Growth of ultrathin Bi₂Se₃ films by molecular beam epitaxy

$Bi2Se3$ is a widely studied 3D topological insulator (TI) with linearly dispersing surface states that form a single Dirac cone at the $Γ$ point within the bulk bandgap. The surface states are spin-momentum locked, reducing back-scattering that can result in long electron lifetimes and, thus, high mobilities.^1–3 Typical TI thin films show promise in a variety of applications, including terahertz photodetectors and emitters,^4–6 spintronic devices,^7–10 gas sensing,¹¹ and quantum computing.¹² When a TI film has a thickness less than approximately 6 nm, the wavefunction of the electrons on the top surface interacts with the wavefunction of the electrons on the bottom surface, opening a gap at the Dirac point.^13–17 In addition, an ultrathin TI film may act as a 2D quantum spin Hall (QSH) insulator.^14,18,19 In this case, the material only hosts two counterpropagating spin-polarized conducting channels around the perimeter of the 2D slab. This is in contrast to a 3D TI, which hosts spin-polarized conducting channels on all surfaces of the material. A QSH insulator has quantized spin Hall conductance at zero fields and is a promising candidate as a component in a system hosting Majorana fermions, which could lead to the realization of fault-tolerant quantum computers.^20–22 

In order to explore the properties of ultrathin TI films, we must first be able to synthesize them at the wafer scale. Thin films of $Bi2Se3$ have been synthesized using molecular beam epitaxy (MBE) for many years.^23–26 These films have a hexagonal R$3¯$m crystal structure with the unit cell along the [0001] direction. In this crystal structure, bismuth and selenium are arranged in successive layers with the stacking order $Se1−Bi−Se2−Bi−Se1$ where the superscripts 1 and 2 represent two inequivalent positions of the selenium atoms. This five-layer repeated structure is known as a quintuple layer and is approximately 1 nm thick. The atoms within the QLs are covalently bonded, but the consecutive QLs are connected to each other via weak van der Waals bonds. These van der Waals bonds make it possible to grow $Bi2Se3$ thin films on a variety of substrates with relaxed constraints on lattice matching.^27–30 

However, the weak interaction between the film and substrate makes it difficult to grow coalesced ultrathin films with defined thicknesses. In general, $Bi2Se3$ films nucleate as small domains. These domains grow in size as the film thickness increases, eventually fusing together to form a coalesced film.^31–33 However, for films with a thickness of a few nanometers (equivalent to a few QLs), these domains may or may not coalesce. This makes it difficult to study the properties of ultrathin TI films at a wafer scale, as this requires coalesced films of a uniform thickness across the wafer. In this paper, we explore the growth of ultrathin (4 nm) films of $Bi2Se3$ on c-plane sapphire using molecular beam epitaxy. We show that substrate pretreatment (growth and decomposition of a few layers of $Bi2Se3$ before the actual film deposition) is critical to the growth of coalesced films. In addition, we find that a low growth temperature and a high growth rate lead to better-coalesced films, contrary to conventional epitaxy.^32,34 We explain our results using substrate wetting and adatom diffusion. Recipes to grow ultrathin $Bi2Se3$ films will allow a more detailed exploration of two-dimensional TI states.

$Bi2Se3$ thin films were grown using a Veeco GENxplor MBE system on $1×1cm2$ sapphire (0001) substrates. Our MBE system is equipped with a selenium valved cracker cell to improve the incorporation of selenium.³⁵ The Se:Bi flux ratio is kept between 80 and 100 as measured by beam equivalent pressures using a flux gauge in the substrate position. Substrates were outgassed in the load lock at $200°C$ for 12 h before being transferred to the growth chamber whereupon they were heated to $650°C$ for 5 min. The latter step ensures any impurities adsorbed on the surface have been removed. They were then cooled to the desired temperature for film deposition or surface treatment. All substrate temperatures were measured using a noncontact thermocouple located at the position of the substrate heater. To grow good-quality 4 nm $Bi2Se3$ films, we tried different growth methods, substrate temperatures, and growth rates on treated and untreated sapphire substrates as discussed in detail below. After the growth, samples were cooled to $200°C$ under a selenium flux and then removed from the growth chamber. Room-temperature Hall effect measurements were taken in the van der Pauw configuration within 15 min after the samples were removed from the MBE. Atomic force microscopy (AFM) scans and Raman spectroscopy measurements were performed on the films to assess film coalescence.

