Graphene as reusable substrate for bialkali photocathodes

Since their discovery in the 1950s, bialkali antimonide photocathodes have been widely utilized as photosensitive materials in radiation detectors and camera tube technologies.^1,2 Cesium potassium antimonide (CsK₂Sb) is one of the highest performing bialkali antimonide photocathodes,^3–7 achieving a quantum efficiency (QE) that exceeds 10%^3,4 by a green laser (532 nm). This is supported by the demonstration of a record-high beam current of 60 mA in a DC injector with 30 h 1/e lifetime.⁵ The established knowledge is that CsK₂Sb photocathode performances are greatly affected by the substrate material, including its crystallinity,^1,3 surface state (contamination, roughness, and surface orientation),^1,8–12 and dopant types.¹³ As such, substrates are generally not reusable because the surfaces suitable for photocathode deposition are lost once the photocathode is deposited. If there were a reusable substrate, we may be able to reduce the number of times that one needs to break the vacuum of accelerator systems and electron microscopes for time-consuming photocathode replacements. Use of vacuum suitcases to transport photocathodes from elsewhere could prevent breaking the vacuum of systems for replacements, however, achieving a high yield in the process is generally not trivial.

In this study, we deposited CsK₂Sb photocathodes onto graphene, silicon (Si), and molybdenum (Mo) substrates to evaluate the feasibility of graphene as a reusable substrate. Specifically, we deposited photocathodes on the above mentioned substrates and measured their QE values. We then reused the same substrates five times by thermally cleaning them and redepositing the photocathodes. We followed up QE measurements for graphene, Si, and Mo with surface analysis characterizations using x-ray photoelectron spectroscopy (XPS) and angle-resolved photoelectron spectroscopy (ARPES). Density functional theory (DFT) was also performed to gain insights into a possible origin of different effects by thermal cleaning of graphene and Si as reusable substrates for high-performing photocathodes.

The detailed experimental setup for CsK₂Sb photocathode deposition is described in Ref. 3. Briefly, the photocathodes were fabricated by sequential thermal evaporation of potassium (K), cesium (Cs), and antimonide (Sb) sources in a vacuum chamber with a base pressure of 1.0 × 10⁻⁸Pa. Antimonide was deposited first for ∼10 nm, followed by ∼60 nm of potassium, and finally ∼120 nm of cesium. The exact thicknesses of potassium and cesium were determined by the saturated QE values monitored during the growth using a 532 nm green laser. Silicon (100) and Mo polycrystalline substrates were half coated with graphene so that we could evaluate the effect of graphene coating on simultaneously deposited photocathodes (inset of Fig. 1). Graphene consisting of mainly monolayer regions with some multiple-layer regions¹⁴ was synthesized via chemical vapor deposition (CVD)¹⁵ on copper (Cu) substrates and transferred onto Si and Mo using an established polymer supported method that involves distilled water exposure.¹⁶ The half graphene-coated Si and Mo substrates were intentionally not cleaned with hydrogen fluoride (HF) as a means to prevent degradation of the graphene.¹⁷ We evaporated the CsK₂Sb on new and reused substrates, and we measured their QE values. Specific experimental procedures are as follows. The substrates were thermally cleaned at 500–600 °C for an hour prior to photocathode deposition. CsK₂Sb was then evaporated onto the new substrates of half-coated graphene on Si or Mo, and their QE values were measured in a vacuum using the 532 nm green laser. The laser spot size was 0.5 mm², and the photocathodes were biased at −100–−200 V to ensure collection of photoelectrons. The laser power was adjusted to 0.6 mW, and the typical photo-current level was <20 μA. The resolution of photo-current in our measurement system was 10 nA. The photocathodes were then exposed to air and returned to the vacuum to repeat the above procedures for the re-depositions. The entire re-deposition procedure was repeated five times on the same substrates of half graphene-coated Si and Mo as well as on several substrates for each type to check the reproducibility.

