RESEARCH ARTICLE | AUGUST 06 2001

Picosecond superconducting single-photon optical detector


G. N. Gol’tsman; O. Okunev; G. Chulkova; A. Lipatov; A. Semenov; K. Smirnov; B. Voronov; A. Dzardanov;
C. Williams; Roman Sobolewski

Appl. Phys. Lett. 79, 705–707 (2001)

[Link]


View Export
Online Citation

Articles You May Be Interested In

Scanning tunneling microscope light emission spectroscopy with picosecond time resolution
Appl. Phys. Lett. (May 2000)

Optimization of tapered semiconductor optical amplifiers for picosecond pulse amplification

Appl. Phys. Lett. (December 2005)

Picosecond time-resolved imaging by nonscanning fluorescence Kerr gate microscope

Appl. Phys. Lett. (September 2005)

25 February 2025 [Link]

APPLIED PHYSICS LETTERS VOLUME 79, NUMBER 6 6 AUGUST 2001

Picosecond superconducting single-photon optical detector

G. N. Gol’tsman,a) O. Okunev, G. Chulkova, A. Lipatov, A. Semenov, K. Smirnov,
B. Voronov, and A. Dzardanov
Department of Physics, Moscow State Pedagogical University, Moscow 119435, Russia
C. Williams and Roman Sobolewskib)
Department of Electrical and Computer Engineering and Laboratory for Laser Energetics,
University of Rochester, Rochester, New York 14627-0231
共Received 22 January 2001; accepted for publication 1 June 2001兲
We experimentally demonstrate a supercurrent-assisted, hotspot-formation mechanism for ultrafast
detection and counting of visible and infrared photons. A photon-induced hotspot leads to a
temporary formation of a resistive barrier across the superconducting sensor strip and results in an
easily measurable voltage pulse. Subsequent hotspot healing in ⬃30 ps time frame, restores the
superconductivity 共zero-voltage state兲, and the detector is ready to register another photon. Our
device consists of an ultrathin, very narrow NbN strip, maintained at 4.2 K and current-biased close
to the critical current. It exhibits an experimentally measured quantum efficiency of ⬃20% for 0.81
␮m wavelength photons and negligible dark counts. © 2001 American Institute of Physics.
关DOI: 10.1063/1.1388868兴

Superconducting devices are the natural choice for fast thermalization length 2␭ T 关Fig. 1共a兲兴. After the initial ther-
and ultrasensitive optical detection, because of their quantum malization, the resistive hotspot size grows 关Fig. 1共b兲兴 as hot
nature and low-noise, cryogenic operation environment. The electrons diffuse out of its center. At the same time, the su-
superconducting energy gap 2⌬ is two to three orders of percurrent is expelled from the hotspot volume and is con-
magnitude lower than in a semiconductor, thus, photon ab- centrated in the ‘‘sidewalks’’ between the hotspot and the
edges of the film 关Fig. 1共c兲兴. If the bias current I bias is suffi-

25 February 2025 [Link]

sorption in a superconducting detector creates an avalanche
electron charge two to three orders of magnitude higher for cient to exceed the critical current in the sidewalks, the phase
the same photon energy. This results in an enhanced resolu- slip centers are sprung7 and a nonsuperconducting barrier is
tion in energy-resolving devices, such as superconducting formed across the entire width w of the device 关Fig. 1共d兲兴,
tunnel junctions,1 and extends the range of detectable ener- giving rise to a voltage signal, due to a collaborative effect of
gies well into the infrared for photodetectors.2 In addition, as the bias current and the radiation quantum. For a given ex-
we have recently demonstrated, energy relaxation time con- periment, the response magnitude is proportional to the bar-
stants of excited electrons in superconductors are in the pi- rier resistance, however, in general, the current-assisted
cosecond range for both the low-temperature3 and hotspot process creates a nonlinear, multidimensional space
high-temperature4 superconductors, assuring the gigahertz of operating parameters, such as w, I bias , ប␻, and T.
repetition rate for superconducting photon counters. The hotspot formation process competes, of course, with
the cooling process, as electrons diffusing out of the hotspot
The dynamics of the hotspot formation in a supercon-
lose their energy through electron–phonon scattering. Thus,
ductor at temperature T below its critical temperature T C , at
after the time depending on both the diffusion rate and the
the position where the photon is absorbed has been described
quasiparticle relaxation dynamics,6 the hotspot heals itself,
before5 and the supercurrent-assisted mechanism experimen-
leading to the restoration of the superconducting path along
tally demonstrated in this work was theoretically studied in
the microbridge. As a result of the hotspot creation and re-
Ref. 6. Therefore, we only mention that the absorption of a
laxation processes, the NbN device switches temporarily be-
photon with energy ប ␻ Ⰷ2⌬ creates, through electron–
electron and electron–Debye–phonon interactions, a local
nonequilibrium perturbation with a large number of excited
hot electrons 共above 300 in the case of NbN, excited with
790 nm wavelength light兲,2 and an increase of the average
electron temperature above T C . This initial thermalization
phase for ultrathin NbN films is characterized by the ther-
malization time ␶ T ⫽6.5 ps 共Ref. 3兲 and results in the forma-
tion of a hotspot—a local nonsuperconducting region of the

