Conduction-band engineering of polar nitride semiconductors with wurtzite ScAlN for near-infrared photonic devices Open Access

The incorporation of rare-earth metal scandium (Sc) into wurtzite III-nitrides has added exciting new functionalities to this intensively studied III-V semiconductor system opening avenues for novel photonic and electronic technologies.¹ The wide-bandgap III-nitride semiconductors have been driving advances in several technological applications.² The highly tunable bandgaps of (Ga,In,Al)N alloys are employed in light emitting and laser diodes spanning the visible to ultraviolet wavelength range.^3–5 Nitride high electron mobility transistors are successful in power electronics.^6,7 Nevertheless, a perennial challenge of nitride devices is the lattice-mismatch between different layers that leads to large defect densities and ultimately degraded device performance.^8–10 Even though In_0.18Al_0.82N is lattice-matched to GaN, it suffers from compositional inhomogeneity that affects its electronic and optical properties.^10–14 Wurtzite Sc_xAl_1−xN has emerged as an alternative barrier material for nitride devices due to its unique properties.^15–47 As Sc_0.18Al_0.82N is purportedly lattice-matched to c-plane GaN, it has the potential to overcome the longstanding challenge of nitride lattice mismatch. Moreover, it promises to expand the spectral range of nitride photonics into the near infrared and beyond.

To reach long wavelengths, infrared optoelectronic devices typically employ intersubband (ISB) transitions between quantized energy levels in the conduction band of heterostructures formed by materials of different bandgaps. Prime examples include the quantum cascade lasers (QCLs)^48,49 and the quantum well infrared photodetectors (QWIPs).^50–53 The ISB transition energies are widely tunable by engineering the conduction band profile (i.e., layer compositions and thicknesses) and are constrained primarily by conduction band offsets (CBOs).⁵⁰ The large CBOs achievable in III-nitride heterojunctions (1.8 eV for GaN/AlN) and sub-picosecond relaxation times make them ideal candidates for ultra-fast devices in the technologically relevant near infrared range.^50–57 Moreover, the large energy of GaN longitudinal optical phonon (92 meV) opens prospects for high-temperature ISB devices spanning the 5–10 THz band.⁵³ Unique to the nitrides, the ISB energies also depend on strong built-in polarization fields. This effect can be eventually eliminated by using nonpolar heterostructures such as the m-plane (In)AlGaN/InGaN multi-quantum well structures we have recently reported.⁵⁴ Due to the large thicknesses required by infrared wavelengths, intersubband devices utilizing III-nitrides have been severely constrained by lattice-mismatch issues, and therefore stand to benefit from novel lattice-matched ScAlN/GaN heterostructures.

Epitaxial growth of wurtzite ScAlN on c-plane GaN has recently enabled nitride devices with new functionality. Most of the research to date on wurtzite ScAlN has focused on its applications for piezoelectric, electronic, and ferroelectric devices. The ultra-wide bandgap, high dielectric constant along with large piezoelectric coefficients and spontaneous polarization coefficients¹⁶ make it a suitable alternative barrier material for the next-generation high electron mobility transistors (HEMTs) and heterostructure field effect transistors (HFETs).^1,17–21 Recently, a two-dimensional electron gas was reported in AlScN/AlN/GaN heterostructures, giving an electron mobility of 6980 cm²/V s and record low sheet resistance of 57 Ω/□ at T = 10 K.⁴² Furthermore, ScAlN's spontaneous polarization, polarization switching, and long polarization retention make it a promising material for memory storage devices as well.¹ ScAlN/GaN heterostructures have exhibited stable ON/OFF states that lasted for months at room temperature and remained stable even at high temperatures (≈670 K) comparable to Curie temperatures of conventional ferroelectric materials.⁵⁸ ScAlN has also demonstrated strong nonlinear optical properties, which could be used to develop nonlinear optoelectronics.⁴⁷ Finally, the ScAlN polarization can be controlled lithographically, a property that may prove advantageous for ferroelectric storage and nonlinear optical devices.⁵⁹ 

This paper investigates the quantized electronic states in the conduction band of ScAlN/GaN multi-quantum well structures (MQWs) using near-infrared ISB absorption and computational modeling. ScAlN/GaN MQWs have demonstrated surprisingly strong ISB absorption around 2 μm⁶⁰ that highlights their potential for infrared optoelectronic devices. However, conduction-band engineering in nitride heterostructures for practical applications requires in-depth knowledge of fundamental material parameters and precise control of atomic structure. By examining the detailed dependence of the ISB transition energy on quantum well width and barrier composition, we are able to provide insight into ScAlN polarization. We also evaluate the potential of ScAlN/GaN material systems for next-generation infrared technologies.

