Performance of Al-Mn Transition-Edge Sensor Bolometers in SPT-3G

TL;DR

This paper evaluates the performance of Al-Mn transition-edge sensor bolometers in the SPT-3G receiver, highlighting their electrical, thermal, and optical characteristics, and addressing calibration systematic errors in CMB measurements.

Contribution

It presents the first in-lab and on-sky performance results of Al-Mn TES bolometers in the SPT-3G, demonstrating their viability for CMB observations.

Findings

Al-Mn TES bolometers show comparable or improved sensitivity.

Calibration systematic errors are identified and corrected.

Optical efficiency and noise performance are characterized.

Abstract

SPT-3G is a polarization-sensitive receiver, installed on the South Pole Telescope, that measures the anisotropy of the cosmic microwave background (CMB) from degree to arcminute scales. The receiver consists of ten 150~mm-diameter detector wafers, containing a total of 16,000 transition-edge sensor (TES) bolometers observing at 95, 150, and 220 GHz. During the 2018-2019 austral summer, one of these detector wafers was replaced by a new wafer fabricated with Al-Mn TESs instead of the Ti/Au design originally deployed for SPT-3G. We present the results of in-lab characterization and on-sky performance of this Al-Mn wafer, including electrical and thermal properties, optical efficiency measurements, and noise-equivalent temperature. In addition, we discuss and account for several calibration-related systematic errors that affect measurements made using frequency-domain multiplexing readout electronics.

Equations

$${\hat{R}}_n=\frac{R_{\text{total}}}{c(i)}-R_p(f_b),$$

$${\hat{P}}_{\text{sat}}=\frac{\sqrt{R_{sh}^2+(2\pi f_b{L}_{sh}{)}^2}}{R_{sh}}\frac{R_{\text{total}}-R_p}{R_{\text{total}}}{P}_{\text{sat}}^{\text{nominal}},$$

$$\begin{array}{rl}{P}_{\text{sat}}^{\text{nominal}}(T_{cl})=& P_0+⟨C⟩T_{cl}\\ & -\eta \frac{R_{sh}}{\sqrt{R_{sh}^2+(2\pi f_b{L}_{sh}{)}^2}}\frac{R_{\text{total}}}{R_{\text{total}}-R_p}\\ & \times \int \frac{h\nu }{\exp (h\nu /k{T}_{cl})-1}{T}_{\text{triplexer}}(\nu )T_{\text{filter}}(\nu )d\nu ,\end{array}$$

Full text

††journal: Journal of Low Temperature Physics

1 Introduction

Observations of the cosmic microwave background (CMB) by telescopes with arcminute resolution are useful in a wide range of measurements impacting cosmology and particle physics. On small angular scales, better measurements of the CMB power spectrum and gravitational lensing will improve constraints on the sum of the neutrino masses and light relic particles Henning:2017nuy; Wu:2019hek; Sherwin:2016tyf. While maps of gravitational lensing can be used to remove lensing contamination from low-resolution CMB surveys targeting inflationary B-modes Manzotti:2017net; Ade:2018gkx, high-resolution surveys can also search for inflationary B-modes directly in their own data. Finally, high-resolution CMB maps can be used to discover massive clusters of galaxies Bleem:2014iim; Hilton:2017gal, dusty star-forming galaxies Vieira:2009ru; Gralla:2019jyc, and new transient sources Whitehorn:2016njg.

SPT-3G is a new receiver for the 10-meter South Pole Telescope (SPT) that was deployed in 2017 targeting these science goals Benson:2014qhw; Bender:2018dwl. Its focal plane consists of $\sim 16,000$ transition-edge sensor (TES) bolometers split between ten 150 mm-diameter silicon wafers. The detectors that were deployed during the 2017 and 2018 observing seasons used a 4-layer titanium/gold TES design fabricated at Argonne National Laboratory Carter:2018unr; however, in parallel we also developed an aluminum-manganese TES design similar to the ones developed by NIST for SPTpol Henning:2012fz and Advanced ACTPol Li:2016. During the 2018-2019 austral summer, we replaced one of the deployed detector wafers with an Al-Mn wafer that we had characterized extensively in the lab. In this paper, we describe the fabrication and lab characterization of the Al-Mn detectors, as well as their on-sky performance, which is comparable to that of the Ti/Au detectors used in SPT-3G.

