Copper migration and surface oxidation of CuxBi2Se3 in ambient pressure environments

Chemical modifications such as intercalation can be used to modify surface properties or to further functionalize the surface states of topological insulators (TIs). Using ambient pressure x-ray photoelectron spectroscopy, we report copper migration in CuxBi2Se3 , which occurs on a timescale of hours to days after initial surface cleaving. The increase in near-surface copper proceeds along with the oxidation of the sample surface and large changes in the selenium content. These complex changes are further modeled with core-level spectroscopy simulations, which suggest a composition gradient near the surface which develops with oxygen exposure. Our results shed light on a new phenomenon that must be considered for intercalated TIs—and intercalated materials in general—that surface chemical composition can change when specimens are exposed to ambient conditions.


Introduction
Topological insulators (TIs) have attracted significant interest from the quantum materials community for their surface electronic structures, notably surface Dirac fermions with spin-momentum locking that are robust to disorder and defects [1][2][3][4]. Chemical intercalation of TIs (i.e. the insertion of atoms or molecules between quintuple layers) can further modify the material properties [5][6][7][8][9][10][11][12] or create additional electronic phenomena, such as superconductivity with Cu intercalation that has a maximum superconducting critical temperature of T c,max ≈ 3.8 K [10,[13][14][15][16][17][18][19]. The Cu-intercalated TI Bi 2 Se 3 (Cu x Bi 2 Se 3 ) is a known superconductor, with T c,max and the superconducting shielding fraction strongly depending on the Cu content and preparation method [14][15][16]20]. Controlling the Cu content and its variation during and after synthesis is thus important for realizing ideal superconducting properties in Cu x Bi 2 Se 3 .
Intercalated materials may be exposed to air during sample preparation or real world usage, such as in the operation of devices, which may affect composition. For instance, the intercalate guest can diffuse within battery cells left at a fixed potential, and it is known that intercalants can diffuse within and between layers in various compounds [12,[21][22][23]. The effect of ambient conditions on intercalant chemistry is not fully understood due to experimental difficulties, despite its practical importance.
In this work, we use Cu x Bi 2 Se 3 as a representative material to establish the effects of controlled ambient environments on evolving chemistry in the near-surface region in intercalated TIs. Using x-ray photoelectron spectroscopy (XPS), the near-surface Cu composition was found to increase under both controlled O 2 dosing and ambient air exposure, coincident with the formation and growth of an oxide layer that is strongest in full atmosphere. Lastly, core-electron spectroscopy simulations show that our XPS observations are consistent with a sample developing vertical gradient distributions of Cu and Se upon exposure to a controlled O 2 environment.

Materials and methods
Synchrotron ambient pressure X-ray photoelectron spectroscopy (AP-XPS) experiments were performed at the Advanced Light Source Beamline 9.3.2 in a photon energy range of E ph = 230-900 eV, with an X-ray spot size of d ∼ 1 mm 2 . The core level spectra were collected with a Scienta R4000 HiPP electron analyzer with differential pumping, allowing the sample to remain at near-ambient pressures during data acquisition [24]. The analysis environment was initially at high vacuum (∼1 × 10 −7 mbar), and then with O 2 at near-ambient pressure (0.133 mbar). The measured partial pressure of O 2 is equivalent to the total ambient pressure in the AP-XPS experiment. Longer timescale experiments were performed using a Kratos AXIS Supra + with a monochromated, unpolarized E ph = 1486.6 eV Al-Kα source. Samples were cleaved in situ with the top post method in the AP-XPS experiment, and cleaved ex situ with Scotch tape in the Kratos XPS experiment, keeping air exposure under several minutes prior to the initial measurements. Angle-resolved photoemission spectroscopy (ARPES) spectra were collected at the Advanced Light Source Beamline 4.0.3 (MERLIN) with E ph = 39.2 eV. Core level peaks were identified using binding energy reference values [25]. Bi 2 Se 3 samples were grown using the Bridgman method, and were then intercalated with Cu to form Cu x Bi 2 Se 3 crystals using a solution-based process [26]. Due to the variation of Cu and Se content highlighted in our study, our convention is to use the nominal bulk stoichiometries after synthesis (Cu 0.15 Bi 2 Se 3 and Cu 0.3 Bi 2 Se 3 ) to distinguish between samples in our analysis. No indications of superconductivity were found in these samples, which depends on the details of the synthesis and the intercalation [20,27] (see supplemental material for further details).  Figure 1(a) shows a survey spectrum of the core levels accessible with E ph = 650 eV in the AP-XPS experiment. The survey confirms the quality of the in situ cleaved sample, showing peaks for Cu, Bi, and Se. Figure 1(b) shows an ARPES spectrum from an in situ cleaved sample, clearly showing the topological surface state and Dirac point at E B ∼ 0.3 eV, demonstrating the maintenance of crystallinity and topological electronic features after Cu intercalation. Figure 1(c) shows the three core levels present in the shallow binding energy region in figure 1(a): Cu 3p, Se 3d, and Bi 5d, none of which show initial oxidation or hydroxylation. This set of three shallow core levels is sufficient to determine the chemical composition of Cu 0.15 Bi 2 Se 3 and Cu 0.3 Bi 2 Se 3 throughout the oxidation process; the relative peak intensities are proportional to the elemental composition after correcting for the relative sensitivity factor (RSF) for each photon energy [28,29]. To quantify the Cu content, we monitor the binding energy region around E B ∼ 76 eV where the Cu 3p doublet is present. All Cu 3p data are measured above the loss feature-containing background (figures 1(d) and (e)).

