Organic synapses with programmable linearity for neuromorphic computing

Organic synaptic devices offer a route to flexible and biocompatible neuromorphic computing and human-machine interfaces. However, electrical signal transmission is often nonlinear and poorly reproducible because of interfacial effects and non-uniform electronic processes that can increase energy consumption. Organic all-photonic synapses circumvent electrical transmission but remain limited by nonlinear photochemical and photoisomerization processes. Here we develop linearity-programmable organic all-photonic synapses based on a charge-separated-buffered adaptive luminescence mechanism. Systematic engineering of guest molecular structures modulates charge-separation kinetics, allowing precise control over synaptic linearity. The resulting devices exhibit a linearity parameter, v, of 0.0093, 99% uniformity, 97% repeatability, an optical trigger energy of 59 zJ per synaptic event and a response time of 1.39 ns. An all-photonic sensor system integrating linearity-programmable organic all-photonic synapses enables high-quality image acquisition and high image-classification accuracy. These results establish a molecularly programmable photophysical platform for neuromorphic signal processing and provide a potential route towards low-energy human-machine interfaces.