A Memristor-Based Winograd Fast Fourier Transform Computation Circuit

Fast Fourier Transform (FFT) is an important technology in signal processing and digital analysis. As an optimized FFT algorithm, the Winograd FFT algorithm can reduce the number of multipliers, but the multiplication calculation in the circuit implementation is still the power consumption and speed bottleneck of the traditional processor. Memristor is an effective method to solve this problem because of its low power consumption and unique advantages in multiply accumulation operation. Therefore, in this paper, a Winogard FFT operation circuit based on memristors is proposed. The proposed circuit can support 8 different point of FFT operation, up to 16-point. The circuit can also be used as the basic unit circuit to realize FFT operation with larger points. Compared with the implementation of DFT with a memristor crossbar array, it avoids the weight mapping of a complex transformation matrix and only needs to map the two weights of 0 and 1. The simulation results show that the proposed circuit has good performance under different FFT operations. In addition, we also analyze the influence of the fluctuation of the memristor resistance on the 16-point FFT operation performance, and the SPICE simulation verifies that the signal-to-noise ratio can be greater than 20 dB when the resistance fluctuation is less than 10%.