Numerical modeling and explainable machine learning-based multi-objective analysis of graphene nanoribbon infrared phototransistors
摘要
In this work, we present a multi-objective performance analysis of a graphene nanoribbon (GNR)–germanium (Ge) infrared phototransistor using quantum-transport modeling combined with an explainable ensemble learning approach. Device operation is described by self-consistent Schrödinger–Poisson solutions within the non-equilibrium Green’s function (NEGF) formalism. Key figures of merit (FoMs), including off-state current, Ion/Ioff ratio, responsivity, and maximum drive current, are investigated as functions of channel length, germanium thickness and doping, and HfO2 gate oxide thickness. The proposed GNR–Ge phototransistor achieves responsivity exceeding 4.6 × 104 A/W, Ion/Ioff ratios above 50 dB, and enhanced drive current at channel lengths down to 10 nm. Explainable Random Forest analysis identifies channel length as the dominant factor controlling leakage current, current ratio, and responsivity, while germanium doping primarily governs the drive current. The results provide physics-based design guidelines for low-power, high-performance infrared phototransistors.