Hardware Implementation of a CNN Digit Recognizer using Basys 3 and ESP8266

This paper presents a lightweight, hardware-accelerated framework for real-time handwritten digit recognition using a Convolutional Neural Network (CNN) deployed on a Basys 3 FPGA board. The system captures user-drawn digits through an ESP8266-powered web interface on a mobile device, which are converted to grayscale 28×28 format before being sent to the FPGA over UART. The CNN core, designed in Verilog using Q1.7 fixed-point arithmetic, performs convolution, ReLU activation, and a fully connected classification stage. The recognized digit is displayed on the Basys 3’s 7-segment display or sent back to the mobile web server. This approach demonstrates the efficient hardware deployment of deep learning models using resource-constrained FPGA platforms without external memory or processors.