Sputtered ferroelectric aluminum scandium boron nitride (Al_1−x−yB_xSc_yN)/n-GaN heterostructures Open Access

Ferroelectricity has recently been demonstrated in nitride and oxide wurtzite solid solutions.^1–3 When compared to traditional ferroelectric (FE) oxides, wurtzites are chemically compatible with complementary metal-oxide-semiconductor (CMOS) or III-nitride technologies, making them interesting for device integration.³ However, wurtzite FEs exhibit large coercive fields (>2 MV/cm), and thickness scaling below ∼50 nm results in reduced insulation resistance and endurance. Prevailing knowledge suggests that the degraded ferroelectric properties observed in thinner films may be related to reduced material quality at the substrate interface. More specifically, ferroelectric films are typically deposited on elemental metal bottom electrodes (e.g., W or Pt), which likely contribute to the following issues: (i) crystallographic and morphological transitions during initial wurtzite nucleation on the metal electrode, (ii) chemical inhomogeneity and point defects at the wurtzite–electrode interface, and (iii) defective interfaces that propagate defects such as nitrogen vacancies during electrical testing.^4–6 

In principle, using a wurtzite substrate should alleviate challenges associated with film growth at the wurtzite–metal electrode interface. Doped GaN is a particularly attractive isostructural material that exhibits a modest lattice mismatch conducive to epitaxial growth. Multiple groups have developed lattice matched Al_1−xSc_xN/GaN heterostructures. Casamento et al. reported reduced coercive field values, Wang et al. reported a coercive field increase, and Wolff et al. reported nanostructured backswitching at the GaN interface.^4,7,8 To date, it is not entirely clear what impact epitaxy has on the ferroelectric nitride response. The present manuscript contributes to this body of work by exploring quaternary compositions of Al_1−x−yB_xSc_yN that contain both B and Sc alloying elements. The addition of two chemically unique impurities (B and Sc) expands the range of possible compositions that can be epitaxially strained to the GaN substrate.⁹ Moreover, quaternary compositions offer greater flexibility from an electrical perspective. For example, films high in Sc content can have coercive fields as low as 2 MV/cm, yet suffer from large leakage currents, while Al_1−xB_xN films have low leakage currents, but exhibit relatively large coercive field values (≥5 MV/cm). Therefore, compositions containing both Sc and B alloys may exhibit unique ferroelectric properties otherwise unobtainable through ternary formulations.

The first step to developing an electrically robust, ferroelectric all-wurtzite heterostructure for technological applications is a GaN template with suitable crystallinity and topography. The available methods employed for GaN fabrication include metal organic chemical vapor deposition (MOCVD), molecular beam epitaxy (MBE), and sputtering. GaN growth on non-native substrates via MOCVD is a well-established commercial process.¹⁰ However, the extent to which microstructure, crystallinity, and conductivity can be independently tuned has finite limits. For example, when film thicknesses exceed 2.0 μm, Si-doped MOCVD GaN exhibits significant tensile cracking at Si concentrations of 2.0 × 10¹⁹ cm⁻³.^11–13 In addition, Si has limited solubility in GaN, making it difficult to achieve high carrier concentrations when using high temperature (>1000 °C) near-equilibrium techniques like MOCVD. While Ge is another common n-type dopant, it is less frequently used in MOCVD because high germane flow rates are needed, which often results in unfavorable processing outcomes.^14,15 In contrast to MOCVD, MBE growth occurs at considerably lower temperatures (∼700 °C). Due to the improved dopant solubility experienced at these temperatures, MBE doped GaN films can support significantly greater carrier concentrations.^16,17 Consequently, MBE is a common method for fabricating highly conductive ohmic contacts in device structures.^18,19 However, like MOCVD, these films can be highly strained and prone to cracking, which limit the films to modest thickness values.

