The InGaAs/InP single-photon detector (SPD) stands out for its simplicity and low power consumption, offering a reliable solution for near-infrared single-photon detection. Its practicality extends across various applications, including quantum key distribution (QKD) [1–3], laser ranging [4,5], optical time-domain reflectometry [6,7], and biomedical imaging [8]. However, the presence of dark counts and afterpulse effects presents challenges for the InGaAs/InP SPD, leading to erroneous counts.

To mitigate these errors, InGaAs/InP SPDs typically operate in gated mode, where avalanche occurs only during the gate opening time, significantly reducing dark count rate and afterpulse probability. Nevertheless, due to the capacitive response of the single-photon avalanche diode (SPAD), the gate applied to the SPAD introduces spike noise, which will obscure avalanche signals and make them difficult to distinguish [9–12]. Another method to reduce dark count rate and afterpulse probability is by lowering the avalanche gain [13,14]. For dark counts, at low avalanche gain, the number of dark carriers excited by the tunneling effect will decrease, and the probability of dark carriers initiating an avalanche will also be reduced. Regarding afterpulse effects, as the average avalanche charge decreases with lower avalanche gain, the number of carriers captured by defects in the SPAD multiplication layer will be reduced, thereby reducing the probability of afterpulse. However, lower avalanche gain results in weaker avalanche signals, which will further increase the difficulty of discriminating the avalanche signal from the spike noise.

In quantum optics experiments, as experimental conditions and application demands diversify, system frequency typically varies [15,16]. A low noise, wide gate frequency tunable InGaAs/InP SPD can significantly enhance system performance and experimental flexibility, reducing equipment dependence and costs. For instance, in large-scale deployments of QKD systems, users select different devices based on system performance and cost [17–19]. The frequency differences among devices from various manufacturers range from megahertz to gigahertz, posing significant interoperability challenges [20]. A low noise, wide gate frequency tunable SPD can adapt to these frequency variations, greatly improving interoperability among different systems and providing a better balance between performance and cost, making QKD systems more cost-effective in large-scale deployments. In laser ranging systems, multi-ranging techniques using multiple measurement standards can significantly increase measurement distance while maintaining accuracy [5,21,22]. However, the maximum measurement distance is determined by the minimum gating frequency of the SPD, and measurement accuracy increases with higher SPD frequencies [21]. A low noise, wide gate frequency tunable SPD can provide optimal performance under various ranging conditions, meeting the needs of diverse applications. Therefore, a low noise, wide gate frequency tunable SPD has significant advantages in quantum optics experiments and is crucial for advancing the progress and application of quantum optics experiments.

To discriminate weak avalanche signals from spike noise, various schemes have been proposed, including the sine-wave gating scheme [10,23–25], self-differencing scheme [11,26], dual-SPAD balancing scheme [27,28], and so on [14,29–31]. In the sine-wave gating scheme, a band-pass filter is adopted to ensure the gating signal has a single frequency, resulting in spike noise with simple frequency components. This allows for effective filtering of the spike noise and the discrimination of weak avalanche signals. The self-differencing scheme splits the SPAD output into two equal signals: one is undelayed, and the other is delayed by an integer multiple of the gating signal period. Since the spike noise is identical in both signals, subtracting them filters out the spike noise while retaining the avalanche signal. While both the sine-wave gating scheme and the self-differencing scheme can effectively eliminate spike noise, the band-pass filter in the sine-wave gating scheme and the delay cable in the self-differencing scheme limit their gating frequency tunability [11,25].

The dual-SPAD balancing scheme eliminates spike noise by subtracting the outputs of two SPADs. This scheme does not require a specific repetition frequency for gating signals, allowing for tunable gating frequencies. However, due to the differences between the two SPADs and the high demand for impedance matching, it is difficult to completely filter out the spike noise [28]. Even with two closely integrated SPADs, significant residual noises can remain at higher gating frequencies, limiting the discrimination of weak avalanche signals. Furthermore, the capacitive cancellation scheme can also discriminate avalanche signals [14,30,31]. This scheme is similar to the dual-SPAD scheme, where the spike noise is eliminated by generating an auxiliary signal through a variable capacitor. Compared to the dual-SPAD scheme, the capacitive cancellation scheme has a larger residual noise owing to the greater differentiation between the SPAD and the capacitor. As the gating frequency increases, the difference between the SPAD and the variable capacitor becomes more pronounced, diminishing the technique’s effectiveness in eliminating spike noise. Consequently, at high gating frequencies, it cannot effectively detect avalanche signals [12,14].

