Toward AlGaN channel HEMTs on AlN: Polarization-induced 2DEGs in AlN/AlGaN/AlN heterostructures

Due to its high breakdown electric ﬁeld, the ultra-wide bandgap semiconductor AlGaN has garnered much attention recently as a promising channel material for next-generation high electron mobility transistors (HEMTs). A comprehensive experimental study of the effects of Al composition x on the transport and structural properties is lacking. We report the charge control and transport properties of polarization-induced 2D electron gases (2DEGs) in strained AlGaN quantum well channels in molecular-beam-epitaxy-grown AlN/Al x Ga 1 − x N/AlN double heterostructures by systematically varying the Al content from x = 0 (GaN) to x = 0.74, spanning energy bandgaps of the conducting HEMT channels from 3.49 to 4.9 eV measured by photoluminescence. This results in a tunable 2DEG density from 0 to 3.7 × 10 13 cm 2 . The room temperature mobilities of x ≥ 0.25 AlGaN channel HEMTs were limited by alloy disorder scattering to below 50 cm 2 /(V.s) for these 2DEG densities, leaving ample room for further heterostructure design improvements to boost mobilities


ARTICLE
scitation.org/journal/apm barrier layer can lead to large gate leakage currents. Therefore, for high voltage switching applications where reverse-bias gate leakage is a pressing issue, the AlGaN channel composition needs to be carefully optimized. Aluminum nitride (AlN) is an emerging platform for mm-wave integrated circuits (MMICs) 7 due to its ultrawide and direct bandgap (6 eV), high thermal conductivity (∼340 W/mK), and high piezoelectricity. AlN/GaN/AlN HEMTs 8-12 on this platform, which consist of thin and strained GaN quantum wells (QWs) hosting 2D electron gases (2DEGs) surrounded by AlN buffer and barrier layers, provide enhanced thermal management and improved breakdown compared to conventional GaN HEMTs. These AlN HEMTs on SiC have recently demonstrated high mm-wave output powers of >3 W/mm at 10 GHz 13 and 2 W/mm at 94 GHz. 14 In this work, we wish to combine the benefits offered by the AlN platform with the merits of an AlGaN channel to realize AlN/AlGaN/AlN double heterostructures and study the physics of polarization induced charges in them. The study of the crystal and electrical transport properties of such structures is needed to determine their potential for applications in RF and power electronics. What is interesting about this AlN/AlGaN/AlN heterostructure is that the whole material stack, consisting of a conductive AlGaN quantum well active region embedded in an insulating AlN matrix, has a very high bandgap, thus enabling access to the highest breakdown electric fields and operating temperatures, beyond the reach of GaN HEMTs. 15,16 Attaining an ultra-wide bandgap device structure while still having a highly conductive AlGaN channel constitutes a heterostructure design trade-off. For metal-polar AlN/AlGaN/AlN double heterostructures grown along the polar c-axis, the discontinuous electrical polarization P(z) across the AlN/AlGaN top interface gives rise to a polarization sheet charge qσπ = − P(z) ⋅ẑ at that interface. Here, P(z) is the polarization discontinuity at the heterojunction interface, q is the absolute value of the electron charge, andẑ is the unit vector along the growth direction, which in this case is the positive c-axis. AlGaN grown over AlN is under a compressive strain and, thus, has a piezoelectric component of polarization opposing the spontaneous one. Because the top AlN has a higher polarization than AlxGa 1−x N (where x is the Al mole fraction), the net unbalanced bound polarization charge qσ π (x) at the AlN/AlGaN interface is positive and decreases with x. 17,18 This positive bound charge creates a high electric field and energy-band bending such that a mobile 2DEG forms at the AlN/AlGaN heterointerface. Similarly, a 2D hole gas (2DHG) should be formed at the bottom AlGaN/AlN interface due to the negative bound charge −qσ π (x) present there. Because of its polarization-generated nature, the 2DEG density should electrostatically depend on the thicknesses of the layers and the composition (x) of the AlGaN channel and not on temperature. Furthermore, the conduction band offset EC(x) between AlN and AlxGa 1−x N, which confines the 2DEG, also decreases with x. 18 Consequently, for fixed thicknesses of the AlN barrier layer t b and AlGaN channel layer tQW, the 2DEG density should decrease with increasing Al composition x of the AlGaN channel. This tunability offered by the Al composition for the polarization control of the 2DEG density is of extreme importance in the design space of AlGaN channel polar heterostructures for transistor applications. Abid et al. 19 and Maeda et al. 20 recently demonstrated the epitaxial growth and promising HEMT device performance of AlN/Al 0.5 Ga 0.5 N/AlN double heterostructures grown by metalorganic chemical vapor deposition (MOCVD) and pulsed sputtering deposition, respectively. The compositional dependence of 2DEG properties in such AlGaN channel HEMT heterostructures has not been examined yet.
