Neuromorphic multisensory inference using mixed-plasticity artificial synaptic cluster

The majority of existing neuromorphic platforms focus on sensory fusion or recognition and lack the ability to identify relationships between multimodal inputs. Here, we developed an artificial synaptic cluster to realize an artificial neuromorphic multisensory inference system. This system comprises two laterally integrated synaptic transistors with shared electrolyte gates. The polycrystalline InZnO channel facilitates ionic modulation for short-term plasticity (STP), while the Au nanocrystal-based channel introduces discrete charge-trapping sites for long-term plasticity (LTP), enabling simultaneous multi-timescale synaptic processing. This mixed-plasticity architecture emulates the transient and persistent memory behaviors of biological synapses and evaluates the spatiotemporal alignment of infrared and ultrasonic sensory signals for hardware-level causal inference. Without external training or neural networks, the system achieves 95.3% and 94.2% accuracy in multisensory object classification and causal inference, respectively.