Highly Linear Perovskite Synapses for Neuromorphic Computing

Neuromorphic computing offers a compelling route to data-intensive applications with boosted computing power and energy efficiency. However, its accuracy for edge vision tasks is critically bottlenecked by the inherent non-linearity and asymmetry of conductance updates in analog memristive synapses. Here, we report a highly linear memristive synapse based on vertically aligned alternating cations in the interlayer space phase halide perovskite. A synergistic regulation strategy involving additive engineering and solvent atmosphere annealing is introduced to fabricate GAMA₅Pb₅I₁₆ halide perovskite films with vertical crystal orientation. The resulting films facilitate controllable ion migration, allowing the fabricated devices to achieve an ultra-low energy consumption of 5.5 aJ alongside 100 discrete conductance states with non-linearity as low as 0.027 and 0.064 and asymmetry of 0.037. These attributes enable the synapses to perform object detection tasks with an mAP50 of 57.6%, retaining 98.1% of the floating-point baseline performance. This work provides a promising biomimetic solution for efficient visual perception in intelligent transportation, paving the way for high-precision neuromorphic systems.