Sc_xAl_1−xN/GaN 50 quantum well samples, also referred to as superlattices hereon, were grown using plasma-assisted molecular beam epitaxy (MBE) on commercially available Fe-doped semi-insulating c-plane GaN templates on sapphire (0001) substrates (dislocation density of 8 × 10⁸ cm⁻²). MBE allows unmatched control over the thickness of grown individual layers and has been instrumental in the development of nitride ISB devices.⁵⁰ The substrates were coated with a 1 μm layer of tungsten silicide on the backside to improve thermal coupling with MBE heater. Prior to loading in the MBE chamber, the substrates were cleaned using trichloroethylene, acetone and methanol to remove any organic impurity and then acid etched in HCl for 10 min to remove any excess metal. After acid etch, the substrates were thoroughly cleaned with DI water and dried with N₂. The substrates were outgassed overnight (>12h) at 550 °C in an ultra-high vacuum chamber attached to the MBE chamber. High purity Ga, Al, and Sc (purity 99.999%) metal sources were used to minimize the adverse effects of various impurities in the superlattices. A Veeco Unibulb radio-frequency (RF) plasma source operated at 305 W forward power was used to supply active nitrogen. The N₂ flow rate was maintained at 0.5 SCCM for all the samples, corresponding to a GaN growth rate of ∼8.2 nm/min. Prior to superlattice growth, a 150 nm GaN buffer layer is grown at 720 °C substrate temperature under Ga-rich growth conditions. This results in a smooth GaN surface with root-mean square (RMS) roughness of 0.7 nm over an area of 5 × 5 μm². The superlattices were grown at 550 °C under N-rich conditions for ScAlN, according to optimal conditions identified by our previous work.⁶⁰ The total metal to nitrogen ((Sc + Al)/N) ratio was approximately 0.8 for the ScAlN barriers. For the growth of the GaN wells, the Ga/N ratio was 1.2. The samples were doped with silicon using two δ-doping sheets inserted in the barriers 1 nm away from each interface while the growth was paused. The doping density is determined by the duration of the δ-doping sheet. The Si cell temperature was 1375 °C which provides a bulk Si atom density of 2.94 × 10¹⁹ cm⁻³ as measured by secondary ion mass spectrometry (SIMS) in GaN grown at 8.2 nm/min. This results in a nominal 10 s δ-doping sheet corresponding to a Si density of 4.0 × 10¹² cm⁻².

The structure of the samples was characterized with atomic force microscopy (AFM), x-ray diffraction (XRD), and transmission electron microscopy. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were used to determine individual layer thicknesses (Fig. 1). The HAADF-STEM samples were prepared using a Thermo Scientific Helios G4 UX Dual Beam Focused Ion Beam (FIB) system, after which they were cleaned using a Fischione Ar Plasma cleaner to remove any FIB or air contamination. The HAADF-STEM imaging and analysis were performed using a Thermo Scientific Themis Z Double Aberration Corrected S/TEM operating at 300 kV acceleration voltage and 0.12 nA screen current.

Energy-dispersive x-ray (EDX) spectroscopy was used to determine the Sc composition in the Sc_xAl_1−xN barriers. EDX spectra were obtained from the HAADF-STEM samples at 300 kV accelerating voltage and 1.5 nA screen current. The intensities of the K-α lines in the background-subtracted EDX spectra were corrected to account for ionization cross section, transition probability, fluorescence yield, and detector efficiency to determine Sc composition. For thick ScAlN, EDX produces Sc compositions that are consistent with Rutherford backscattering spectrometry (RBS) measurements within the experimental error of ±2%.

