Demonstrating Single Photon Counting with Kinetic Inductance Detectors from 3.8 to 25 $μ$m

Abstract

One of the primary objectives of modern astronomy is the atmospheric characterization of Earth-like exoplanets at visible and infrared wavelengths. Achieving this goal requires extremely sensitive detectors capable of measuring faint signal of the exoplanet at the single-photon level while maintaining near-zero dark count rates. In the infrared, however, conventional semiconducting detector technologies struggle to meet these stringent requirements. In this work we demonstrate single-photon counting with superconducting Microwave Kinetic Inductance Detectors at the wavelengths 3.8, 8.5, 18.5, and 25 $μ$m and measure resolving powers ($E/δE$) of 9.9, 5.9, 3.2, and 3.3, respectively, with corresponding dark count rates of 4, 8, 34, and 48 mHz. Our membrane-based devices reach phonon-loss limited performance at 3.8 $μ$m, more than doubling the performance attainable with comparable solid-substrate devices. These results showcase the detector technology in the mid-infrared and the intricate measurement setup required for these sensitive detectors. We discuss how the detector design and measurement setup can be further optimized to increase the detector performance in the mid-infrared.

Figures (9)

• Figure 1: MKID architecture and cryogenic setup. (a) Geometry of a single detector. The microwave resonator is made of a NbTiN IDC in parallel with aluminium CPW (THz line) with a highly inductive central line. The THz line is at the focus of a Si lens which is also the feed point of a leaky-slot antenna. The THz line is suspended on a SiN membrane to decrease phonon loss to the substrate. The detector design is originally optimized for the far-IR at 200. (b) Cross-sectional view of the detector assembly. The figures in panels a and b are reproduced from Ref. baselmans2022ultra. (c) Cross section of the DR as configured for the experiments at [list-units=single]3.8;8.5 and (d) at 18.5. The detectors are mounted at the sample stage. Indicated are the temperature shields as well as the Cryophy and Niobium magnetic shields. The key difference between the configurations is the source of radiation; (c) an external QTH lamp and (d) a cryogenic radiator at 3K. Optical filters can be mounted at the indicated positions.
• Figure 2: Characterization of the measurement setups. (a) Estimated spectral radiance at the detector and (b,c) measured fractional frequency noise at [list-units=single]3.8;8.5;18.5;25 (blue, yellow, orange, purple, respectively) and in the dark (black). The 25 experiment was done in (c) the ADR, the others in the (b) DR. In all panels the solid lines represent the radiance and noise from just the background, while the dashed lines also include the radiation source.
• Figure 3: Single-photon detection across the mid-IR. Shown are phase response time streams in the dark and at [list-units=single]3.8;8.5;18.5;25 in panels a-e, respectively. At [list-units=single]3.8;18.5;25 we show the absence of pulses when the radiation source is turned off (gray line, offset by -0.5rad for clarity). At 8.5 the radiation source is the every present thermal background of the lab. The data is smoothed with an exponential window with a lifetime of 50. We rescale the response at a wavelength $x$ by a factor $Q_\mathrm{l}^x/Q_\mathrm{l}^\mathrm{dark}$.
• Figure 4: Energy resolving power in the mid-IR. Optimally filtered noise (gray shaded) and pulse-height distributions at (a, blue shaded) 3.8 (c, yellow shaded) 8.5 (d, orange shaded) 18.5 and (e, purple shaded) 25. We have annotated three different pulse contributions in panel a which are explained in main text. The pulses are detected with a 3σ threshold in the smoothed data and then cut from the raw data before being optimally filtered. The optimal filter is constructed from the pulses in the primary distribution which is identified by the KDE (black, dashed line) and obtained by increasing the detection threshold to the level in the legend. The resolving power is then determined for the same selection. We rescaled the pulse heights at a wavelength $x$ by a factor $Q_\mathrm{l}^x/Q_\mathrm{l}^\mathrm{dark}$. (b) The average pulse shape of the directly absorbed 3.8 photons (indicated by the KDE in panel a), both on a linear and semi-logarithmic scale (inset). The shaded region indicates the $\pm1$ standard deviation from the average pulse shape. An exponential decay fitted to the tail of the pulse (black, dashed line) gives a $\tau_\mathrm{qp}=\qty{55\pm1}{\unit{\upmu\second}}$.
• Figure 5: Energy resolving powers and dark count rates at [list-units=single]3.8;8.5;18.5;25. (a) Plotted are the energy resolving power $R$ and the signal-to-noise $R_\mathrm{SN}$ and intrinsic $R_\mathrm{i}$ contributions (blue, yellow and orange markers, respectively) that were obtained at [list-units=single]3.8;8.5;18.5;25. We plot the Fano limit from \ref{['eq:Fano']} (black, solid line, $J=0$) and the phonon-loss limited resolving powers for solid-substrate ($J=3.1$; black, dotted line) and membrane devices ($J=0.38$; black, dashed line). We extrapolate the $R_\mathrm{SN}$ at 25 to the other wavelengths by scaling with $\lambda$ (yellow, dash-dotted line). We also plot $R$ and $R_\mathrm{i}$ obtained by Ref. day202425 at 25 (purple markers). (b) Dark count rates obtained from a 10000s measurement, see also Appendix \ref{['app:B']}. The dark count rate are calculated per wavelength using the pulse-height distributions in \ref{['fig:resolving power']} and a 99.73 ($\mu\pm 4\sigma$ or $\bar{H}\pm 1.7\delta H$) confidence interval. The shaded area gives the 95.45 ($\mu\pm 3\sigma$) to 99.99 ($\mu\pm 5\sigma$) confidence range.