1. Introduction

A single epitaxial quantum dot (QD) is one of the most promising synthetic nano-systems for the realization of efficient emitter of single photons on-demand [1–6]. Quantum dots can be fabricated from numerous compound semiconductors that usually combine group III/V or II/VI elements [7–9]. Due to the quantum size effect and atomic-like density of states, the optical transitions between discrete QD energy levels allow for generation of single photons, which can be used as flying qubits playing a significant role in quantum communication protocols. Such a quantum emitter to become practical and competitive to other solutions should be ideally an isolated system characterized by high radiative rate and good single-photon emission purity (low multiphoton events probabilities), whereas at the same time enabling high radiation collection efficiency. This is however not an easy task to achieve, especially when combined with emission wavelengths to fall into the second or third telecommunication windows, which is still a holy grail as it promises better characteristics in many respects than the parametric lasers sources currently used for realization of the quantum communication protocols in both the free space as well as in the fiber networks.

The epitaxial self-assembled growth very often results in too high areal density of nanostructures which makes the measurable spectra dense, thus disallowing for operation on a well-defined exciton state. An efficient way to reduce the number of usable quantum emitters, ideally down to a single QD, relies on the processing of the sample surface to fabricate a microstructure of a few microns (or even submicron) in diameter, preferably in a deterministic way [10–13], namely, in a specific position followed by a preselection procedure realized by surface luminescence imaging, where a suitable QD can be identified. Although indeterministic processing of microstructures randomly selects a few tens of QDs from a large ensemble, it is still possible to find a functioning example.

In this work we focus on the use of a plasma focused ion beam technique with double beam system (SEM/Xe-PFIB), which is an advanced tool for micro- and nanofabrication of precise structures [14]. The modern SEM/FIB instruments enable simultaneous sample imaging and processing by its dual-beam configuration that consists of both electron and ion columns. Such systems are widely used for a large variety of applications (for example TEM lamella preparation or failure analysis [15,16]) including rapid prototyping of photonic structures being a mask-less technique, as compared to chemical or reactive ion etching techniques which require mask development. Next, both FIB processing accuracy as well as fabrication yield depends strongly on several key parameters such as acceleration voltage, beam current, and dwell time, therefore for specific purposes preliminary studies in this regard are required. In our case, when a quantum dot emitter is integrated inside photonic microstructures, we must pay further attention on the possible and unavoidable amorphization and damage of the crystal structure due to high-energy ion and target sample interactions (e.g. implantation, redeposition) which may induce change in transmission at specific wavelengths, refractive index and, most importantly, it may reduce the internal quantum efficiency of QD by boosting non-radiative recombination rate due to generated crystal defects. The latter is crucial when optimization of the photonic performance of the device requires having the cap layer thin (few tens or single hundreds of nm) above the QD layer. Then it should be realized with the least possible acceleration voltage to minimize the implantation depth. If the FIB processing is used to perform the photonic structure with preselected QDs (i.e., micropillar or circular Bragg grating), the size of the structure must be bigger than the possible degradation length of ∼100 nm in the lateral direction which is typically comparable to the implantation depth, to sustain sufficient distance from the optically active QD.

The Xe-PFIB technology allows for a reduced implantation depth as compared to the more commonly used FIB system with Ga ions, which was used for realization of micropillars cavities with CdTe [17] or InAs [11] quantum dots materials systems covered, however, by a more than micron thick multilayer structure forming the top distributed Bragg reflector (DBR). In such case, the thick capping layer protects the optically active QDs from ion induced defects. In contrast to that, our goal is to enable FIB processing also for significantly thinner capping layers, for instance, when the top DBR is not needed. The reduced implantation depth is related to the fact that the xenon ions are inert elements and thus neither permanent chemical bonds are formed, nor any active chemical reactions occur. In addition, the Xe ions are larger than the Ga ones, so they penetrate crystal structure to a lesser extent. Another consequence of the Xe ion size is the faster milling and structure formation. On the other hand, however, a pronounced redeposition of the material often takes place which is typically undesirable result and might be reduced by using the lowest possible beam currents in the dual-beam microscopes. Then, the theoretical resolution for the Xe-PFIB is lower as compared to the gallium equivalent system, which for 4 pA current the plasma ion beam spot size is around 20 nm. We note here that for the photonic microstructures of few micrometers in diameter of their in-plane size, the Xe-PFIB resolution in terms of beam spot size around 150 nm for 1 nA current [18] is fully sufficient.

