1. Introduction

The single photon emitter (SPE) is a device crucial to the development of many emerging technologies in the areas of quantum metrology, quantum computing, and quantum information, with continuous-variable quantum key distribution being a noteworthy application [1–3]. There have been many proposed material platforms for SPEs, including carbon nanotubes [4], quantum dots [5], diamond nitrogen vacancy centers [6], trapped calcium ions [7], transition metal dichalcogenides (TMDCs) [8], and more recently, defect centers in hexagonal boron nitride (hBN) [9,10]­­. Among these materials, hBN has emerged as one of the most promising [1,2,11].

Much of the popularity of hBN is due to (a) its bright emission, (b) the ro­om-temperature operation of its SPEs, and (c) its ability to withstand aggressive fabrication techniques. SPEs in hBN also possess good photostability (though not a unique property) and high quantum efficiency [2,12,13], in-plane linearly polarized emission [14], and optically addressable spin states [15,16]. The optical transition energies corresponding to defect-originated emission are typically embedded deep within the bandgap, meaning that zero phonon lines are detectable above the ambient thermal noise floor without cryogenic cooling [17]. The exact origin of defect emission in hBN is still a point of discussion, with several proposed models circulating [2,9]. Recently, however, there have been significant advances including the discovery of mechanically decoupled electronic transitions in some defect centers [18], quantum emission from the ${N_B}{V_N}$ nitrogen-boron substitution in which a neighbouring nitrogen atom is missing [9], and unique spin properties from the ${V_{B}^{-}}$ defect [19,20]. The rapid pace of such advances means that the origins of defect-based emission within hBN are becoming increasingly well understood. Reliable fabrication techniques at the wafer-scale will soon be in demand as progress is made towards device implementation rather than first-principle investigations.

While emission from hBN has been studied using mechanically exfoliated flakes prepared in the style of graphene [10,21,22], investigations involving thin, continuous films have been rising in popularity. There has already been some success in devising methods of polymer-assisted thin film transfer for hBN as well as many of the TMDCs [11,23]. Such transfer methods are scalable and therefore highly suited for integration with existing­ silicon-based technology such as waveguides, microdisks, and optical cavities. In the case of hBN, wet transfer methods provide the potential for the deterministic activation of defect-based emitters by introducing strain on the crystal lattice [9,11,24]. While the outlines of several proposed hBN thin film schemes have surfaced [25,26], none so far have provided a fully detailed process flow. This is problematic since such transfer processes involve many non-trivial steps. Additionally, studies are typically more concerned with small-scale transfers, rather than a scalable process useful for integrated devices [11,23].

One promising way to construct arrays of SPEs is to drape a thin hBN film over nanostructures patterned on a silicon substrate, as demonstrated by Proscia et al. in their modified wet transfer technique [11]. They theorized that charge carrier trapping occurs near the regions with highest strain, causing defect centers to function as potential wells. More recently, work by Li et. al. has disputed the role of strain in hBN defect emitter activation. Rather, they reported localized single photon emission only from hBN films deposited directly via CVD onto prepatterned substrates [25]. The conflict between these findings leaves this important topic openly debated in the current literature. Our contribution to this discussion is two-part. First, we present an hBN transfer process flow and describe common problems one might encounter. Then, using this method, we transfer hBN films to a substrate containing nanoscale-height features and verify that the brightest emission is localized to the underlying features rather than unintentional wrinkles.

2. hBN transfer process

We began by fabricating pillar-patterned substrates from SiO₂ which were eventually used as the target substrates for our hBN transfer. The details of this process can be found in Supplement 1. The first step in our transfer process was to cut ∼13 nm thick CVD-grown multilayer hBN/Cu foil (Graphene Supermarket CVD-2X1-BN-ML) into ∼1 cm by ∼1cm squares using scissors cleaned with isopropyl alcohol. The squares were individually coated with 950K molecular weight PMMA using a Brewer spinner/hotplate combo. Note that a sanitized miniature gasket seal was used to hold the foil in place on the substrate vacuum. The spin settings were 10 seconds at 100 RPM for the spread step and 40 seconds at 2200 RPM for the spin step. The coated films were then immediately transferred to a 180° C hotplate for 120 seconds to bake the resist. A schematic representation of the device at this point in the process is shown in the first step of Fig. 1.