Figure 1 shows the QE values of photocathodes deposited on new and reused graphene, Si, and Mo substrates. Silicon was chosen as an atomically flat reference substrate for follow up surface characterization, and Mo was chosen to demonstrate a case for practical substrates, such as those in accelerator facilities. Error bars were calculated as the standard deviation for five specimens or depositions for each data point of “New” and “Used,” respectively. The QE values for the photocathodes on new graphene/Si and graphene/Mo substrates were 6.8 and 7.0%, respectively, at 532 nm with a relatively small error bar. The QE remained largely unchanged, even for the photocathodes deposited on reused substrates (6.2 and 6.1% for used graphene/Si and graphene/Mo substrates, respectively). Note that the error bar also remained relatively narrow. The Si and Mo substrates, on the other hand, showed a drastic decrease in QE values for the photocathodes deposited on reused substrates (from 5.2 down to 1.2% and from 4.6 down to 0.9%, respectively), with a significant increase in the error bar. This indicates not only that QE values decrease for photocathodes on reused Si and Mo substrates, but shows they also lose their consistency in materials' quality. A possible origin of higher QE values for photocathodes on the new graphene substrates compared to those of Si and Mo is reported in our previous study.⁸ The QE values of photocathodes on new Si (100) and Mo substrates were similar to our previous reports (without HF cleaning prior to the growth for Si).^3,13 These results clearly demonstrate that graphene is a good candidate for reusable substrates for CsK₂Sb photocathodes, whereas Si and Mo are not.

To gain insights into the origin of QE degradation difference between the photocathodes on used graphene, Si, and Mo substrates, we applied surface analysis techniques of XPS and ARPES. Measurements were performed at the UVSOR beamline (BL6U) at the Institute for Molecular Science (IMS), Japan.^18,19 Specifically, we performed XPS on substrates that are thermally cleaned at 400 °C and 500 °C after the five repeated photocathode depositions to analyze the elemental compositions on the surfaces. The sensitivity resolution of our XPS is better than 0.1 at. %. We also performed ARPES on the same graphene substrates after thermal cleaning at 500 °C to investigate its material integrity.

Figure 2 shows the XPS results for the graphene, Si, and Mo substrates after thermal cleaning at 400 °C and 500 °C (hυ = 350 eV). The information depth of the analyzed region is about 0.5–1 nm, which is determined by the electron inelastic mean-free path length and the takeoff emission angle of 90°. The spectra were fit by the Doniach–Sunjic function after subtracting a Shirley background, as indicated in the black lines below the spectra. We used the CasaXPS software for the analysis after converting the kinetic energy of data to binding energy. For the graphene substrate after thermal cleaning at 400 °C, we only observed small peaks of Cs 4d. Peaks for Sb 4p at a binding energy of ∼96 eV as well as for K 2p at ∼295 and ∼298 eV were not observed. The peaks at binding energies of ∼101, ∼97.5, and ∼97 eV are for oxidized Si 2p and non-oxidized Si 2p 1/2 and 3/2 of the Si substrate underneath, respectively.²⁰ In addition to a clear 1–2 eV energy difference between the peaks for Sb 4p and non-oxidized Si 2p 1/2 and 3/2, they are also distinguishable by a lack of spin split for the Sb 4p peak. The energy difference between Si 2p 1/2 and 3/2 is 0.6 eV, which is consistent with our peaks, while there is only a negligible binding energy difference between Sb 4p 1/2 and 3/2.²¹ On the other hand, the Si and Mo substrates cleaned at the same temperature, had additional peaks of slight K 2p for both and Sb 4p for the Mo. These results demonstrate that graphene is inert and thus suitable for a reusable substrate. Striking results were obtained for the graphene substrate that was thermally cleaned at 500 °C. The only remaining residue of Cs was also removed to below the detection limit of 0.1%. The fact that all of the residues were removed has an important practical implication, such as for accelerator applications, as any particle contaminations outside the photocathode area could be detrimental in high-current operations. The decrease in non-oxidized Si 2p 1/2 and 3/2 peak intensities compared to those of after thermal cleaning at 400 °C could be due to the non-uniform thickness of native oxide for the Si substrate underneath. The thickness of our native oxide is expected to be 1–2 nm, which is close to the information depth of our measurements. Our measurements after thermal cleaning at 500 °C could have been at thicker SiO₂ regions as compared to the 400 °C counterpart, which we observed as no or decreased intensities of Si 2p 1/2 and 3/2 peaks. This could also explain a variation in the relative peak intensities of non-oxidized Si 2p 1/2 and 3/2 to oxidized Si 2p between thermal cleaning at 400 °C and 500 °C in Fig. 2(e). Cesium and K remained for the Si substrate and all of Cs, K, and Sb remained for the Mo substrate after thermal cleaning at 500 °C. Table I summarizes the at. % of Si, C, Sb, K, Cs, and Mo on the graphene/Si, Si, and Mo substrates after thermal cleaning at 400 °C and 500 °C. The intent of content values is to provide relative values rather than absolute values as we have excluded the oxygen content. The oxygen content is not included due to smaller photon energy used in our measurements compared to the binding energy for O 1s (532 eV). However, we performed separate measurements using a higher photon energy of 580 eV and confirmed that there is no change in its content within our experimental conditions. This is most likely due to the majority of the oxygen in our specimen being native oxide on Si substrates, which does not dissociate at temperatures below ∼800 °C. The results support our QE observations in that the graphene is suitable for a reusable substrate for the photocathodes, but the Si is not.