Building from established techniques for thicker $Bi2Se3$ film growth, 4 nm films were initially synthesized with a two-step growth method at a rate of 0.60 nm/min on an untreated substrate.^26,36 In the two-step method, the first few layers of the film are grown at a lower temperature and then the rest of the film is grown at a comparatively higher temperature. For sample 1, we started by codepositing bismuth and selenium for 1 min and annealing for 80  s with continual selenium flux at $325°C$⁠. This grow-anneal process was repeated until the desired seed layer thickness of 3 nm was reached. At this thickness, the streaks associated with $Bi2Se3$ start to appear on the reflective high energy electron diffraction (RHEED) screen.²⁶ Then, the substrate temperature was raised to the second set point (⁠$100°$C higher than the first growth temperature unless mentioned otherwise) under a continual selenium flux to grow the remaining 1 nm of the film by simply codepositing bismuth and selenium. Growth details for this sample—and all samples in this study—are summarized in Table I. Hall measurements on this sample indicated that the films were insulating. Figure 1 displays a $5×5$ $μ$m² AFM image of this sample where discrete small islands are observed and the typical triangular morphology for $Bi2Se3$ synthesis is not observed. The comparatively large value of surface roughness at 1.6 nm suggests a noncoalesced film. The insulating nature of the films as determined by Hall measurements further supports this conclusion.

In an attempt to realize fully coalesced films, sample 2 was grown using the same procedure as sample 1 but with an additional annealing step in a selenium atmosphere at $425°$C for 1 h after the growth was concluded. The goal of this process was to provide enough time for the bismuth and selenium atoms to diffuse across the substrate and result in a coalesced film. Contrary to our expectation, the extra annealing did not result in coalescence but instead facilitated growth of regions which are even more separated, as is evident in the AFM image of sample 2 shown in Fig. 2. Quantitatively, the surface roughness of the sample increased to 2.5 nm and the Hall measurement showed insulating behavior. Taken in aggregate, we, therefore, conclude that sample 2—even with the additional annealing step—did not coalesce.

Based on the results of the first two growths, we hypothesized that $Bi2Se3$ was not wetting the sapphire substrate, causing island-type growth and precluding film coalescence upon annealing. To overcome this issue, we decided to change the chemistry of the substrate surface through pretreatment. Pretreatment of sapphire substrates has been used to improve the growth of other van der Waals materials, including $Bi2Se3$ , $Bi2Te3$⁠, $Ga2Se2$⁠, and $WSe2$⁠.^30,37,38 Although the mechanism is not entirely clear, it is suspected that the pretreatment may passivate dangling bonds and step edges, making it easier for the film to wet the substrate.³⁹ 

For all subsequent samples, a 5 nm $Bi2Se3$ thin film was grown on the sapphire substrate by first codepositing bismuth and selenium at a single growth temperature of $325°$C with no annealing. The samples were then heated in a selenium flux to $470°$C, which is higher than the $Bi2Se3$ thermal decomposition temperature for our MBE. We kept the samples at $470°$C for 30 min to completely desorb the $Bi2Se3$ film, as was confirmed by monitoring the RHEED pattern. Specifically, at the end of this step, the RHEED pattern only showed lines associated with the sapphire substrate. After $Bi2Se3$ decomposition, the substrates were cooled to the required growth temperatures. Two sets of films were then synthesized on the pretreated substrates. First, samples A–D were grown using direct codeposition of bismuth and selenium at a given temperature with no annealing. Substrate temperatures of $300$ or $325°$C and growth rates of 0.25 or 0.6 nm/min were employed for this direct growth process. Second, samples E–I were grown using the two-step growth technique analogous to sample 1. Initial growth temperatures of $300$ and $325°$C were tried and growth rates of both 0.25 and 0.6 nm/min were employed. The specific growth rates and growth temperatures for samples A–I along with their transport properties are shown in Table I.

