Using GDI Structure in Hardware Implementation of Convolution Operation in Deep Neural Networks

In this paper, GDI structure is used to implement convolution operation for Convolutional Neural Networks (CNN). Given the potential to design circuits with fewer transistors using the GDI structure compared to the standard CMOS structure, the proposed design can be more suitable for applications facing challenges such as increased area, losses, and reduced speed. To optimize transistor dimensions in the GDI structure and prevent capacitive chaining, a combination of CMOS and GDI structures is used. For the design chosen for the implementation, the proposed structure achieves a speed increase of 2.08 times and a 16.5% improvement in energy efficiency compared to the equivalent CMOS structure. The proposed CNN is implemented in 65 nm technology.