 

Fig. 1. A visualization of the hBN transfer process. PMMA is spun onto CVD hBN/Cu in step 1. The backside hBN is roughened in step 2, followed by a ferric chloride etch to remove the copper in step 3. The film is cleaned and lifted from solution using a target substrate in steps 4-6, followed by removal of the PMMA layer in step 7. The hBN film conforms to the substrate surface topology during the transfer process.

Download Full Size | PDF

It is important to remove the exposed hBN film on the side of the foil opposite the PMMA, so that ferric chloride etchant used in subsequent steps comes into direct contact with the copper. We found that hBN is somewhat resistant to the ferric chloride, and that islands of copper can become encapsulated between hBN layers, resulting in gaps in the final devices as well as long etching times that could otherwise be avoided. We have had success roughening the exposed hBN film using an aggressive O₂ plasma etch in a reactive ion etching (RIE) system. Exact etching parameters are given in Table 1. The samples were placed PMMA-side down in the processing chamber without worry of damaging the target hBN film, which was sandwiched between copper and PMMA layers (step 2 in Fig. 1).

Table 1. Etching parameters for backside hBN roughening.

View Table

Next, the films were floated copper side down via surface tension on a room-temperature ferric chloride bath (MG Chemicals 415) to etch the copper (step 3 in Fig. 1). It was crucial for the film to remain floating on the surface of the solution. We noted that if the film became submerged, it was difficult to transfer between baths. Additionally, the orientation of the film could easily change, making it difficult to determine which side was PMMA-coated. Once all the copper had been etched, leaving a transparent hBN/PMMA film (endpoint determined visually, ∼24 hours), the films were lifted from the surface of the etchant using a quartered Si wafer piece held with wafer tweezers, which were rinsed between each step. The films were immediately transferred between three sets of deionized water beakers in an attempt to dilute any leftover ferric chloride, again being careful to ‘float’ the films on the surface (step 4 in Fig. 1). Finally, the films were lifted from the final deionized water bath using piranha-cleaned pillar-patterned substrates (step 5 in Fig. 1) and allowed to dry overnight (step 6 in Fig. 1). It was important to allow the devices to dry before continuing, otherwise excess moisture trapped between the substrate and hBN film boiled out in following steps, resulting in areas where the hBN appears to be missing. Water between layers also negatively impacts transfer success rate, resulting in “patches” of hBN being lifted away in later wet processing steps. Drying or cleaning with pressurized nitrogen gas was avoided since the hBN/PMMA films could be blown off of the substrate.

Before removing the PMMA layer, each chip was heated to 200°C using a hotplate to remove any trapped gas between the hBN and underlying substrate. This has been shown to lead to better film conformity [11]. Heating at this point in the process flow did not damage the hBN film since the hBN/substrate interface was dry. Annealing times ranging from 10 min to 30 min were tested based on current literature [11,27]. However, there was no discernable difference in film conformity between devices annealed for different times. The devices were then placed in room-temperature acetone overnight to remove the PMMA layer before being submerged in IPA, rinsed with DI water, and allowed to dry (step 7 in Fig. 1).

