Toward neuromorphic computing using longitudinal pulses in a fluid near phase transition

Abstract

Longitudinal waves propagate information about the stimulus in multiple dimensions, including the medium density and pressure. Pulses that reversibly cross a phase transition have a nonlinear response that resembles properties of neuronal signaling. This multidimensionality suggests that longitudinal pulses may be harnessed for in-materio computation, mimicking biological or artificial neural algorithms. To explore a feedforward physical neural network using longitudinal pulses, we demonstrate the implementation of (1) a complete set of logic gates, (2) classification of data, and (3) regression of a mathematical function. Our results illustrate the potential of harnessing nonlinear longitudinal waves—common in a plethora of materials—for the purpose of computation.