In this work, we demonstrate the optimization of micro-mesa structures with QDs buried in GaAs emitting close to the second telecom window, i.e., from 1200 to 1260 nm. The photonic yield of the mesa is first approximated by using FDTD numerical studies, taking into account various capping layer thicknesses, mesa diameters and etching depths. Next, we use two different approaches to fabricate the mesas (see Fig. 1). In the first approach, we determine the mesa geometry entirely by Xe-FIB processing, while in the alternative approach we use first a wet-chemical etching for capping layer thickness reduction, and then the Xe-FIB is used for shaping its lateral geometry. Eventually, in the framework of high-resolution micro-photolumienscence (µPL) studies, we investigate the QD emission performance to conclude about the yield of the proposed processing methods.

 

Fig. 1. Schematic approach to the fabrication of photonic microstructures. This diagram shows schematically two approaches to process mesa structure. The first approach relies on Xe-FIB processing of the mesa, where the ion beam acting from above decreases the capping layer thickness and then forms a cylindrical mesa structure. Another proposed approach is a two-step process, where a wet chemical etching is used first to remove the top GaAs layer and then by Xe-FIB the mesa is formed.

Download Full Size | PDF

2. Fabrication of QD-microstructures

In our studies, we use a GaAs based sample grown by metalorganic chemical vapor deposition (MOCVD) with a self-organized InGaAs/GaAs quantum dots (QDs) embedded layer and a GaAs/AlGaAs distributed Bragg reflector (DBR) underneath. The DBR is composed of 23 pairs of GaAs/AlGaAs layers on top of the undoped GaAs buffer (300 nm) to obtain a high reflectivity plateau centered around 1.3 µm wavelength (see Fig. S5 of the Supplementary with the reflectivity spectrum). The DBR is followed by 630 nm thick GaAs layer on which the dots are deposited and the covered by 650 nm GaAs cap. The thicknesses were planned so, to locate the QD layer in one of the field distribution maxima (see Fig. S4 of the Supplement). The QDs are grown utilizing the Stranski-Krastanov transition from a 2D thin film (or a wetting layer) into nucleation of 3D smooth nanostructures. The estimated surface density of the QDs is approximately 10⁹cm^-2 (10 per µm²). Due to an inevitable strain occurring in the InGaAs/GaAs, the ground state energy of the optical transition is typically below 1 µm wavelength. In our case, the QD layer is covered with additional low indium content In_0.20Ga_0.80As strain-reducing layer to tune the emission wavelength towards the second telecommunication window. The QD morphological parameters can be expected to be similar to the dots investigated previously [19], i.e. where the in-plane extension of the dot is about 20-30 nm, the height approx. 6 nm and the In content inside the QD ∼70%. The sample layout is shown schematically in Fig. 1, whereas more details have been described elsewhere [19–22].

We use the MOCVD grown sample containing optically active QDs processed by the Xe-PFIB method. The key challenge here is to limit the unwanted effects of the ion-target interaction which might cause a crystal damage or amorphization in the vicinity of the active layer that influence the QD radiative recombination rate, or a sputtering of the material which can cause a redeposition on the sample surface and influence the light scattering at the mesa edges.

In case of additional in-plane optical confinement provided by the fabricated micromesa structures we can expect, in general, pronounced variations of the far-field pattern and the photon extraction efficiency within the given numerical aperture, with the change of both, the mesa diameter and the thickness of the top GaAs layer. It also means that by tuning these parameters carefully with a properly adjusted fabrication process, we should be able to optimize the mesa geometry for the highest collection efficiency. Our recent preliminary studies [23] showed experiments in which we performed a series of mesa structures of various diameters from 2 µm to 15 µm and fixed height of approximately 1 µm. Those results indicated that mesas of 4-5 µm in diameter are suitable to observe sparse spectra of excitonic emission from only few QDs and assure good isolation of the centrally located dots from the outer parts exposed to the ion-induced damages. Therefore, here we perform all processes for the fixed in-plane geometry where the mesa diameter is 4 µm and the outer etched region along the radii is set to 10 µm.