All samples grown on the pretreated sapphire substrate were electrically conductive, suggesting a degree of coalescence for these films that at least reached the percolation threshold. AFM and Raman images provided in Fig. 3 support this conclusion when analyzed in tandem. False color Raman images were created by assigning color scaled to the strength of the $A1g1$ mode of $Bi2Se3$ (for details, see supplementary material⁴⁰). Shown in the rightmost column of Fig. 3 and indicated by the sample letter with a prime, brighter regions correspond to stronger $Bi2Se3$ signals while darker regions are indicative of the inverse. Importantly, modes stemming from $Bi2Se3$ are present in all spectra—both bright and dark regions—consistent with a coalesced film. By comparing the AFM images and Raman maps in Fig. 3, the number and distribution of dark spots found in the Raman images are seen to roughly correlate with the distribution of columnar regions observed in the AFM images (i.e., “bright regions” in topographic images). Simply put, columnar regions seem correlated with regions of lower Raman signal. Taken in aggregate, we, therefore, conclude that $Bi2Se3$ is present over the entirety of the scanned range and not just within columnar “islands.”

Despite the common film coalescence, samples grown on pretreated substrates exhibit significantly different morphologies. Changes in morphology, in turn, provide insight into the growth mechanisms at play. We can first compare pairs of films with different growth rates by looking across the rows in Fig. 3: sample A versus B, C versus D, E versus F, and G versus H. We generally observe a lower surface roughness and larger domains for the samples grown with a higher growth rate, contrary to expectations. Next, we can compare films grown at different temperatures by looking down the columns in Fig. 3: sample A versus C, B versus D, E versus G, and F versus H. We see a lower surface roughness for the samples grown at a lower substrate temperature, again contrary to expectations. To determine the optimal substrate temperature, the first set point temperature in the two-step growth was further reduced to $275°$C. However, this did not noticeably improve the surface roughness and resulted in smaller domains. The AFM image for this sample is given in Fig. S2 in the supplementary material.⁴⁰ Finally, we can compare samples grown with the direct growth method and those grown with the two-step growth method: sample A versus E, B versus F, C versus G, and D versus H. Although the surface roughness is similar for both methods, the morphology of the films grown with the two-step method is more typical of what is observed for thicker $Bi2Se3$ films. For the films grown with the two-step method, we see the typical triangular domain morphology, which we interpret to be indicative of better-quality films.

To investigate if we could further optimize the growth by adjusting the temperature difference between the first and second steps of two-step growth, another sample with a growth rate of 0.6 nm/min and first and second-step growth temperatures of $325$ and $400°$C was grown. The AFM image of the resulting sample I is shown in Fig. 4. We can compare this sample with both sample F (same initial step temperature) and sample H (same final step temperature). Sample I has a lower RMS roughness than sample F but a higher RMS roughness than sample H. It can also be seen that in addition to small RMS differences, the ratio of taller columnarlike growth features is higher in films with higher RMS values. We, therefore, conclude that the final step temperature has more influence on the surface roughness than the initial step temperature.

As described above, we used room-temperature Hall effect measurements to understand the transport properties of the films. The sheet densities are similar for all samples A–I regardless of the growth methods and conditions with only small changes as the film morphology changes. We attribute this consistency to defects at the film/substrate interface which dominate the transport properties. These defects can be reduced by growing a buffer layer between the sapphire substrate and the $Bi2Se3$ thin films.^28,41,42 Unlike the sheet density, the mobility changes with growth method. The mobility in these ultrathin films will always be inherently limited by scattering from the top and bottom surfaces, but it can also be limited by surface roughness and scattering from grain boundaries. We find that the mobility increases with increasing growth rate, which is consistent with the reduction in roughness observed in the AFM images. The effect of the growth temperature on the mobility is smaller and not consistent across all pairs of samples. Samples B and C show a relatively high mobility despite their columnar structure. In these samples, the electrons may be traveling through the thin coalesced “background” film rather than through the tall islands.