Results
To accurately quantify the Cu content in figure 1, which is present on top of a relatively large background, we utilize a background correction procedure incorporating the electron energy loss features in the Cu 0.3 Bi 2 Se 3 XPS spectrum. The loss features near E B ∼ 72 eV (green and cyan curves in figures 1(d) and (e)) originate from bulk plasmon resonances in Bi 2 Se 3 [30]. Since this background is also present in a Bi 2 Se 3 reference sample from the same batch (figure 1(d)), we use the Bi 2 Se 3 spectrum from a newly cleaved sample as a reference for fitting to the plasmon losses in the Cu 0.3 Bi 2 Se 3 spectrum (figure 1(e)). After fitting Voigt peaks to the Se 3d peaks and the loss features in the reference spectrum, the result is a background containing the loss features in the vicinity of the Cu 3p doublet (dashed curve). Fitting this background to the Cu 0.3 Bi 2 Se 3 spectrum allows us to isolate the Cu 3p signal intensity, shown as a separate Voigt doublet (brown curve). The areas of the fitted Cu, Se, and Bi peaks then correspond to the total XPS intensity for these elements. Additional information on the Cu quantification, RSF assumptions, and plasmon loss feature correction can be found in the supplemental material. Figure 2 shows the main results of the AP-XPS experiment, which track the evolution of the Cu and Se composition from an initial condition with no O 2 dosingand later times in 0.133 mbar O 2 to simulate ambient atmosphere. The chemical compositions are expressed in terms of the atomic ratios of Cu/Bi and Se/Bi for each time and photon energy, with the Cu fraction x = 2 × Cu/Bi. To determine the atomic ratios, the XPS intensities determined from the peak fitting were weighed by the RSF for each element and photon energy (see supplemental material). The atomic ratios are normalized to Bi since Bi changes relatively less compared to Cu and Se in our experiments.