2 Design and Fabrication

TESs fabricated with Al-Mn alloys have several attractive features. Since Al-Mn is an alloy, it can be deposited in single step without requiring multiple layers to achieve the target critical temperature ($T_{c}$), thus simplifying the fabrication process. The $T_{c}$ and the normal resistance ($R_{n}$) of the devices can also be tuned relatively independently of each other. The $R_{n}$ of the film can be tuned by adjusting the thickness and geometry, while the $T_{c}$ can be grossly tuned by adjusting the Mn doping and then finely adjusted by controlled heating of the film after deposition Li:2016. These features make Al-Mn a promising TES material for future CMB experiments, such as CMB-S4 Abazajian:2016yjj.

Except for the features on the bolometer island, the fabrication of this Al-Mn detector wafer at Argonne was similar to previous SPT-3G detectors, as detailed in Posada:2016. The layout of the SPT-3G pixel and the Al-Mn bolometer island are shown in Fig. 1. The TES films are 75 $\mathrm{nm}$ thick, sputtered from aluminum doped with 850 ppm manganese by atomic percentage, capped with 15 $\mathrm{nm}$ Ti and 15 $\mathrm{nm}$ Au. The 15 $\mathrm{nm}$ Au cap layer protects the TES film after deposition, and the 15 $\mathrm{nm}$ Ti layer prevents the formation of a compound between the Au cap and the Al in the TES film. The TES film was patterned using lift-off to define 10 $\mathrm{\SIUnitSymbolMicro m}$ by 70 $\mathrm{\SIUnitSymbolMicro m}$ devices. After TES film deposition, the wafer was baked at 180 C for 10 min to adjust the $T_{c}$ to the target of 420 $\mathrm{mK}$. The Au cap is a normal metal, so it reduces $T_{c}$ through the proximity effect, but the change in $T_{c}$ is small because the Au layer is thin. A 600 $\mathrm{nm}$ palladium film was also deposited on the island, overlapping with the TES, to increase the total heat capacity and improve the electrothermal stability of the bolometer. This Pd film was thinner than the 850 $\mathrm{nm}$ films used previously in SPT-3G, in order to reduce the thermal time constant of the detectors.

3 Detector Performance

3.1 Electrothermal properties

The yield of functional detectors on the Al-Mn detector wafer is very similar to the rest of SPT-3G Bender:2018dwl; Bender:2019. Of 1572 wired detectors, 91.7% have nominal resistances measured at 300 $\mathrm{K}$ and 89.9% correspond to measured resonances in the readout electronics. Losses are due to a combination of on-wafer opens and shorts, defects in wirebonding, and open channels in the multiplexing readout electronics ($<2$%). During routine observing during May 2019, an average of 79.4% of all detectors were actually operated in the transition. Nearly all of the additional losses (9.6% of the 10.4% difference) are due to a minimum requirement on optical responsivity.

The TES $R_{n}$ and $T_{c}$ were measured in the lab prior to deployment on approximately one third of the detectors on the wafer. The measured $T_{c}$ of $441\pm 3$ $\mathrm{mK}$ was acceptably close to the 420 $\mathrm{mK}$ target and had excellent uniformity of better than 1% across the wafer, as shown in Fig. 2. This difference in $T_{c}$ is typical of the batch-to-batch variation of devices fabricated at Argonne. The $R_{n}$ of each bolometer in the deployed system is estimated as

[TABLE]

where $R_{\textrm{total}}$ is the total resistance measured by the readout system in the normal state, $R_{p}(f_{b})$ is the average parasitic impedance as a function of bolometer bias frequency111The parasitic impedance measured in the superconducting state is dominated by reactances that produce a characteristic frequency-dependence across all detector wafers and readout modules. $f_{b}$, and $c(i)$ is an empirical correction factor accounting for stray capacitance as a function of resonator index in the LC multiplexer chips as described in Dutcher:2018fod. An $R_{n}$ of $1.41\pm 0.15$ $\mathrm{\SIUnitSymbolOhm}$ was measured in the deployed wafer, also shown in Fig. 2. Of the variance in $R_{n}$, 0.06 $\mathrm{\SIUnitSymbolOhm}$ is attributable to using a model for $R_{p}$ rather than bolometer-by-bolometer measurements, and much of the remainder is likely due to residual calibration uncertainties in the readout electronics; DC four-wire measurements of $R_{n}$ typically show much lower scatter than measurements made with our frequency-domain multiplexing (fMux) readout electronics. The variance in $R_{n}$ in Fig. 2 should therefore not be interpreted as entirely due to physical variation in $R_{n}$ across the wafer.

