The MSP Edge AI microcontroller delivers leading performance in low latency and power efficiency by integrating a dedicated hardware accelerator called the TinyEngine™ NPU. This cost-effective design supports a wide range of neural network layers and mixed data types (including 8-bit, 4-bit, and 2-bit weights), while featuring flexible CPU and DMA integration to seamlessly offload unsupported layers back to the main CPU.

The MSP microcontrollers support three primary machine learning modalities to process sensor data efficiently at the edge: classification, fault and anomaly detection, and regression.

• Classification aims to categorize unseen 1D, multi-dimensional, or 2D image data into predefined classes, such as identifying motion types from a 3-axis accelerometer or recognizing shapes.
• Fault and anomaly detection focuses on identifying abnormal system behaviors within 1D and multi-dimensional time-series data, enabling applications like ECG rhythm analysis, arc fault detection, and predictive maintenance for motors.
• Lastly, regression maps the sensor inputs—including PIR, audio, and voltage data—to a continuous scale rather than discrete categories, allowing systems to predict future continuous states based on real-time input data.