First, to realize the mesa series with various capping layer thickness (CLT) only by Xe-PFIB, the sample surface is prepared by using an ultrasonic cleaner with isopropanol. The process requires smooth etching of planar sample from 145 to 545 nm with 100 nm step to obtain CLT of 500 nm, 400 nm, 300 nm, 200 nm and 100 nm. This step is performed with the beam energy of 10 keV, current of 1 nA and dwell time of 1 µs settings. Then, we process mesas with 30 keV, 1 nA and 1 µs resulting in roughly 1 µm mesa height (see Fig. S1a in the Supplement). We note here that all the processed mesas are realized with the ion beam scanning mode along the radial direction starting from the inside to the outside.

Next, we performed another mesa series in a two-step process, in which a sequential wet chemical etching (WCE) is followed by Xe-PFIB mesa processing. The WCE process is utilized only to reduce the CLT. After sample cleaning (cold concentrated H2SO4, hot acetone and isopropanol), a photoresist (Microposit S1813 G2 positive) is deposited on the surface, exposured (h-line lithography, 96 mJ/cm2) and developed (Microposit 351:H2O, 1:5, 30 s) giving a sample covered in a half with the resist (see Supp. Fig. S2a)). The etching is realized by H3PO4:H2O2:H2O (1:1:40) which allows to etch about 100 nm of GaAs per minute. First etching is done for 30 seconds which reduces the CTL by 50 nm. Then, the photoresist is removed (Microposit Remover 1165) and orthogonal half of the sample (Supp. Fig. S2b)) was covered by the resist, as previously. The second etching holds for 1 minute which reduces the CTL from the exposed surface of the sample by another 100 nm. Last, we exchange the photoresist again covering less than a quarter of the sample (a reference, not etched sample piece), and the third etching holds for 4 minutes and 30 seconds thinning CTL by another 450 nm (Supp. Fig. S2c)). Such an etching procedure results in a sample profile as demonstrated in Fig. 2(a)). The surface topography of the etched structure was investigated by white light interferometry (WLI) method. The measurements were taken with a Talysurf CCI optical profiler fitted with 1024 × 1024 pixel camera, 10x/NA = 0.30 Mirau objective and quartz halogen light source illuminator. The measured data was analyzed with TalyMap software. The obtained profile shows the remaining CLT series of approximately 50 nm, 100 nm, 150 nm, 200 nm, 500 nm, and 650 nm.

 

Fig. 2. a) Diagram obtained from the profilometer (white light interferometry method) after wet etching showing the layout and the approximate cap layer thicknesses of all regions are given. b) SEM image showing Xe-PFIB fabricated mesa structures situated in different regions. In the inset we focus on the two mesas, the one with characteristic defects formed by the redeposition effect, and the second with a protective layer of platinum applied electronically with a GIS system, where a majority of the defects are reduced.

Download Full Size | PDF

The mesa fabrication procedure by Xe-PFIB for the fixed in-plane geometry is then realized on all regions defined in the WCE processes. First, the sample is processed with a fixed settings of 30 keV, 1 nA and 1 µs, such that in each sample region we fabricate two identical mesa structures (see Fig. 2(b)). Although 1 nA beam current values is not the lowest possible in the used microscope, it results in high quality structures already, realized within less than 5 minutes. We note here that further optimization, like polishing of microstructure sidewalls, is still possible with lower beam currents, but with a cost of nevertheless the processing time extended to approx. 60 minutes. In our case, therefore, we are looking for optimum parameters, i.e. the lowest possible energy and current, so that the quantum dot radiative decay is preserved and a reasonable fabrication time on the order of several minutes.

Naturally, the higher the beam energy, the deeper the ions can penetrate into the crystal structure. For typical maximum beam energies of the dual beam SEM/FIB systems, the beam energy level is set to 30 keV. At this energy, the maximum depth of ion penetration is at the level of tens of nanometers for amorphous materials. In case of a crystal structure, the depth of ion penetration into the structure is higher due to the channeling effect, which depends on the orientation of the beam relative to the crystal planes of the material being treated. In the case of using a beam with high energy and dose, which in a typical Xe ion plasma beam voltage ranges from 8 kV to a maximum value of 30 kV, the penetration depth of a few micrometers is expected. For example, when the beam of 30 kV acts on silicon and its orientation is parallel to <111 > crystallographic plane, the maximum range is approx. 0.4 µm [24]. Therefore, to reduce the destructive impact on the optically active QDs of the ion implantation causing point or line defects, it is possible to impose a tilt of the ion beam by 3.8 degrees, which also improves sidewalls quality of the mesa [11] . However, it should be noted that the higher voltage settings are crucial for higher resolution of the milling, as the spot of the ion beam is smaller.