It is worth noting that rather broad C 1s spectra of the Si and Mo substrates [Figs. 2(f) and 2(i), respectively] are attributed to hydrocarbon contamination and the sharper peak for the graphene/Si substrate [Fig. 2(c)] is attributed to sp² bonds of graphene. These indicate that removal of hydrocarbon contamination by thermal annealing is also more effective on the graphene compared to the Si and Mo substrates in addition to the case of Sb, Cs, and K. The shoulder component of the C 1s spectrum on the higher binding energy side obtained for the graphene/Si substrate after thermal cleaning at 400 °C [Fig. 2(c)] is attributed to hydrocarbon contamination, although the amount is clearly smaller compared to those on the Si and Mo substrates after thermal cleaning at the same temperature.

We then performed ARPES on a graphene substrate that was thermally cleaned at 500 °C after five repeated times of CsK₂Sb photocathode deposition. The purpose was to ensure that graphene remained intact even after those rigorous processes of repeated depositions and thermal cleanings. The result indicated that graphene indeed remained intact. Figure 3 shows the ARPES results of the graphene substrate with ±10° and ±15° of the capture angle for vertical and horizontal directions, respectively. The photon energy was 101 eV, and a bias voltage V_b of −380 V was applied to the specimen. 10° from the surface normal direction corresponds to the in-plane momentum $k∥=0.5123Ekinsin10°$⁠, which is 0.866 Å⁻¹ for a kinetic energy E_kin of 94.8 eV. The acceptance angle is extended up to $k∥=0.5123Ekin−eVbsin10°$⁠, which corresponds to 1.94 Å⁻¹. The yellow-dotted hexagon in Fig. 3(a) indicates the expected Brillouin zone of graphene in the photoelectron angular distribution. A sixfold symmetric signature pattern of the π band around the K symmetric point in graphene was observed and is indicated by the red-dashed curves (E_kin = 94.8 eV). Broadening of the spots is most likely due to the presence of multi-layer regions in our graphene. Figure 3(b) shows the integrated photoelectron spectra over the entire Brillouin zone. The shoulder structure at the nominal kinetic energy E_k – eV_b of 473.5 eV corresponds to the density of states at the M points. Figure 3(c) shows the band dispersion along the blue- and red-dotted arrow directions in Fig. 3(a). The expected band dispersions along ΓΜ and ΓΚ directions are indicated by the red-dashed lines. A clear parabolic band dispersion of graphene is observed. These results confirm that our graphene remains intact,^18,19 even after the photocathode depositions and thermal cleanings at 500 °C, which supports our hypothesis of a dangling-bond-free inert surface of graphene as an origin of its reusability as a substrate for the photocathodes. The result also excludes the possibility of a “contaminated” graphene top layer being removed after each thermal cleaning.

