Photoemission characterization of N-polar III-nitride photocathodes as candidate bright electron beam sources for accelerator applications

The search for robust photocathode materials that can provide bright electron beams with high efficiency, minimal transverse energies, and able to withstand the harsh environment of electron guns is an active topic for the community involved in the fundamental and applied research of electron sources for accelerator applications. Progress in the realization of electron sources with improved performances will have a direct impact on a number of existing electron-beam-based facilities by allowing expansion of the scope and scientific reach of experiments, which can be performed and can enable the realization of new facilities with unprecedented electron beam quality.¹ 

Photocathode materials that have demonstrated successful operation in electron guns for accelerators belong to two main categories: metals, with copper,² magnesium,³ and yttrium⁴ being the only ones that have found a practical application, and semiconductors. The number of semiconducting photocathodes used in electron guns is much larger than that of metals. Semiconducting cathodes can be further classified in terms of Positive Electron Affinity (PEA) and Negative Electron Affinity (NEA) materials. For the PEA materials, the electron affinity level at the interface with the vacuum lies above the bulk conduction band minimum while it is located below that for NEA materials. Alkali-antimonide photocathodes (⁠$Cs3$Sb, $CsK2$Sb, $Na2$KSb, etc.) with an emission threshold in the visible range of the spectrum and alkali-telluride photocathodes (⁠$Cs2$Te, CsKTe, etc.) that have the onset of the emission in the UV region of the spectrum belong to the PEA family.⁵ 

The NEA condition is not naturally found in any known bulk semiconductor material, but it is usually enabled by the formation of a strong electric dipole at the vacuum surface. The most widespread method to generate such a dipole makes use of a sub-monolayer of cesium evaporated over the pristine surface of a semiconductor. The strength of such a dipole can be enhanced by exposing the cesium-covered surface to an oxidizing agent like oxygen or $NF3$⁠. This method has been successfully applied to achieve NEA condition at the vacuum interface of several semiconductors belonging to the III-Vs,⁶ III-nitrides,⁷ and, more recently, also to lower the work function of ordered oxide materials.⁸ Due to the extreme sensitivity of the activating layer with respect to surface contamination, photocathodes based on effective NEA have a limited lifetime when operated as electron sources for accelerators and require stringent vacuum conditions (vacuum levels better than $1×10−11$ Torr) to achieve sufficiently long lifetimes to perform meaningful experiments. Furthermore, the activation layer can also be damaged by moderate increases of sample surface temperatures⁹ and by the energetic chemical species ionized in the residual gas by the primary electrons back-streaming toward the cathode surface because of the electric field applied between cathode and anode.¹⁰ 

With all these limitations, the practical operation of NEA activated photocathodes has been possible only in high voltage DC photoelectron guns. The large volume required to maintain a sufficiently large distance between the negatively biased cathode electrode and the walls of the vacuum vessel allows for massive vacuum pumping, enabling the required vacuum levels in the XHV range (below the $1×10−11$ Torr level). Unfortunately, the intensity of the accelerating electric DC fields at the cathode is at least one order of magnitude lower than the typical accelerating gradients possible in radio frequency (RF) guns. Nonideal vacuum conditions prevent the use of NEA photocathodes in high field RF guns and fully leverage their combined large Quantum Efficiency (QE) and low Mean Transverse Energy (MTE) near the photoemission threshold. This limits the production of higher brightness electron beams from these cathodes.^11,12

GaN and its ternary alloys and structures with Al and In allow exploration of new opportunities for photocathode applications. In their wurtzite phase, III-nitrides exhibit extremely intense internal fields due to self-polarization and piezoelectric effects, which can be leveraged to engineer photocathode structures that do not require cesium to achieve effective NEA, making such structures robust-to-vacuum NEA photoemitters.¹³ 

III-nitrides in their wurtzite structure can be grown in a number of orientations including Ga-polar (⁠$c$-plane along the 0001 direction), N-polar (⁠$c¯$-plane along the 000$1¯$ direction), non-polar (m- and a-plane respectively along 1100 and 11$2¯$0 directions), and in a variety of other semi-polar planes. The magnitude of the polarization charge is dependent on the polar angle, with the largest out of plane polarization charge present in c-plane grown III-nitrides.¹⁴ Ga-polar p-type GaN photocathodes have previously been studied with the use of a cesium-activation layer¹⁵ and with novel cesium-free architectures employing a Si delta-doped surface layer.¹⁶ The presence of the Si delta-doped layer and thin n+GaN cap required to stabilize the surface creates a narrow depletion region and increases downward surface band bending. In the Ga-polar orientation, the polarization charge is negative at the surface, which compensates the positive depletion charge induced by the structure. In the N-polar orientation, however, the polarization charge at the surface is positive and compounds with the positive depletion charge to increase the downward surface band bending and lower the vacuum level relative to bulk conduction band energy level.¹³ Furthermore, by utilizing the N-polar orientation, the n-type concentration in the cap layer can be reduced while maintaining a narrow depletion width, which increases the material quality and efficiency of N-polar photocathode devices.