Over the course of the four day experiment, the Cu fraction x increases from x = 0.13 to x = 0.40 after dosing O 2 for the E ph = 900 eV data, with smaller changes seen at lower, more surface sensitive E ph (figure 2(a)). The growth of the Cu/Bi ratio is consistent with Cu migrating vertically to the probed-surface region of the sample, since the only additional Cu present is deeper within the sample, with no external Cu  deposition. This is best seen when plotted versus the inelastic mean free path (IMFP) for each photon energy in figure 2(b), which provides a length scale for the observed Cu migration. The measured depth (3 × IMFP) at each photon energy includes depths below one IMFP, which still contribute to the XPS measurement. A similar plot of the Se/Bi ratio is shown in figure 2(c), showing a steady decrease in Se/Bi at all E ph after dosing O 2 , decreasing from the ideal Se/Bi = 3/2 ratio of Bi 2 Se 3 . There are large differences in the initial and final Se/Bi ratios for the different E ph , implying a large variation in the Se distribution, with the bulk being rich in Se and the surface being deficient in Se (figure 2(d)). Between one and four days in the AP-XPS chamber, the changes to Cu/Bi and Se/Bi are less pronounced than those that happen immediately after dosing O 2 . Similar trends have also been observed in another AP-XPS experiment in a mixed O 2 /H 2 O environment (see supplemental material). Figure 3 shows the initial and final Bi 5d peaks for all E ph before and after dosing O 2 , quantifying the growth of bismuth oxide. The initial state peaks are well separated, showing little to no signs of Bi oxidation ( figure 3(a)). After four days in the AP-XPS chamber, chemically-shifted peaks are present for  E ph = 650 eV, 370 eV, and 230 eV ( figure 3(b), dashed colored lines). These changes to the Bi levels directly show the formation of Bi 2 O 3 at the surface, which is most dominant at lower E ph . The noise in the initial E ph = 230 eV spectrum (blue, left) is due to the low photon flux at this particular E ph , which was improved for the final state (blue, right) by extending the measurement time. The large width of the oxide components in the Bi 5d spectra suggest a convolution of several electrostatically distinct (i.e. band bent) or chemically distinctive states, which are most pronounced in the spectrum measured with highest surface sensitivity (E ph = 230 eV). Figure 3(c) quantifies the growth of the Bi 2 O 3 peaks relative to the total Bi peak intensity, showing that oxidation starts after dosing O 2 , with a larger oxide fraction seen with the more surface-sensitive E ph = 370 eV. When compared to the E ph = 650 eV data, the overall detected oxide is less and with a more gradual onset due to the lower surface sensitivity of E ph = 650 eV. Interestingly, most of the oxidation process occurs within the first day and only little change is observed thereafter, as is the case with the Cu and Se changes.
To gain insight into changes in the Cu 0.15 Bi 2 Se 3 composition over longer timescales (t > 20 h), we continue by performing a different set of lab-based XPS measurements on an exsitu cleaved Cu 0.15 Bi 2 Se 3 sample after air exposure in full atmosphere, with increasing periods of air exposure. The loss feature-corrected spectra for Cu 3p are shown in figure 4(a), showing large changes in the Cu/Bi atomic ratio after air exposure for multiple days ( figure 4(b)). In figure 4(c), we again see the oxidation of the surface, with the Bi oxide strongly growing over several weeks. The evolution of Bi 2 O 3 in figure 4(d) shows an initial jump after exposure to air for 0.9 days, and continues to grow afterward. Oxidation of Se is also seen (see supplemental material), and we note that this sample had a smaller initial Cu concentration than the sample measured with AP-XPS.

Discussion
Our interpretation of the present data is guided by prior studies of pristine Bi 2 Se 3 exposed to air and controlled ambient environments. The effects of ambient conditions on regular Bi 2 Se 3 have been studied in the following ways with the following conclusions. Exposure to ambient environments has been shown to alter measured ARPES spectra and the surface composition in Bi 2 Se 3 [31][32][33][34]. The topological surfaces states can be modified by forming 2D quantum well states in ambient conditions [32], and can present band-bending and controlled charge-doping after dosing H 2 O [31] and UV irradiation [35]. In most cases, the robust topological surface states are still present in Bi 2 Se 3 despite air exposure and oxidation [32,36]. The thinness of the oxide layer and the robustness of the topological surface states are common themes in Bi 2 Se 3 , including the persistence of surface states in intercalated samples [23,37], with some works not reporting any surface reactivity [38,39]. Still, the surface chemistry in Bi 2 Se 3 remains an open question, particularly after intercalation.
To begin the discussion of our results, we first turn to the observations seen in the AP-XPS experiment. The main result in figure 2 shows that the increase in the Cu 3p peak intensity over the course of the experiment is coincident with the introduction of O 2 gas, showing that Cu migrates to the surface region during measurement. This behavior has not been previously reported or quantified in a TI with XPS, although some indications of Cu near the surface have been reported with scanning tunneling microscopy imaging after cleaving [14].
The probing depth in XPS is mainly limited by the IMFP of escaped electrons at each E ph , so measurement with several different E ph to vary IMFP allows one to obtain a depth profile of the elements in the sample. In the E ph = 230-900 eV photon energy range in Cu 0.3 Bi 2 Se 3 , the IMFP of Cu 3p photoelectrons increases from 7.6 to 18.9 Å, calculated in QUASES using the TPP2M algorithm [40].
