Advertisement

Journal of Low Temperature Physics

, Volume 167, Issue 3–4, pp 121–128 | Cite as

Implications of Weak Link Effects on Thermal Characteristics of Transition-Edge Sensors

  • C. N. Bailey
  • J. S. Adams
  • S. R. Bandler
  • R. P. Brekosky
  • J. A. Chervenak
  • M. E. Eckart
  • F. M. Finkbeiner
  • R. L. Kelley
  • D. P. Kelly
  • C. A. Kilbourne
  • F. S. Porter
  • J. E. Sadleir
  • S. J. Smith
Article

Abstract

Weak link behavior in transition-edge sensor (TES) microcalorimeters creates the need for a more careful characterization of a device’s thermal characteristics through its transition. This is particularly true for small TESs where a small change in the bias current results in large changes in effective transition temperature. To correctly interpret measurements, especially complex impedance, it is crucial to know the temperature-dependent thermal conductance, G(T), and heat capacity, C(T), at each point through the transition. We present data illustrating these effects and discuss how we overcome the challenges that are present in accurately determining G and T from I–V curves. We also show how these weak link effects vary with TES size. Additionally, we use this improved understanding of G(T) to determine that, for these TES microcalorimeters, Kaptiza boundary resistance dominates the G of devices with absorbers while the electron-phonon coupling also needs to be considered when determining G for devices without absorbers

Keywords

Transition edge sensor (TES) Thermal conductance Weak link effects Kapitza resistance Electron-phonon coupling 

1 Introduction

To achieve the optimal performance in a transition-edge sensor (TES), it is critical to understand the physical properties of the superconducting-to-normal transition. Recent measurements show that our TESs act as weak superconducting links [1, 2, 3, 4], where the behavior of the critical current, I c , with temperature, T, depends on the device size, L; I c (T,L)αe L/ξ(T), where ξ(T) is the temperature-dependent coherence length [1] (Fig. 1e). Therefore, weak link effects become more prominent as the TES size is decreased, making device characterization more complex. Hence, a better understanding and method of TES characterization must take effect for continued detector improvements.
Fig. 1

(Color online) (a) 35 μm TES with no absorber. (b) Complete TES microcalorimeter pixel with absorber overhanging a 12 μm TES. (c) Typical I–V curve. (d) Typical resistance vs. voltage curve showing the curved ‘normal’ state. (e) Critical current vs. temperature for a 12 μm (red) and 50 μm (blue) TES

We have developed high fill-factor, high quantum efficiency, kilo-pixel arrays of TES microcalorimeters that provide high spectral resolution for x-ray astrophysics and solar physics applications. These devices [5, 6, 7, 8] are made of square Mo(45 nm)/Au(200 nm) bilayers (Fig. 1a) and have transition temperatures of ∼100 mK. Au or Au/Bi absorbers overhang the TES (Fig. 1b), supported through a small area attachment stem, to provide sufficient x-ray stopping power. Different applications have varying requirements, such as count rate and spatial scales. To meet these various specifications we are developing devices which have the same basic geometry as those developed for x-ray astrophysics applications [9], but with different TES sizes (to satisfy the needs of varying spatial scales), normal-metal features (to reduce noise at each size), absorber stem attachments (to maintain superior energy resolution at each size), and fabricated on the silicon substrate [10], instead of on silicon nitride membranes (speeding the device response to allow higher count rates). Here we focus solely on these ‘solid-substrate’ devices and compare characterization techniques for TESs varying in size, from square 12 μm Mo/Au bilayers to devices that are 20, 35, 50, and 140 μm square.

2 Determining G Considering Weak Link Effects

Our typical microcalorimeters that we have developed for x-ray astrophysics applications [9] use TESs that are 140 μm square and have thick absorbers, which dominate the heat capacity. In these devices, the physical properties such as G, C, and T c , are constant through the transition. While developing smaller TESs we have entered a new regime in which weak link effects are more evident, causing these devices to be highly current dependent where the operating temperature, and therefore G(T), change based on the bias current. For example, for a 12 μm square TES the operating temperature can vary by ∼20% throughout the superconducting transition while for a 50 μm square TES it varies by ∼10% and for a 140 μm square TES it only varies by ∼2.5%.

