Nonlinearity in H4RG-10 Near-Infrared Detectors at Elevated Temperatures: Characterization and Data-Driven Correction Method

Abstract

We report a newly identified nonlinearity in H4RG-10 near-infrared detectors operating under moderately elevated-temperature conditions (114 K). This component, that potentially arises from illumination-independent defect currents, introduces additional nonlinearity not captured by conventional correction models. To address this issue, we propose a data-driven nonlinearity correction (NLC) method that models the nonlinear behavior of both the classical response and the defect currents, using dual-illumination measurements and a dark exposure. Applied to H4RG-10 detectors on the PRIME telescope, the method significantly improves signal linearity, especially for pixels with high defect current, while maintaining comparable performance elsewhere. By selecting the optimal correction model per pixel, reliable NLC is achieved across the full array. This study characterizes a nonlinearity intrinsic to H4RG-10 detectors and demonstrates that data-driven post-processing can effectively restore linearity in the presence of large defect currents. Although these effects are unlikely to be significant under nominal operating temperatures, the approach may provide a practical calibration framework for future warm-operation scenarios.

Figures (4)

• Figure 1: Left: A part of an H4RG-10 image observed by the PRIME telescope toward the Galactic bulge at a detector temperature of 114 K, processed with superbias subtraction, reference pixel correction, and ramp fitting, but without applying any nonlinearity correction. Middle: The same dataset processed with superbias subtraction, reference pixel correction, conventional nonlinearity correction (CNLC), ramp fitting, and dark subtraction, shows a cross-hatched pattern in certain regions. This pattern is potentially caused by elevated dark current in some pixels, unaccounted for in conventional nonlinearity correction. The left and middle panels share a common color scale, with a stretch applied using a method similar to the DS9 zscale + sinh scaling. Right: A dark image obtained at the same detector temperature (114 K) calculated from the slope of ten frames from a dark dataset (exp3), excluding the first frame (frame 0), extracted from the same spatial region as in the left panel.
• Figure 2: Left: Measured signal $S_\mathrm{meas}$ in analog-to-digital units (ADU, or DN) as a function of frame number for two illuminated exposures (blue and orange) and one dark exposure (green) taken with a PRIME H4RG-10 detector. Right: Corresponding per-frame signal slope from finite differencing (Eq. \ref{['eq:a.2']} ) as a function of $S_\mathrm{meas}$ (Eq. \ref{['eq:a.1']}). The data corresponds to a representative pixel exhibiting a significant spurious dark current. These behaviors suggest that the signal in this pixel without illumination is not constant but evolves nonlinearly with the accumulation of charge, especially under elevated detector temperatures.
• Figure 3: Per-pixel differences in measured slopes, $\Delta\dot{S}_\mathrm{meas}$, plotted as a function of $S_\mathrm{meas}$ for a representative pixel exhibiting large defect current. The weighted mean of the five points at lower $S_\mathrm{meas}$ (shown in red) is used to estimate the factor of $(A_1 - A_2)$ in Eq. \ref{['eq:10']}.
• Figure 4: The blue points with error bars represent the correction factor $R_\mathrm{meas}$ defined in Eq. \ref{['eq:Rmeas']}, plotted as a function of the measured ADU value $S_\mathrm{meas}$. The orange line shows the fitted function $R_\mathrm{fit}: S_\mathrm{meas} \rightarrow S_\mathrm{lin+def}$ with polynomial expansion of $S_\mathrm{meas}$.