1. Introduction

InGaAs single photon avalanche detectors (SPAD) find important applications in optical communications, three dimensional imaging, and quantum cryptography [1,2]. InGaAs SPADs are required for the high gain demanded from these applications, as other single photon detectors such as photomultiplier tubes are not efficient at 1550nm wavelengths. Traditionally, Geiger-mode SPADs are operated in gated mode or with active quenching to reduce the thermally generated dark counts, afterpulsing effects, and excess noise. Alternatively, a self-quenching and self-recovering epitaxial design based on heterojunction barriers can be used to simplify device usage, increase device performance, and improve device fabrication [3–5].

Using bandgap engineering, a conduction band offset is created next to the multiplication region to momentarily stop any avalanche generated electrons. The collected charge at this barrier reduces the electric field in the multiplication region, stopping the avalanche process and thus the device is self-quenched. This process is fast (sub-nanoseconds) and limits the total avalanche gain of the device. After quenching, the electrons escape from the barrier via tunneling or thermionic emission, returning the multiplication field to the original field the device experienced before the avalanche occurred. The device is then self-recovered. A detailed explanation of the self-quenching mechanism can be found in [5]. This design allows for free-running Geiger-mode operation of the SPAD, while removing the need of fabricating additional quenching components such as resistive or active sensing circuits. Another benefit is the simplification of creating large SPAD arrays, due to the fact that each device is self-quenching.

2. Design of zinc diffused SPAD

To reduce the dark counts and afterpulsing arising from the surface states of a mesa-etched device, a Zn-diffused p-well needs to be formed to isolate the multiplication p-i-n junction from the mesa edge [6]. For most Zn-diffused SPAD structures, ensuring a uniform electric field throughout the multiplication region is necessary. Due to the high bias above the breakdown level (overbias) required for Geiger-mode detection of single photons, any slight variations in the electric field would create premature breakdown in that local area and greatly reduce the device's overall detection efficiency. To counteract the field non-uniformity of a typical Zn-diffusion profile, which experiences strong field crowding effects near the diffusion edge, techniques such as double diffusion profiles, guard rings, and trench isolations [7–10] have been used. Although these techniques have been widely used in conventional APD photoreceivers (operating at a gain between 10 and 30), they show only varying success in Geiger-mode SPADs due to the much higher gain (typically 10^5-6).

Opposite to the conventional design concept, which aims at achieving an uniform E-field and minimizing the field crowding effect, in this paper we exploit a new design to purposely populate the field-crowded regimes over the entire device active area and maximize the fill factor of such high-field regions. The intent is to fundamentally eliminate the complicated process to control the Zn-diffusion profile that becomes super critical during Geiger-mode single photon detection. In our design, many thin diffusion wells are placed in proximity to each other, and are electrically connected to collectively form a single electrode. The simulated electric field profile of a reverse biased p-i-n diode with this novel diffusion pattern is shown in Fig. 1(a) . Compared to the conventional single diffusion well structure in Fig. 1(b), which has the field crowded regions only at the edges, our design has a large proportion of high-field regions over a large part of the active area.

 

Fig. 1 Electric field contour plots of a simulated reversed bias p-i-n diode, at room temperature, with various diffusion patterns to form the p-region. The simulation is cylindrically symmetric about x = 0. (a) Multiple diffusion ring structure, demonstrating a larger areal fill factor of high field regions. All p-regions are electrically connected to form the anode. (b) Single p-well structure.

Download Full Size | PDF

Concentric circular Zn rings, as seen in Fig. 2 , are used to realize this design. By using rings instead of straight lines, the problem of high-field effects at the termination of the straight lines is removed. Independent of the diffusion process, each ring edge would experience the same field-crowding effects. The major advantage is that we no longer have to form the exact Zn-diffusion profile that can eliminate the field-crowded areas, thus making the fabrication process less critical, much easier to implement, and more reproducible.

 

Fig. 2 Micrograph of silicon dioxide mask used for Zn diffusion.