Another advantage of the dual-beam system is that it gives us the ability to directly prototype the structures. In general, additional protective layers can be sputtered on the sample surface using the electron mode (typically tens of nanometers) and the ion mode (much thicker layers in a micrometer scale). The penetration depth of the ions into the sample structure depends on the accelerating voltage in the ion column, while the proximity of the ion induced defects is also related to the cap layer thickness. When the penetration depth is similar to the cap layer thickness, the issue of the crystal defects in the proximity of the QDs must be avoided. We propose covering the sample surface with a protective layer of e.g. platinum or carbon. The protective layer can be sputtered by using the gas injection systems (GIS). In this manner, we use both carbon- or platinum-based precursors for deposition of about 80 nm of the layer thickness. Besides the expected decrease of the ion penetration depth, i.e. both carbon and platinum acts positively by trapping the ions and by stabilizing the trajectory of the ion beam at the interface. In addition, we also expect improved mesa shape. Another advantage is expected in the context of a reduced redeposition effect [25]. The redeposition effect caused by Xe-PFIB sputtering is typically observed as characteristic oval-shaped nano-droplets on the surface of the mesa [see Suppl. Fig. S3], which could in principle influence the light emission and therefore should be avoided by using one of the protective layers, if possible. However, the disadvantage of depositing additional layer on the surface is its problematic removal. If the layer is not removed or it is too think, the efficiency of the device with single QDs could be limited. Naturally, the improved quality of the Xe-PFIB processed mesa structure by depositing the protective layer can be verified by analyzing the QDs photoluminescence intensity.

3. Numerical results – FDTD simulations

In this section, we demonstrate the numerical investigation of the dipole emission properties from the mesa microstructure. We use the finite difference time domain (FDTD) method implemented in commercially available software Ansys Lumerical. Our simulations allow for evaluation of collection efficiency of light emitted by a dipole-like emitter towards the first lens, i.e., a detection system in the free space with the microscope objective used for laser excitation and detection. The modelled structure is a 3D system composed of GaAs cylindrical mesa of 1-7 µm in diameter on AlGaAs/GaAs bottom DBR. The dipole is located in the position of the QD layer and centered in the mesa plane. We evaluate the field distribution with the FDTD engine in a wavelength range of 1.2-1.4 µm for a series of structures in function of the capping layer thickness (CLT) from 50 to above 600 nm. The analysis of the far-field above the mesa allows us to determine the collection efficiency η within the angle of 40 degrees which corresponds to the numerical aperture of 0.6, and which is consistent with the microscope objective we use in the experiment. In Fig. 3(a)) we show the results for η in function of CLT and mesa diameter at 1240 nm wavelength. The acquired data show that the highest extraction efficiency is achievable for d = 1 µm and CLT = 600 nm and for d = 2 µm and CLT = 400 nm. The latter exhibits approx. 10% extraction efficiency - the 2D cross-sectional visualization of the dipole emission, as well as evaluated far-field profile taken above the mesa structure, are demonstrated in Fig. 3(b)) showing a clear directional emission upwards. In addition, in Fig. 3(a)) we observe a characteristic oscillatory behavior of η versus CLT, which could be related to the change in the local density of states at the dipole position when varying the distance to the upper semiconductor-air interface [26] or to the optimum trade-off between a direct outcoupling from the dipole and outcoupling of the back-reflected light by DBR structure [27]. For the case of d = 4 µm (see inset in Fig. 3(a) on the right of the colored map), i.e. the mesa size investigated then more in detail experimentally, we observe two maxima at CLT = 400 nm and CLT = 200 nm of 5.4% and 5.2%, respectively. An increased efficiency is related to the dominating upward outcoupling, as it is visualized for the latter case in Fig. 3(c)) showing the field distribution. In both cases, a high reflectivity of the DBR around 1240 nm wavelength on the order of about 90% is crucial for observing increased extraction efficiency (see Supplementary showing the measured and calculated reflectivity spectra in Fig. S5).