Symmetric $ω−2θ$ XRD scans were measured using a Panalytical X'Pert³ Materials Research Diffractometer (MRD) system. To estimate Sc composition in the barriers and confirm layer thicknesses, the experimental XRD scans were modeled in the Panalytical Epitaxy software package assuming the composition dependence of lattice constants established in our previous work⁶⁰ as shown in Fig. 2. Reciprocal space mapping (RSM) of the $( 10 1 ¯ 5)$ reflection was performed using a Panalytical Empyrean diffractometer to examine the strain state of the superlattices.

To estimate the charge density of the Sc_0.18Al_0.82N/GaN MQWs, Hall measurements were carried out using the Van Der Pauw geometry in a Quantum Design DynaCool Physical Properties Measurement System (PPMS). The Hall samples were approximately 3 × 3 mm². Indium contacts were made at the corners of the samples and annealed at 450 °C to ensure good ohmic contacts.

Direct Fourier transform infrared spectroscopy (FTIR) was used to measure the near-infrared intersubband absorption in each of the samples. The measurements were performed using a Thermo Scientific Nicolet 8700 spectrometer, equipped with an InSb detector and a white light source. The samples were polished to have parallel 45° facets to ensure multiple passes into the active region of the sample (see the supplementary material for experimental geometry). The s- and p-polarized transmission spectra were recorded along with their backgrounds. The ISB absorption spectra were obtained from the ratio of the background-normalized p- to s-polarization transmission spectra.

The band structure calculations of the ScAlN/GaN superlattices were done using the nextnano simulation package. The software computes the electronic band structure using the 8-band k⋅p model and Poisson equation self-consistently.⁶¹ For GaN and AlN, nextnano uses generally accepted material parameters from Ref. 62. However, the parameters for ScAlN are not well established. The model assumes Sc_0.18Al_0.82N is in-plane lattice-matched to GaN. According to recent high-resolution x-ray photoelectron-spectroscopy measurements, the Sc_0.18Al_0.82N/GaN heterostructure exhibits a staggered (type-II) band alignment, with a CBO of 2.3 eV, VBO of 0.2 eV, and Sc_0.18Al_0.82N bandgap of 5.5 eV.⁶³ There is a crossover from type-I to type-II alignment as the Sc composition increases roughly beyond 0.11.⁶⁴ To simulate the band structure of wurtzite Sc_xAl_1−xN/GaN heterostructures, the spontaneous and piezoelectric polarization for a hypothetical wurtzite ScN binary are needed and were taken from numerical simulations.¹⁶ Caro et al. fit the composition dependence of the spontaneous polarization in wurtzite Sc_xAl_1−xN to a quadratic function $P S P=−0.874x−0.089( 1 − x)+0.741x( 1 − x) C/ m 2$⁠.¹⁶ The resulting spontaneous polarization of a hypothetical wurtzite ScN (P_sp= −0.874 C/m²) is considerably larger in magnitude than the values for GaN (−0.034 C/m²) and AlN (−0.09 C/m²) but has not yet been confirmed experimentally. All material parameters used in the simulations are given in the supplementary material.

We calculated the energy eigenvalues, electronic wavefunctions, and charge density in the conduction band of single GaN quantum wells with ScAlN barriers using periodic boundary conditions to simulate the superlattice. The ISB energies were then corrected for many-body effects in the local density approximation (LDA) and Hartree–Fock approximation (HFA).⁵⁴ Our previous work on nonpolar InGaN/AlGaN MQWs indicated that the LDA approximation is better suited than the HFA for reproducing ISB transition energies in nitride heterostructures.⁵⁴ 

Table I summarizes the design parameters and main experimental results for the samples discussed in this paper. All samples exhibit strong and narrow ISB absorption lines in the near infrared range consistent with our previous findings.⁶⁰  Figure 3(a) shows the ISB spectra for a series of lattice-matched Sc_0.18Al_0.82N/GaN superlattices with the same charge density but different QW widths. The spectra were fitted with Lorentzian functions to extract transition energy and linewidth. The peak wavelength increases from 1.94 to 2.38 μm (ISB energies from 640 to 520 meV, respectively) for QW widths of nominal thickness of 2.5–8 nm. The absorption peaks are narrower than previously reported in any nitride superlattices in this energy range.⁶⁵ The linewidth increases from 50 to 65 meV as the QW width decreases likely due to increased interface roughness. As expected from previous reports, the ISB linewidth also depends on charge density and reaches the lowest value of 48 meV for the lowest doped sample E [Fig. 3(b)].