a兲
Also at: Department of Electrical and Computer Engineering and Labora-
tory for Laser Energetics, University of Rochester, Rochester, New
York 14627-0231. FIG. 1. Schematics of the supercurrent-assisted hotspot formation mecha-
b兲
Author to whom correspondence should be addressed: also at the Institute nism in an ultrathin and narrow superconducting strip, kept at temperature
of Physics, Polish Academy of Sciences, PL-02904 Warszawa, Poland; far below T C are shown. The arrows indicate direction of the supercurrent
electronic mail: sobolewski@[Link] flow.

0003-6951/2001/79(6)/705/3/$18.00 705 © 2001 American Institute of Physics

706 Appl. Phys. Lett., Vol. 79, No. 6, 6 August 2001 Gol’tsman et al.

with a 1 kHz repetition rate at 0.4, 0.81, 1.55, and 2.1 ␮m

wavelengths, with the bulk of the measurements performed
using 0.81 ␮m photons. During our experiments, the fluence
per pulse reaching the device plane inside the dewar was
approximately J in⬵1 fJ/ ␮ m2, and could be further attenu-
ated using banks of neutral density filters, giving the total
attenuation of 10⫺7 .
The actual fluence per pulse absorbed by our SPD, J abs ,
can be estimated according to the relation J abs⫽J inS d ␩ ,
where S d is the active area of the device and ␩ is the radia-
tion absorption coefficient of a metallic film, given by10
␩ ⫽4 共 R s /Z 0 兲 / 关共 R s /Z 0 兲共 n sub⫹1 兲 ⫹1 兴 2 , 共1兲
where n sub is the index of refraction of the SPD substrate, R s
FIG. 2. I – V characteristics of a 0.2 ␮m wide and 1.2 ␮m long NbN super-
is the surface resistance of the NbN film measured just above
conducting microbridge are shown. Point A denotes the initial detector bias T C , and Z 0 ⫽377 ⍀ is the free-space impedance. For our
level in the superconducting state and point B corresponds to the switched sapphire substrate (n sub⫽1.72), ␩ max⫽37%. ␩ is frequency
state upon absorbing a photon, leading a voltage pulse generation, before independent as long as n sub remains frequency independent
relaxing back to point A.
and the film is much smaller than the radiation skin depth,
and can be regarded as the intrinsic quantum efficiency 共QE兲
tween the superconducting and resistive states on a time of our device.
scale of ⬃30 ps. For a device biased near, but below I C 共point A in Fig.
We have developed simple to manufacture, easy to op- 2兲, photon absorption instigated the supercurrent-assisted
erate, superconducting single photon detectors 共SPDs兲 using hotspot formation leading to a temporary switch from the
nominally 0.2 ␮m wide and 1 ␮m long microbridges pat- superconducting state to the hotspot resistive state 共point B
terned from ultrathin 共5 nm thick兲 NbN films deposited on in Fig. 2兲 along the 50 ⍀ load line. As a result, an output