The method described above produced ∼80% successful transfer, measured by area in the regions where contact was initially made between the hBN and substrate. A typical final transfer product can be observed on a large scale in Fig. 2(a). Most imperfections were located near the edges of the transferred films. These results have been verified over multiple trials, with the expectation that even higher yield is possible through additional process refinement such as further dilution of the leftover ferric chloride or a more delicate way of transferring the thin film stack between baths. Figure 2(b) shows an optical microscope image of a 1 mm by 1 mm unit cell that has been successfully transferred, while Fig. 2(c) shows the edge of the transferred film. More hBN coated unit cells representing an entire transfer can be seen in Fig. S2 in Supplement 1. All prepatterned substrates were fabricated from a thermally oxidized wafer (Supplement 1) and features ranged in height from 150 nm to 170 nm. It should be noted that our transfers were done using 1 cm by 1 cm substrates, though the technique could be easily scaled up for single transfers to entire wafers, with the limiting factor being the availability of appropriately sized CVD hBN on copper foil. Additionally, only one brief device heating step is required during the process, leaving plenty of room in the thermal budget of silicon for additional processing steps.

 

Fig. 2. Images of the hBN transfer process product. (a) A photograph of a 1 cm by 1 cm substrate coated with hBN. (b) A microscope image of a 1 mm by 1 mm unit cell completely coated with hBN. (c) A unit cell from the same wafer that contains the edge of the transferred film. Note that the areas near the edge of the hBN film have more defects compared to the interior region.

Download Full Size | PDF

Figure 3 shows an array of images taken with a scanning electron microscope (SEM). It is important to note that discontinuities and edges were specifically presented to provide contrast between coated and uncoated regions. These images clearly show the film making contact with the structures before bending back in contact with the substrate, such as in Fig. 3(e), where the edge of a film falls on a bullseye structure.

 

Fig. 3. Scanning electron microscopy images of transfers to various structures. (a) An array of ∼0.6 μm diameter pillars near the hBN film edge. (b) An uncoated ∼2 μm pillar. (c) A coated ∼2 μm pillar with a wrinkle. (d) An array of coated bullseyes clearly showing varying layer thickness and so called “natural wrinkles”, and a missing patch above the middle bullseye in the top row. (e) An hBN film conforming to the topology of a bullseye feature.

Download Full Size | PDF

It also appears that wrinkles in the film were transferred to the SiO₂ substrate. This is visible in Fig. 3(a) and 3(d) as lines that run diagonal to the array of pillars, as well as in Fig. 3(c) where a wrinkle crosses over a pillar. Some have theorized that these “natural wrinkles” are formed during the hBN CVD growth process rather than the transfer process [11], though we were unable to verify this claim. In similar hBN transfer methods, others noted that nanopillars over certain heights (>155 nm) resulted in “piercing” of the hBN film [11,25]. Piercing was not observed in our devices which typically ranged from 150 nm to 170 nm in height. We believe that this was due to the decreased aspect ratio of our micro/nanostructures, which were not patterned with electron beam lithography.

The method presented is highly suited for integration with optical cavities because of the potential to position defects at resonant antinodes paired with the robustness of hBN. However, the etched surface of the substrate appears to be somewhat rougher than the top surface which could affect optical quality. Other etching techniques, such as a buffered oxide etch could be used to improve uniformity, while still preserving the sharp edges of mask features. The topic of surface roughness characterization is left for future work.

3. Optical results and discussion

All photoluminescence images and spectra were obtained using a modified Zeiss Axioscope, as shown in Fig. 4. The excitation source was a 405 nm Pangolin laser (LDX-405NM-200MW) aligned to a 150 μm pinhole. The beam was then passed through a 450 nm short pass filter and a 468 nm short pass filter (Thorlabs FESH0450 and Semrock FF01-468, respectively). The filtered pump beam was reflected off of a dichroic mirror (Thorlabs DMLP425R) towards an infinity-corrected objective, ranging in magnification from 5x to 100x. Measurement of individual sites was performed with a 0.9 NA 100x objective lens (Zeiss EC Epiplan) and ∼4 mW of laser power incident on the sample. Light from the sample was then collected through the same objective and passed through the dichroic mirror. A subsequent 450 nm long pass filter (Thorlabs FELH0450) was used at the input of a 90:10 beamsplitter (Thorlabs BS025) to remove pump light. The collected light was split between an 8.9 MP CMOS camera (Thorlabs CS895MU) and a 50 μm core pickup fiber (Thorlabs FG050LGA). The fiber was mounted to a FC/PC collimator with a focal length of 34.74 mm (Thorlabs F810FC-543) at the 90% terminal of the beamsplitter. The output of the fiber was coupled directly into an Ocean Optics USB 4000 visible range portable spectrometer. The integration time on the spectrometer was typically set to the maximum value of 10 seconds. This put a clear limitation on the achievable signal-to-noise ratio, as well as the minimum resolvable signal. The relatively small core size of the pickup fiber functioned as a pinhole, limiting the FOV seen by the spectrometer to a ∼5 μm window on the sample. The experimental setup prohibited the study of individual SPEs, which will be the subject of future work. The main focus of the following study was to verify that the brightest hBN emission is localized to the underlying features of a target substrate.