The band structure of the ScAlN/GaN superlattices was modeled using the nextnano package. Figure 4(a) shows the conduction and valence band edge profiles together with the electron probability profile for the ground and first excited electronic states for a single 4-nm GaN QW with nominally lattice-matched 6-nm Sc_0.18Al_0.82N barriers corresponding to sample B. The ground state of the QW is populated with electrons provided by the Si δ-doping placed in the barriers 1 nm away from each interface. ISB absorption is due to transitions of these electrons from the ground to the first excited state.

The band edge profiles are also sensitive to the spatial separation of electrons from Si ions (Hartree potential). Theoretically, all Si impurities placed in the barriers are expected to be activated if the activation energy is smaller than approximately 2 eV. However, dopant activation in Al-containing nitrides has been found previously to be lower than expected and to depend on growth conditions. For this reason, we performed Hall measurements to determine the experimental sheet charge density for several representative doping levels. We found that the average sheet charge density is consistently lower than expected. For example, sample D had an average sheet charge of 4.5 × 10¹² cm⁻² per QW, or a factor of about 1.78 lower than calculated for two 10 s δ-doping sheets. Under the circumstances, the distortion of the band edge profile due to δ-doping is negligible, as can be seen in Fig. 4(a).

Figure 4(b) shows the theoretical dependence of the ISB energies on charge density calculated for the 4-nm GaN QW with 6-nm Sc_0.18Al_0.82N barriers shown in Fig 4(a). The bare ISB energy calculated in nextnano with the 8-band k⋅p model (labeled with k⋅p) is extremely sensitive to the choice of the spontaneous polarization constant (P_sp) as shown by the black curves. The bare energy decreases with increasing charge density due to the Hartree potential (i.e., charge redistribution). Many-body corrections have been found instrumental in quantitatively reproducing ISB energies in nitride QWs.⁵⁴ The many-body corrections increase the ISB transition energies proportionately with the charge density, with the HFA contributing a larger correction than LDA.⁵⁴ Therefore, the bare transition energies calculated in nextnano were corrected for many-body effects using LDA and HFA. However, the increase in transition energy due to many-body corrections is partially compensated by the energy decrease due to charge redistribution, resulting in a diminished net effect of the charge density on ISB energy.

Figure 4(b) also shows the measured ISB energies for three samples containing 4-nm GaN QWs with 6-nm Sc_0.18Al_0.82N barriers (samples B, E, and F) that were extracted from the ISB spectra shown in Fig. 3(b). Interestingly, the experimental transition energies appear to be relatively insensitive to charge density and are best reproduced by theoretical calculations using the spontaneous polarization proposed by Caro et al.¹⁶ without any many-body corrections. Nevertheless, given the measured charge density, it is reasonable to assume that the many-body corrections are not negligible and therefore the magnitude of the spontaneous polarization needs to be reduced accordingly to better match the experimental ISB values. Figure 4(b) shows that a P_sp= −0.814 N/m² for wurtzite ScN produces quantitative agreement with the LDA corrected energies, while P_sp= −0.754 N/m² produces an acceptable agreement with the HFA corrected energies. The bowing parameter for the spontaneous polarization was maintained at the theoretically proposed value of 0.741 N/m² in both cases. The decrease in the net alloy spontaneous polarization magnitude is similar, i.e., 9% lower for LDA and 18% lower for HFA.

Figure 5 shows the theoretical dependence of the ISB energy on QW width as calculated with the k⋅p model as well as corrected for many-body effects using HFA and LDA. The ISB energy in QWs wider than 3 nm is dominated by the tilt of the conduction band at the bottom of the QW due to intrinsic polarization fields. For lattice-matched ScAlN/GaN heterostructures, only the spontaneous polarization mismatch contributes to the steep tilt of the bands that essentially forms a triangular well. As a result, all curves reproduce the experimental relative decrease of the transition energy with increasing QW width (i.e., slope of the energy vs QW width). Figure 4(b) indicates that both HFA and LDA overestimate the transition energy, with the LDA doing so by a smaller amount. Nevertheless, both models can be improved by adjusting the spontaneous polarization parameter for Sc_0.18Al_0.82N as described above to obtain excellent agreement between experimental and theoretical ISB energies for QWs wider than 3-nm.