N-polar GaN photocathodes reported here were grown by metal organic chemical vapor deposition (MOCVD). Nominally, on-axis 2–sapphire wafers with $0.2°$ mis-cut toward the m-plane were used as substrates. An N-polar unintentionally doped (UID) GaN template layer was grown on the sapphire substrate using growth conditions reported elsewhere.¹⁷ The N-polar photocathode structure was then overgrown on the template layer consisting of nominally 450 nm p-GaN with 10 nm UID GaN cap layer.¹⁸ A simulated energy band diagram of the device is shown in Fig. 1 assuming a hole concentration of $3×1017cm−3$ in the p-type absorbing layer, and an electron concentration of $1×1016cm−3$ in the UID GaN cap layer. Based on this simulation, the structure aims to achieve a narrow depletion width at the surface and effective NEA.

Trimethylgallium (TMGa) and ammonia (⁠$NH3$⁠) were used as the precursors for the p-GaN film and bis(cyclopentadienyl)magnesium (⁠$Cp2$Mg) as the dopant source with a V/III ratio of 13 000, TMGa flow of 65 $μ$mol/min and $Cp2$Mg flow of 360 nmol/min. The UID GaN cap layer was grown with the same precursors at a V/III ratio of 3000 and TMGa flow of 80 $μ$mol/min. Following the completion of the photocathode structure growth, the p-type dopants were activated in situ by annealing in nitrogen ambient at $775°$C for 15 min. This is done to remove the hydrogen from the Mg–H complex.

Hexagonal pyramid or “hillock” structures are common on the N-polar surface. The size and areal density of hillock structures can be varied based on the growth conditions,¹⁷ and substrate off-cut.¹⁹ For the samples studied here, the hillock density varied modestly across the 2 in. wafer with approximately 1500 hillocks/$cm2$ near the wafer flat and 1000 hillocks/$cm2$ on the opposite side. Previous studies have shown improved Mg incorporation efficiency within the sidewalls of the hillock structures leading to improved p-type optical and electrical characteristics for samples with a high density of hillock structures.²⁰ Additional details regarding the growth, impurity depth profiles, and optical and electrical characteristics of similar structures have been reported by Rocco et al.^18,20 Two samples were created by cleaving $10×10mm2$ pieces from the 2 in. wafer, with sample A taken near the wafer flat with high hillock density and sample B from the opposite side with lower hillock density.

Prior to photoemission measurements, both samples A and B were etched in a 35% HCl solution for about 30 s, rinsed with de-ionized water, dried with $N2$⁠, connected to our sample holder and double bagged in nitrogen atmosphere. The HCl etch is completed to remove native oxide and other contaminants on the N-polar surface.¹³ Both samples were loaded in the load lock of the main photocathode UHV chamber within 15 min from the initial etch. The load lock was pumped down immediately and after 12 h, $8×10−8$ Torr vacuum level was reached. Measurements were performed in a UHV system, which comprises multiple interconnected vacuum chambers. QE and MTE characterizations were performed in two different vacuum chambers. MTE measurements were performed after QE and lifetime measurements. Samples were transferred from one chamber to the other in a vacuum level lower than $1×10−11$ Torr. A sample transfer usually requires about 15 min. A similar amount of time is usually required to align the light source on the sample surface and the generated electron beam on the detector.

Sample A was moved into the UHV analysis chamber with a base pressure lower that $1×10−10$ Torr, heated for about 12 h to a temperature of 150 $°$C, and then cooled to room temperature. Initial measurements of quantum efficiency with the surface illuminated with the light produced by a 300 nm LED revealed a very poor QE, below the detection limit of our measurement system, which is estimated at $5×10−8$⁠. The very low QE possibly indicates that the surface was still contaminated or the threshold energy of the photocathode exceeds the photon energy of 4.1 eV. Hence, an attempt was made to activate the surface with cesium to see if the atoms deposited at the surface could produce a sufficient lowering of the work function so that a measurable photocurrent could be extracted. Cesium atoms were produced using Cs dispenser from SAES Getters. Upon cesium evaporation, the photocurrent measured from the sample increased, yielding an equivalent QE value of 4% at 300 nm (see Fig. 2).