Download Full Size | PDF

3. Device fabrication and experimental setup

A self-quenching epitaxial design was used to test the performance of the zinc diffusion rings. Figure 3 shows the corresponding band diagram of the device at equilibrium. An electron barrier was created next to the n-region of the multiplication layer, which quenches the device after an avalanche. Zn-diffusion into the undoped InP top layer defined the p-i-n structure as the multiplication region. To form the Zn diffusion mask, silicon dioxide was deposited by PECVD onto the epitaxial wafer and patterned with buffered HF etching solution. After Zn diffusion in an OMCVD reactor using dimethyl zinc as the source, the oxide mask was stripped and fresh oxide was deposited and patterned to define the contact pad areas. Mesas for device isolation were etched with a HBr solution, and polyimide was deposited and patterned for sidewall passivation. Final device contact was made with Ti/Au metallization using lift-off process.

 

Fig. 3 Band diagram of the self-quenching SPAD with an electron barrier, at equilibrium. In Geiger-mode, electrons generated by impact ionization in the p-i-n drift towards the cathode (InP substrate) and are temporarily stopped at the InAlAs/InP conduction band barrier, reducing the voltage across the multiplication region and quenching the avalanche pulse.

Download Full Size | PDF

Devices with varying Zn-diffused ring widths and spacings were fabricated. The widths and spacings were varied from 1.0 to 2.0 μm, with 3-4 periods to achieve device active areas of 15-30 μm in diameter. The device mesa edge is typically 3-5 μm away from the active area. The current-voltage plot of a ring structure at 160K is shown in Fig. 4 .

 

Fig. 4 Current-voltage characteristics of a 1.0μm/1.5μm rings device at 160K in dark (solid lines), and 1550nm illumination (dotted lines) conditions.

Download Full Size | PDF

The single photon detection efficiency (SPDE) and dark count rate (DCR) of the device is measured using time correlated single photon counting. An external pulse generator triggers a high speed counter and a pulsed picosecond 1550nm laser simultaneously, while the counter records the delay between the laser trigger and the SPAD's avalanche pulse signal. Any counts that are not correlated with the laser pulses are considered as dark counts or afterpulses. The SPDE is obtained from the ratio of the correlated pulse counts and the number of laser pulses, after taking into account the Poisson characteristics of the laser pulses. During the SPDE measurement, the laser is focused to a 20μm diameter spot size on top of the device and attenuated to ~0.3 photons/pulse.

4. Results and discussion

The results of the SPDE measurements are shown in Fig. 5 for the various device geometries, with the first and second numbers referring to the widths of and the spacings between the Zn-diffused rings. For comparison, we also include in Fig. 5 the characteristics of a conventional Zn-diffused device, which has the simulated E-field profile in Fig. 1(b). Because the highest avalanche probability occurs only at the diffusion edge in the conventional Zn-diffused structure, it is not surprising that the device shows very low detection efficiency and large dark count rate. Among the devices in our design, the devices with 1.5μm wide diffused rings have too wide rings to produce dense enough high field regions within the photosensitive area (i.e. relatively low fill factor of the high-field regions). Thus such devices show only modest improvements over the conventional structure. On the other hand, the devices with 1.0μm spacing between Zn-diffused rings have their rings too close to each other to produce large enough field crowding effects. Nonetheless, very significant performance improvements over the conventional structure have been achieved with the above devices. Within the limitations of our current process, the 1.0/1.5μm (ring width/spacing) devices give rise to the highest performance, attributed to their high fill factor of high-field regions and the pronounced field crowding effect utilized in our design to enhance single photon detection. In such a design, we have achieved both high single-photon detection efficiency and low dark count rate. Figure 5 also shows the characteristics of such devices measured at 140K, 160K, and 180K, showing the trade-off between SPDE and DCR under different bias voltage.