The typical procedure for experimentally measuring the thermal conductance, G, of a TES involves measuring the joule heating power as a function of bath temperature, which is achieved by recording how the sensor current varies with applied voltage (an I–V curve, see Fig. 1c) at various bath temperatures. Then, we perform a fit to the data of the form:
$$ P(T) = K (T^{n} - T_{b}^{n})$$
(1)
to find the coefficient K and exponent n, where T is the TES temperature and T b is the bath temperature. The thermal conductance, at a given temperature, is then equal to:
$$ G(T) = n K T^{(n-1)}$$
(2)
While this is a widely used and straightforward method, several challenges can arise. First, the I–V curve usually needs to be centered on zero. This is normally done by selecting a region of data where the TES resistance is above the abrupt onset of superconductivity (at high applied voltage), finding the slope, and then forcing the line to go through zero. However, this resistance has a slow temperature dependence (Fig. 1d). To accurately center the curves, one must use the region above the abrupt onset of superconductivity on both sides of the I–V curves. Even though this is well known, in some measurement set-ups and/or conditions it can be difficult, if not impossible, to measure both sides of the I–V curve without SQUID jumps or other instabilities. Additionally, it is only possible to correctly center using both sides of an I–V curve if the curve was measured with no trapped magnetic fields, as these may introduce asymmetries in the curve shape. In the past we have often assumed that both centering techniques result in the same inferred G. We have found that improperly centering I–V curves introduces systematic errors up to 10%.

A second challenge in determining the thermal conductance of a TES is in dealing with weak-link effects. This typical procedure used to fit for G assumes the device’s critical current is constant with temperature for T b <T c . While this is a fair approximation for the 140 μm square TESs, it is no longer appropriate for small devices where weak-link behavior becomes stronger as the length scale, i.e. TES size, decreases. This highly current-dependent transition shape makes the determination of G challenging [11], as we no longer expect our data to be consistent with (1) globally. To accurately interpret measurements, it is crucial to calculate G at many discrete points through the TES transition by iterating the procedure described earlier in this section for many TES resistances.

Since our data is no longer consistent with (1) globally, due to the current-dependence in the transition (e.g., see Fig. 1e), at the highest bath temperatures the measured power is higher than (1) would predict (Fig. 2). Therefore, one must select a subset of points in which to do the curve fitting. Since it can be very difficult to know which points should be included in the fit, a more accurate method, if the dominant mode for the thermal conductance is clear, is to fix the exponent in the fit. When determining what value to fix the exponent, n, we used input from previous data and devices of varying sizes (see Sect. 3). Based on this study we concluded that the Kapitza coupling dominates the thermal conductance between the TES and substrate. This conclusion only holds for data from devices with absorbers and fabricated on solid substrates (as opposed to membranes).
Fig. 2

(Color online) Power versus temperature plots for a 12 μm (left) and 50 μm (right) TES. The points are derived from I–V curves, at 10% R/R n (red), 50% R/R n (green) and 80% R/R n (blue). The dotted lines are fits to the data, of the form given in (1), where the exponent is allowed to float while the solid lines correspond to fits with a fixed exponent, n=4 (consistent with dominant contribution to G coming from Kapitza coupling). The derived G (100 mK) value is given for each of the n=4 curves. The best-fit effective transition temperature varies by ∼40 mK (or ∼25%) on the 12 μm device but are much more constant in the 50 μm device. Also note how the data trails off the fit curve at high temperatures for all 12 μm curves, but only the 50 μm curve corresponding to points lowest in the transition