The saturation powers were measured both in the lab and with optical load in the field, shown in Fig. 3. The detectors are operated under a voltage bias by placing a $30$\mathrm{m\SIUnitSymbolOhm}$$ shunt resistor in parallel with the TES. Saturation powers are calculated assuming that there is an additional 0.9 $\mathrm{nH}$ stray inductance in series with the shunt resistor. Although we do not have a complete model of all parasitic impedances in the readout circuit, we find that this value is similar to previous measurements of this stray inductance, and including this parasitic impedance alone is sufficient to eliminate the spurious dependence of the saturation power on bias frequency222Note that saturation powers reported in Dutcher:2018fod do not include this correction.. Since the saturation power depends on the thermal conductivity of the SiNx legs supporting the bolometer island, its value should be independent of the frequency at which it is biased by the readout electronics. Relative to the saturation powers measured using nominal values of components in the readout circuit, the saturation powers are increased by

[TABLE]

where $P^{\textrm{nominal}}_{\textrm{sat}}$ the saturation power assuming the nominal value $R_{sh}$ of the shunt resistor, $f_{b}$ is the TES bias frequency, and $L_{sh}$ is the inductance in series with the shunt resistor. The target saturation power for this wafer, as for previous SPT-3G detectors, is 10 / 15 / 20 $\mathrm{pW}$ for 95 / 150 / 220 $\mathrm{GHz}$. Operated at a base temperature of 280 $\mathrm{mK}$, the median dark saturation powers were measured to be 10.6 / 14.8 / 19.1 $\mathrm{pW}$, in excellent agreement with the target. During 2019 we began operating the SPT-3G receiver at a higher base temperature of 315 $\mathrm{mK}$ to improve the 3He/4He refrigerator cycle efficiency, so the saturation powers shown Fig. 3 are smaller but still have sufficient margin to avoid saturation.

3.2 Optical properties

Optical time constants are measured regularly during observing using a chopped thermal source in the center of the SPT-3G secondary mirror. The median time constants by bolometer are shown in Fig. 4 and average 4.4 / 3.6 / 1.7 $\mathrm{ms}$ for 95 / 150 / 220 $\mathrm{GHz}$ detectors, when measured with atmospheric loading at an elevation of 67.5 deg. The electrical power is highest for the 220 $\mathrm{GHz}$ detectors, which results in the smaller optical time constants for this band. At the SPT-3G scan speed of 1 deg/sec on the azimuth bearing (less on the sky), the 3 $\mathrm{dB}$ frequencies of these TESs correspond to angular multipoles between 30,000 and 90,000 depending on observing elevation and band. Since the 3 $\mathrm{dB}$ point of telescope beam at 150 $\mathrm{GHz}$ is $\ell\sim 8,000$, the detectors are fast enough to preserve good sensitivity to point sources.

To measure the optical efficiency, the detectors were illuminated by a blackbody source, and saturation powers were measured at eight temperatures of the blackbody between 8 $\mathrm{K}$ and 15 $\mathrm{K}$. The measured saturation powers, $P_{\textrm{sat}}^{\textrm{nominal}}$, as a function of cold load temperature, $T_{\mathrm{cl}}$, are fit to the model

[TABLE]

where the fit parameters $P_{0}$ and $\eta$ represent respectively the saturation power with no optical load and the optical efficiency. Several additional factors are included to correct for systematic biases in the measurement:

• •

Wafer heating: The cold load aperture illuminates only one triangular hextant of the wafer, through three low-pass metal-mesh filters (cutoff of 300 $\mathrm{GHz}$), and other detectors are blocked by an aluminum plate maintained at the base temperature near 250 $\mathrm{mK}$. The saturation powers of blocked detectors show a small linear dependence on the cold load temperature, suggesting that the radiative load is heating up the wafer slightly. We fit data from each blocked detector to a linear model of the form $P_{\textrm{sat}}^{\textrm{nominal}}(T_{cl})=P_{0}+CT_{cl}$, compute the average $\left<C\right>$ of the slopes across all bolometers in each observing band, and then assume that the illuminated bolometers experience the same thermal effects as the blocked detectors by including the term $\left<C\right>T_{cl}$ in Eq. 3. The net effect of this term is to decrease the estimated optical efficiency.

