A reconfigurable self-biased optoelectronic transistor for integrated high-fidelity sensing and neuromorphic signal recovery

Space-ground optical links enable ultra-high-bandwidth and secure long-distance communication but are severely limited by atmospheric-turbulence-induced signal distortion and fading, requiring both high-fidelity linear optoelectronic conversion and nonlinear signal recovery. Neuromorphic computing provides a promising framework for processing temporally complex signals with reduced training cost and improved efficiency. However, its practical deployment in optical communication systems is hindered by the absence of compact and reconfigurable devices capable of simultaneously supporting sensing and processing within a unified hardware platform. Here, we report a reconfigurable self-biased optoelectronic transistor that enables tightly coupled high-fidelity sensing and nonlinear processing at the device level. By leveraging gate-drain coupling and electrostatic doping, the built-in electric field can be dynamically modulated, resulting in reconfigurable carrier transport with continuously tunable linear and nonlinear responses. For high-fidelity optoelectronic conversion, the device exhibits symmetric, bidirectional responses with exceptional linearity (R² = 0.99) across various wavelengths and illumination intensities. For accurate signal recovery, we present a novel reservoir computing design that employs expansive neurons in place of compressive ones and is directly hardware-realized via the device’s nonlinear characteristics, substantially reducing the bit error rate from 10⁻² (typical of conventional equalizers) to 10⁻⁴ while lowering the trainable-parameter count to 45.1% and training time to 1.4%. This work establishes a unified, reconfigurable optoelectronic device platform, enabling the practical deployment of neuromorphic computing for robust space-ground optical links.