Measured XPS intensities generally follow an exponential form for attenuation [41]: with a total emitted photoelectron intensity I 0 , vertical depth z, atomic fraction X(z) at each depth z, the IMFP λ, and the photoelectron collection angle θ from the surface normal. From equation (1), the measured photoelectron intensity is I ≈ 0.95I 0 within three IMFPs below the surface, providing an upper limit for the probing depth. Figure 2(b) shows the initial and final distributions of Cu/Bi as a function of IMFP and the measured depth (3λ) in the AP-XPS experiment (black and gray solid lines), showing the growth of Cu from deeper within the sample. The greatest Cu increase is at an IMFP of 18.9 Å, which corresponds to Cu migrating into the top 6 nm surface region of the sample. We note that the observed Cu migration is very small when measured with E ph = 230 eV, which indicates that the topmost layer of the sample is dominated by the surface bismuth oxide. The beamline photon flux is also considerably lower at E ph = 230 eV, which can contribute to measurement error.
Another notable observation is the decrease in Se content relative to Bi in figure 2(c) over the course of the AP-XPS experiment. Initially after cleaving (that is, prior to the first data points in figure 2(c)), the Se 3d intensity is already reduced, which continues to drop over the course of the experiment from the ideal Se/Bi = 3/2. The initial distribution of Se when expressed versus IMFP in figure 2(d) shows that the uppermost 2 nm (two quintuple layers) is deficient in Se, while the uppermost 6 nm (six quintuple layers) is richer in Se and representative of bulk stoichiometric Bi 2 Se 3 . After four days in O 2 , Se/Bi decreases at all depths.
The observation of a long Se gradient with minimal Se at the surface is consistent with the well known volatility of Se in Bi 2 Se 3 . Bi 2 Se 3 generally has selenium vacancies which make samples naturally n-type without further chemical compensation [31,32,[42][43][44][45][46][47][48]. However, our results indicate that these selenium vacancies may be more concentrated near the surface, and the deeper bulk is closer to nominal stoichiometry. The pure Bi 2 Se 3 reference sample also shows a decrease in Se after dosing O 2 , and with the decrease more significant at the surface. See the supplemental material for a comparison to the Se/Bi ratio in Bi 2 Se 3 . The Fermi level E F in the ARPES spectrum in figure 1(b) intersects the bulk conduction band, confirming that this specimen is n-type at the surface region, with an IMFP of 4.9 Å for E ph = 39.2 eV calculated in QUASES. Cu doping is also known to shift the chemical potential further into the bulk conduction band [46].
A prior XPS/atomic force microscopy study has observed small Bi islands that appear within 1 h after cleaving [33], which is also consistent with decreasing Se/Bi in figure 2(c), suggesting that some Bi migration could also be occurring along with Cu. However, the larger increases in Cu/Bi we see suggest that any Bi migration would be very small, below the sensitivity of a standard XPS instrument [33] and indistinguishable from Bi oxide at the surface.
While Cu migration can be observed on shorter timescales, the process continues and is more easily seen at the longer timescales (t > 25 h) in figure 4(a). On these timescales the oxidation of the near-surface Bi is evident with new oxide peaks, consistent with prior work [33,44] and the oxide peaks we see during the AP-XPS experiment in figure 3. The oxidation is stronger and continues for longer in full atmosphere (P = 1013.25 mbar) than in the AP-XPS experiment, even when considering the deeper probing depth of the E ph = 1486.6 eV Al-Kα source (IMFP = 28.9 Å). The oxidation is accompanied by a steady increase in Cu 3p over several days ( figure 4(b)), greater than what was observed in AP-XPS. The link between oxidation and Cu migration is clear when looking at the trends in figures 2-4: both Cu migration and Bi 2 O 3 formation start right after dosing O 2 , and when oxide growth slows between one and four days, Cu growth also slows, changing only slightly. This suggests that surface oxides establish the conditions needed for Cu to diffuse towards the surface. Our observations in figure 4 suggest that surface oxidation and Cu migration continues slowly for months in air, until the oxidation becomes self-limiting.
There are several microscopic mechanisms that can promote Cu migration, such as a surface work function mismatch between Bi 2 O 3 and Cu x Bi 2 Se 3 that can drive the Cu to the surface with a built-in E-field. The work function difference between Bi 2 O 3 and Bi 2 Se 3 is estimated to be ∼1.6 eV [49,50], which would create a sufficiently large E-field near the surface. Cu is able to occupy five different sites in the van der Waals gap and in interstitial vacancies [45,46], and some migration could be driven by Cu diffusion among these sites. Occupation of surface Se vacancies by Cu could also influence the Cu migration to the surface, which is possible due to the amphoteric character of Cu impurities in Bi 2 Se 3 [51]. It is likely that Cu migration can happen for different intercalation and synthesis methods, which has been previously observed in electrochemically intercalated Cu x Bi 2 Se 3 [12], but may vary depending on the synthesis method.