We calculate the device heat capacity, C, by measuring the intrinsic pulse fall time, τ, above the transition temperature, where there is minimal influence from electrothermal feedback due to the readout circuit, where C(T)=τ(T)G(T). Therefore we can verify that we determine the correct thermal conductance by comparing the calculated and expected heat capacity, which is calculated based on the measured absorber and TES volume and material properties. When fixing the exponent, the calculated heat capacity is consistent with expectation, for all but the smallest sized devices (Fig. 3). Thus, in the presence of weak-link effects, the best G estimate is made by fixing the exponent, when the coupling physics is known, and only using the points measured at the lowest bath temperatures.
Fig. 3

(Color online) The graph shows the ratio of the calculated to expected heat capacity as a function of sensor resistance for a 12 μm (red), 20 μm (green), and 50 μm (blue) TES. The closed circles correspond to fits that allowed the exponent to float while the open squares correspond to fits that fixed the exponent to be 4. When the exponent is allowed to float, for the 20 μm TES, the calculated heat capacity is lower than expected. When fixing the exponent, the calculated heat capacity is consistent with expectation

3 Kapitza vs. Electron-Phonon Coupling for Various Microcalorimeter Device Designs

To fully understand our devices, we want to know what physics dominates the thermal resistance for our solid substrate devices—electron-phonon coupling or Kapitza boundary resistance between the metal film and the substrate. Previously we conducted a limited study of 140 μm square devices, with T c ’s between 50–120 mK, and found that the devices with absorbers were consistent with the cubic temperature dependence, or an exponent n=4, expected of a Kapitza resistance, with a boundary conductivity of 217T 3 W/m2 K, roughly 3 times lower than expected [9]. The limited data from devices with no absorbers suggested a significant electron-phonon contribution to the total thermal conductance [9].

Recently we have measured G of many more solid-substrate devices, including TESs with and without absorbers and with T c ’s up to approximately 275 mK. The inclusion of this additional data allows us to further understand the contribution of the Kapitza resistance and electron-phonon coupling to the total thermal conductance. Since this data is from many differently sized devices, to directly compare them, G needed to be scaled to account for these size differences. Since, for Kapitza resistance, G is proportional to area, the total thermal conductance measured for each device was divided by area. The area used is the area of the Mo/Au bilayer plus the area of the stem extended past the bilayer minus the area under the bias leads. A separate analysis of test devices showed that the area of the bilayer under the bias leads does not contribute to the measured G.

The new data from TESs with absorbers is consistent with a cubic dependence on temperature (Fig. 4), but with a boundary conductivity of about 500T 3 W/m2 K, closer to the theoretical value of ∼633T 3 W/m2 K [12]. Taking a closer look at the data, the 140 μm data alone does not go to very high temperatures so it is difficult to determine if different sized devices have different boundary conductivities or if the fit is skewed.
Fig. 4

(Color online) Plot of the measured thermal conductance, divided by area, as a function of \(T_{0}^{3}\) (where T 0 is the TES temperature at a specific bias point in the transition). The two lines correspond to linear fits to data with absorbers (black) and data from devices without absorbers (grey). For the small devices, where G(T 0) measurably changes throughout the transition due to weak link effects, we have calculated G(T 0) and T 0 at 8 points through the transition, at sensor resistances corresponding to 10% R n to 80% R n . These points are included in the fit and connected together in the plot

The measured G from devices without absorbers is lower than expected from a Kapitza resistance inferred from the devices with absorbers (Fig. 4). Currently we understand this difference to be due to the different volumes, roughly 100×, associated with devices with and without absorbers. Since the electron-phonon conductance is proportional to volume, instead of surface area, as is the case with Kapitza boundary resistance, the data indicates that the added volume of the absorber causes the electron-phonon conductance to become much higher than the Kapitza component, so that the latter then determines the total thermal conductance for devices with absorbers. However, without absorbers, the electron-phonon conductance is weak enough to impact the overall G, therefore lowering it. In order to obtain a good estimate of the electron-phonon coupling, and determine if superconducting effects change this coupling in the TES (depending on its SNS phase) [13], we need additional data points, especially at the higher temperatures.

4 Conclusions

The current dependence of the transition due to weak leak behavior, which is most pronounced in small devices, makes standard TES characterization more challenging, in particular the TES temperature, and therefore the thermal conductance, is not constant through the transition. To fully characterize a TES exhibiting weak link behavior, one must determine G at each point through the transition.

