High-performance InGaAs/InP single-photon avalanche diodes (SPADs) in near-infrared range between 1.0–1.7 μm are required in many fields, such as quantum key distribution (QKD), ^{[ 1 ]} time-of-flight (TOF) measurements, fluorescence lifetime imaging (FLIM), and fluorescence correlation spectroscopy (FCS). ^{[ 2 ]} For single photon detection, an SPAD is usually dc biased a few volts below its breakdown voltage, and is periodically pulse biased above breakdown voltage for a very short time, which is called gated operation. ^{[ 3 ]} This requires the synchronization between the gates and the coming photons, thus limits its application. ^{[ 4 ]} For a variety of situations, photons do not come in a known pattern, where the SPADs are required to operate in the free-running regime. ^{[ 5 ]}

Due to its relatively narrow bandgap ( E _g = 0.75 eV for In _0.53 Ga _0.47 As), it usually exhibits more dark counts. ^{[ 2 ]} Moreover, carriers produced by previous avalanches, can be trapped by the deep levels in the multiplication layer. They can release at a later time and lead to another avalanche. Therefore, in practice, we must apply a very long deadtime, during which the SPAD is biased blow its breakdown voltage, and wait for these carriers to release. ^{[ 6 ]} This severely limits detection rate. On the other hand, the more carriers an avalanche produces, the more likely an afterpulse happens; and the number of carriers is proportional to avalanche duration. Consequently, to reduce the afterpulsing phenomena, one needs to quench the avalanche rapidly, as soon as it is sensed. ^{[ 7 ]}

Many techniques have been developed to achieve this goal. With the traditional passive quenching method, the passive reset stage takes a rather long time. Active quenching method actively quenches and resets SPAD, so the detection rate can be increased, and the deadtime is well controlled. ^{[ 8 ]} Integrated active quenching circuits ^{[ 9 , 10 ]} have been developed and also applied to the InGaAs/InP avalanche photodiodes (APD). ^{[ 11 , 12 ]} Discrete passive quenching active reset ^{[ 13 ]} is another method to reduce the afterpulsing effects, however it cannot be integrated with the APD. Negative feedback avalanche diodes (NFADs) ^{[ 5 , 14 ]} utilizing integrated resistors achieved good performance and are promising for array integration, but they only achieved passive quenching and reset.

In this paper, we demonstrate a passive quenching and active reset integrated circuit, and its operation principle. Then, we characterized the free-running InGaAs/InP single photon detector, using this integrated circuit, and extracted its afterpulsing probability, de-trapping time, dark count rate and photon detection rate through deadtime and afterpulse correction. At the same time, we compare the DCR and PDE between free-running regime and gated regime.

It is believed that, the intrinsic passive quenching is swift, while the passive reset takes a rather long time. ^{[ 15 ]} Here, we developed a passive quenching active reset integrated circuit with SMIC 0.18 μm CMOS process. The schematic is presented in Fig. 1(a) , and the photomicrograph of the single cell of the circuit in Fig. 1(b) . The pixel layout area is within 150 μm × 150 μm, easy for future array integration. As can be seen, the APD is connected with the IC at its anode, and it is biased with VBias at the cathode. The electronic model of the APD is based on the model in the article. ^{[ 16 ]} The parasitic resistor and capacitance is measured as described following the procedure in it.

Fig. 1.

	Figure Option View Download New Window
	Fig. 1. (a) Schematic of the passive quenching active reset IC. (b) Photomicrograph of the single cell of the circuit.