To connect the proposed changes in chemical composition to the measured XPS intensities, we model our experimental results with core-electron spectroscopy simulations using the National Institute of Standards and Technology Simulation of Electron Spectra for Surface Analysis (SESSA) software/database [52,53]. SESSA can accurately simulate XPS spectra and peak intensities for different experimental conditions, geometries, and sample compositions using database reference values. As strongly suggested by our experimental observations, Cu and Se form a compositional gradient in the near-surface region of the material. A gradient structure consisting of several discretized, homogeneous layers with varying Cu, Bi, and Se compositions models a sample with Cu and Se gradients, shown in figures 5(a) and (b). The initial Cu distribution is assumed to be constant. The gradients have a X(Z) = Ae −Z/L + B falloff when approaching the surface, with fitting parameters A, L, and B chosen to match the boundary conditions observed in experiment. Due to the oxidation present after O 2 dosing, the final simulated structure is capped with a thin 2 Å overlayer of Bi 2 O 3 . Modeling the topmost 10 Å of Cu x Bi 2 Se y is motivated by the fact that the topmost 10 Å contributes predominately to the measured XPS intensities, as it is on the order of one IMFP. Below the topmost 10 Å, we model the Cu and Se distributions as constant, equal to the maximum of the gradient. Additionally, 10 Å is the approximate thickness of one quintuple layer in Bi 2 Se 3 , and the van der Waals gap can temporarily serve as a barrier to deeper oxidation of the material [33]. Thus, it is likely that the oxidation is limited to the uppermost quintuple layer in the AP-XPS experiment, with lesser contributions below.
The red points in figures 5(c) and (d) show the SESSA-calculated peak intensities for the initial and final structures, expressed in terms of the Cu/Bi and Se/Bi intensity ratios versus IMFP and the measurement depth. The SESSA-calculated intensity ratios are compared to the experimental XPS intensity ratios to confirm agreement between the model and the experimental data. In the initial state, a structure with a single homogeneous Cu composition with no Cu gradient matches closely with the observed constant dependence versus IMFP in figure 2(b). Initially, no Cu has migrated into the surface regions and no oxide has formed yet. In the final state after oxidation, the dual Cu, Se gradient structure agrees best with our measured XPS intensities in figures 5(c) and (d), capturing both the increase in Cu and decrease in Se near the surface. We note that there is greater error when comparing the gradient to experiment at the smallest depth, due to the low photon flux at E ph = 230 eV.
There are other factors that are not present in this model which can also affect the evolution of Cu x Bi 2 Se 3 surface chemistry. These include imperfect cleaves that can form step edge sites for oxidation [34], lingering Cu remaining at the surface after the cleave, different Bi, Se surface terminations [42], as well as nanosheet morphology [44,54]. Other Cu, Se compositional distributions are possible and consistent with our experimental observations, such as ones with discontinuous step edges or sigmoidal distributions. Because of the exponential photoelectron falloff at greater depths, compositional changes deeper in the sample are more difficult to detect. Thus there may also be less significant compositional changes occurring below the topmost 10 Å, but still within the overall measurement depth, which our model does not include.

Conclusion
In summary, we have observed an increase in the surface Cu content in the intercalated TI Cu x Bi 2 Se 3 with AP-XPS measurements. Our results show that Cu migrates to the surface and is enhanced by the appearance and growth of a bismuth surface oxide over several days, with the most pronounced changes seen in full atmosphere. Modeling Cu migration concomitant with Se depletion and oxidation matches our depth-selective XPS observations for a wide range of photon energies.
These findings show that oxidation can be used as an approach for driving chemical species towards the surface of layered intercalated materials, and add additional chemical complexity that must be considered at TI surfaces exposed to ambient conditions. Chemically tailoring the surfaces of topological materials will be needed for realizing real-world environmental applications in chemical sensing, catalysis, and electronics. Most intriguingly, the proximity of the topological surface state to the observed chemical changes in Cu x Bi 2 Se 3 points to further study of the effect of intercalants on the surface states of TIs, particularly on timescales that allow environmental changes to influence them.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.