 

Fig. 4. A schematic of the setup used to perform photoluminescence measurements. Where DIC stands for dichroic mirror, BS stands for beamsplitter, MMF stands for multimode fiber, and SPF/LPF stand for short pass filter and long pass filter, respectively. The components enclosed by a dotted outline were built into a Zeiss Axioscope that was used as a starting point for the setup.

Download Full Size | PDF

A wide-field photoluminescence (PL) image of hBN-coated nanostructures is shown in Fig. 5. The entire FOV was illuminated by aligning the pump laser to the sample at a glancing angle with an irradiance of ∼0.2 W/cm². Camera images were collected using a 10 second integration time. The large dark strip near the top right corner of the PL image shows an area where the hBN transfer failed, leaving the SiO₂ underneath exposed. As expected, there is little emission in this area, except for trace amounts of hBN that was left behind when the bulk of the film was ripped away. The area containing a defect in the hBN transfer was specifically chosen so that a control region would be present in the image. Additionally, there are many bright lines running horizontally across the coated areas in the PL image, which could either be an artifact of the CVD growth process or regions of the film that were strained as they came into contact with rough areas of the substrate during the transfer process. These results clearly show that the hBN emission is highly correlated to the underlying structures, as verified by the corresponding bright field (BF) image shown in the inset of Fig. 5.

 

Fig. 5. A wide-field PL image showing arrays of emitters with the corresponding BF image as the inset. Note that the area at the very top right of the image is coated with hBN, though the spread of the microscope illuminator makes it difficult to see from the BF inset.

Download Full Size | PDF

Figure 6 shows spectra collected by focusing the pump beam near the edges of various structures using a 0.9 NA 100x objective lens (Zeiss EC Epiplan). We noted that larger features typically resulted in a greater number of peaks from distinct emitters shown in Fig. 6(a)-(c), which show spectra collected from a 1.6 μm pillar, a 3.6 μm pillar, and an 8 μm bullseye feature. The large density of peaks in Fig. 6(c) can be attributed to the underlying bullseye feature, where strain is introduced along the inside and outside circumference of the ring feature, in addition to the center pillar. The wavelength distribution of distinct emitters in these figures falls within the well-documented range of hBN emission [9–11,25]. Sharp, intense spectral features were not present in regions without underlying structures. There are two potential causes of the broad background emission accompanying the sharp features present in these spectra. The first is that spectral diffusion resulting from the time-dependent nature of SPEs in hBN might result in what is perceived as broadening [11]. This theory is further supported by ‘blinking’ diagrams, such as the one presented in Fig. 7(d), where the respective intensities of individual emitting sites vary greatly over time. This undoubtably impacts the spectra seen in Fig. 6(a)-(c), since long integration times were necessary to achieve a desirable signal-to-noise ratio. The second potential explanation is the existence of inhomogeneous broadening resulting from the pumping and collection of out-of-plane defects within the multilayer hBN [28,29]. In this case, spectral features from many emitters closely spaced in wavelength could appear as a broad background feature.