• •

Filter transmission: The blackbody illuminates optically active detectors through a stack of three low-pass metal-mesh filters. While we do not have high-quality measurements of all filters over the full observing band, a measurement of one filter over the 150 and 220 $\mathrm{GHz}$ bands has an average transmission of 96%. For simplicity, we therefore assume each filter has a fixed frequency-independent transmission of 96%, represented by the factor $T_{filter}(\nu)$ in Eq. 3.

• •

Triplexer response: $T_{triplexer}(\nu)$ represents the frequency response of the inline triplexer for each band, which is taken from Sonnet simulations that assume a dielectric thickness typical of deployed wafers.

• •

Inductance in series with shunt resistor: As discussed in Section 3.1, there is a wiring inductance $L_{sh}\sim 0.9~{}$\mathrm{nH} in series with the $R_{sh}=30~{}$\mathrm{m\SIUnitSymbolOhm} shunt resistor. This gives rise to a difference between the physical power incident on the TES, which is accounted for by the first factor in front of the integral in Eq. 3.

• •

Parasitic resistance in series with TES: As discussed in Section 3.1, there is an impedance in series with the TES, so that the voltage bias produced by the shunt resistor is divided between the series resistance and the TES. This effect is accounted for by the second factor in front of the integral in Eq. 3.

Fig. 5 shows the total optical efficiency as a function of bias frequency, estimated from the fit of Eq. 3. The mean efficiencies are 81% / 83% / 73% for 95 / 150 / 220 $\mathrm{GHz}$. The correction for the bias inductance has removed most of the spurious bias-frequency-dependence of optical efficiency within each band. Some slight dependence may remain in the 220 $\mathrm{GHz}$ detectors, which could explain the lower estimate efficiency in that band.

In addition to the lab measurements, we also routinely measure the noise-equivalent temperature (NET) of the detectors during observing. While NET is not an equivalent metric to optical efficiency, poor optical efficiency will degrade the NET. The median NET during June 2019 for this Al-Mn wafer was 675 / 523 / 1919 \mathrm{\SIUnitSymbolMicro K}$\sqrt{$\mathrm{s}$}$ for 95 / 150 / 220 $\mathrm{GHz}$ detectors, versus 640 / 511 / 1815 \mathrm{\SIUnitSymbolMicro K}\sqrt{\mathrm{s}} for all other detectors. The small difference in NET between the Al-Mn wafer and the other Ti/Au SPT-3G wafers is comparable to the variation among the Ti/Au wafers, suggesting that overall performance and optical efficiency of the Al-Mn is similar to the Ti/Au.

4 Status and Outlook

We have fabricated an array of Al-Mn TES bolometers for the SPT-3G experiment. Extensive lab characterization of this array indicates that it meets our design specifications for normal resistance, saturation power, critical temperature, and optical efficiency, with high uniformity in all of these parameters. It was installed in the SPT-3G receiver and has been used in the 2019 observing season, where on-sky yield and NET are very similar to the other detector wafers. The excellent overall performance indicates that Al-Mn TES bolometers produced by Argonne would be suitable for a next-generation CMB experiment such as CMB-S4.

Acknowledgements.

The South Pole Telescope is supported by the National Science Foundation (NSF) through grant PLR-1248097. Partial support is also provided by the NSF Physics Frontier Center grant PHY-1125897 to the Kavli Institute of Cosmological Physics at the University of Chicago, and the Kavli Foundation and the Gordon and Betty Moore Foundation grant GBMF 947. Work at Argonne National Laboratory, including Laboratory Directed Research and Development support and use of the Center for Nanoscale Materials, a U.S. Department of Energy, Office of Science (DOE-OS) user facility, was supported under Contract No. DE-AC02-06CH11357. We acknowledge R. Divan, L. Stan, C.S. Miller, and V. Kutepova for supporting our work in the Argonne Center for Nanoscale Materials. Work at Fermi National Accelerator Laboratory, a DOE-OS, HEP User Facility managed by the Fermi Research Alliance, LLC, was supported under Contract No. DE-AC02-07CH11359. NWH acknowledges support from NSF CAREER grant AST-0956135. The McGill authors acknowledge funding from the Natural Sciences and Engineering Research Council of Canada, Canadian Institute for Advanced Research, and the Fonds de recherche du Qubec Nature et technologies. JV acknowledges support from the Sloan Foundation.