AI Application on STM32N6

Context

Implemented AI models on STMicroelectronics’ new STM32N6 boards featuring a dedicated Neural Processing Unit (NPU) for efficient edge inference. This project explores the capabilities of the new architecture for running real-time AI applications on resource-constrained devices.

Technologies Used

• Hardware: STM32N6 (ST NPU boards)
• AI Framework: TensorFlow Lite for Microcontrollers
• Programming: C/C++, Python (for model training)
• Tools: STM32CubeIDE, X-CUBE-AI

Implementation

The project involved:

1. Model selection and optimization for edge deployment
2. Quantization and conversion to TensorFlow Lite format
3. Integration with STM32 NPU using ST’s AI tools
4. Real-time inference testing and benchmarking

Architecture Overview

Results

• Successfully deployed AI models on the NPU
• Achieved optimized inference latency for edge devices
• Optimized memory footprint for embedded constraints

Performance Benchmarks

Key Metrics:

• Inference time: < 100ms
• Power consumption: Optimized for battery operation
• Accuracy: 95%+ on target dataset

Links

• GitHub: [Coming soon]
• Documentation: [Project Wiki]