for flexible adaptation to different installation environments and unit-to-unit variations, even within the same equipment model.
Equipped with ROHM’s proprietary AI accelerator “AxlCORE-ODL,” these MCUs deliver approximately 1,000 times faster AI processing compared to ROHM's conventional software-based MCUs (theoretical value at 12MHz operation), enabling real-time detection and numerical output of anomalies that “deviate from the norm”. In addition, high-speed learning (on-site) at the point of installation is possible, making them suitable for retrofitting into existing equipment.
These AI MCUs feature a 32-bit Arm Cortex-M0+ core, CAN FD controller, 3-phase motor control PWM, and dual A/D converters, achieving a low power consumption of approximately 40mW. As such, they are suited for fault prediction and anomaly detection in industrial equipment, residential facilities, and home appliances.
The lineup will consist of 16 products in different memory sizes, package types, pin counts, and packaging specifications.
Mass production of 8 models in the TQFP package began sequentially in February 2025. Among these, two models with 256KB of Code Flash memory and taping packaging are available for purchase, along with an MCU evaluation board, through online distributors.
ROHM has released an AI simulation tool (Solist-AI Sim) on its website that allows users to evaluate the effectiveness of learning and inference prior to deploying the AI MCU. The data generated by this tool can also serve as training data for the actual AI MCU, supporting pre-implementation validation and improving inference accuracy.