The operation principle is illustrated in Fig. 2 . At quiescent state, the bias voltage is higher than its breakdown voltage; and when triggered the avalanche current flows into R _A and R _B , and raises the voltage A. As soon as the voltage A bypasses the REF voltage, which is provided outside the pixel, the fast comparator turns low at node P. It triggers a monostable to provide a fixed deadtime, which is controlled outside the chip. The deadtime is adjustable from 10 ns to 500 μs allowing for desired optimization of the trade-off between peak count rates and low afterpulsing. The falling edge passes through an AND gate and a small delay to node Q, switching the quench transistor M _Q off swiftly. As a result, the current path is cut off. Then it charges the parasitic capacitor and the voltage continues to grow until the voltage across the APD drops near the breakdown voltage. This achieves passive quenching. The monostable output also transmits to node S through a relatively long delay, thus switching the transistor M _S off. The comparator input voltage B drops quickly, and the comparator turns high again. After the deadtime ends, the monostable output turns high, and transmits to node Q and S to close the switches M _Q and M _S successively. The time difference of signal Q and S, caused by different path delays, produces a reset pulse. It closes M _R , which rapidly decreases the voltage A to ground. This achieves active reset. At this moment, the single photon detector is back to the quiescent state and ready for another photon. The output pulse can either be sent outside the chip or be sent to an 8-bit linear feedback shift register (LFSR) to be counted. Gate signals can be applied to the chip through the GATE pin, which we called the gated mode. Although it differs from the normal gated operation slightly, where at the gated-off period the APD is biased below breakdown voltage, in this scenario the APD is cut off from the circuit, where it still cannot trigger avalanches. Consequently, we still consider this operation as the gated operation and further compared the performance in detail below. The measured anode signal waveform is shown at overbias of 0.9 V and deadtime of 1 μs in Fig. 3 , which is consistent with the analysis. It can been seen in Fig. 3 , the resistive passive quenching time is about 150 ns; the passive charging time is 350 ns. As a result, the total quenching time is about 500 ns. While the fast reset stage takes about 30 ns.

Fig. 2.

	Figure Option View Download New Window
	Fig. 2. Waveform illustration of the passive quenching active reset IC.

Fig. 3.

	Figure Option View Download New Window
	Fig. 3. Measured waveform at the anode of APD at overbias of 0.9 V and deadtime of 1 μs.

With this passive quenching active reset integrated circuit, we can characterize the free-running single photon detector performance, such as dark count rate (DCR), photon detection efficiency (PDE), and afterpulsing probability (AP). We utilized the APD, fabricated in separate absorption graded charge multiplication (SAGCM) structure. ^{[ 17 ]} The 1550 nm pulsed laser (id300, id Quantique) is fed to the APD through a single mode fibre, and attenuated to 0.1 photon/pulse. The APD is cooled to 219 K by a Peltier cooler (RMT Ltd, 2MDX04-138-0816).

When the laser is switched off, the APD’s internal thermally generated carriers, direct band-to-band tunneling (BBT) and trap-assisted tunneling (TAT) can also contribute to count rates, which is dark count rate. The photon detection efficiency is the product of external optical coupling efficiency on the photodiode, the electric carriers generation probability by absorbed photons and the avalanche triggering probability by the photo-generated carriers. ^{[ 18 ]} Afterpulsing probability is the probability that trapped charge carriers created during a previous avalanche release and induce spurious avalanches.

We applied a deadtime to relieve the afterpulsing effect. The double-pulse method ^{[ 19 ]} is adopted to extract the afterpulsing probability by applying a train of short pulse of 100 ns to the GATE pad. By tuning the time difference between light gates and subsequent dark counts, we can obtain the afterpulsing probability P _ap with

We plot afterpulsing probability dependence on deadtime at different bias voltages in Fig. 4 . As can be seen, the afterpulsing probability decreases with deadtime following exponential pattern. With the bias voltage increasing, the avalanche current becomes bigger. As a result, the afterpulsing probability grows with the bias voltage.

Fig. 4.

	Figure Option View Download New Window
	Fig. 4. Afterpulsing probability dependence on deadtime at different bias voltages and their exponential fit.

It indicates that the deep levels in the APD can be modeled as a single type of trap. ^{[ 19 ]} Therefore, we can extract its de-trapping time parameter through the procedure described in Ref. [ 20 ].

First of all, we can express the total measured count rate R _tm as

Fig. 5.