In comparison, sputtering is a widely available and scalable growth technique that generally enables lower-thermal-budget synthesis. While there are concerns regarding sputtered GaN purity and overall quality,^20–22 literature demonstrates highly conductive GaN with a variety of dopants (Si, Ge, and Sn) on GaN templates with step-and-terrace surface morphology and narrow x-ray linewidths.^23–25 However, sputtering GaN on industry standard sapphire typically results in films with rough surfaces and lower crystalline fidelity.^21,26 Due to the widespread use of sputtering for piezoelectric and ferroelectric nitrides, there is incentive to process fully epitaxial nitride stacks with integrated ferroelectric layers using a single sputtering tool directly on c-plane sapphire.⁴ In this report, we utilize the recently developed sputtering methods to deposit highly crystalline, smooth Si-doped n-GaN with a carrier concentration of 6.0 × 10¹⁹ cm⁻³ directly on sapphire templates. This sputtered n-GaN then acts as a bottom electrode for epitaxial, ferroelectric Al_1−x−yB_xSc_yN (AlBScN) grown at ∼350 °C. The ferroelectric AlBScN retains the underlying GaN layer morphology and exhibits ferroelectric switching with square, saturated loops.

To explore epitaxial trends that influence wurtzite nitride switching, we present several AlBScN film series where compositions are varied on a common substrate to engineer epitaxial strain, and where common compositions are prepared on a spectrum of substrates to vary both crystallographic disorder and film microstructure/growth morphology. It is well known that interfacial strain and strain relaxation significantly impacts nitride device performance.²⁷ In many thin film systems, small amounts of epitaxial strain can be tolerated, or, in the case of quantum-well lasers, even beneficial to III–V semiconductor applications.²⁸ However, extensive strain often results in unwanted defect generation, lattice relaxation, and cracked films—all of which can be detrimental to material functionality.²⁹ Therefore, strain engineering is critical to developing high performance ferroelectric wurtzite III–N materials for a broad range of electronic, optoelectronic, and photonic applications.

Epitaxial registry and ferroelectricity have been observed for both AlBN and AlScN thin films grown on GaN templates.^4,7,30 However, the ferroelectric Al_1−x−yB_xSc_yN/n-GaN heterostructure proposed in this study expands the range of feasible ferroelectric wurtzite III–N compositions in two new ways. First, the co-substituted Al_1−x−yB_xSc_yN system offers an additional degree of freedom when fine tuning lattice parameters. Increasing B content decreases the in-plane lattice parameter, while adding Sc expands it.^2,31 This creates an opportunity to develop novel AlBScN compositions with unique electrical properties, bandgaps, and polarization values, while still maintaining an in-plane lattice match to GaN. Second, with sputtered n-type GaN, it is possible to further refine lattice parameters through energetic bombardment and plasma chemistry, thus expanding the formulation space at constant strain or the strain space at constant formulation.

In this work, AlBScN/n-GaN heterostructures are fabricated by depositing 200 nm AlBScN films with varying amounts of B and Sc on the n-GaN substrate. To include both ternary and undoped sample comparisons, 200 nm AlBN/n-GaN and 100 nm AlN/n-GaN heterostructures were also fabricated. Sample choice is predicated on designing ferroelectric wurtzite layers with differing degrees of lattice mismatch to the n-GaN template. A recent study by Hayden et al. discusses a- and c-lattice parameter trends in the AlBScN system for a range of compositions, and this information provides a starting point for lattice parameter prediction and formulation development.⁹ As shown in Table I, the selected AlBScN compositions exhibit a range of predicted in-plane lattice parameters (3.08–3.19 Å), while the sputtered n-GaN lattice parameter, as determined by XRD, is 3.15 Å.

The AlBScN composition series includes in-plane lattice parameters above or below the sputtered GaN lattice parameter (3.15 Å). More specifically, the Al_0.85B_0.06Sc_0.09N composition (3.10 Å) is under tension with a predicted lattice mismatch of −1.59%, while the Al_0.78B_0.04Sc_0.18N (3.19 Å) sample is under compression with a predicted lattice mismatch of +1.27%. In contrast, the Al_0.76B_0.06Sc_0.18N (3.15 Å) sample has an in-plane lattice parameter equivalent to the n-GaN template, which should result in an unstrained, epitaxial heterostructure. Both the AlN (3.11 Å) and ternary Al_0.94B_0.06N (3.08 Å) formulations are tensilely strained to the n-GaN substrate and have a predicted lattice mismatch of −1.27% and −2.22%, respectively.