At present, there is no effective scheme that simultaneously offers low noise and wide tunability in gating frequency, which is necessary for QKD systems, laser ranging, and other quantum optics experiments.

In this paper, we propose a discrimination method and experimentally demonstrate a low noise InGaAs/InP SPD with wide gating frequency tunability. To enhance the sensitivity of avalanche detection, we eliminate the spike noise using an auxiliary signal generated by a variable capacitor. Subsequently, by adjusting the delay cable, we overlay the avalanche signal onto the peak of the residual noise, overcoming the limitation of residual noise. This method enables single-photon detection at lower avalanche gains, significantly reducing dark count rate and afterpulse probability. Since the spike noise and the auxiliary signal are derived by differentiating the gating signal, they are primarily related to the shape of the gating signal. Consequently, the residual noise obtained after their subtraction will also depend on the shape of the gating signal. We can effectively discriminate avalanche signals for gate signals of different frequencies while maintaining a consistent gate width. Finally, we achieve a DC to 1 GHz continuously tunable gate frequency InGaAs/InP SPD. Across the entire tuning range, this InGaAs/InP SPD offers consistently low dark count rate and afterpulse probability, providing an ideal solution for quantum optics experiments.

Figure 1 illustrates the experimental setup of the InGaAs/InP SPD. The pulse generator, triggered by the clock source, outputs a fixed width gate signal (amplitude 8.5 V peak-to-peak, width 0.5 ns). As shown in Fig. 1(a), when the gate signal is applied to the SPAD (Mini-Flat Type, Wooriro), the capacitive response of the SPAD causes spike noise (blue line) to occur at the rising and falling edges of the gate signal, which can obscure weak avalanche signals (red line), making them difficult to discriminate. To reduce the impact of spike noise and improve the sensitivity of avalanche signal detection, we generate an auxiliary signal with a shape similar to the spike noise using a variable capacitor (adjustment range: 0.14–0.43 pF), as depicted in Fig. 1(b). The output signal of the SPAD and the auxiliary signal are respectively inputted to the 0 and π inputs of balun, where the balun functions as a unity-gain differential amplifier. However, due to the differences in the properties of the SPAD and the variable capacitor, the spike noise cannot be completely eliminated, resulting in residual noise. The residual noise limits the ability to discriminate weak avalanche signals. To further improve detection sensitivity and overcome the limitation of residual noise, we overlay the maximum amplitude position of the avalanche signal on the peak of the residual noise by adjusting the delay cable (length: 6 cm), as shown in Fig. 1(c). The residual noise maintains a constant amplitude, whereas the avalanche signal, when superimposed on the residual noise, exceeds the peak amplitude of the residual noise. This enables effective discrimination of weak avalanche signals by setting a threshold in the comparator slightly above the peak voltage of the residual noise. The avalanche signal outputs a pulse with a width of 10 ns after shaping, which introduces a dead time of 20 ns.

The spike noise generated by the SPAD and the auxiliary signals generated by the variable capacitor are caused by the charging and discharging effects induced when the gate signal is applied. This is described by dQdtR=dUdtRC, where Q, U, and C represent the charge, voltage, and capacitance of the SPAD (or the variable capacitor), and R is the terminal resistance, typically 50 Ω. According to this equation, spike noise and auxiliary signals are primarily influenced by the rate of gate signal voltage, i.e., by the shape of the gate signal. They are independent of the repetition frequency of the gate signal. The pulse generator can generate gate signals with essential unchanged shapes for different gate frequencies. Thus, weak avalanche signals can be effectively discriminated without additional hardware adjustments for gate signals of different frequencies. This residual noise assisted discrimination method offers a simple and easily implementable solution for an InGaAs/InP SPD with tunable gate frequencies.

