Design and Implementation of the STM32N6-Based Modular Embedded Edge AI Teaching Platform for Engineering Education and Competition Practices

This paper presents a modular embedded edge-AI teaching platform built around the STM32N6 microcontroller, designed to meet demand for low-power, real-time, deployable edge intelligence in engineering education. The platform uses a heterogeneous architecture combining an ARM Cortex-M55 core with a dedicated Neural-ART NPU, enabling efficient on-device inference for both classroom projects and vision-based competition tasks. To improve stability across multi-peripheral setups, a multi-power-domain supply architecture combines switched-mode power supplies with low-noise LDO regulators, plus dual-input power switching, reverse-current and reverse-polarity protection, overcurrent limiting, and soft-start control. High-speed modular peripherals are integrated on board—MIPI-CSI camera input, RGB display output, high-speed NOR Flash, and SPI, I2C, UART, and TIMER expansion interfaces—with an 80-pin board-to-board connector for flexible extension. A four-layer PCB layout improves signal integrity for reliable high-speed operation. Deploying a custom lightweight vision algorithm (PEPoseNet), the prototype achieves an inference-only latency of 18.4 ms and 28.31 FPS/Watt energy efficiency within a sub-3 W power budget. These results confirm the platform’s reliability as an educational training platform for embedded edge-AI computing.