Epitaxial AlBScN on GaN, especially at modest substrate temperatures, should promote wurtzite phase formation, high crystalline fidelity, smooth surfaces, and abrupt internal interfaces. To explore this possibility, all AlBScN/n-GaN heterostructure formulations are analyzed via XRD. Figure 1(a) shows symmetric θ–2θ scans for the AlBScN/n-GaN stacks, with each Al(B,Sc)N film showing 002 diffraction peaks and a uniform c-axis orientation, regardless of composition. Both residual strain and composition are known to significantly impact a film's 2θ value. In this instance, as the in-plane lattice parameter decreases as a function of composition, we observe an upward shift in Al(B,Sc)N 2θ peak position. It is worth noting that only the lattice matched Al_0.76B_0.06Sc_0.18N composition exhibits extended Pendellösung fringes, which indicate a particularly high degree of crystallinity and an abrupt interface. In comparison, the compressively strained Al_0.78B_0.04Sc_0.18N film shows subtle Pendellösung fringes that taper off rapidly at higher 2θ angles, while the tensile-strained Al(B,Sc)N heterostructures show no fringing. The Al_0.94B_0.06N film has the greatest degree of lattice mismatch, is tensilely strained, and shows relatively pronounced Al_0.94B_0.06N 002 peak broadening.

Rocking curve measurements were conducted on the n-GaN and Al(B,Sc)N 002 peaks of each sample as shown in Figs. 1(b)–1(g). The lattice matched Al_0.76B_0.06Sc_0.18N/GaN sample has the narrowest rocking curve (FWHM = 0.08°), a value comparable to, or narrower than AlScN heteroepitaxial films reported in the literature.^7,32 The tensile Al_0.85B_0.06Sc_0.09N and compressive Al_0.78B_0.04Sc_0.18N samples show more mosaic spread with FWHM values of 0.25° and 0.20°, respectively. In general, sample formulations with compressive epitaxial registry to GaN have less mosaicity than samples with comparable tensile registry. We speculate that the compressive growth strains known to occur in sputtered films at low homologous temperatures are responsible for this difference because the effective lattice mismatch is smaller.

In comparison, the 100 nm AlN rocking curve FWHM is 0.45°—the largest value in the sample series. Based only on the lattice parameter, we would predict a FWHM similar to Al_0.85B_0.06Sc_0.09N. We note that the AlN film is 50% thinner than the other compositions, and while texturing typically diminishes in thinner AlN films, a more narrow rocking curve is possible for epitaxially grown AlN that has not yet reached a critical thickness conducive to relaxation.³³ However, a more probable contributing factor is that B impurities have a surfactant-like effect on AlN crystal growth that can impact morphology and improve texture.^9,34,35 This boron effect is potentially consistent with the narrower-than-expected rocking curve (FWHM = 0.20°) for the Al_0.94B_0.06N film, which should be wider than AlN formulation when based on the lattice constant mismatch values alone. Alternatively, the previously mentioned Al_0.94B_0.06N θ–2θ measurements depict a broad shoulder at higher 2θ values, suggesting possible film relaxation or a minor second crystallite population with more disorder. The possibility that these rocking curves also sample an interfacial layer that is coherently strained and aligned with the GaN substrate cannot be excluded—additional skewsymmetric scans are needed to clarify this point.

Reciprocal space map (RSM) measurements containing the n-GaN and AlBScN-105 peaks were performed to explore in-plane lattice registry [Figs. 1(h)–1(l)]. The compressively strained Al_0.76B_0.06Sc_0.18N and Al_0.78B_0.04Sc_0.18N films exhibit sharp peaks that are aligned to n-GaN in Q_x, denoting pseudomorphic growth. In contrast, the tensilely strained AlN and Al_0.85B_0.06Sc_0.09N have in-plane lattice parameters of 3.10 Å and exhibit −105 peaks that are subtly offset in Q_x, suggesting finite relaxation. Last, as indicated by diffuse RSM data, the Al_0.94B_0.06N film retains a finite level of coherence, but is almost fully relaxed with an experimentally observed lattice parameter of 3.08 Å.

