ARTICLE IN PRESS

Nuclear Instruments and Methods in Physics Research A 520 (2004) 296–299

*Corresp

E-mail a

0168-9002/$

doi:10.1016

Study of a Mo–Au TES deposited directly ona freestanding membrane

J.E. Olsena,*, E.C. Kirka, Ph. Lercha, M.E. Huberb, K. Arznera,W. Hajdasa, A. Zehndera, H.R. Ottc

aPaul Scherrer Institute, CH-5232 Villigen, SwitzerlandbUniversity of Colorado, Denver, CO 80217, USAc ETH-H .onggerberg, CH-8093 Z .urich, Switzerland

Abstract

Transition edge sensor devices made with Mo–Au proximity bilayer thermometers have been fabricated. The bilayers

were DC-sputter deposited directly on 250 nm thick pre-etched Si3N4 membranes. Mo was deposited at 700C; whereas

Au was deposited at room temperature in the same pumpdown in order to allow for a good proximity effect. The

superconducting transition temperature of devices with 43 nm Mo and 200 nm Au was around 190 mK: The transition

width of bilayers deposited on membranes was substantially larger than those measured on samples deposited on the

bulk substrate in the same fabrication run ðE2 mKÞ: We consider this difference to be the result of a large temperature

gradient across the membrane during the deposition process. Devices were tested with a single stage SQUID array kept

at 2:2 K: The resolution of a device that has a transition width of 30 mK is 30 eV at 6 keV: A detailed study shows that

the detector noise roughly scales with aI0; confirming the importance of thermal fluctuation noise in the thermometer.

r 2003 Published by Elsevier B.V.

PACS: 85.25.Oj

1. Introduction

TES microcalorimeters [1] require an absorberin good thermal contact with a sensitive thermo-meter which in turn has a weak contact with athermal bath. The thermometer is a superconduct-ing film deposited on a thin membrane of Si3N4:The membrane structure is usually obtained byetching of the underlying Si from the originalSi3N4=Si substrate. One approach is to firstdeposit the thermometer on a Si3N4=Si substrateand then to etch away Si. This route imposes

onding author.

ddress: john.olsen@psi.ch (J.E. Olsen).

- see front matter r 2003 Published by Elsevier B.V.

/j.nima.2003.11.242

severe measures to prevent the thermometer frombeing etched. Future designs will include severalpixels and reduced pitch. Protection may becomemore difficult to realize. We explore the oppositeroute and fabricate membranes first and followwith the device.

2. Fabrication

First, a Si wafer is coated on both sides with250 nm thick amorphous Si3N4: We define squarewindows and cleaving grooves into one layer ofSi3N4 and etch through the Si using a standardKOH solution. This step yields single 12 12 mm2

ARTICLE IN PRESS

0.12

0.10

0.08

0.06

0.04

0.02

0.00

Res

ista

nce

[Ω]

0.300.280.260.240.220.200.18Substrate Temperature [K]

(1) Thermometer on window, 1µA(2) Thermometer on bulk, 1µA

(1)(2)

Fig. 1. Several superposed resistance vs. temperature sweeps of

two Mo/Au bilayers.

J.E. Olsen et al. / Nuclear Instruments and Methods in Physics Research A 520 (2004) 296–299 297

substrates with freestanding membranes. A100 nm thick layer of Nb is DC-sputtered in orderto form the electrical contacts. They are shapedusing one lithographical step followed by wetetching ðH2O:HNO3:HF 6:6:1Þ:

Mo and Au compose the proximity bilayer andare DC-sputtered in one single pumpdown. The 43nm Mo layer is deposited at 700C: The substrateis allowed to reach room temperature ð4 hÞ andcleaned for 3 min with a fast atom beam (FAB).The deposition of the 200 nm thick Au layer onthe cleaned interface begins B15 s later. Aphotolithographical step followed by wet etchingis used to shape 150 400 mm2 thermometersðCv ¼ 0:3 pJ=KÞ: The Au etch used ðKI:I2:H2O4:1:40) is found to etch Mo as well when in thepresence of Au. We observe that the etching ratefor Mo is even higher than for Au, this produces aslight underetch at the edges of the device and isused to favor normal metal boundary conditions[2]. On each chip there is one series of samples onsuspended Si3N4 membranes and one on Si3N4=Si:

3. Characterization

In Fig. 1 we present the variation of resistancewith temperature for two bilayers. The devicedeposited on Si3N4=Si has a much sharpertransition than its companion made on Si3N4:This difference cannot be explained by self-heating, ðmax 0:1 pWÞ since the tail of the transi-tions occur almost at the same temperature.Again, both bilayers were processed together onthe same wafer. We suppose that the severedifference in transition width is due to a tempera-ture gradient created across the membrane duringthe sputter deposition. This gradient is likely tocreate mechanical stress which may induce a non-homogeneous film growth. We thermally cycledthe sample 15 times during a period of severalmonth and did not notice any changes, however.

4. Experiment

The detector and a 1:4 mO thin film Au shuntare mounted in a custom made dilution refrig-

erator reaching 50 mK: The TES is connected inseries with the input coil of a single stage DC-SQUID array using a 10 cm long twisted pair ofNbTi/Cu. Al wire bonds connect to the devicewhereas Nb wire bonds [3] connect to the inputcoil. A bulk Nb can placed at 2:2 K shields theSQUID. The feedback line includes a 2 kO resistorthermally anchored at 2:2 K and connected to afeedback unit. An uncollimated source of 6 keVphotons illuminates the device.