Below 3-nm QW width (sample A), the measured transition energy is lower than calculated. This may be due to increased interface roughness that leads to uncertainty in the QW width estimate by XRD. HAADF-STEM of this sample indeed indicates a wider QW width (3.8-nm) than targeted (Table I). Additional MBE growth development is needed to control GaN thickness accurately below 3-nm.

Decreasing the barrier Sc-composition to 0.13 has only a modest effect on ISB transition energies (samples G, H, and I in Fig. 5). We can infer from this that the change of net polarization fields is also relatively small in the probed 0.13–0.18 composition range with possibly slightly less net polarization mismatch for Sc_0.13Al_0.87N/GaN interfaces. The distribution of polarization between the spontaneous and piezoelectric components is not fully understood, though. Our theoretical calculations based on current understanding of piezoelectric polarization constants¹⁶ predict a consistent increase of the ISB transition energies (>50 meV) for all Sc_0.13Al_0.87N/GaN QW widths due to rapid increase of the piezoelectric fields away from the lattice-matched conditions. Our experimental results do not validate these predictions and suggest different polarization dynamics.

It is important to note that several recent reports in the literature identified the composition of ScAlN lattice-matched to GaN at lower values than the previously accepted 0.18 (0.09⁶⁶ and 0.12⁶⁷). We believe the discrepancies are due at least in part to the method used for determining Sc-composition (RBS vs x-ray photoelectron spectroscopy (XPS) vs energy-dispersive x-ray spectroscopy (EDXS)). If the exact lattice-matched conditions end up being closer to 0.13, the spontaneous polarization parameter can be refit to reproduce data for samples G–I. In that case, the piezoelectric polarization will need to be taking into acount in the calculation of the ISB energies for our samples A–F. Considerably more experimental work needs to be done to pinpoint the lattice-matched composition and to refine the composition dependence of the piezoelectric constants. This work is currently under way in our laboratory.

Wurtzite Sc_xAl_1−xN/GaN (x = 0.13–0.18) multi-quantum wells grown by molecular beam epitaxy on c-plane GaN exhibit exceptional near-infrared intersubband absorption in the technologically important 1.8–2.4 μm range. The measured near-infrared ISB linewidths are the narrowest reported in the literature for nitride semiconductors in this energy range and indicate strong potential for practical applications. Band structure simulations including many-body corrections were used to understand the effects of charge density, alloy composition, and quantum well width on the ISB transition energies. For GaN wells wider than 3 nm, the quantized energies are dominated by the steep triangular well due to intrinsic polarization fields. Consequently, ISB absorption provides unique experimental access to fundamental ScAlN polarization parameters. To reproduce the dependence of measured absorption energies on quantum well width, the spontaneous polarization difference at the nominally lattice-matched Sc_0.18A_0.82lN/GaN interface needs to be decreased in magnitude from theoretically proposed values. The model using LDA many-body corrections requires a smaller decrease in magnitude of the spontaneous polarization of wurtzite ScN (7%) than the model using HFA correction (14%). The intersubband transition energies are relatively insensitive to the barrier alloy composition indicating negligible variation of the net polarization field in the probed 0.13–0.18 Sc composition range. More experimental work is needed, though, to unravel the contributions of spontaneous and piezoelectric polarization away from the lattice-matched conditions.

See the supplementary material for the geometry of the ISB absorption measurement and material parameters used in the nextnano simulations of ScAlN/GaN band-structures.

We acknowledge support from the National Science Foundation (NSF). G.G., T.N., and O.M. acknowledge partial support from NSF Award No. DMR-2004462. All HAADF-STEM imaging and analyses were performed at the Electron Microscopy Facility at the Birck Nanotechnology Center, Purdue University.

The authors have no conflicts to disclose.

Govardan Gopakumar: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – original draft (equal). Zain Ul Abdin: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – original draft (equal). Rajendra Kumar: Investigation (equal). Brandon Dzuba: Investigation (equal). Trang Nguyen: Investigation (equal). Michael J. Manfra: Conceptualization (equal); Funding acquisition (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal). Oana Malis: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available within the article and its supplemental material.