Neuromorphic metamaterials for mechanosensing and perceptual associative learning

The work reports a class of neuromorphic metamaterials that combine mechanosensing, memory, and learning. It uses mechanical instabilities in a multistable metamaterial to filter, amplify, and convert external mechanical inputs across large areas into simple electrical signals via piezoresistivity. Those electrical signals are recorded in non-volatile flexible memristors, which retain changes in state after sequential mechanical inputs. The authors describe how accumulated memristance can physically encode a Hopfield network, enabling the system to learn a series of spatially distributed input patterns. The learned patterns can be retrieved from the final accumulated memristor state, and the system is described as learning without supervised training while retaining spatially distributed inputs with minimal external overhead. The authors position this as a route to synthetic neuromorphic metamaterials for touch-like sensing in robotics, autonomous systems, wearables, and morphing structures.