multi-pixel detector, I investigated series and parallel structures, their potential and limitations, which are discussed in section 3.3. I then proposed a solution that mitigates issues found in parallel designs and enabled latch-free operation at high detection rates, which is presented in section 3.4. All measurements presented in this chapter are performed with continuous light at 1550 nm, at a temperature of 0.82 K, except for the photon-number resolving data presented in section 3.5 which was taken using a 33 ps pulsed laser at 1550 nm.
3.2 Multi-Pixel SNSPDs
3.2.1 Design
Multi-pixel SNSPDs consist in an array of nanowires, which together cover the same surface area as a single conventional SNSPD meander. As explained above, among other advantages such design allow for higher detection rates. We created 4-pixels designs and fabricated them following the same process described in appendix B for conventional SNSPDs (figure 3.1).
One disadvantage of this design is the need for multiple coaxial lines (one for each pixel) which are required to read the individual signals. As seen on fig. 3.2, each pixel is fully independent from its neighbours and a cryo amplifier is used for each readout. Hence four coaxial cables are used from the 0.8 Kelvin stage to the 40 Kelvin stage where the amplifiers are placed. Additional coaxial cables are used from this stage to the outside of the cryostat where the pixels signal are further amplified and discriminated, each with its specific electronics. This additional load on the cyrostat cooling power, as well as additional readout complexity, prevents easy scaling of the solution to high number of pixels. Nevertheless, impressive results have been obtained with this solution. Recent work on 16 pixels interleaved arrays demonstrated SDE of 50% at 500 MHz and reached rates of 1.2 GHz with a SDE of ∼12% [52]. Large arrays up to 64 pixels have even been demonstrated [57] but the practical cost of such system is much higher than what is usually considered standard for single-photon detector applications.
3.2.2 Results
We tested the multipixel devices with the same characterization setups used with conven-tional SNSPDs. Results presented here were obtained with a device similar to the one on fig. 3.1b. The detector has four pixels separated from each other by 633 nm. Each pixel is made of a 6 nm thick, 100 nm wide (90 nm on the device imaged on fig 3.1b) nanowire. The filling factor of each pixel is 60%. All pixels were illuminated with CW light at 1550 nm through a single SMF fiber. The four outputs were each discriminated
(a) Squared pixels design. The length of each nanowire is equal, so that the recovery time is consistent for each pixel. No additional spacing was added between the pixels.
(b) Round multipixels design. Here the two central pixels have a shorter length than the top and bottom ones, and are expected to have a slightly higher maximum detection rate. Ad-ditional spacing between the pixels was added to avoid thermal crosstalk (see section 3.2.3).
Figure 3.1: SEM images of two multipixels SNSPDs designs. Colors are added on the picture to highlight the position and contacts of the four different pixels. The expected optical mode of the fiber is represented by the white circle. In these two cases, the exposed area of each pixel has been calculated so that the average number of photon absorbed is equal. Hence the central pixels have a reduced exposed area. We will refer to the pixels with numbers 1 to 4 in the text, from bottom to top.
Figure 3.2: Schematics of a multipixel detector readout. Each nanowire is connected individually to its own amplification electronic via a coaxial line.
3.2. Multi-Pixel SNSPDs separately using the four inputs of our ID900 Time Controller. We obtained saturated efficiencies of 5.5%, 14.9%, 19.1%, 13.3% for each pixel respectively from top to bottom.
Figure 3.3a shows the SDEvs. bias current curve for each pixels, with colors matching the representation on fig 3.1b. It is interesting to note that the efficiencies are not equal on the two inner or outer pixels, as one could expect. This issue has been observed on several devices and might indicate a small offset created by the self-alignment method used when coupling the optical fiber on the detector. Hence a lower efficiency is observed on the top pixel (pixel 1). The dashed purple line gives the calculated sum of each measurement, which corresponds to the global efficiency of the detector. It is important to note that, in this case, the sum was performed after the independent measurement on each pixel. A final setup would require to sum the output of the four different discriminators in real time.