 

Fig. 3. a) Calculated extraction efficiency of the dipole emission at 1240 nm wavelength for the photonic microstructure defined by a cap layer thickness (CLT) and mesa diameter (“d”). b) and c) show a near field profile of the dipole emission and the far-field profile in the inset of two cases which result in 10% and 5.4% of light extraction, respectively.

Download Full Size | PDF

4. Optical characterization – microphotoluminescence spectroscopy

Micro-photoluminescence measurements were performed at cryogenic temperature (5 K) using a liquid-helium continuous-flow cryostat, where the samples were mounted on a copper cold finger. Non-resonant optical excitation was realized by a semiconductor laser diode emitting continuous wave at 660 nm. Both, the excitation and the emission are transmitted via a long working distance microscope objective with 0.6 numerical aperture. The QD emission is further spectrally resolved by a 500-mm focal-length monochromator with a 600 grooves/mm grating (blaze at 1200 nm) and detected using a multichannel liquid-nitrogen-cooled InGaAs linear detector (Princeton Instruments). The setup provides effectively a spectral resolution of about 25 µeV, and diffraction-limited spatial resolution and light spots below 2 µm. These allow to excite precisely one pillar at a time and to resolve spectrally single QD emission lines.

In Fig. 4(a)) we show results of measurements performed on a planar, non-processed sample in function of the excitation power. The µPL spectra give an insight into the QDs ensemble, in which we observe broad distributions of emission centered at 1190 nm and 1220 nm, with pronounced optical transitions from multiple single QDs. Next, we investigated the reference mesa structure processed by Xe-PFIB with no modifications of the capping layer, as shown in Fig. 4(b)).

 

Fig. 4. Micro-photoluminescence spectra in dependence on the excitation power for a) a planar reference ample and b) a Xe-PFIB processed mesa reference sample with a diameter of 4 µm – no change of the cap layer thickness.

Download Full Size | PDF

Obviously, the number of emission lines is fewer indicating less optically active QDs excited by the laser in the mesa as compared to the planar sample. The reason can be manifold. First, the laser spot size of the Gaussian-like beam is in fact larger than the diffraction limited size, with just significantly lower intensity when going further from the spot center, which however, can still be sufficient to excite the dots. Secondly, due to the distribution of hot carriers generated with non-resonant excitation the electrons and holes can diffuse out of the excitation spot even on micrometer ranges [28,29]. Fabricating the mesa of e.g. 4 µm in diameter physically limits both these factors. In addition, the dots which are located close to the mesa sidewalls may be exposed to nonradiative processes due to carrier traps on the sidewall surfaces, or due to defects located deeper from the sidewall generated by ions-induced crystal damage (expected there to be more significant than in the mesa center). So, the number of optically active dots detected in the spectra from the mesas can be even smaller.

Next, it is worth to notice that the maximum emission intensity of particular lines is very similar to the ones observed from the planar sample, which implies insignificant negative influence of the ion beam. Furthermore, we have also compared the PL linewidths for unprocessed samples with the mesas and they are similar in the range of 120-150 µeV, and significantly larger than the setup resolution. They are dominated by spectral diffusion due to illumination-induced fluctuations of the electrical environment of charges trapped in the vicinity of the dots [30–32]. As we do not see clear fingerprints of additional spectral diffusion caused by the mesa sidewalls, we can assume that its source are charge traps in the bulk surrounding the dots. It can also partly be related to the fact that the dots located closer to the mesa sidewalls are more vulnerable to ion-induced material deterioration generating non-radiative recombination centers. Hence, they such QDs are very often optically inactive and their response, which could potentially be additionally broadened, might be absent in the spectra.

Next, we examine the mesa samples processed by Xe-PFIB with additional cap layer milling. Here, we are able to observe single QD emission lines for the reduced cap layer thickness by 100 nm from the original thickness (see Supp. Fig S1b), while further CLT reduction revealed no PL from QDs probably due to ion beam induced defects.

