Energy-Efficient ReRAM-Based High-Precision Discrete Fourier Transform for Image Reconstruction.

In this work, we implemented the discrete Fourier transform (DFT) using a Pt/Al 2 O 3 /AlO x /W resistive random-access memory (ReRAM) for high-precision signal processing. By introducing the bit-slicing method, the ReRAM devices are capable of overcoming the imprecise analog computing limitation and can achieve software-comparable processing precision. Validated through the image reconstruction tasks, the ReRAM technology is demonstrated to achieve reconstruction quality virtually equivalent to 8-bit integer precision. Furthermore, this work is evaluated to achieve an energy efficiency of 622 GOPS/W, which is 22× higher than the state-of-the-art digital processors. We expect the ReRAM-technology can be further employed to perform more complex signal processing tasks with high efficiency in practical applications.