 

Fig. 6. Typical PL results for various features with BF insets shown for reference. Raw data is the blue dotted lines and smoothed data is the black lines (a) Emission from hBN draped on a 1.6 μm diameter pillar. (b) Emission from hBN draped on a 3.6 μm diameter pillar. (c) Emission from hBN draped on an 8 μm (outer) diameter bullseye feature. (d) A BF image of a bullseye feature separate from the spectra shown in (c). (e) A PL image showing brightness distribution across the bullseye feature in (d). Emission originates from regions corresponding to the strained film edges.

Download Full Size | PDF

 

Fig. 7. Time dependence of emission in hBN. (a) Spectrum of a particular 3.6 μm diameter pillar site showing both raw (blue dotted) and smoothed (black) data. (b) BF image of the 3.6 μm diameter pillar site. (c) PL image of the pillar site. (d) Spectral trace diagram illustrating blinking nature of emitters in hBN. (e) An example time vs intensity plot for a well-isolated emitter at λ = 527 nm (black) showing random fluctuations in emission intensity. A time vs intensity plot representing the noise floor (red) has been added to show that intensity fluctuations corresponding to emitters are much greater than the noise floor. The location of the emitter in (e) is circled in (a).

Download Full Size | PDF

Figure 6(d)-(e) show a comparison of BF and PL images for a bullseye feature. In this case, the pump beam was expanded (by passing the beam through the lens contained in the microscope illuminator pathway) such that it was larger than the bullseye feature (∼8 μm) so that the brightness distribution could be visualized on a small scale. A majority of the bright emission (red) is located at the edges of ring and pillar structures, which is consistent with previous findings [11,25] and supports the theory that emission from hBN is the result of strain-activated defect centers.

To further elucidate the origin of emission in our devices we performed a spectral trace measurement on a collection of emitters located on a single 3.6 μm diameter pillar. A PL spectrum of the region is presented in Fig. 7(a) and shows a number of emitters distributed in the wavelength range 500 nm to 600 nm. The corresponding BF and PL images in Fig. 7(b)-(c) show that the “hotspots” or brightest regions fall on or near the circumference of the pillar. The corresponding spectral trace measurement in Fig. 7(d) shows random intensity fluctuations across all emitters that do not appear to be correlated. An intensity line scan for a single emitter at λ = 527 nm is also plotted against the detector noise floor in Fig. 7(e) to show that intensity fluctuations are greater than measurement noise. The noise floor was sampled from the same measurement data at a wavelength far below the measurement window determined by the optical filters in the setup (λ ∼ 250 nm). This type of spectral blinking is well documented within hBN [9,11,25] and is one indicator that single photon emission could be the dominant mechanism at play.

Next, we performed a brightness analysis across multiple arrays of emitters using the same wide-field pumping arrangement as in Fig. 5. It is important to note that the 0.4 NA 20x objective lens (Zeiss EC Epiplan) used to capture these images has a depth of field on the order of several microns, so that the light collection efficiency (at fixed focus) is not impacted by the difference in height between the etched and non-etched regions of the samples. Figure 8(a) shows a set of 12 arrays which have been assigned X and Y coordinates. Figure 8(b) shows the total distribution of pixels not capturing emission (or “dark pixels”) centered around 3.9% of the maximum camera signal before saturation. To see where the brightest pixels were located relative to the sample, we applied a filter that highlighted pixels within a certain 8-bit brightness range and set all other pixel values to zero. The minimum value for which a pixel was considered to be capturing emission (a “bright pixel”) was qualitatively set to 24% of the maximum camera signal before saturation. Similarly, the maximum value for which a pixel was determined to be capturing emission was set as 78% of the maximum camera signal before saturation. The upper threshold was applied to remove detector noise, which was naturally present due to the long integration times used to capture images.

