Analysis and Prediction of Dark Current Mechanisms in Si:P Blocked Impurity Band (BIB) Infrared Detectors

TL;DR

This work addresses nonlinear dark current phenomena in Si:P BIB detectors, including negative differential resistance (NDR) and current oscillations. It develops a hybrid physical model combining hopping conduction and current-path clustering due to dopant nonuniformity with space-charge limited current (SCLC) theory, capturing field-dependent transport via $\sigma(E) \approx \sigma(0) \exp(- e E L_0 /(2 k T))$ and related trapping dynamics under $ e E L_0 \gg k T$. A six-feature TransformerRegressor is trained on tens of thousands of dark-current measurements to predict log current, achieving RMSE 2.11, MAE 1.36, and R^2 0.60 on a held-out test set, with BT and LA identified as key design levers. The study demonstrates a principled path to rapid detector optimization by integrating physics-based understanding with data-driven prediction, offering design guidance for reducing dark current while maintaining high quantum efficiency.

Abstract

We investigated the nonlinear phenomena observed in the dark current of BIB (blocked-impurity-band) infrared detectors, including negative differential resistance (NDR) and current oscillations. Our analysis systematically elucidated the intrinsic transport mechanisms in optimized devices, revealing that these anomalies arise from current path clustering induced by structural disorder and impurity band conduction dynamics. Notably, the simulated current-voltage (I-V) characteristics demonstrated strong agreement with experimental measurements across a wide bias range, confirming the validity of our proposed physical model.Furthermore, we developed a transformer-based predictive model using experimental dark current datasets. The model achieved robust performance metrics and this framework enables rapid prediction of dark current trends under varying operational conditions, providing actionable insights for detector optimization.

Figures (8)

• Figure 1: A simple model of Si:P BIB which is a cuboid silicon doped with phosphorus. The dopping design is at the concentrations of lower than $1\times 10^{15}cm^{-3}$ at block layer and $1 \times 10^{18}cm^{-3}$at absorption layer. The length of the blocking layer is typically around 5 $\mu m$, the absorbing layer is typically around 20 $\mu m$, and the width of the cross-section can be adjusted in the range from several micrometers to 200 $\mu m$.
• Figure 2: A small number of devices exhibit dark current oscillation phenomena, which may indicate particularly inhomogeneous doping of the silicon in this batch. Such distinctive dark currents, manifested under particularly prominent conditions, suggest that they also exist in ordinary situations but are just less pronounced. The unit of the ordinate is the logarithm of the dark current I to the base 10 (i.e., lg(I)), with the unit of ampere (A). The abscissa represents the applied voltage of the device, with the unit of volt (V).
• Figure 3: Microscale current network. The figure omits the background silicon atoms and only highlights the local current paths and partial current blocking phenomena. The orange spheres in the figure represent phosphorus atoms. The black lines denote the paths, which may also pass through silicon during the hopping process of phosphorus atoms, however, this physical process is omitted in the figure.
• Figure 4: Dataset exhibiting negative differential resistance phenomena in dark current measurements.
• Figure 5: Overview of Data Distribution Characteristics.The current I here refers to lg(I), with the unit of ampere (A). Both WT and BT have the unit of absolute temperature, Kelvin (K). LB, LW, and LA have the unit of micrometer ($\mu m$). The voltage V has the unit of volt (V).