Topological acoustic synapse for high-dimensional neuromorphic computing

neurons while consuming ~20 watts of power. Neuromorphic computing seeks similar efficiency, but current devices face bottlenecks in bandwidth, energy, wiring, footprint, and reliability that limit scalability. Here, we introduce the topological acoustic synapse (TAS), an acoustic-wave neuromorphic device that circumvents these limits by mapping information in multivariate state spaces. A single TAS generates and manipulates numerous computing channels that operate independently and in parallel. The TAS leverages nonlinear interactions to emulate biorealistic neuromorphic functionalities, including reconfigurable synaptic plasticity, neuromodulation, and hybrid analog-digital control. In classification tasks, a TAS handles multiple inputs simultaneously and generates various outputs, converging 20% faster while using 60% fewer parameters and at least an order of magnitude less power than state-of-the-art electrical devices. This work establishes the first acoustic synapse with parallel HD computing capabilities, presenting a scalable paradigm for neuromorphic hardware with high computational density.