In this work, we report a systematic study of the structural, optical, and electronic transport properties of AlN/AlGaN/AlN double heterostructures with the AlxGa 1−x N channel composition ranging from x = 0 to x = 0.74, grown using molecular-beam-epitaxy (MBE). Photoluminescence spectra show that the bandgap of the AlxGa 1−x N channel increases with x. Charge control observed in these heterostructures by AlGaN composition tuning agrees well with the polarization model, with 2DEG densities varying from ∼3.7 × 10 13 cm 2 in a GaN channel to no measurable 2DEG in an Al 0.74 Ga 0.26 N channel of thickness 24 nm. In addition, temperature-dependent electronic transport measurements reveal that alloy disorder scattering is the dominant scattering mechanism in the AlGaN channel 2DEGs, with an alloy fluctuation potential of ≥1.8 eV for electrons in AlGaN.
AlN/AlGaN/AlN double heterostructures were grown on the Si-face of semi-insulating 6H-SiC substrates using a Veeco GEN10 MBE system equipped with one standard effusion cell for elemental Ga, two cells Al1 and Al2 for elemental Al, and a radio frequency plasma source for the active N species. N plasma power of 400 W, corresponding to a growth rate of 0.42 m h −1 , was used. A 1 m thick unintentionally doped (UID) AlN buffer layer was epitaxially grown on SiC with a nucleation layer of 50 nm that was grown in a slightly N-rich condition with Al (ϕ Al ) to an active N flux (ϕ N ) ratio ϕ Al : ϕ N of 0.9, while the remaining 950 nm AlN buffer layer was grown in metal-rich conditions with a ϕ Al : ϕ N of 1.1. The Nrich nucleation layer blocks the Si up diffusion from the SiC surface and prevents it from incorporating in the active region. 21,22 This was followed by the growth of tQW = 24 nm thick strained AlGaN quantum well to host the 2DEG channel. The Al composition (x) in the AlxGa 1−x N channel was systematically varied from x = 0 to x = 0.74 (calibrated by x-ray diffraction as discussed later) across six samples by changing the ϕ Al with respect to ϕ N while keeping the Ga flux (ϕ Ga ) constant. The relative flux conditions of ϕ Al + ϕ Ga > ϕ N , ϕ Al < ϕ N were maintained to ensure N-limited (or metalrich) AlGaN MBE growth conditions. An AlN layer of thickness t b = 15 nm was grown on the AlGaN channel as the top barrier. The AlN buffer layer and the top AlN barrier layer were grown using the Al1 cell, whereas the AlGaN channel layer was grown using Al2 cell to avoid growth interrupts. The entire heterostructure was grown at a constant substrate thermocouple temperature of 750 ○ C. Figure 1 X-ray diffraction (XRD) was used to determine the structural quality of the epitaxial structures and the average Al composition x of the AlGaN channel. Figure 1(b) shows the measured symmetric (002) reflection XRD spectra of the six samples with increasing Al content x, along with the simulated diffraction patterns based on the layer structure shown in Fig. 1(a). The AlGaN peaks are clearly resolved to the left of the substrate SiC peak for all the samples. The clearly visible thickness fringes suggest abrupt hetero-interfaces in the structure. Fitting measured spectra with the simulation APL Materials x as well as the thickness of the AlGaN QW layer tQW. Quantification of strain in the AlGaN channel layer is critical for HEMT heterostructure design as it dictates the maximum thickness of the AlN barrier layer that can be grown without cracking. 12 Moreover, it influences the carrier transport properties as it not only determines the piezoelectric polarization, which, in turn, affects the 2DEG carrier density, but also modifies the electron effective mass and, therefore, the mobility.
