ROHM launches the first MCUs with on-device AI learning and inference for real-time predictive maintenance—redefining edge intelligence in factories.

ROHM Co., Ltd., a Japanese electronics heavyweight, unveiled a groundbreaking pair of AI-equipped microcontrollers (MCUs), the ML63Q253x-NNNxx and ML63Q255x-NNNxx series. These are the industry’s first MCUs capable of performing both learning and inference right on the device, with no network or cloud connection needed, according to ROHM’s research (June 4, 2025). Designed to predict equipment anomalies and forecast degradation using sensor data, these chips are set to revolutionize predictive maintenance in industrial settings. Powered by ROHM’s proprietary AI accelerator, “AxlCORE-ODL,” and their Solist-AI™ technology, these MCUs are fast, secure, and efficient, making them a big deal for factories and beyond. Let’s dive into why these chips are about to change the game.

What Are the ML63Q253x-NNNxx and ML63Q255x-NNNxx Series?

ROHM’s new MCUs bring artificial intelligence to the edge, meaning they can learn from data and make predictions without relying on external systems. These chips are built to analyze sensor data from industrial equipment, like motors, to detect anomalies (e.g., unusual vibrations or temperature spikes) and predict when components might fail. Unlike traditional MCUs that need cloud servers or high-performance CPUs for AI tasks, these MCUs handle everything on-device, making them the first of their kind in the industry (ROHM Co., Ltd., 2025).

The ML63Q253x-NNNxx and ML63Q255x-NNNxx series use ROHM’s proprietary Solist-AI™ technology, which employs a compact 3-layer neural network to perform both learning and inference independently. This setup delivers AI processing speeds about 1,000 times faster than ROHM’s conventional software-based MCUs at 12MHz operation, enabling real-time anomaly detection (ROHM Co., Ltd., 2025). Available in packages like TQFP48 (9x9mm) and WQFN64 (9x9mm), with memory options of 128KB or 256KB Code Flash, these MCUs are versatile enough to fit into various industrial systems.

The Technology Behind the Breakthrough

At the heart of these MCUs is the “AxlCORE-ODL” AI accelerator, a custom hardware component that powers their on-device learning and inference. Unlike cloud-based or edge AI systems that rely on network connectivity, AxlCORE-ODL enables these MCUs to process sensor data locally using a 3-layer neural network. This design allows the chips to learn from data like motor vibrations or temperature readings and predict anomalies or degradation without sending data to a server, reducing latency and boosting security (ROHM Co., Ltd., 2025).

The Solist-AI™ framework is what makes this possible. Inspired by the term “soloist,” it’s designed for standalone edge devices, delivering AI capabilities with low power consumption—around 40mW—making it ideal for energy-constrained environments. The MCUs also feature a 32-bit Arm® Cortex®-M0+ core, CAN FD controller, 3-phase motor control PWM, and dual 12-bit A/D converters, enabling precise control and data processing for industrial equipment (ROHM Co., Ltd., 2025). Compared to traditional endpoint AI, which requires cloud training, these MCUs adapt to unique equipment variations on-site, making them perfect for retrofitting existing systems.

In performance tests, ROHM reports that these MCUs achieve high-precision anomaly detection and numerical outputs for conditions that “deviate from the norm.” Their ability to learn on-site means they can adapt to specific machines without needing massive, curated datasets, unlike cloud-based AI models. This efficiency, combined with their speed, makes them a practical solution for real-time predictive maintenance.

Why This Matters

These MCUs address major challenges in industrial predictive maintenance. Equipment failures can cost companies millions in downtime, and traditional AI solutions often rely on cloud connectivity, which introduces latency, security risks, and high costs. ROHM’s AI MCUs eliminate these issues by performing all AI tasks on-device, ensuring real-time monitoring without network dependency. This is a huge win for factories, where early detection of issues like motor bearing damage can prevent costly line stoppages (ROHM Co., Ltd., 2025).

The chips’ low power consumption and compact design make them ideal for a range of applications, from factory automation to home appliances. Their network-free operation also enhances security by keeping sensitive data local, a critical feature for industries like manufacturing or defense. By enabling condition-based maintenance (CBM), these MCUs allow companies to shift from reactive repairs to proactive strategies, boosting efficiency and extending equipment lifespan.

Real-World Applications

The ML63Q253x-NNNxx and ML63Q255x-NNNxx series are built for industrial use but have broad potential. In factory automation, they can monitor motors, pumps, or batteries, using sensor data to predict failures. For example, combining an accelerometer with an AI MCU can track vibration levels to implement tailored CBM for specific machines. Acoustic emission (AE) sensors paired with these MCUs can detect ultra-early mechanical anomalies by analyzing sound indicators like peak amplitude or energy (ROHM Co., Ltd., 2025).

Beyond factories, these MCUs can enhance residential facilities and home appliances by detecting anomalies in HVAC systems or predicting maintenance needs for smart devices. In robotics, they can monitor components and optimize adjustments without relying on a central CPU. Their versatility also makes them suitable for power tools, remote sensors in energy systems, or secure environments where operational downtime is unacceptable.

Support and Ecosystem

ROHM has made adoption easy with a robust ecosystem. Their Solist-AI™ Sim tool lets users simulate AI operations before deployment, generating training data to improve inference accuracy. The Solist-AI™ Scope provides real-time visualization of AI performance, and the chips are compatible with standard Arm® development tools like Keil® MDK. ROHM also offers an MCU evaluation board (e.g., RB-D63Q2537TB48) for testing, available through distributors like DigiKey™, Mouser™, and Farnell™, with sample prices around $20/unit (ROHM Co., Ltd., 2025).

Mass production of eight TQFP models began in February 2025, with two 256KB models already available for purchase. ROHM’s partnerships with companies for model development and integration ensure users get comprehensive support, from training data creation to implementation.

ROHM’s AI-equipped MCUs are a landmark in edge AI, offering a fast, secure, and efficient solution for predictive maintenance. As industries demand smarter, more reliable systems, these chips—backed by ROHM’s semiconductor expertise that span decades—set a new standard. With their ability to learn and infer on-device, the ML63Q253x-NNNxx and ML63Q255x-NNNxx series are poised to drive innovation in factory automation, robotics, and beyond, making operations smoother and more cost-effective.