25 February 2025 [Link]

sapphire substrate.8 The microbridge was connected to the voltage was generated with a magnitude corresponding to the
external circuit, via much thicker and larger, Au-coated con- voltage level at point B, and that was independent of the
tact pads. Figure 2 presents a current–voltage (I – V) charac- actual photon energy, as long as the photon energy was suf-
teristics of a NbN microbridge, operated at 4.2 K and biased ficient to form a hotspot large enough to trigger the super-
using a voltage source. The characteristics are typical for a current redistribution effect. The response time of the voltage
long superconducting constriction9 and show that the bridge pulses followed the formation and subsequent healing of the
can be operated in either of the two distinct states: the super- resistive state induced by the photon absorption.6
conducting 共flux–flow兲 state or the resistive 共hotspot兲 state. The measured response of our SPDs 共not shown兲 was
The hotspot plateau under dc conditions corresponds to the indeed ‘‘quantum’’ or ‘‘granular,’’ in a sense that the voltage
growing normal-state region, as the voltage across the device pulse amplitude was roughly the same 共⬎400 mV after am-
is increased, eventually reaching the bridge normal-state re- plification, with a signal-to-noise ratio above 100:1兲 for all
sistance, which in our case is approximately 500 ⍀. The tested laser wavelengths. The response pulse width was
thick, dashed line represents a 50 ⍀ load line, when the ⬃100 ps, limited by the bandwidth of our chain of output
device is connected to the output microwave transmission amplifiers, and with negligible shot-to-shot jitter.
line. From Fig. 2, we see that the device I C is approximately True single-photon counting requires that the photon de-
78 ␮A. tection probability has a linear dependence on the number of
For our experiments, a NbN SPD was mounted on a cold photons incident on the device. For a mean number of m
plate (T⫽4.2 K) inside an optical liquid-helium cryostat. We photons per pulse, the probability P(n) of absorbing n pho-
used two cold glass filters 共inner glass window was at ⬃4.2 tons from a given pulse is P(n)⬃(e ⫺m m n )/(n!). When m
K兲 to block thermal radiation longer than 2.5 ␮m from the Ⰶ1 共achieved by drastically attenuating the flux of photons
sample. The sample was dc biased through a bias tee and incident on the SPD兲, the probability P(n) simplifies to
mounted on a rigid, 50 ⍀ coplanar transmission line with the mn
ac output connected through a stainless-steel, semirigid co- P共 n 兲⬃ . 共2兲
n!
axial cable to a cryogenic low-noise amplifier 共placed inside
the dewar兲, characterized by 30 dB gain and 1 to 2 GHz Consequently, the probability of absorbing one photon is
bandwidth. The noise temperature of our cryogenic amplifier proportional to m, the probability of detecting two photons is
was below 15 K, which yielded voltage fluctuations below 7 proportional to m 2 , and so on.
␮V—several orders of magnitude below our signal levels. Figure 3 shows the probability of the detector producing
Outside the dewar, the signal passed through an isolator and an output voltage pulse as a function of the average number
a second broadband power amplifier 共9 GHz; 20 dB gain兲 关 J inS d /ប ␻ 兴 of 0.81 ␮m wavelength photons in a 100 fs
before going to a 6 GHz bandwidth single-shot oscilloscope pulse, incident on the device area, for two different values of
for display, or to a 200 MHz voltage-level threshold counter I bias . Since all photons arrive within the 100-fs-laser-pulse
for real-time event counting and statistical analysis. We window, only spatial correlations 共number of photons per
worked with 100 fs wide, ⬃50 ␮m diameter optical pulses device area兲 are important in the experiment. The left vertical
Appl. Phys. Lett., Vol. 79, No. 6, 6 August 2001 Gol’tsman et al. 707

age oscillations in the metastable region兲. Lowering I bias , on

the other hand, led to the two-photon detection. Solid squares
in Fig. 3 correspond to the same device, operated under the
same conditions as discussed, but with I bias⫽0.8 I c . We note
immediately that our experimental data points now follow a
quadratic dependence of detection probability showing the
two-photon operation. The two-photon absorption, appar-
ently, must form a larger hotspot size, needed to compensate
for the smaller I bias , but in this case, QE is significantly
lower and is ⬃1%. At the same time, we do not see the
laboratory photon background since the joint probability of
two stray photons hitting the device area within the required
space and time is negligibly small. Further reduction of I bias
共not shown in Fig. 3兲 resulted, unsurprisingly, in a cubic
FIG. 3. Number of counts per second recorded by the NbN SPD versus the 共three-photon detection兲 dependence of detection probability
average number of photons per pulse incident upon the device, for two to the number of photons per pulse.
different bias current levels is shown. The solid lines correspond to the Eq.
共4兲 theoretical predictions. The incident photon wavelength was 0.81 ␮m.
In conclusion, we have demonstrated that a supercurrent-
assisted, hot-spot-formation mechanism can be implemented
using an ultrathin NbN strip for ultrafast single-photon de-
axis in Fig. 3 shows the experimental data i.e., the number of tection and counting of visible and infrared photons with an
detector counts per second 共equivalently, per 1000 laser experimentally measured 20% QE for 0.81 ␮m photons and
pulses兲, based on the average number of counts detected by negligible dark counts. The bandwidth-limited measured re-
the SPD over a 10 s counting period for the highest photon sponse time was ⬃100 ps, corresponding to a 10 GHz pho-
doses, and up to 1000 s for the lowest. The counter threshold ton counting rate. Already identified applications for our su-
was adjusted to minimize spurious counts. The right vertical perconducting SPDs range from sensing ultraweak
axis corresponds to the probability P of detecting an optical electroluminescence from submicron complementary metal–