Fitting the power vs. temperature to calculate thermal conductance can be especially challenging for these small devices. In particular, for a good polynomial fit one must remove points at the highest temperatures, near T c . However, it can be difficult to determine which should be removed. We found that, especially for the smallest devices, the best technique for fitting the power vs. temperature curves was to fix the exponent.

When comparing thermal conductance data between many devices, with different transition temperatures and with and without absorbers, we see that devices with absorbers have Gs that are Kapitza dominated while devices without absorbers have Gs that are electron-phonon dominated.

Notes

Acknowledgements

This research was supported in part by an appointment to the NASA Postdoctoral Program at Goddard Space Flight Center (C.N. Bailey), administered by Oak Ridge Associated Universities through a contract with NASA.

References

  1. 1.
    J.E. Sadleir et al., Phys. Rev. Lett. 104, 047003 (2010) ADSCrossRefGoogle Scholar
  2. 2.
    S.J. Smith et al., J. Appl. Phys. (2011, submitted) Google Scholar
  3. 3.
    J.E. Sadleir et al., J. Low Temp. Phys. (2012). Special issue LTD14 Google Scholar
  4. 4.
    S.J. Smith et al., J. Low Temp. Phys. (2012). Special issue LTD14 Google Scholar
  5. 5.
    J.E. Sadleir et al., Phys. Rev. B 84, 184502 (2011) ADSCrossRefGoogle Scholar
  6. 6.
    S.R. Bandler et al., Proc. SPIE 7732, 773238 (2010) CrossRefGoogle Scholar
  7. 7.
    S.J. Smith et al., J. Low Temp. Phys. (2012). doi: 10.1007/s10909-012-0574-y. Special issue LTD14 Google Scholar
  8. 8.
    M.E. Eckart et al., J. Low Temp. Phys. (2012). doi: 10.1007/s10909-012-0514-x. Special issue LTD14 Google Scholar
  9. 9.
    C.A. Kilbourne et al., J. Low Temp. Phys. 151, 223–228 (2008) ADSCrossRefGoogle Scholar
  10. 10.
    F.M. Finkbeiner et al., IEEE Trans. Appl. Supercond. 21, 223–226 (2011) ADSCrossRefGoogle Scholar
  11. 11.
    M.A. Lindeman et al., J. Low Temp. Phys. 151, 180–184 (2008) ADSCrossRefGoogle Scholar
  12. 12.
    E.T. Swartz, R.O. Pohl, Rev. Mod. Phys. 61, 605 (1989) ADSCrossRefGoogle Scholar
  13. 13.
    T.T. Haikkila, F. Giazotto, Phys. Rev. B 79, 094514 (2009) ADSCrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media, LLC 2012

Authors and Affiliations

  • C. N. Bailey
    • 1
  • J. S. Adams
    • 2
    • 3
  • S. R. Bandler
    • 2
    • 4
  • R. P. Brekosky
    • 2
    • 5
  • J. A. Chervenak
    • 2
  • M. E. Eckart
    • 2
    • 3
  • F. M. Finkbeiner
    • 2
    • 6
  • R. L. Kelley
    • 2
  • D. P. Kelly
    • 2
    • 7
  • C. A. Kilbourne
    • 2
  • F. S. Porter
    • 2
  • J. E. Sadleir
    • 2
  • S. J. Smith
    • 2
    • 3
  1. 1.NASA Postdoctoral Program FellowNASA Goddard Space Flight CenterGreenbeltUSA
  2. 2.NASA Goddard Space Flight CenterGreenbeltUSA
  3. 3.CRESST and University of MarylandBaltimore CountyUSA
  4. 4.CRESST and University of MarylandCollege ParkUSA
  5. 5.Northrop Grumman Information TechnologyMcLeanUSA
  6. 6.Wyle Information SystemsMcLeanUSA
  7. 7.Muniz Engineering Inc.SeabrookUSA

Personalised recommendations