Dry-state single-atom Pt engineering on crystalline carbon nitride for integrated hydrogen evolution and neuromorphic computing

Journal article, 2026

Precise construction of high-density single-atom active centers on polymeric semiconductors, together with concurrent regulation of their interfacial charge-transfer behavior, remains a central challenge for both photocatalytic energy conversion and neuromorphic electronics. Yet conventional wet photodeposition routes suffer from solvent-induced coordination distortion, defect formation, and limited metal dispersion. Here, we report a solvent-free dry-state in situ photoreduction strategy that anchors atomically dispersed Pt onto highly crystalline carbon nitride (AD-Pt-HCCN), achieving a Pt precursor conversion efficiency of 70.5%, which is 5.5 times higher than that of wet photodeposition. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM), X-ray photoelectron spectroscopy (XPS), and X-ray absorption fine structure (XAFS) collectively confirm uniformly distributed Pt single atoms coordinated in a quasi-fivefold configuration within triazine-heptazine frameworks. This coordination environment suppresses the formation of a classical nanoparticle-induced Schottky-type barrier and promotes ultrafast interfacial charge extraction, as supported by femtosecond transient absorption (fs-TA), photoluminescence (PL), time-resolved PL (TRPL), and electrochemical impedance spectroscopy (EIS) analyses. As a result, a photocatalytic H2 evolution rate of 5.8 mmol & centerdot;g-1 & centerdot;h-1 is achieved, outperforming the counterpart prepared by conventional wet photodeposition (3.8 mmol & centerdot;g-1 & centerdot;h-1), owing to the synergistic contributions of the increased Pt loading efficiency and the enhanced interfacial charge transfer induced by atomically dispersed Pt sites. Remarkably, the same atomic Pt sites serve as efficient charge-modulation centers in neuromorphic transistors, enabling pronounced excitatory postsynaptic current (EPSC)/inhibitory postsynaptic current (IPSC) responses, robust long-term potentiation/depression (LTP/LTD), and linear, hardware-relevant synaptic weight updates. Integrating experimentally extracted conductance states into an artificial neural network (ANN) framework yields high recognition accuracy of 98.6%, highlighting the broad potential of AD-Pt-HCCN as a multifunctional building block for energy-intelligence convergence.