Another approach to obtain a 3D shaping of the mesa structure relies on the two-step processing, i.e. capping layer etching by WCE and mesa processing by Xe-PFIB. The µPL results using both the planar sample and the mesas in dependence on the cap layer thickness is demonstrated in Fig. 5. A change in the CLT for the planar sample (Fig. 5(a)) results in variation of the integrated intensity in the two spectral ranges, which could be related to the variation of the local density of states or to a possible inhomogeneity of the QDs found in different sample regions. In case of the mesa structures (Fig. 5(b)) we observe sparse spectra with well isolated single optical transitions of high intensity as compared to the planar sample. We note here that both, the planar sample and the mesa structures with cap layer thickness of 50 nm and 100 nm are optically inactive, which is probably related to the close proximity (and/or too high concentration) of the non-radiative recombination centers.

 

Fig. 5. Micro-photoluminescence spectra for different cap layer thickness realized by hybrid approach, a two-step process, in which wet chemical etching is used for cap etching and Xe-PFIB is used for mesa processing, for a) a planar sample and b) a mesa sample with a diameter of 4 µm.

Download Full Size | PDF

The improved quality of the mesa structure can be achieved by using a protective layer made from carbon or platinum. We performed few more mesa structures on the sample region with capping layer of 200 nm using a decreased beam energy of 20 keV and with a platinum protective layer (80 nm). As it is shown in Fig. 6(a)) we observe bright QD excitonic states emission from such structures, with a high intensity optical transition labeled as “A” at 1244 nm, which could be used in principle as a good single photon source. Although we have not measured the emission extraction efficiency, we could roughly estimate it by comparing the signal levels with previously investigated dots for which the extraction efficiency had been determine [33,34]. It gives about 2-4% in the current case which is not far from the calculated value of above 5%.

 

Fig. 6. a) Micro-photoluminescence spectra for Xe-PFIB processed mesa protected with 80 nm platinum layer and 200 nm cap layer. In the inset SEM image of the microstructures is presented. B) Excitation power series and c) double logarithmic plot of emission intensity of lines “A” and “B” showing linear and superlinear dependence suggesting exciton and biexciton origin of emission.

Download Full Size | PDF

In order predict on the lines’ origin, we performed the excitation power series of µPL spectra shown in Fig. 6(b)) and plotted the PL intensity vs excitation power in Fig. 6(c)) (in a double logarithmic scale). For low excitation, a linear intensity increase with a slope of 0.98 ± 0.05 is obtained for line “A”, with an evident saturation for higher powers, while a superlinear increase with a slope of 1.49 ± 0.05 corresponds to the redshifted line “B”. It could suggests that the observed lines are associated with the exciton and biexciton emission, respectively, which is of practical-relevance due to possible generation of cascaded, polarization entangled photon pairs [35,36]. However, no fine structure splitting (FSS) could be detected for these lines in the polarization dependence (not shown here) within the setup resolution, which can suggest that the investigated dots reveal rather symmetrical in-plane confinement potential. But, emission related to trions exhibiting no FSS cannot be excluded. Therefore, a decisive argument on the origin of the lines could be only given by performing cross-correlation experiment, which however, is beyond the scope of this work.

5. Conclusions

We use the Xe-plasma focused ion beam, which is a maskless processing technique utilized for direct prototyping of the surface, for a precise shaping of the photonic GaAs-based microstructures with QDs emitting near the second telecom window. By choosing appropriate ion beam parameters, geometry of the structure consistent with the FDTD numerical simulations, and by applying an additional protective layer to reduce the ion implantation induced defects, we achieved bright emission related to excitonic complexes confined in single QDs, which proves that this technological approach is promising for further quantum optical experiments. The most successful fabrication process was realized by a two-step hybrid approach including wet chemical etching used for cap layer control and Xe-PFIB for mesa processing. We showed that despite degradative influence of the ion beam, the process is still efficient for QD cap layers below 200 nm even, which has never been shown for processing the QD structures with FIB. We are convinced that there is still room for improvement of the next generation of Xe-PFIB processing of microstructures by using lower currents, by tilting the ion beam angle or by applying different protective layers. The proposed framework can also be analogically verified for the InP-based quantum dots emitting in the range of the third telecom band, making it even more application-relevant for quantum technologies.