Figures 2(a)–2(f) show AFM tapping mode images of the sputtered n-GaN substrate (a), an AlN film on the as-grown substrate (b), and Al(B,Sc)N/n-GaN films on the as-grown substrate (c)–(f). As illustrated in Fig. 2(a), the ∼500 nm thick sputtered n-GaN template exhibits a leaf-like morphology with c/2 steps and an overall smooth surface (RMS = 2.4 nm). AFM scans of the AlBScN/n-GaN heterostructures reveal that the ferroelectric layer follows the morphology of the GaN substrate, similar to the previous reports of AlBN on GaN.⁴ Regardless of the sample composition or lattice strain, each Al(B,Sc)N film is smooth (RMS < 1.8 nm) and crack-free. The structural integrity displayed by the 100 nm AlN/n-GaN stack is particularly noteworthy, considering the significant thermal and lattice mismatches between the two materials. For context, prior studies indicate that the AlN films on GaN frequently develop tensile cracks at thickness values above 7 nm.³⁶ 

To further explore the residual stress trends in sputtered nitride films on different GaN substrates, 100 nm of sputtered AlN was deposited on commercially bought bulk GaN and lab grown Si-doped MOCVD GaN.¹⁰ The bulk GaN [Fig. 3(a)] exhibits a smooth, featureless microstructure (RMS = 0.2 nm), while the MOCVD GaN [Fig. 3(b)] has a typical step-and-terrace morphology (RMS = 0.6 nm). Figures 3(c)–3(d) show AFM images for 100 nm sputtered AlN on bulk and MOCVD GaN templates; tensile cracking is prominent on both. These findings align with prior AlN/n-GaN strain-relaxation studies and highlight a unique opportunity to better mitigate cracking in all-sputtered AlN/n-GaN heterostructures. Regardless of the synthesis technique, the interaction between Ga vacancies and dislocations produces a mechanism for tensile strain in n-type GaN, particularly for high carrier concentrations, which makes lattice matching to AlBScN overlayers even more challenging.³⁷ However, as previously mentioned, sputtering is an energetic process that leads to compressive growth strain that resists, in part, the tensile mechanism. In the present case, sputtering produces an in-plane lattice parameter of 3.15 Å, while the bulk and MOCVD GaN lattice values are 3.18 and 3.17 Å, respectively. The sputtered n-type GaN from this study has a smaller lattice mismatch with AlN (3.10 Å), which contributes to the structural robustness of sputtered AlN/n-GaN heterostructures even at large thickness values. Broadly speaking, these findings suggest that sputtered GaN templates could provide new compositional and thickness flexibility for designing crack-free heterostructures via strain engineering.

Last, to verify that pseudomorphic growth can extend to GaN substrates with differing film morphologies, 200 nm of Al_0.94B_0.06N was grown on a 500 nm sputter deposited GaN with a step-flow morphology and an RMS roughness of 1.4 nm [Figs. 4(a) and 4(b)]. As observed in sputtered n-GaN, the Al_0.94B_0.06N film maintains a smooth surface (RMS = 1.5 nm) and retains the underlying morphology of the GaN template. These results demonstrate that even at reduced growth temperatures of 350 °C, exceptional surface morphologies can be obtained for a variety of III–V sputtered heterostructures.