To experimentally investigate the strain in the AlGaN QW as a function of its Al composition, reciprocal space mapping (RSM) using XRD was conducted on the AlN/AlGaN/AlN heterostructures around the SiC (1 1 12 QW are also included. It is clearly seen that both the strain and the relaxation of the AlGaN QW decrease as x increases. The strain shows a monotonic reduction, while the relaxation drops to 0 for x ≥ 0.44. The dashed gray line in Figs. 3(a) and 3(b) show the theoretical dependence of εxx and R with x if the AlGaN layer is grown pseudomorphically to the AlN buffer. These results indicate that higher composition (x ≥ 0.44) AlGaN QW can be grown even thicker than 24 nm over AlN without significant relaxation, opening up opportunities for further improvements in the heterostructure design. In addition, the screw threading dislocation densities in the samples were estimated from the full width at half maximum (FWHM) values of the x-ray rocking curves (XRCs) along the AlGaN(0002) plane. The GaN QW sample has a screw dislocation density of ∼1.6 × 10 9 cm 2 , which decreases to ∼1.1 × 10 8 cm 2 for the Al 0.74 Ga 0.26 N QW sample.

ARTICLE
scitation.org/journal/apm taking into account a bowing parameter of 1 eV. 23 The difference between the measured and predicted values, specifically the PL peak broadening, intensity suppression, and peak shifts, are attributed to properties including a variability in optical absorption coefficient, quantum confinement, strain, compositional fluctuation, and polarization-induced quantum-confined Stark effect (QCSE). Thus, the AlN/AlGaN/AlN samples have the desired structures in which the electrical transport properties of 2DEGs can be studied systematically as a function of Al composition x. As discussed previously, the sheets of charge ±qσ π (x), located at the AlN/AlGaN heterointerfaces, induce a rearrangement of free carriers. Theoretically, this rearrangement generates a 2DEG at the top interface and a 2DHG at the bottom one. A self-consistent 1D Schrödinger-Poisson energy band diagram simulation for the heterostructures under study is shown in Fig. 5(a) for three samples with GaN, Al 0.44 Ga 0.56 N, and Al 0.74 Ga 0.26 N QW channels, with a surface barrier height qϕ B = 2.2 eV. With increasing Al concentration x, the bandgap inside the QW widens, thereby pushing up the conduction band edge EC with respect to the Fermi level EF. The 2DEG density decreases commensurately with increasing x. Eventually, as shown in Fig. 5(a), EF should move into the bandgap, as is the case for Al 0.74 Ga 0.26 N QW channel. At and beyond that x, there should be no 2DEG in an as-grown heterostructure for the given layer thicknesses.