	Figure Option View Download New Window
	Fig. 5. Raw count rate versus photon flux at 62.4 V bias voltage at 10 μs, 20 μs, 40 μs deadtime. The cps is for counts per second.

Fig. 6.

	Figure Option View Download New Window
	Fig. 6. Extracted pre-factor C (square, right axis) and de-trapping time τ tr (circle, left axis) versus photon flux.

Here, we can see that, the pre-factor C is almost linear with the photon flux in this log–log plot. It confirms that C ₀ is proportional to avalanche current. The de-trapping time does not vary much at different photon flux, indicating that the de-trapping time constant is about 16.8 μs.

In order to prevent afterpulsing, the free-running regime applies long deadtime while photons come randomly. Thus, the device is nonlinear and its performance depends on deadtime and afterpulses. We can perform deadtime correction ^{[ 21 ]} and afterpulse correction ^{[ 14 ]} to the count rates and obtain intrinsic DCR and PDE. Also, we performed the gated operation by applying 100-ns-width pulse to the GATE pad and compared the performance between free-running regime and gated regime. By scanning the laser-to-gate delay with a step of 5 ns, we obtain the photon counts distribution. We normalize the counts and obtain the effective gate width of 95 ns as shown in Fig. 7 .

Fig. 7.

	Figure Option View Download New Window
	Fig. 7. Normalized photon counts versus laser-to-gate delay.

We measured the dark count rate with laser off as a function of bias voltage for different deadtimes. Depending on deadtime, the laser repetition frequency is f _laser = 1/ T _d . The intrinsic dark count rate R _di for deadtime and afterpulse correction is expressed as

With the free-running DCR and gated DCR at different deadtime in Fig. 8 , we can see that the DCR increases exponentially with bias voltage in both regimes. The free-running DCR is almost one order higher than the gated DCR. At different deadtimes, the gated DCR is almost the same; while the free-running DCR increases as the deadtime decreases, whereas they are still on the same order.

Fig. 8.

	Figure Option View Download New Window
	Fig. 8. Dark count rate versus bias voltage at different deadtimes with respect to gated regime and free-running regime. For example 10 μs_ f and 10 μs_ g denotes 10 μs deadtime free-running regime and gated regime, respectively.

By taking into account the deadtime and afterpulsing, the total measured counts R _tm is modeled as

In Fig. 9 , the free-running PDE is consistent with the gated PDE at different deadtime, and both increase linearly with the bias voltage. However, at higher voltage, the free-running PDE begins to saturate, which is caused by the high DCR. Despite the little divergence, it is also concluded that PDE does not depend on the deadtime. The maximum free-running PDE is about 23%, where the DCR is 10 ⁻⁵ ns ⁻¹ .

Fig. 9.

	Figure Option View Download New Window
	Fig. 9. Photon detection efficiency versus bias voltage at different deadtime with respect to gated regime and free-running regime. For example, 10 μs – f and 10 μs – g denotes 10 μs deadtime free-running regime and gated regime, respectively.

Finally, the DCR versus PDE relationship at 20 μs deadtime is plotted in Fig. 10 , where it follows almost the same exponential principle as the DCR versus bias voltage.

Fig. 10.

	Figure Option View Download New Window
	Fig. 10. Dark count rate versus photon detection efficiency at 20 μs deadtime.

In conclusion, we present a passive quenching active reset integrated circuit by taking advantage of the inherent fast passive quenching and active feedback reset. Combining this IC with an InGaAs/InP APD, we develop a free-running near-infrared single photon detector. Then, we comprehensively characterize the performance of this detector and compare with gated operation. We find that, the afterpulsing probability decreases with deadtime exponentially and the de-trapping time constant is 16.8 μs. The free-running DCR is higher than the gated DCR because of narrow band gap, and also follows the same principle. The free-running PDE is consistent with the gated PDE and does not depend on the deadtime. The maximum free-running PDE is about 23%, where the DCR is 10 ⁻⁵ ns ⁻¹ .