QE measurements were performed using a discrete set of UV LEDs emitting at wavelengths of 365, 375, 385, 340, 300 nm and are reported in Fig. 3 (data reported in green dots). In these and other QE measurements, the light spot on the sample surface had a diameter of about 5 mm and was located near the center of the photocathode accessible area.

After efficiency measurements were performed, sample A was heated to 600 $°$C for about 2 h to remove Cs and other contaminants from the surface. The heating process was repeated twice at 24 h separation to allow the vacuum to recover, and QE was measured again after the second heating cycle. The second set of measurements on sample A with Cs-removed (data reported in blue dots in Fig. 3) is well aligned with the data obtained for similar samples (N-polar GaN reference-1 and reference-2) with identical targeted structure and grown under similar conditions (continuous black and red lines in Fig. 3) that were never exposed to Cs vapors. N-polar GaN reference-1 and reference-2 were sent to Jet Propulsion Laboratory and there etched in HCl solution prior to QE evaluation, performed under illumination with a continuous spectrum of wavelengths ranging from 200 to 450 nm with 2 nm step size as described in Ref. 13. Dataset for samples reference-1 and reference-2 are shown here as a comparison to the measurements of samples A and B at finite wavelengths as they were obtained with identical growth procedures and contain spectral responses collected over an extended number of wavelengths, which we could not access with the setup used in the present set of experiment. While we expect that most of the cesium has been removed by the prolonged heat treatment at 600 $°$C, the completed evaporation of Cs atoms from the surface cannot be unequivocally ascertained. Previous work revealed traces of metallic Cs still present on GaN surfaces unless temperatures larger than 700 $°$C are used.²¹ 

Sample B was moved into the UHV analysis chamber held in a base pressure lower than $1×10−10$ Torr. There, it was heated for about 2 h to a temperature of 500 $°$C and then left to cool down to room temperature (for sample A the first heat cycle was done at 150 $°$C for 12 h). Unlike sample A, sample B was found to be photo-emitting after the heat cleaning and the QE was estimated at wavelengths of 265, 300, 340, 365, 375, and 385 using a set of UV LEDs (data reported as purple dots in Fig. 3). The obtained values for the QEs are about one order of magnitude larger than the values obtained for sample A but still in the range of efficiency measurements conducted on reference samples, at least at the shortest wavelengths. The lifetime of samples, defined as the time required for the QE to decay to 1/e of the original value, was estimated by measuring the photocurrent over a span of just over 20 h. In a base pressure of about $3×10−10$ Torr, the estimated lifetime of sample A has a value of about 2 weeks with an initial QE at 265 nm just above the $1×10−4$ level as reported in Fig. 4.

The same Fig. 4 reports the measured QE of sample B during a longer than 24-h test. After an initial fast decrease (roughly a factor 3 from about $3×10−3$ to $1×10−3$⁠) QE stays essentially stable, if not slowly increasing until the end of the run in a vacuum level of about $3×10−10$ Torr.

Photoelectrons’ MTE was measured using the transverse energy meter (the TEmeter—a low energy photoelectron gun) installed at the photocathode lab of Cornell University using the method of the voltage scan with electron beam energies in the range of 4–10 keV.²² To perform MTE measurements, a 100 $μ$m diameter pinhole illuminated by UV light from an LED was imaged on the sample surface on a subset of the area used for QE measurements. Due to the low average electron beam current achievable using an LED light source and the relatively low QE of the sample at these wavelengths, MTE measurements for sample A were only possible at the wavelengths of 265 and 300 nm. The results are reported in Fig. 5. An MTE of about 100 meV was measured at the wavelength of 265 nm and about twice this value at 300 nm. These results are somehow unexpected as usually lower MTEs are associated with lower photon energies. The MTEs of photoelectrons from sample B were characterized at the three wavelengths of 265, 300, and 340 nm. While values similar to sample A were observed at 265 nm, the set of measurements on sample B revealed a local minimum at the wavelength of 300 nm (Fig. 5).

As the laser beam was scanned over the surface of sample A during the initial stages of alignment of the instrument, it was possible to pinpoint some local areas where the electron beam as observed on the Ce:YAG screen of the TEmeter appeared to have a larger divergence along with reduced intensities. By scanning the light spot over one of these regions with a light spot having a 25 $μ$m RMS diameter, the sequence of beam images reported in Fig. 6 has been observed. It was not possible to identify, while scanning the laser spot on the photocathode surface of sample B, a location yielding similar electron beam images with a hollow center as it was the case for sample A. This seemed to indicate more uniform properties of photoemission over the accessible area.

