Year: 2026
Type: article
Abstract: This study presents the design and implementation of a trainable artificial neural network (ANN) on a Field-Programmable Gate Array (FPGA). The proposed architecture consists of artificial neurons with two inputs and one output, employing a 16-bit floating-point representation for arithmetic operations and the Sigmoid activation function. The network is trained using the backpropagation algorithm to iteratively update the connection weights and improve prediction accuracy. A bottom-up hardware design methodology is adopted, where the ANN is constructed from modular neuron units and their interconnections. The entire architecture is implemented in VHDL using finite state machine (FSM)-based control to achieve a scalable and efficient hardware realization. The proposed model is validated using the XOR classification problem, and the experimental results demonstrate successful convergence of the network outputs to the expected values after training. The findings indicate that the proposed FPGA-based ANN architecture effectively exploits hardware parallelism while providing flexibility and energy efficiency, making it a promising solution for real-time intelligent hardware applications.
Source: DergiPark (Istanbul University)
Language: Turkish
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7224466767