 

Fig. 8. Brightness distribution for an array of ∼3.6 μm pillars. (a) Arrays of emission corresponding to an array of pillars that have been assigned arbitrary names. (b) Distribution of “dark pixels” (i.e. pixels not capturing emission) across all arrays. (c) A single array from (a) that has been filtered to show the brightest emission or “bright pixels”. (d) Distribution of bright pixels across all arrays in (a), with the inset showing the distribution of bright pixels for individual arrays.

Download Full Size | PDF

The distribution of “bright pixels” across all arrays is given in Fig. 8(d) where the number of pixels was plotted against the relevant binning range of the 8-bit detector. The inset of Fig. 8(d) shows the same range of data, but for each individual array. From this, we note that the brightness distribution between neighbouring arrays is relatively consistent, providing a method of determining which arrays are likely to contain many high-quality emitters, such as array X3Y3, where there are a large number tightly distributed bright pixels.

We were then able to apply this brightness filter directly to the camera images themselves to see where the brightest pixels were occurring. Figure 8(c) shows the same X3Y3 array as Fig. 8(a), but with all “dark pixels” removed, and a uniform color applied to the remaining “bright pixels”. The most interesting part of this analysis is that the emission originating from pillar sites appears to be appreciably brighter than emission originating from film wrinkles, given that the outline of the wrinkles is not present in the filtered image. This supports the theory of strain activated defect emission, since this phenomenon cannot easily be explained by preferential nucleation during CVD growth as some have theorized [25].

A similar brightness filter was applied to a large set of arrays on a different area of the sample, shown in Fig. 9(a), that was entirely coated with hBN. This figure is a combination of BF (red) and PL (cyan) data, with only the brightest pixels coloured. Again, we found that the brightest emission was correlated to pillar sites rather than wrinkles, and furthermore, that this observation held true across different regions of the transfer. Figure 9(b)-(d) shows the effect of increasing the bright pixel floor on the emission from arrays of different hBN coated features. We chose to sweep from 3.9% to 35.3% as this range showed both emission from unintentional wrinkles as well as emission from pillar sites being gradually filtered out. Emission originating from wrinkles was generally filtered out at a lower brightness floor compared to emission originating from substrate features. Additionally, we found that emission from larger pillars (diameter ∼ 3.6 μm) was generally more uniform in brightness between individual features compared to smaller pillars (diameter ∼1.6 μm) or bullseyes. The difference in circumference and therefore potential for emitters to form might explain the difference between small and large pillars. In the case of the bullseye features, the brightness is far less consistent between emitting sites, which might be due to the comparatively low aspect ratio resulting in less uniform strain from feature to feature. For example, Fig. 3(e) shows the edge of an hBN film partially conforming to a bullseye feature in a somewhat unique way compared to the coated pillars in Fig. 3(a).

 

Fig. 9. Filtered PL images overlayed on BF images of the same region. These images were generated using post-processing similar to that described in Fig. 8. (a) Wide-field PL measurement showing pixels in the range 24% to 78% brightness before saturation. (b) Varying brightness floor applied to an array of 1.6 μm pillars. (c) Varying brightness floor applied to an array of 3.6 μm pillars. (d) Varying brightness floor applied to an array of 8 μm bullseyes. Emission corresponding to wrinkles is filtered out at a lower brightness floor compared to emission corresponding to patterned features. This observation holds true for all feature types.

Download Full Size | PDF

4. Conclusion

In summary, we have presented a polymer-assisted hBN transfer method that reliably produces defect-based emission in strained areas. This method achieved ∼80% successful transfer by area. Our wide-field PL results show not only that areas of hBN emission correspond to underlying patterned structures, but also that the brightest emission occurs at these areas rather than at unintentional wrinkles in the film. Our findings reinforce the theory that strain is one of the mechanisms responsible for emitter activation in hBN. We also presented emission spectra collected near the strained edges of hBN-coated structures and showed that the number of distinct emitters scales with feature size. Future work will involve a more complete optical characterization with the overall goal of realizing arrays of hBN-embedded optical cavities functioning as SPEs.