N2CMU (Neural Network Coprocessing Microcontroller Unit)
N2CMU is a project aimed at implementing a neural network coprocessing microcontroller unit (MCU) for STM32F103C8T6 Bluepill development board using the TinyGo (Go programming language for embedded systems). This MCU enables offloading feedforward neural network computations to a dedicated hardware unit, allowing for efficient and low-power feedforward neural network training and inference on small hardware embedded systems.
Features
- Neural Network Coprocessing: N2CMU allows offloading feedforward neural network computations to a dedicated coprocessing unit, freeing up the main microcontroller for other tasks.
- UART Communication: Communication with N2CMU is facilitated via UART, enabling external devices to control and interact with the coprocessing unit.
- Flexible Configuration: N2CMU supports dynamic configuration of neural network parameters such as input, hidden, and output neuron counts, as well as weights, biases, and gradients.
- Training and Inference: The coprocessing unit supports both training and inference modes, allowing for the optimization of neural network models directly on the microcontroller.