The electron beam images in Fig. 6 were recorded using identical settings for the camera. Beam profiles have been obtained by integrating the signal in the beam images along the vertical direction [see Fig. 7(a)]. From those beam profiles, relative intensities and distribution widths have been obtained and reported in Fig. 7(b). The data in Fig. 7(b) confirm that as the laser beam is moving along the vertical path, it encounters a location characterized by both reduced quantum efficiency (as deduced by the relative decrease of the integrated signal on the scintillator screen) and increased MTE (as inferred by the relative increase of the width of the signal distribution). Furthermore, 2D electron beam intensity distributions reported in frames 4 and 5 of Fig. 6 seem to indicate that a correlation on initial position and momentum of electrons must be present at the photocathode surface so that a hollow electron beam is imaged on the screen once the laser beam spot is centered on the low efficiency emitting area.

The surface morphology of open air transferred samples was investigated using a white light interferometry profilometer (Zygo Zescope). Two-dimensional maps of the representative scanned area belonging to samples A and B are reported in Fig. 8. Hillock structures, crystallites with a characteristic hexagonal shape, are observed to have developed at the surface of both samples. While the density of hillock structures is within the same order of magnitude on both samples, hillock structures on sample A in general have larger diameters and heights as compared to sample B, and sample A has a modestly higher hillock density on average. These structures have been observed to develop during the MOCVD growth of N-polar samples and have been associated with increased p-type characteristics resulting from improved Mg incorporation efficiency inside the GaN lattice (just above $1×1019cm−3$⁠).²⁰ 

The photoemission properties of the two N-polar photocathode structures are extremely interesting for application as electron beam sources for accelerators. In the case of sample B, the relatively large QEs measured at the largest photon energies, above the $1×10−3$ level, coupled with the measured MTE of 100 meV and the long lifetime associated with the relative inertness of the surface are surely competitive and outperform many of the metals used in high gradient RF guns. Magnesium metal operated with UV photon energies of about 4.6–4.8 eV has QEs on the same range (1–$3×10−3$⁠) but measured MTEs are much larger (130–800 meV) depending on the roughness induced on the photocathode surface by the laser cleaning procedure required to remove the oxide layer.^23,24 As the photon energy decreases, approaching the threshold of photoemission at 4.1 eV, the efficiency lowers to the $10−5$ level, which is typical of metals like copper. However, electrons’ MTEs measured for the N-polar photocathode are a factor 2 smaller than the ones measured for copper cathodes with extremely low roughness.²⁵ On the other hand, the rapid increase of MTEs observed at the largest wavelengths does not allow for a straightforward interpretation and will require more detailed investigations: a similar behavior was recently observed for Cs–Te operated at near threshold and attributed to crystal phases of Cs–Te characterized by a lower emission threshold and lower QE.²⁶ In that study, an unexpected increase was observed in photoelectrons’ MTEs extracted with photon energies in the visible range using the light of a supercontinuum light source covering the spectral range with QEs in the range between $1×10−5$ and $1×10−7$⁠. Unfortunately, at the time, the present photoemission experiments were performed, the lack of a continuously tunable light source in the UV yielding a photon flux sufficient to perform MTE measurements with a fine wavelength scan, only allow us to speculate on the origin of the non-monotonic MTE behavior observed for sample B.