Before being identified by the comparator, we observe the avalanche signals using an oscilloscope (OSC, WaveRunner640zi, LeCroy, bandwidth 4 GHz, 20 GSample/s). Each time, the oscilloscope can acquire and save 1000 waveforms, including about 20 avalanche signals, while the rest are residual noises composed of spike noise and auxiliary signals. Figure 2 illustrates these waveforms in a color temperature mode. Before adjusting the delay, the presence of residual noises can obscure some weak avalanche signals, making them undetectable, as shown in Fig. 2(a). To enhance the ability to discriminate weak avalanche signals, we adjust the delay cable so that the avalanche signal is overlaid on the peak of residual noises. From Figs. 2(b)–2(e), it is evident that the avalanche signal is overlaid onto the residual noise. The weak avalanche signal can be easily identified with the comparator set to an appropriate threshold. Thanks to this avalanche signal discrimination method, weak avalanche signals can be effectively discriminated for different gate frequencies.

To verify the photon counting performance, the InGaAs/InP SPD is illuminated with a 1550 nm distributed feedback laser with a pulse width of 26 ps. The optical pulses are attenuated to an average of 0.1 photons per pulse using a variable optical attenuator before incident on the InGaAs/InP SPD. The SPAD is cooled to −35°C by a built-in thermo-electric cooler during testing. We investigate the variation of dark count rate and afterpulse probability at different detection efficiencies, where the detection efficiency is varied by adjusting the bias voltage. The dark count rate is determined by measuring the probability of dark counts for each gate period, while the afterpulse probability is measured for each photon count. The afterpulse measurement method follows the standard procedure proposed by Yuan et al. [11]. Afterpulse probability PA can be expressed as PA=(INI−ID)RIPh−INI,(1)where IPh represents the total count of gates with light input, INI denotes the total count of another gate, and ID indicates the total count when the laser is off, evenly distributed across each gate. These values IPh, INI, and ID can be obtained from the raw data recorded by the time-to-digital converter (TDC, HydraHarp 400, PicoQuant). R represents the ratio of gate frequency to laser repetition frequency.

Figures 3(a) and 3(b) respectively depict the relationship between dark count rate, afterpulse probability, and detection efficiency under different gate frequencies. Both dark count rate and afterpulse probability increase with detection efficiency for all gate frequencies. The increase in dark count rate can be attributed to two main factors. (1) As the reverse voltage of SPAD rises, the probability of avalanche increases, leading to a higher likelihood of free carriers generated by thermal excitation or tunneling excitation triggering avalanche counts [32]. (2) The probability of tunneling excitation increases with the reverse voltage, further contributing to the heightened dark count rate [33]. As for the afterpulse phenomenon, the number of avalanche carriers increases with the reverse voltage. This leads to a higher probability of carriers being trapped by defects in the SPAD multiplication layer, thus increasing the afterpulse probability. However, at the same detection efficiency, the probability of dark count remains relatively constant, indicating the effectiveness of this avalanche signal discrimination method across different gate frequencies. Regarding the afterpulse phenomenon, the afterpulse probability increases with the gate frequency, as expected. With an increase in gate frequency, the number of gate openings per second increases, providing more opportunities for carriers captured by defects to be released and generate afterpulse. As shown in Fig. 3, at a 20% detection efficiency, the dark count rate for gate frequencies spanning from DC to 1 GHz is approximately 5.0×10−7 per gate, and the afterpulse probability is less than 1.0%.