Robust ferroelectric reconfigurability is a necessity when developing III–N wurtzite heterostructures for applications in high electron mobility transistors (HEMTs) or light emitting diodes (LEDs). In this work, electrical properties of the heterostructure stacks are explored via 100 Hz polarization hysteresis measurements. Figure 5 shows polarization electric field (P–E) loops for the quaternary AlBScN/n-GaN heterostructures. Each P–E loop is square in shape and shows distinct pinch off regions in positive and negative field directions, indicating full polarization saturation. More specifically, the lattice matched and compressively strained AlBScN/n-GaN stacks exhibit remanent polarization values <135 μC/cm², while the coercive fields remain constant at ∼6.1 MV/cm. For the tensilely strained, partially relaxed AlBScN/n-GaN heterostructures, we observe remanent polarization and coercive field values of ∼185 μC/cm² and 6.7 MV/cm, respectively. In contrast, the ternary Al_0.94B_0.06N/n-GaN heterostructure exhibits electrical breakdown prior to ferroelectric switching. The observed ferroelectric behavior strongly correlates with RSM measurements—the quaternary, epitaxially strained AlBScN/n-GaN heterostructures switch, while the relaxed, ternary AlBN/n-GaN stacks break down. The authors further note that the n-type conducting GaN films used as bottom electrodes have a carrier concentration of 6 × 10¹⁹ cm⁻³ and are 200 nm thick. The total carrier density in that film, were it compressed into a sheet of charge, would be 1.2 × 10¹⁵ cm⁻². In comparison, a 150 μC/cm² electric polarization will produce 9 × 10¹⁴ cm⁻² of surface charge. In this case, sufficient charge is present to compensate polarization reversal; however, scaling epitaxial nitride ferroelectric capacitors to arbitrarily small lateral and vertical dimensions will likely require GaN films with substantially higher carrier concentrations.

A sputtered n-GaN template with an in-plane lattice constant of 3.15 Å, a leaf-like morphology (RMS = 2.4 nm), a doping concentration of 6.0 × 10¹⁹ cm⁻³, and a 0.025° GaN 002 rocking curve was successfully used to develop all-wurtzite ferroelectric Al_1−x−yB_xSc_yN/n-GaN heterostructures. As observed by XRD, lattice matched (Al_0.76B_0.06Sc_0.18N) and compressively strained (Al_0.78B_0.04Sc_0.18N) compositions are epitaxially strained to the GaN template, while the tensilely strained Al_0.85B_0.06Sc_0.09N film shows signs of partial relaxation. In comparison, the ternary Al_0.94B_0.06N film is almost fully relaxed, though conflicting XRD results suggest that a coherently strained crystallite layer may be present at the Al_0.94B_0.06N/n-GaN interface. Depositing Al(B,Sc)N films on sputtered n-GaN templates lends to isostructural growth of the ferroelectric Al(B,Sc)N layer, regardless of composition. Moreover, growing 100 nm of AlN on a sputtered n-GaN template results in a structurally robust heterostructure, while prevalent tensile cracking is observed when AlN is grown on bulk or MOCVD fabricated GaN. Quaternary Al_1−x−yB_xSc_yN compositions that are epitaxially strained and partially relaxed exhibit robust ferroelectric polarization reorientation, while the fully relaxed Al_0.94B_0.06N composition exhibits breakdown before polarization reorientation is observed. These results support the notion that a wide range of functional wurtzite heterostructures can be obtained through purposeful strain engineering of both the GaN template and/or ferroelectric layer by (i) compositionally modifying the ferroelectric III–N film or (ii) utilizing the flexible processing space provided by sputter deposition.

Sputtered n-GaN thin films were grown on c-sapphire (0001) substrates at a temperature of 750 °C. Prior to deposition, sapphire substrates were cleaned by annealing in air at 1000 °C. A liquid Ga target (99.9999%, Indium Corporation) was held in a stainless-steel dish. Doping was performed by co-sputtering from a 2-in. sputter cathode using a Si wafer (University Wafer) as a sputter target. GaN templates were developed using plasma drives at optimized Ar–N₂ gas ratios. More specifically, high power impulse magnetron sputtering (HiPIMS) sputtered GaN was fabricated at 10 mTorr with pulses of 50 μs at 120 Hz with a time-average current of 0.23 A, resulting in a power density of 3.30 W/cm³ with gas injection. The Si carrier concentration values for the n-GaN films were determined via Hall-effect measurements (HMS-3000) in the van der Pauw configuration with a 0.545 T magnet. To define the sample area, samples are cleaved post-deposition to a 7 × 7 mm².