Additionally, we note that the uncapped films like those described in Table I and Fig. 3 showed severe aging effects. To reduce this effect, an additional sample was grown using conditions identical to sample G and capped with selenium.⁴³ To do this, before taking the sample out of the growth chamber, we cooled it from $200$ to $65°$C in vacuum and then exposed it to the Se flux for 40 min, resulting in a crystalline Se capping layer as shown in Fig. S3 in the supplementary material.⁴⁰ It can be seen in Fig. 5 that the selenium capping layer resulted in much more reproducible transport measurements over time compared to the uncapped sample. The increase in sheet density for the capped sample is likely caused by a change in band bending due to the capping layer, while the increase in mobility is likely due to a decrease in oxidation and surface adsorbates.

Overall, we found that for growth on pretreated substrates, a faster growth rate and a lower growth temperature produced smoother films. We also found that the two-step growth method resulted in the usual triangular domain film morphology. These results are consistent with the aforementioned hypothesis that $Bi2Se3$ does not wet the sapphire substrate well. Deposition at a high growth rate and a low substrate temperature reduces adatom mobility. In growth of normal covalently bonded materials, this would lead to rougher surfaces. However, for $Bi2Se3$ grown by van der Waals epitaxy on sapphire, the low adatom mobility prevents the film from forming islands to minimize its contact with the substrate. When adatom mobility is high, the bismuth atoms are able to diffuse along the substrate, find an existing $Bi2Se3$ domain, diffuse up the domain sidewalls, and incorporate on top, leading to islands or columns with reduced substrate coverage. By reducing the adatom mobility, we can limit this behavior and induce the film to nucleate across the entire substrate. The two-step growth method further improves the film morphology by increasing adatom mobility once the film has nucleated on the substrate. This is, of course, all predicated on using the substrate pretreatment to improve the wettability of $Bi2Se3$ on the substrate.

We remain unsure of precisely how the substrate pretreatment is changing the surface chemistry. However, we can make an educated guess by looking at similar systems.^37,39 In previous experiments, x-ray photoemission spectroscopy (XPS) was used to investigate sapphire substrates that were pretreated with GaSe in a similar manner to our procedure. They found that after the Ga–Se film was desorbed from the substrate, both gallium and selenium peaks were still visible in XPS measurements. They hypothesized the existence of a reacted layer in which gallium and selenium were bonded to the sapphire substrate. In this reacted layer, selenium may have replaced oxygen in the substrate, perhaps making the surface less polar, less reactive, and, therefore, more easily wetted. It is also possible that the adatoms are incorporated at step edges and passivating surface dangling bonds without atomic replacement. It is likely that similar reactions are happening in our samples, but further work is needed to fully understand the surface chemistry of the substrate after pretreatment.

In summary, we grew ultrathin $Bi2Se3$ films using a variety of different growth recipes and conditions on treated and untreated sapphire (0001) substrates. We demonstrated that pretreatment of the sapphire substrate results in better substrate coverage and improvement in domain coalescence. We also demonstrated that the two-step growth method on the pretreated substrate results in the typical triangular domain morphology of $Bi2Se3$⁠. We observed an improvement of the surface roughness and the film morphology with higher growth rates and lower substrate temperatures, contrary to that usually observed for epitaxial growths of conventional materials. By capping the thin films with crystalline selenium, their surface quality can be preserved. It makes these materials reliable for the study of thickness-dependent optical or electronic properties around the critical thickness of 6 nm and can provide us with useful information for device applications.

S.N. and S.L. acknowledge funding from the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences, under Award No. DE-SC0017801. The authors acknowledge the use of the Materials Growth Facility (MGF) at the University of Delaware, which is partially supported by the National Science Foundation Major Research Instrumentation under Grant No. 1828141 and UD-CHARM, a National Science Foundation MRSEC under Award No. DMR-2011824.

The authors have no conflicts to disclose.

Saadia Nasir: Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Writing – original draft (lead); Writing – review & editing (lead). Walter J. Smith: Investigation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Thomas E. Beechem: Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Supervision (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Stephanie Law: Conceptualization (lead); Funding acquisition (lead); Investigation (supporting); Methodology (supporting); Project administration (lead); Resources (lead); Supervision (lead); Writing – original draft (supporting); Writing – review & editing (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.