25 February 2025 [Link]

pulse. Open squares correspond to the SPD performance oxide–semiconductor very large scale integrated circuits,11
when it was biased at 0.92 I c . For high incident photon to quantum communication systems.
fluxes, the detector managed to count all 1000 laser pulses in
each second ( P⫽1), without actually resolving the number The authors thank Aleksandr Verevkin, Kenneth Wilsher,
of photons. For smaller fluxes, however, our experimental Steven Kasapi, and Gerald Gilbert for helpful discussions
data show that for over four orders of magnitude, the detec- and comments. This work was supported by Schlumberger
tion probability decreases linearly with the decrease of the SS, U.S. Office of Naval Research under Grant No. N00014-
average number of incident photons, unambiguously demon- 00-1-0237, and the NATO Linkage Grant No. CR-
strating the single-photon detection mechanism. Since our G.LG974662, and the Award No. RE-2227 of the U.S. Civil-
experiment was performed in an optically unshielded envi- ian Research and Development Foundation for the Indepen-
ronment 共the detector was inside the dewar, but not in a dark dent States of the Former Soviet Union.
box and only the main laboratory lights were off兲, at very
low photon doses our experimental data points leveled off at 1
0.4 s⫺1 counts ( P⫽4⫻10⫺4 兲, which must be regarded as the A. Peacock, P. Verhoeve, N. Rando, A. van Dordrecht, B. G. Taylor, C.
Erd, M. A. C. Perryman, R. Venn, J. Howlett, D. J. Goldie, J. Lumley, and
laboratory single-photon background noise, where the back- M. Wallis, Nature 共London兲 381, 135 共1996兲; R. J. Schoelkopf, S. H.
ground is essentially stray photons uncorrelated to the laser Moseley, C. M. Stahle, P. Wahlgren, and P. Delsing, IEEE Trans. Appl.
pulses. The intrinsic dark count rate of our SPD was below Supercond. 9, 2935 共1999兲.
2
0.001 s⫺1 ( P⬍10⫺6 兲, which corresponded to zero number of
K. S. Il’in, I. I. Milostnaya, A. A. Verevkin, G. N. Gol’tsman, E. M.
Gershenzon, and R. Sobolewski, Appl. Phys. Lett. 73, 3938 共1998兲.
detector responses over the time of 1000 s when its input was 3
K. S. Il’in, M. Lindgren, M. Currie, A. D. Semenov, G. N. Gol’tsman, R.
completely blocked. Sobolewski, S. I. Cherednichenko, and E. M. Gershenzon, Appl. Phys.
From Fig. 3, we can estimate that the QE of our NbN Lett. 76, 2752 共2000兲.
4
M. Lindgren, M. Currie, C. Williams, T. Y. Hsiang, P. M. Fauchet, R.
microbridge is 20%, as the value corresponding to the prob- Sobolewski, S. H. Moffat, R. A. Hughes, J. S. Preston, and F. A. Heg-
ability of detecting a pulse containing an average of one mann, Appl. Phys. Lett. 74, 853 共1999兲.
photon incident upon the device. The practical detection ef- 5
A. M. Kadin and M. W. Johnson, Appl. Phys. Lett. 69, 3938 共1996兲.
6
ficiency of our SPD is, of course, much lower because of the A. D. Semenov, G. N. Gol’tsman, and A. Korneev, Physica C 351, 349
共2001兲.
very low S d , as compared to the optical spot size. The maxi- 7
M. Stuivinga, C. L. G. Ham, T. M. Klapwijk, and J. E. Mooij, J. Low
mization of detection efficiency is, however, strictly an engi- Temp. Phys. 53, 633 共1983兲.
8
neering issue. Using proper coupling optics, the incident S. I. Cherednichenko, P. Yagoubov, K. S. Il’in, G. N. Gol’tsman, and E. M.
Gershenzon, in Proceedings of the Eighth International Symposium on
photon beam could be focused to a diffraction-limited spot
Space Terahertz Technology 共Harvard University, Cambridge, MA, 1997兲,
size. At the same time, various geometrical configurations of pp. 245–252.
the detector can be implemented, including a large-active- 9
W. J. Skocpol, M. R. Beasley, and M. Tinkham, J. Appl. Phys. 45, 4054
area meander-type design. 共1974兲.
10
M. Born and E. Wolf, Principles of Optics: Electromagnetic Theory of
The 0.92-I c detector bias was selected to achieve the Propagation, Interference, and Diffraction of Light 7th edition 共Cam-
single-photon detection with negligible dark counts 共larger bridge University Press, Cambridge, UK, 1999兲, pp. 752–758.
values of I bias , resulted in dark counts associated with volt- 11
J. C. Tsang and J. A. Kash, Appl. Phys. Lett. 70, 889 共1997兲.