Unlike GaAs photocathodes activated to NEA using cesium and oxygen showing a sharp rise of the QE as the photon energy used to allow for electrons to transition from the valence to conduction band,²⁷ cesium activated GaN samples show sensitivities extending to photon energies much lower than the bandgap at 3.5 eV.²⁸ Similar to what was observed for GaAs,⁶ we were expecting very low MTEs for photoelectrons generated with photon energies close to the bandgap energy. The spectral response data reported in Fig. 3 for sample B show that the photoemission is characterized by a low efficiency shoulder located in the range between 300 and 385 nm and by a higher efficiency region for wavelengths shorter than 300 nm. At this time, we do not have sufficient information, allowing us to pinpoint the origin of photoemission generating a low efficiency shoulder to a specific mechanism (e.g., surface states, deep defects level, etc.). A simplified model allows us to estimate the expected photoelectrons’ MTE as 1/3 of the excess energy, which is the difference between the photon energy and the photoemission threshold.²⁹ This simplified model has been used to describe the photoemission result from Cs–Te near the onset of photoemission reporting an unexpected non-monotonic behavior of the MTE as a function of the photon energy.²⁶ These studies on Cs–Te lead to the conclusion that the MTE non-monotonic behavior, which is seen to increase, rather than decrease, as the photon energy falls below a certain threshold was indeed due to the presence of two distinct populations in the electron beam and to an abrupt change of the ratio of them as the photon energy approaches the threshold separating the high and low quantum efficiency spectral regions.²⁶ Similarly to the Cs–Te case, the spectral response of the N-polar GaN seems to be characterized by two regions showing a low efficiency shoulder at longer wavelengths and a high efficiency region at shorter wavelengths (see Fig. 3). Again, the origin of the shoulder at longer wavelengths is not yet unequivocally determined but, if we assume that similarly to the case of Cs–Te we are in the presence of two distinct electron populations in the electron beam, we can expect that when the wavelengths are longer than the threshold for the high QE photoemission region the MTE will progressively increase as the threshold of the high QE spectral region is approached. Once this threshold is reached and eventually overcome, the fraction of electrons emitted from the low QE shoulder will become less and less important with respect to the total number of electrons and the MTE of the electron beam should initially decrease due to the fact that the largest fraction of photoelectrons now has a smaller excess energy and MTE because of the larger emission threshold of this second population. As the illuminating wavelengths get shorter, the MTE will increase again because of the photoelectrons’ larger excess energy. The formation of a hollow electron beam is observed on the screen when the laser spot illuminates some specific locations on the surface of sample A and it is accompanied by an apparently decreased quantum efficiency and larger momentum spread, as inferred from the projected beam profiles reported in Fig. 7(a). Morphological analysis had revealed that surfaces of both samples are partially covered with structures having a characteristic size comparable with the laser spot diameter used on our experiments (see Fig. 8). The structures appear as 3D hexagonal based pyramids with angled, sloping sidewalls. A detailed study on the effect that these structures might have on the electron MTEs will require more accurate measurements, possibly performed with a laser spot size smaller than the one used in our experiment allowing us to better correlate the position on the cathode surface with the MTEs and QE. It can be expected that such structures will increase the divergence of the electron beam and hence affect the MTE of electrons.³⁰ The decrease in QE deduced by the reduced integrated intensity on the beam images indicates that the QE performances measured in these samples have large margins for improvement. Despite the still limited data collected, the overall performance of N-polar GaN photocathodes is already noteworthy and demonstrates that these engineered structures can potentially fulfill the requirement for applications like high repetition rates FELs. Lifetimes in excess of 15 days have been measured, indicating the robustness of these advanced structures. The possibilities for further improvement and new explorations are vast and can leverage decades of R&D investments and development in photon detectors and in the LED industry: emission wavelength tuning can be achieved using ternary alloys with Al and In, efficiency at near threshold can be improved by using resonating Fabry–Pérot structures, and furthermore, by reducing the density of defects using newly available substrates based on single crystal GaN rather than growing on sapphire substrates.³¹ Future studies of N-polar III-nitride photocathodes should aim at understanding the origin of the low efficiency shoulder in the photoemission response and identifying pathways for its mitigation so that lower MTEs can be achieved. Growth procedures that can produce smoother surfaces at the vacuum interface are likely to be needed as the hillock structure rich surfaces might generate increased MTEs and possibly problematic levels of field emission currents in very high electric field gradients. Investigations of the robustness against known surface contaminants and of the possibility of rejuvenating the surfaces only by heating, and the measure of response time of the photoemission, will also be beneficial in identifying possible practical applications of this new technology.

A new class of photocathode structures, N-polar GaN, has been investigated for the first time as a possible candidate to produce bright electron beams. These preliminary results are promising but the range of required characterizations needs to be further expanded to gather more insight from the point of view of both material science and the electron beam production. Simultaneous QE in the range of $10−3$⁠, with reasonably low MTEs of about 100 meV and long lifetimes already make these materials very interesting. The use of transparent substrates makes them appealing for operation in transmission mode. Furthermore, the concept of N-polar structures can be extended to other III-nitride structures with reduced threshold for photoemission.

This work was supported by the U.S. Department of Energy (DOE) under Contract Nos. DE-SC0012704, DE-SC0021092, and DE-SC0020517 and the U.S. National Science Foundation under Award No. PHY-1549132, the Center for Bright Beams. This research was carried out in part at the Jet Propulsion Laboratory, California Institute of Technology, and was sponsored by the National Aeronautics and Space Administration (No. 80NM0018D0004). We acknowledge the use of facilities within the Eyring Materials Center at the Arizona State University supported in part by Grant No. NNCI-ECCS-2025490.

The authors have no conflicts to disclose.

Ethics approval is not required.

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