We performed density functional theory (DFT) calculations to support our hypothesis. Specifically, we calculated binding strengths of Cs, K, and Sb adatoms to graphene, Si (100), and oxidized Si (100) surfaces to compare if they are smaller for graphene. The oxidized Si represents a simplified case of native oxide SiO₂ on Si. Calculation for the Mo was not performed as our specimens were polycrystalline and it is generally challenging for DFT to predict such electronic properties. The result indicated that binding strengths of the adatoms to graphene are smaller compared to the Si and oxidized Si counterparts, supporting our hypothesis. Our calculations were performed with the use of DFT and the projector augmented-wave (PAW) method²² as implemented in the Vienna ab initio Simulation Package (VASP).²³ The generalized gradient approximation (GGA) of the Perdew–Burke–Ernzerhof (PBE)²⁴ functional was used to represent the exchange-correlation interaction. The substrates were represented with slab models with a vacuum gap in the direction normal to the surface to eliminate the interactions between the replicas due to the periodic boundary conditions. The binding strengths of adatoms on graphene, Si (100), and oxidized Si (100) are defined as $Eb=Esubstrate+adatom−(Esubstrate+Eadatom)$⁠, where $Esubstrate+adatom$⁠, $Esubstrate$⁠, and $Eadatom$ are the total energies of the complex system, the substrate, and the adatom, respectively. A negative binding energy indicates the adatom can adsorb on the substrate. The reconstructed Si (100) 2 × 1 structure and its oxidized surface were used to model the Si and oxidized Si substrates as we used Si (100) with native oxide for our experiments. To get the atomic structure of the oxidized silicon substrate, we performed the ab initio molecular dynamics (AIMD) simulation. This approach has been previously used to study the oxidation of silicon.²⁵ Our AIMD simulation was performed with a time step of 2 fs at 500 K, with 10 oxygen atoms on an Si (100) surface. The structure at 2 ps was used to mimic the native oxide layer on silicon. For graphene, the most stable binding site for Cs, K, and Sb is at the hollow site, as shown in Fig. 4(a). On the reconstructed Si (100) 2 × 1 substrate, two different binding sites were considered, as shown in Fig. 4(b). The site for oxidized Si (100) is shown in Fig. 4(c). The calculated binding energies are summarized in Table II. Cesium, K, and Sb adatoms all had much stronger binding strengths on Si (100) and oxidized Si (100) substrates than on graphene. The strengths were 2.3–2.7 times higher for Cs and 2.4–2.7 times higher for K, and for Sb, it was as high as ∼16 times for the Si and 3.4 times for the oxidized Si (100). These results indicate that the graphene surface is inert to Cs, K, and Sb adatoms compared to Si (100) and oxidized Si (100) surfaces and are consistent with our hypothesis of dangling-bond-free inert surface of graphene as an origin of its reusability for photocathode growths. The results are also in good qualitative agreement with the amount of residues observed in our experiments for both graphene and oxidized Si (100). Note a much lower binding energy of oxidized Si (100) between Sb compared to that of Cs and K, which could explain why we did not observe the Sb residue in our experiments [Fig. 2(e), Table I]. We expect the same conclusion to be drawn whether the photocathodes are deposited sequentially as in our case or via co-deposition.

To conclude, we investigated the reusability of graphene, Si, and Mo substrates for CsK₂Sb photocathode depositions. Our findings indicate graphene is a good candidate as a reusable substrate. Specifically, the repeatedly deposited CsK₂Sb photocathodes after thermal cleanings demonstrated reproducible QE values with minimal degradation, whereas those on Si and Mo did not. We quantitatively evaluated the residues of photocathodes by XPS and found that Sb, K, and Cs were removed on the graphene substrate by thermal cleaning at 500 °C. Silicon and Mo, on the other hand, had all or the majority of Cs, K, and Sb residues that were not removed even after thermal cleaning at 500 °C. These results demonstrate that the graphene substrates can be reused via a simple thermal cleaning. The difference in the residue amount after the thermal cleaning can be attributed to inertness of graphene compared to Si and Mo due to a lack of dangling bonds on the surface, as we confirmed by DFT calculations for graphene and Si cases. Our ARPES indicated that graphene possesses high crystallinity even after repeated deposition and thermal cleaning processes, thus remaining dangling-bond free. Our results provide a foundation for graphene-based reusable substrates for high QE semiconductor photocathodes.

The work was financially supported by the High Energy Accelerator Research Organization (KEK), Japan for the Japanese team and the U.S. Department of Energy (DOE) Office of Science for the U.S. team under the U.S.-Japan Science and Technology Cooperation Program in High Energy Physics.