Al_0.94B_0.06N (AlBN) and Al_1−x−yB_xSc_yN (AlBScN) thin films were grown on sputtered GaN (using a custom magnetron sputtering system featuring three 2 in. Kurt J. Lesker MagKeeper cathodes with Al (99.9995% Kurt J. Lesker), Sc (99.9%, QS Advanced Materials), and BN (99.5% Kurt J. Lesker) targets) on ammonothermal GaN single crystals (MSE supplies) and on MOCVD n-type GaN grown on (0001) sapphire.¹⁰ Al and BN targets are powered by RF-VII RF power supplies, while the Sc target is powered by an ENI RPG50 pulsed dc supply in the dc mode. The magnetrons are disposed in a confocal configuration with a 22.5° attack angle with respect to the substrate normal and the target-to-substrate distance for each gun is ∼10 cm. The substrate heater (AJA International) is modified to allow for deposition on 4-in. wafers with direct radiative heating of the wafer backside using two 1000 W halogen lamps. Ultrahigh purity Ar and N₂ process gases are passed through an in-line Entegris GateKeeper GPU for gas purification before introduction into the chamber. The GaN substrates are outgassed in the growth chamber for 30 min at 425 °C, before allowing an additional 30 min to stabilize at AlBN and AlBScN growth temperatures of 350 °C. The chamber base pressure before deposition is <2 × 10⁻⁷ Torr. The AlBN and AlBScN films are sputtered in a mixed nitrogen and oxygen atmosphere at 1.9 mTorr flowing 20 SCCM Ar and 20 SCCM N₂. The Al target power is maintained at 300 W for all depositions and the BN target power for the AlBN film is 100 W. To modify the B and Sc content in AlBScN films, the BN target power is 75 or 100 W (e.g., 100 W results in ∼6 mol. % BN), while the Sc target power is 100 or 200 W (e.g., 200 W results in ∼18 mol. % ScN). The B and Sc concentrations in each quaternary film were inferred from individual source flux calibration of the AlBN, BN, and ScN films. We note that no constituents are volatile and that the measured thickness values of the quaternary films agree with that predicted from the individual fluxes. Deposition times are varied to achieve films with thicknesses of ∼200 nm. The films are removed from the vacuum chamber when they have cooled to room temperature. To prepare samples for electrical testing, W top electrodes (∼100 nm thick) with a 100 μm diameter are deposited on the AlBN and AlBScN films through shadow masks.

Crystalline phase, orientation, and lattice parameters were measured with a Malvern Panalytical Empyrean x-ray diffractometer using a 2-bounce Ge (220) hybrid monochromator incident optic with one active channel for the θ–2θ scans and the rocking curve. Reciprocal space maps were collected with a “frame-based” area detector. Film thickness and roughness are determined via x-ray reflectivity (XRR) using a single detector channel and the PASS follow mode that matches beam size to the active area of the detector. Surface morphology and microstructure were characterized using an Asylum Research MFP-3D atomic force microscope (AFM) in the tapping mode. Polarization–electric field (PE) measurements were collected with a 1 kHz triangular wave using a PolyK ferroelectric tester with a Trek PZD350 A M/S high voltage amplifier.

C.S. is supported by Army Research Office (ARO) Research (Contract No. W911NF-24-1-0010), which developed methods for pseudomorphic epitaxial nitride growth. J.N. gratefully acknowledges support from the Department of Defense (DoD) through the National Defense Science and Engineering Graduate (NDSEG) Fellowship Program. J.-P.M. and J.H. acknowledge the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences Energy Frontier Research Centers Program under Award No. DE-SC0021118, which conceptualized and developed deposition methods, instrumentation, and optimization of the AlN-BN-ScN ternary system.

The authors have no conflicts to disclose.

Chloe Skidmore: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (equal); Methodology (equal); Validation (equal); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Josh Nordlander: Conceptualization (equal); Data curation (supporting); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (supporting). John Hayden: Conceptualization (supporting); Investigation (supporting); Methodology (supporting). Anthony Rice: Resources (equal). Ramón Collazo: Resources (equal). Zlatko Sitar: Resources (equal). Jon-Paul Maria: Conceptualization (lead); Formal analysis (equal); Funding acquisition (lead); Methodology (equal); Project administration (lead); Resources (lead); Supervision (lead); Validation (supporting); Visualization (supporting); Writing – review & editing (equal).

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