Reconfigurable Analog Computation With Twisted Metasurfaces

Abstract Analog computation based on passive optical metasurfaces and flat‐optics devices holds the promise of reducing latency and energy consumption in information processing. However, most computational metasurfaces demonstrated so far do not allow for dynamic reconfiguration of the optical response, which is instead crucial for any practical implementation of these devices. Here, a route is proposed and numerically demonstrated to realize reconfigurable analog computational devices based on twisted bilayer metasurfaces. In particular, a bilayer device is discussed that performs temporal differentiation on incoming pulses, and whose response can be continuously varied between zeroth‐order, first‐order, and second‐order differentiation by controlling the twist angles of the two metasurfaces. Besides demonstrating the working principle, the practical feasibility of this design is investigated in terms of alignment tolerances, efficiency, and computational accuracy. Moreover, the use of the device is discussed for dual‐band reconfigurable operation. These results pave the way for a new generation of reconfigurable passive devices for all‐optical computation, with potential applications in neural networks and neuromorphic computing.