The normalized efficiencyvs. detection rate of each pixel is plotted on figure 3.3b with the same color code. As expected, each single pixel reaches a higher maximum detection rate than single-meander SNSPDs. Rates of 14.5 MHz, 20 MHz, 24.5 MHz and 20.5 MHz are observed for each pixel respectively. This is around 5 times higher than the rate observed on the single-meander device presented in chapter 2 (figure 2.2.2) and corresponds to the expected reduced kinetic inductance of each pixels. The measurements are taken independently on each pixel consecutively. The sum of the efficiencies gives an indication on what count should be obtained when using such a detector with the 4 simultaneous output summed. Detection rates around 78 MHz could be obtained with a 50% loss of efficiency compared to the value at low detection rate. Increasing the incoming photon flux leads to further drop of the efficiency, but detection rates of more than 200 MHz with an average SDE of∼3% could be obtained in this regime. We also measured the jitter of each pixel independently and obtained values of 35 ps to 38.3 ps FWHM which is consistent with typical values obtained with single meander SNSPDs.
These measurements demonstrate that increasing the detection rate of SNSPDs while minimizing the loss of efficiency is possible with multiple shorter nanowires. However the increased thermal load on the cryostat and the electronic complexity quickly become an issue when scaling up the system to more nanowires.
3.2.3 Thermal Crosstalk
When first testing multipixel detectors, we observed coincidences between adjacent pixels output signals. Output pulses of adjacent pixels had a non-zero probability to be triggered at the same time. This effect is most likely caused by thermal crosstalk between the pixels : the heat generated by the hotspot heats up the neighbouring pixel, the local temperature rises and lowers the critical current density Js(T) of the nanowire which switches to its resistive state. This effect has been described before and can even be used to improve certain characteristics of the detectors by taking advantage of the thermal
15 20 25 30 35 40 45
(a) SDE of each pixels. The total efficiency is lower than that of conventional single-meanders SNSPDs, because of the additional distance between the pixels.
(b) Normalized efficiency vs. detection rate for each pixel. The lower kinetic inductance improves their maximum detection rate of each pixel compared to single-meander devices.
Figure 3.3: Results obtained with a 4-pixels SNSPD. Each pixel was measured indepen-dently. The sum was obtained by adding the results of each pixels. A real-time use of the device requires using four different discriminators and summing theirs outputs in real time.
avalanche [58]. In our case, this effect is unwanted because the neighbouring nanowires must remain superconducting for subsequent photodetections. We investigated the issue using a start-stop timestamp setup which allowed us to display the histogram of coincident detections, as shown on fig. 3.4. This figure shows the histogram of the timing difference between two pixel clicks. We used a low average number of incoming photon per second µ(approximately105 photons per second), which ensures that the coincidence peaks seen on fig. 3.4 are due to interaction between the pixels and are not generated by different photo detection at short time intervals. The localised timing distribution obtained show that the output of these two pixels are indeed correlated. Such correlations occurs only in neighbouring pixels.
The amount of time needed for the hotspot to heat up the neighbour pixel enough so that its critical current decreases belowIb is given by the time difference between the blue pixel click (dashed blue line on the figure), and the measured point. As expected, lowering the bias current increases the coincidence delay, as a higher temperature is needed to lower the critical current density Jc(T) under the bias current density. Interestingly, there are other distributions of coincident detections at later timings. This suggest that the same phenomenon can also be triggered by a photon being absorbed in the second or the third line of the meander, as illustrated on figure 3.4. In such case, the time needed for the nanowire to switch to the resistive state is increased and the distribution becomes less sharp.
3.2. Multi-Pixel SNSPDs
Figure 3.4: Illustration of the thermal crosstalk between pixels. A hotspot in a nanowire close to the edge of a pixel will trigger a hotspot in the neighbour pixel. Different timing can be observed depending on which exact line of the meander absorbed the photon. The blue dashed line indicates the timing of the blue pixel click. The measured points are the green pixel timings for different bias currents.
This information is of great interest when designing multi-nanowire SNSPDs such as those presented in this chapter. The thermal crosstalk might introduce false detections, this can be corrected by introducing a deadtime in the readout electronics of all outputs after a single detection. However this defeats the purpose of the design, which is reaching higher detection rates. Moreover, as single photon can deactivate two pixels at the same time, the possibility for subsequent photon detection at high rates could also be hindered by the thermal crosstalk.
Based on this information, we created designs where pixels are separated by a distance above 600 nm, where no thermal crosstalk was observed (fig. 3.1b). This solution decreases the global fill factor of the detectors (which lowers the total SDE), a trade-off could be found using closer pixels while still limiting the thermal crosstalk to a fairly low probability. It could also be interesting to better thermally insulate the pixels from each other, and use substrates with higher thermal conductivity which would act as an heatsink to help dissipate the energy of the hotspot. The optimization of the mirror cavity could also help to recover high efficiencies with lower fill factors.