Optical logic computing harnesses the speed of light and the high bandwidth of optical signals to achieve ultrafast, highly parallel and energy-efficient operations. In particular, all-optical logic gates can perform Boolean functions using only photons, acting as core elements of future optical computing and communication systems. However, the multifunctional integration and flexible reconfiguration of optical devices remain challenging. Here, we present an electrically reconfigurable all-optical logic processing unit that leverages the Kerr nonlinear effect in conjunction with modulation of high-entropy MXene surface terminations. This architecture enables dynamic switching among seven fundamental Boolean operations — including AND, OR, NOT, NOR, NAND, XOR and XNOR — within a single optical configuration. With our platform we demonstrate handwritten digit recognition on the MNIST dataset, achieving a classification accuracy of 97.7%. By combining reconfigurable nonlinear optics with multifunctional two-dimensional materials, our approach establishes an alternative logic architecture pathway with the potential to enhance throughput and energy efficiency in response to AI workloads. The rapid rise of artificial intelligence pushes the need for more efficient computing. Here, authors propose an electrically reconfigurable, all-optical logic processing unit based on the combination of Kerr nonlinearities and high-entropy MXene surface terminations. They achieve an accuracy of 97.7% for digit recognition on the MNIST dataset.
Ge et al. (Tue,) studied this question.