 

Fig. 5 Effects of diffusion geometry and temperature on SPDE: single 10μm diffusion; 1.0μm/1.5μm (width/spacing) rings; 1.0μm/1.0μm rings; 1.5μm/1.5μm rings. The best performance was obtained with the 1.0μm/1.5μm rings structure. All devices with Zn-diffused rings show improved performance compared to the conventional single diffusion device. The frequency refers to the laser pulse rate.

Download Full Size | PDF

From the DCR vs SPDE plot in Fig. 5, there exists a peak SPDE after which the SPDE starts decreasing as the DCR increases. This saturation effect in SPDE results from the slow self-recovery time of our present devices due to the large conduction band offset between InP and InAlAs (see Fig. 3). If a photo- or thermally-generated carrier arrives in the multiplication region before the device has recovered from the previous avalanche response, this carrier is not able to trigger the single-carrier response as its predecessor does. The high DCR reduces the chance of the device being fully recovered before the next photon arrives, thus limiting the SPDE. By the same token, slow recovery time also limits the SPDE at high optical pulse rates.

The self-recovery time of the device is dependent on the barrier height (conduction band offset) and, to a lesser degree, the width of the barrier (InAlAs layer). Both factors affect the escape time of the electrons from the barrier by thermionic emission or tunneling. As indicated in the characteristics of 1.0μm/1.5μm device in Fig. 5, the corresponding DCRs at the maximum SPDEs increases with temperature, signifying that thermal energy aids electron escape from the InAlAs/InP heterojunction. We have estimated the self-recovery time by measuring the SPDE dependence on DCR and deduced that for the SPDE to peak at a DCR of 10-100kHz, the carrier escape time is 10-100μs with our current devices.

To investigate whether the SPDE measurements in Fig. 5 are affected by the laser pulse rate, due to the long recovery time, we have measured the SPDE as a function of the laser illumination rate. The results for a 1.0μm/1.5μm device at 140K are shown in Fig. 6(a) . At illumination frequencies above 100kHz, the device is unable to fully recover between laser pulses, thus showing a reduced SPDE value. When the laser pulse rate is reduced to 50-100 kHz, the maximum SPDE of 20% is obtained.

 

Fig. 6 Estimates of device recovery time. (a) Saturation effects on SPDE due to slow recovery time. The 1.0μm/1.5μm structure was tested at 140K with variable input laser rates, at a bias of 43.2V corresponding to 8kHz DCR. Single-photon detection efficiency is 20% at 100kHz or lower illumination rates. (b) Recovery time dependence on bias voltage, at 180K, from self-triggered dark counts. The breakdown voltage of the device is 42 V.

Download Full Size | PDF

The device recovery time is also strongly dependent on the amount of overbias applied above breakdown, signifying the importance of tunneling on the escape of electrons from the barrier. By self-triggering the counter with the dark pulses produced by the device, we have found that the dark count measured immediately after the trigger is lower than the normal dark count rate. The duration for the dark count rate to return to its normal value is characterized by the self-recovery time. The self-recovery time measured in this manner is plotted in Fig. 6(b) as a function of applied bias.

It is rather easy to reduce the recovery time by lowering the barrier height via replacing the InP/InAlAs heterojunction with InP/InAlGaAs or similar materials. By adjusting the composition of the quaternary material, the heterojunction barrier can be finely tuned. SPADs with tens of nanoseconds self-recovery time have been demonstrated by our group [4]. The practical limit of the self-recovery time will then be the afterpulse rate and optical cross talk.

5. Conclusion

In summary, a self-quenching and self-recovering SPAD with a novel zinc diffusion design has demonstrated that in a simple single zinc diffusion step, one can achieve both high SPDE and low DCR. This design is particularly attractive to single-photon detection in Geiger-mode operation, which demands extremely critical process control in prevailing Zn-diffused structures such as single and double guard ring structures. When operating at <100 kHz laser pulse rate, 20% SPDE has been achieved at 8 kHz DCR at 140K. By modifying the epitaxial layer design following the same device concept, the self-recovery time can be reduced significantly to accommodate higher data rates.