1. Introduction

It has been shown that efficient quantum computation can be performed using only linear optical elements in combination with single-photon sources and single-photon detectors [1]. For that reason, the on-chip implementation of such elements to obtain a scalable system is strongly under investigation on several platforms, e.g. silica [2–9] and III–V semiconductor based systems [10–17]. Self-assembled semiconductor quantum dots (QDs) are promising candidates as on-chip single-photon sources, due to their practical integration in solid state systems, robustness and high brightness [18]. Furthermore, they can generate entangled photon pairs [19,20] and indistinguishable photons [18,21]. It has been shown that purely resonant excitation enhances the coherence properties of the emitter [22], and pulsed resonant excitation enables triggered photon emission with a high degree of indistinguishability [23]. Coherent optical control of a single QD has been experimentally verified by the measurement of Rabi oscillations [24]. However, clear Rabi oscillations, i.e. more than one period, and triggered single-photon emission from resonantly excited QDs, were not shown up to now for on-chip QDs, emitting in an integrated waveguide circuit. Recent experiments in free-standing [16] and multi-mode [25] waveguides (WG) have shown to suffer from a high laser-background guided in the WG system.

In our work, we demonstrate that this hurdle can be overcome by carefully adjusted laser excitation pulses enabling nearly background free triggered single-photon emission in a rib WG structure. Clear Rabi oscillations over two periods and a cross-correlation measurement between the two output arms of the on-chip beamsplitter (BS) show that our device works at a single-photon level under resonant pulsed excitation, opening exciting new perspectives for fully integrated quantum circuits.

2. Sample structure and Experimental setup

The sample was grown by metal-organic vapor-phase epitaxy using a (100)-GaAs substrate. It consists of a 2 μm thick Al_0.42Ga_0.58As cladding layer below a 380 nm thick GaAs core layer with implemented InGaAs/GaAs QDs. The BSs were subsequently fabricated using electron beam lithography and dry etching. To obtain WGs which support only the fundamental TE and TM modes, the width of the WGs was chosen to be 570 nm. The profile of a coupler region, fabricated similarly to the one used for the optical measurements, can be seen in Fig. 1(a). To reduce the amount of surface defects, which destabilize the electrical environment around the QDs, the structure was not etched down to the AlGaAs layer; instead a GaAs slab of about 30 nm in thickness was left. Due to diffusion limitations the etching rate between the coupled WGs is slower than at the outer edges of the WGs. This leads to a coupler formed by WGs which are not fully separated. More details on the sample are reported in [17] where the BS has been investigated under quasi-resonant continuous wave (CW) laser excitation.

 

Fig. 1 (a): Cross-section SEM picture of a BS similar to the one used in the measurements. It can be seen that the WGs are not completely separated. (b): Excitation scheme: QDs are excited from the top, behind the coupling region; the emitted light propagates along the WG, splits inside the coupler and can be collected from both output arms simultaneously. (c): Spectra collected from the output arms under non-resonant pulsed excitation at 800 nm. The emission line at 878.25 nm shows the desired 50/50 splitting ratio. The second emission line originates from another QD nearby.

Download Full Size | PDF

After the fabrication, the sample was cleaved perpendicularly to the WGs, placed into a liquid helium flow cryostat and cooled to 5 K. The optical setup allowed the excitation of QDs from the top and the simultaneous observation of light transmitted from the two output arms of the BS to the side (Fig. 1(b)).

3. Results

Typical emission spectra are shown in Fig. 1(c), taken from each output arm under pulsed excitation above the GaAs band gap with a laser pulse duration of 3.9 ps and a corresponding linewidth of 101.6 GHz. The emission line at 878.25 nm shows the splitting of triggered photons with the intended ratio of 50/50 which is in good agreement with previous results under quasi-resonant CW excitation [17].

For a pulsed resonant excitation schema, spectrally broad laser emission can result in a high background inside the WG which leads to the necessity of post-filtering techniques to reduce its impact on the visibility of the quantum operation. To overcome this issue, the laser pulses have been spectrally shaped, in order to reduce the amount of unwanted scattered laser light into the WG. Therefore, the light of the excitation laser is sent into a monochromator and coupled into a single mode fiber afterwards. This is a comfortable and straightforward way to change the properties of the laser pulses [26]. Due to the spectral separation of the light due to diffraction, only a part of the frequency spectrum enters the fiber which results in a shaped laser pulse at the output. By using gratings with a different blaze, the duration and the linewidth of the pulses can be varied (Figs. 2(a)–2(c)). Two different gratings were used to find the best conditions for achieving the purest single-photon emission.

 

Fig. 2 (a–c): Temporal and spectral width of the three different laser pulses used in this work for the resonant excitation of the QDs. All curves are fitted with a Gaussian lineshape. (d–f): Spectra of triggered resonance fluorescence of QD 1 under pulsed excitation for laser linewidths shown in (a–c); the QD is located in arm 2 and the light is detected at the output of arm 1. The data points are fitted with a double Gaussian lineshape. (g–i): Integrated QD intensity as a function of the square root of the excitation laser power, for the pulse widths shown in (a–c). The data are plotted with (squares) and without (triangles) laser background subtraction. The data shown in (i) were fitted numerically considering the optical Bloch equations.