Table 1 compares various tunable gate frequency InGaAs/InP SPD schemes. We list the main performances, including detection efficiency ηdet, dark count rate PD, and afterpulse probability PA. In Ref. [14], the authors combined the advantages of the capacitive cancellation scheme and the low-pass filtering scheme to mitigate spike noise. Specifically, at low gating frequencies, the spike noise is primarily eliminated using the capacitive cancellation scheme, while at high gating frequencies, it relies on the low-pass filtering scheme. While this scheme can discriminate avalanche signals, it has relatively high residual noise at low gating frequencies due to the difference between the SPAD and the variable capacitor, leading to significant afterpulse probability. Additionally, at the gate frequency of 700 MHz, where the eliminating effect of both schemes is poor, substantial afterpulse probability is observed. In Ref. [29], they utilized an analog-to-digital converter (ADC) to differentiate between avalanche signals and spike noise for avalanche signal discrimination. However, due to the small voltage difference between weak avalanche signals and spike noise, as well as inherent noise in the ADC itself, this scheme yields suboptimal results in avalanche signal discrimination. Moreover, since the SPAD operates at room temperature, while this aids in reducing the afterpulse probability, it also increases the probability of dark count due to thermal excitation.Table 1.

Comparison of the Present Work with Previously Reported Performance for Tunable Gate Frequency InGaAs/InP SPD Schemes

Our residual noise assisted discrimination method uses only a variable capacitor and a delay cable to effectively discriminate weak avalanche signals over a wide range of gating frequencies. First, our SPD employs a variable capacitor to eliminate the spike noise, enhancing the sensitivity of avalanche signal detection. To further mitigate the impact of residual noises, we adjust the delay cable so that the maximum amplitude of the avalanche signal aligns with the peak of the residual noise, enabling the discrimination of weak avalanche signals. Moreover, this alignment allows our SPD to tolerate differences between the SPAD and the variable capacitor, ensuring effective discrimination of weak avalanche signals across a wide range of gating frequencies, as demonstrated in Table 1.

The exceptional performance of our InGaAs/InP SPD can be attributed to our avalanche signal discrimination method, which accurately identifies weak avalanche signals, enabling us to achieve effective single-photon detection at a low avalanche gain (i.e., a low reverse voltage). A low avalanche gain brings several benefits. First, it reduces the probability of dark count rates [32,33]. Second, lower avalanche gain results in fewer avalanche charges, reducing the number of carriers captured by defects and consequently lowering the afterpulse probability. We monitor the avalanche currents and calculate the average avalanche charge by dividing the avalanche current by the counts. At a detection efficiency of 20%, we measure avalanche charges at different frequencies, with the minimum recorded avalanche charge being 29 fC. The avalanche charge measured in our InGaAs/InP SPD is lower than the previously reported minimum of 35 fC [26]. This explains why our InGaAs/InP SPD achieves such a low dark count rate and afterpulse probability.

For single-photon detectors, both effective gate width and time jitter are crucial parameters. The effective gate width of the SPD is determined to be 165 ps by scanning the relative delay between optical pulses and gates, as depicted in Fig. 4(a). Figure 4(b) shows the time histogram of detection recorded with a 4 ps resolution time-to-digital converter, revealing a time jitter of 140 ps.

We have demonstrated a low noise InGaAs/InP SPD operating at gate frequencies from DC to 1 GHz. By using the residual noise assisted discrimination method, which aligns the maximum amplitude of the avalanche signal with the peak of the residual noise, we can effectively detect single photons even at lower avalanche gain. A lower avalanche gain brings many benefits. First, it reduces the probability of dark count occurrences. For instance, at lower avalanche gain, dark counts caused by tunneling effects are significantly reduced. Second, the avalanche charge decreases with the avalanche gain, leading to a diminished capture of charge by SPD defects and consequently lowering the afterpulse probability. The experimental results demonstrate that our avalanche charge is lower than the minimum charge observed in prior experiments. Moreover, this avalanche signal discrimination method is not limited by the gate frequency. Within the DC to 1 GHz gate frequency tuning range, we can discriminate weak avalanche signals effectively. Across the entire tuning range, at a 20% detection efficiency, the dark count rate is approximately 5.0×10−7 per gate, and the afterpulse probability is less than 1.0%. The InGaAs/InP SPD has the advantages of low noise, continuously tunable gate frequency, and simple structure, making it suitable for quantum optics experiments.

It is worth mentioning that our residual noise assisted discrimination method is universal and can be applied to all discrimination schemes. In this paper, we take the capacitive cancellation scheme as an example to demonstrate this method.