The Industry's First* AI Microcontrollers Capable of Standalone AI Learning and InferenceROHM has developed 32-bit general purpose MCUs with standalone Solist-AI™
that enable real-time failure prediction for electronic devices
at directly at the endpoint, eliminating the need for cloud servers

ROHM has recently announced the development of on-device learning AI MCUs designed for edge computing and IoT endpoints, enabling predictive failure detection in electronic devices equipped with motors and sensors.
The novel design combines an AI accelerator with 32-bit CPU to achieve learning and inference with ultra-low power consumption (~40mW) and processing speeds approximately 1/1000* of conventional software-based methods, providing real-time failure prediction without the need for cloud connectivity. To accelerate adoption, ROHM is actively building an ecosystem in collaboration with partner companies that supports customer product development and provides easily implementable AI solutions.
(*compared to ROHM’s conventional product)

32-bit Solist-AI™ MCU

Solist-AI™ Brand

ROHM’s AI MCU Concept

ROHM's AI MCUs are equipped with the proprietary AxlCORE-ODL AI accelerator capable of independently performing both learning and inference. They excel at anomaly prediction detection using sensor inputs to identify abnormalities by quantifying deviations from the learned normal state.
The optimized design allows for standalone operation at the endpoint, removing cloud server dependency while maintaining ultra-low power consumption – ideal for battery-powered applications. Solist-AI™ is the industry's first MCU solution capable of on-device learning.

Comparing Conventional AIs and Solist-AI™

AI can be categorized into three types: cloud-based, edge-based, and endpoint-based. Cloud AI performs learning and inference on cloud computers, while edge AI processes data within intermediate network devices. In contrast, endpoint AI integrates AI directly into terminal devices such as robots and sensors. ROHM’s AI MCUs, equipped with an AI accelerator, enable independent learning and inference without relying on the cloud, allowing for on-device relearning and additional learning even after deployment. This ensures flexible adaptation to various installation environments and equipment.

AI MCU Applications

ROHM's AI MCUs are designed to detect early signs of anomalies and issue alerts before equipment failure occurs, akin to a health checkup for humans. By analyzing inference data, the AI evaluates the severity of abnormalities, enabling proactive intervention before issues escalate. This technology supports 24/7 automated monitoring, minimizing the risk of production line shutdowns and reducing repair costs. It also addresses customer demands for advanced notifications prior to equipment breakdowns.

Specific Use Cases of AI MCUs

AI MCUs allow for the quantification of complex factors that were difficult to process using conventional software, paving the way for diverse applications in previously challenging fields. Below are some specific use cases.

The AI can learn and infer acceleration signals obtained from accelerometers attached to fan motors to identify abnormalities. When an issue arises in the fan motor, the acceleration changes and the AI assesses the degree of abnormality based on the variation. By analyzing the magnitude of the anomaly, it becomes possible to detect various signs of failure due to component deterioration and/or wear.

Motor bearing damage is sometimes detected through human hearing, but by the time the noise becomes audible, significant damage has often already occurred. By analyzing accelerometer data using an AI MCU, however, it becomes possible to detect signs of damage at an early stage.

Other use cases are as follows, with applications expected to expand across a wide range of fields.
・Equipment requiring failure prediction (anomaly detection) or continuous operation
■Industrial equipment: Devices using motors or sensors (motors, robots, FA sensors, etc.)
■Home appliances: Systems demanding improved prediction accuracy, such as water heaters and refrigerators
・Equipment requiring enhanced prediction accuracy
■End of operation prediction: Washing machines, rice cookers, water heaters (devices with operating cycles lasting for tens of minutes), batteries (degradation monitoring)

No Need to Prepare Training Data

Unlike conventional AI solutions that involve multiple steps, including data collection, labeling, AI model training, software generation, and MCU programming for inference, ROHM AI MCUs equipped with AxICORE-ODL accelerator performs standalone hardware processing that eliminates the need for AI model software generation. The ability to learn and infer directly at the installation site significantly streamlines the process by enabling an immediate return to inference after learning. And while conventional AI systems require software regeneration for retraining, the AI MCUs remove this step, improving both efficiency and ease of implementation.

Performance Comparison of ROHM’s AI MCU vs Various AI Chips

The table below highlights the differences between conventional AI solutions and ROHM’s AI MCUs.

ROHM's AI MCUs offer several advantages:
• No network dependency – operates independently at the endpoint
• High-speed learning and inference – completes tasks in just a few ms
• Ultra-low power consumption – operates at approx. 40mW
• Millisecond-level response time – executes tasks in approx. 1/1000th the time required by conventional software-based AI systems

ROHM AI MCUs

ML63Q2500 Group Overview

The ML63Q2500 Group consists of 32-bit MCUs based on the Arm® Cortex®-M0 core, equipped with a rich set of peripheral blocks including an AI accelerator (AxICORE-ODL) and CAN FD controller.
A standout feature is the ability to run AI accelerator (AxICORE-ODL) processing in parallel with non-AI software tasks. This allows for efficient operation of serial communication and peripheral control while the accelerator is running, ensuring immediate output of the results once inference is completed.
*Arm® and Cortex® are registered trademarks of Arm Limited (or its subsidiaries) in the US and other countries.

Features

• AI Functions:　Quantifies and outputs deviations in vibration, temperature, and other parameters
• CAD FD Controller: CAN2.0B qualified, with a max. transfer speed of 5Mbps
• 3-Phase Motor Control: Supports 6 PWM outputs (3-phase x positive/negative)
• Program Updates: Programs can be rewritten by switching the boot mode
• Communication Support: Compatible with CAN and IO Link/RS485 transceiver connections via UART

Development Support Tools

ROHM offers the following development support tools for its AI MCUs (ML63Q2500 Group). Among these, (1) to (5) and (7) are provided by ROHM, (6) is supplied by ecosystem partners, and (8) and (9) are commercially available products that customers must procure independently.
The accelerator (AxICORE-ODL) handles processing by hardware, eliminating the need for AI model development. ROHM also offers the following utilities to further streamline application software development for customer devices.
■Solist-AITM SIM: A PC-based simulation tool for verifying AI application effectiveness
■Solist-AITM Scope: A high-speed debugger waveform viewer for visualizing AI operations in real time (graph display)

Solist-AI™ Ecosystem

Solist-AI™ Ecosystem Concept

ROHM is building an ecosystem in partnership with industry leaders to promote the adoption of AI MCUs. We envision two deployment scenarios: customers integrating AI directly into their equipment and implementing off-the-shelf products for high-mix, low volume industrial equipment systems. At the same time, ROHM is expanding collaborations with key players across various fields, including hardware/software providers and system integrators, while actively seeking new partners.

Summary

ROHM AI MCUs with Solist-AI™ achieve real-time on-device predictive failure detection while maintaining ultra-low power consumption. The MCUs operate independently without cloud or network dependency, ensuring zero latency. Furthermore, the ability to flexibly adapt to different installation environments and device variations eliminates the need to prepare training data in advance.

Going forward, ROHM will continue to provide a range of solutions, from ready-to-use AI-enabled devices to custom products tailored to specific needs, by expanding its lineup of AI MCUs and building an ecosystem through collaboration with partner companies.