Download Full Size | PDF

Afterwards, the shaped laser light is send into an objective with a NA of 0.45 to resonantly excite a QD. For an effective suppression of laser stray light, the laser polarization is chosen parallel to the WG, as previously shown [25]. The selected QD is named QD 1 from now on and it is found in one of the WG arms, labeled as arm 2. QD 1 is located approximately 50 μm behind the directional coupler, in respect to the cleaved edge. The QD emission propagates through the directional coupler, divides into both WG arms and is collected at the output named arm 1 from the side. For QD 1 an additional very weak off-resonant laser (100 pW) at ≈660 nm was used to stabilize the electrical environment [27]. However, the off resonant laser gives no measurable contribution to the QD emission, in contrast to the resonant laser.

Figures 2(d)–2(f) show resonance fluorescence spectra of QD 1 under π-pulse excitation with different excitation pulse widths. All spectra clearly reveal a sharp emission line on top of a broader laser background signal. The QD emission-to-laser background ratio increases with decreasing linewidth of the excitation laser. The highest ratio of I_QD/I_L = 9.3 is achieved with an excitation pulse duration of 54.6 ps with the corresponding linewidth of 11.5 GHz. These results can be understood considering that the QD linewidth is 4.9 GHz, and therefore closest to the spectrally smallest laser excitation pulse. Decreasing the excitation power to ≈2/3 of the π-pulse power, I_QD/I_L further increases to 16.9.

Figures 2(g)–2(i) show the excitation power dependence of the integrated QD intensity for three different excitation laser linewidths. The power for which the π-pulse is reached shifts to lower values with decreasing laser linewidth and clear Rabi oscillations with more than one period can be observed when the laser is shaped to the smallest linewidth (11.5 GHz). The fact that we cannot observe clear oscillations in Figs. 2(g) and 2(h) might be related to the increased scattered laser light and due to excitation power induced dephasing processes for the corresponding laser pulse types. The data with subtracted laser background were fitted numerically by solving the optical Bloch equations of a two level system with an additional decay channel, similar to [28]. With the fit data it can be estimated, that the QD state is excited with a fidelity of 72.4% ± 3.3%. Therefore, the following experiments have been performed with excitation pulses of 54.6 ps pulse duration and corresponding linewidth of 11.5 GHz.

To verify single-photon operation under pulsed resonant excitation, a cross-correlation measurement between the two output arms of the on-chip BS was performed, which corresponds to an autocorrelation measurement using the on-chip BS. QD 1 was excited with a laser power corresponding to the π-pulse. The emission from one output arm is send to a monochromator where a spectral filtering is done before sending the light to an APD. The purpose of the monochromator in the beam path is the spectral control since an optional access to a CCD camera is given with it. The light from the other output arm is send through a fiber directly to an APD, with just a 865 nm long-pass filter to assure that no light from the off-resonant laser is detected (see Fig. 3, left). The result of the cross-correlation in Fig. 3 reveals a g⁽²⁾(0)-value of 0.18 which is in good agreement with the expected value for I_QD/I_L = 9.3(g⁽²⁾(0) = 0.186).

 

Fig. 3 Left: Sketch of the setup detection path. Right: Cross-correlation measurement of the QD emission between both detection arms via resonant excitation of the QD under pulsed excitation. The numbers are normalized coincidence values for a ±3 ns binning.

Download Full Size | PDF

The absence of the spectrometer in one detection arm shows that the spectral filtering of the QD line is not necessary to achieve a good single-photon emission.

Two additional QDs were also investigated under pulsed resonant excitation, located nearby QD 1 in the WG. Autocorrelation measurements on the output of arm 1 (not shown) reveal g⁽²⁾(0)-values of 0.21 and 0.29, comparable to the previous measured on-chip g⁽²⁾(0)-value. We anticipate that the small fraction of laser background that is still present could be further reduced by using laser pulses with an even smaller linewidth.

It is important to note that the average linewidth of QDs under pulsed resonant excitation is 4.9 GHz measured from two different QDs in the WG structure. That is an indication of broader linewidths of QD emission lines in our structure in comparison to bulk material [29]. The reason for this could be spectral diffusion caused by electrical fluctuations at the sidewalls of the WG in the vicinity of the QD. This could be overcome by the fabrication of a WG with a tapered end for having the side surface far enough from the QD. Moreover, a thin silicon compound layer deposited on top could help to stabilize the QD environment.

4. Conclusion

We demonstrated triggered resonance fluorescence from a QD integrated in an on-chip BS structure. By varying the duration of the excitation pulses, we show that for a sufficiently small laser linewidth (11.5 GHz), in comparison to the linewidth of the QD emission (4.9 GHz), a ratio of I_QD/I_L = 9.3 for a π-pulse and I_QD/I_L = 16.9 for a 2/3π-pulse could be achieved. This led to the observation of Rabi oscillations over two periods, demonstrating the coherent excitation of a single QD state with a fidelity of 72.4%. A cross-correlation measurement between the two output-arms of the BS exhibits the correct operation of the device on the single-photon level (g⁽²⁾(0) = 0.18) under π-pulse excitation. Since one detection path was only spectrally filtered with a short-pass filter at 865 nm, the combination of the present excitation scheme and used WG design has the potential to work without additional off-chip filtering which is an important step towards fully integrated quantum photonic circuits.