ZnO Quantum Dots@CsPbBr3 Poly‐Heterocrystalline Film Enables High‐Performance Floating‐Gate Transistor Arrays for Edge Computing

The ZnO QDs@CsPbBr3 poly‐heterocrystalline film, featuring a semi‐coherent interface, delivers high electrical conductivity and excellent photoelectric properties. Functioning as a floating gate in photosensitive transistors, it enables high hole storage at 1 V and achieves a conductivity ratio of ≈3.57×105. A high‐density array of 25,600 devices (145 ppi) is fabricated, demonstrating a significantly higher recognition precision (87.64%) at the fifth epoch than that of the original images (58.08%) and CMOS‐processed images (77.68%), rivaling CMOS circuits for edge computing applications. Abstract Nonvolatile optoelectronic synapses motivated by the human eye can effectively function as convolutional kernels to preprocess images, demonstrating significant promise for edge computing. Among the optoelectronic synapses, the floating‐gate photosensitive transistor (FG‐PT) is particularly noteworthy due to its rapid response speed and excellent retention. Although some FG‐PTs are reported, they still suffer from high operating voltages, low conductance ratios, and difficulties in array preparation. Here, a ZnO QDs@CsPbBr3 poly‐heterocrystalline (PHC) film to serve as the floating gate layer of FG‐PT is synthesized. The PHC film combines the desirable properties of each phase, exhibiting both high electrical conductivity and excellent photoelectric properties. The high electrical conductivity allows FG‐PT to store a large amount of charge at a low voltage (1 V). While the excellent photoelectric properties facilitate the gradual erasure of these charges under light pulses, resulting in a large conductance ratio (≈3.57×105). Moreover, the excellent film‐forming properties of the PHC film enable the fabrication of an FG‐PT array comprising 25,600 devices on 1 cm2 substrate. Using the FG‐PT array to preprocess images achieves a significantly higher recognition precision at the fifth epoch (87.64%) than that of the original images (58.08%) and CMOS‐processed images (77.68%), indicating great potential for edge computing. The ZnO QDs@CsPbBr 3 poly-heterocrystalline film, featuring a semi-coherent interface, delivers high electrical conductivity and excellent photoelectric properties. Functioning as a floating gate in photosensitive transistors, it enables high hole storage at 1 V and achieves a conductivity ratio of ≈3.57×10 5. A high-density array of 25,600 devices (145 ppi) is fabricated, demonstrating a significantly higher recognition precision (87.64%) at the fifth epoch than that of the original images (58.08%) and CMOS-processed images (77.68%), rivaling CMOS circuits for edge computing applications. Abstract Nonvolatile optoelectronic synapses motivated by the human eye can effectively function as convolutional kernels to preprocess images, demonstrating significant promise for edge computing. Among the optoelectronic synapses, the floating-gate photosensitive transistor (FG-PT) is particularly noteworthy due to its rapid response speed and excellent retention. Although some FG-PTs are reported, they still suffer from high operating voltages, low conductance ratios, and difficulties in array preparation. Here, a ZnO QDs@CsPbBr 3 poly-heterocrystalline (PHC) film to serve as the floating gate layer of FG-PT is synthesized. The PHC film combines the desirable properties of each phase, exhibiting both high electrical conductivity and excellent photoelectric properties. The high electrical conductivity allows FG-PT to store a large amount of charge at a low voltage (1 V). While the excellent photoelectric properties facilitate the gradual erasure of these charges under light pulses, resulting in a large conductance ratio (≈3.57×10 5 ). Moreover, the excellent film-forming properties of the PHC film enable the fabrication of an FG-PT array comprising 25,600 devices on 1 cm 2 substrate. Using the FG-PT array to preprocess images achieves a significantly higher recognition precision at the fifth epoch (87.64%) than that of the original images (58.08%) and CMOS-processed images (77.68%), indicating great potential for edge computing. Advanced Science, Volume 12, Issue 43, November 20, 2025.