We studied the performance of our TESoperated in negative electrothermal feed backmode. The equilibrium TES resistance R0 wasvaried and data sets of 2000 pulses containing 4000points each were recorded at a sampling rate of2:5 106 s1: A typical 6 keV pulse is shown in theinset of Fig. 2. The height of each pulse isestimated using an optimal filter template con-structed for each data set individually, as describedin Ref. [4]. In Fig. 2 (top) we show the best 6 keVpulse height spectrum with a resolution of DE ¼30 eV (FWHM). This kind of performance isachieved for 7oR0o30 mO: The dependence ofDE on R0 is shown in the lower part of Fig. 2. Theperformance loss at high R0 values is due to thepulse amplitude reduction that follows the reduc-tion of a ¼ d log R=d log T : At very low R0 values,however, the detector is less stable as revealed bythe enhanced scattering of the data together withthe general increase of DE: The stability of theSQUID needs also to be taken into account.

ARTICLE IN PRESS

120

100

80

60

40

20

0

Cou

nts

per

bin

(ar

bitr

ary

bin

wid

th)

6.66.46.26.05.85.6Photon Energy [keV]

Pulse height histogramm,optimally filtered pulses

Gaussian fit on Kα peakHeight: 110.9 FWHM: 29.9 eV

Gaussian fit on K peakHeight: 11.0 FWHM: 33.5 eV

40

38

36

34

32

30S

QU

ID c

urre

nt [

µA]

8006004002000Time [µs]

200

150

100

50

0

Res

olut

ion

∆E

FW

HM

[eV

]

706050403020100

TES resistance [mΩ]mOhm

ResolutionDifferent SQUID bias pointsResolution limit for 100 pA/Hz1/2 SQUID noise

β

Fig. 2. Top: Best 6 keV pulse height spectrum measured at

R0 ¼ 15 mO: Bottom: Resolution vs. TES equilibrium resis-

tance R0:

10-12

10-11

10-10

10-9

10-8

Cur

rent

Spe

ctra

l Den

sity

[A

/Hz]

100 Hz 1kHz 10kHz 100kHz 1MHz

Frequency [s-1]

Measured spectra: Modelled noise: i Signal+noise a Phonon noise ii Signal b Johnson noise iii Noise in signal c TFN in TES

d Sum

τ0-1 τeff

-1 τLR-1 τTES

-1a

b

c d

i

ii

iii

700

600

500

400

300

200

100

0Spe

ctra

l den

sity

of n

oise

[pA

/Hz1/

2 ]

3.53.02.52.01.51.00.50.0α × Ieq [mA]

Noise level at intermediatefrequency (~8 kHz)Guide to the eyes

Datasets using a similar TESbias but different SQUID bias

Fig. 3. Top: spectral densities of signal and noise; bottom,

signal noise vs. aIeq:

J.E. Olsen et al. / Nuclear Instruments and Methods in Physics Research A 520 (2004) 296–299298

Minute variations of the bias point of the SQUIDaffect detector performances. The current noise ofthe SQUID is measured E100 pA=

ffiffiffiffiffiffiffiHz

p: The

corresponding resolution limit has been estimatedassuming an ideal signal processing with an infinitebandwidth [5] and is shown as a straight line onthe graph.

We performed a detailed noise study. For eachdata set, we calculated the power spectral density(PSD) of the signal displayed i on the Fig. 3 top.Using an averaged pulse form we estimated thePSD of an ideal signal, ii. The difference of thesePSD represents the signal noise, iii. Surprisingly,no clear correlation between the noise level and theenergy resolution could be extracted. The mainnoise contributions [6] are also displayed.

The thermal fluctuation noise (TFN) in thedevice, which is expected to scale with aIeq;

dominates at intermediate frequencies. This factis confirmed in Fig. 3 (bottom) which shows thenoise level at 8 kHz as a function of aIeq:

5. Conclusion

The advantage of a fabrication route usingmembranes to begin with is compensated by thepresence of a thermal gradient during the deposi-tion of Mo. The superconducting transition widthof Mo/Au bilayers deposited on hot membranesare much wider than for bilayers deposited on hotbut bulk substrates of the same material. Thewider transition width limits the TESs perfor-mances. Best energy resolution ð30 eV–6 keVÞ isachieved if R0 ranges between 7 and 30 mO; Rn ¼120 mO: The scattering observed in the resolution

ARTICLE IN PRESS

J.E. Olsen et al. / Nuclear Instruments and Methods in Physics Research A 520 (2004) 296–299 299

data at low R0 cannot be explained by excess noiseonly.

Acknowledgements

We thank J. Flokstra and J. Sese from Univ. ofTwente (NL) for realizing the Nb bonds connect-ing the SQUID.

References

[1] W.M.B. Tiest, et al., AIP Conf. Proc. 605 (2002) 199.

[2] K.D. Irwin, et al., Nucl. Instr. and Meth. A 444 (2000)

184.

[3] W. Jaszczuk, et al., Meas. Sci. Technol. 2 (1991) 1121.

[4] A.E. Szymkowiak, et al., J. Low Temp. Phys. 93 (1993) 281.

[5] S.H. Moseley, et al., J. Appl. Phys. 56 (1984) 1257.

[6] H.F.C. Hoevers, et al., Appl. Phys. Lett. 77 (2000) 4422.