Hall-effect measurements were performed on the samples using soldered corner indium contacts to the AlGaN QW 2DEGs. The results of the Hall-effect measurements are summarized in Table I. All the samples except the one with Al 0.74 Ga 0.26 N QW showed Hall conductivity, indicating the presence of 2DEGs. As discussed earlier, the polarization induced 2DEG is most likely absent in the Al 0.74 Ga 0.26 N QW sample as the polarization difference between the top AlN and Al 0.74 Ga 0.26 N layers is not large enough to induce the 2DEG for the given thicknesses of the layers. No direct evidence of the 2DHG formed at the bottom AlGaN/AlN interface was observed in the Hall-effect measurements for the samples reported here. Figure 5(b) compares the experimentally measured 2DEG Hall densities in the AlN/AlGaN/AlN samples to the simulated 2DEG densities ns from a self-consistent 1D Schrödinger-Poisson calculation. The 2DEG densities ns(x) are calculated as a function of AlGaN channel composition x for different surface barrier heights qϕ B of 1.2, 2.2, and 3 eV. As the Al mole fraction in the channel increases, the measured 2DEG density decreases from 3.78 × 10 13 cm 2 in the GaN channel to the Al 0.74 Ga 0.26 N channel sample, which shows no 2DEG, in agreement with the simulated results. A surface barrier height of qϕ B = 2.2 eV results in the best fit to the experimental data. The surface Fermi level is not pinned in AlN and a fixed surface barrier height qϕ B may not necessarily explain the experimental data well over the entire range of samples with varying Al composition x in the AlGaN QW channel. For AlGaN/GaN HEMTs, it has been previously observed that qϕ B is dependent on carrier concentration 24 and the AlGaN barrier thickness. 25 The unpinned surface Fermi level in the AlN barrier layer presents an attractive opportunity to employ work function engineering to tune threshold voltages of transistors.

ARTICLE scitation.org/journal/apm
A good electrostatic control over the 2DEG densities is thus demonstrated by varying the Al composition of the AlGaN layer in as-grown heterostructures. The higher the Al composition x, the higher the AlGaN channel bandgap and hence higher the expected electrical breakdown field. However, the 2DEG density decreases with an increase in x. A high 2DEG density is desired in RF HEMTs to obtain high on-currents and low access resistances. Hence, this trade-off between channel breakdown and on-current needs to be carefully optimized for designing an AlGaN-channel HEMT. This study, therefore, provides valuable experimental data for guiding the heterostructure and device design for AlGaN-channel HEMTs for high-power, high-frequency applications.
It is worth mentioning here that even though the as-grown heterostructure with the Al 0.74 Ga 0.26 N channel is highly resistive, it can potentially enable normally off operation with high threshold voltage, which is desirable for power switching and energy conversion applications. Previous reports 26,27 have theoretically simulated the enhancement mode operation in such high composition AlGaN HEMTs; however, the experimental demonstration of such devices remains to be realized yet. Clever device engineering would be needed to alleviate the effects of increased access region resistance originating from the reduced polarization difference between the barrier and channel in such E-mode HEMTs. As an example, selective-area ion implantation of Si in the access region of the AlGaN channel could help solve this problem.
A temperature-dependent electron transport study was performed for all the AlGaN QW channel samples that showed conductivity. Figures 6(a) and 6(b) show the temperature-dependent Hall-effect measurement results from 300 K down to 10 K. In all samples, the 2DEG density does not show significant variation with temperature. This robustness of the 2DEG densities with temperature confirms their polarization-induced origin.
The electron mobility in the GaN and Al 0.1 Ga 0.9 N QW channel samples increases monotonically upon lowering the temperature, as expected due to freeze-out of phonon scattering. However, due to additional alloy scattering, the 10 K mobility in Al 0.1 Ga 0.9 N QW is limited to ∼360 cm 2 /(V.s) compared to ∼630 cm 2 /(V.s) in GaN QW. The rest of the samples with Al 0.25 Ga 0.75 N, Al 0.44 Ga 0.56 N, and Al 0.58 Ga 0.42 N channels show nearly constant mobilities 50 cm 2 /(V.s) over the whole temperature range, indicating that temperature independent alloy scattering is the dominant scattering mechanism. Other scattering mechanisms, such as dislocation scattering, surface charge scattering and interface roughness scattering, were found to be much weaker than alloy scattering in the AlGaN channel for the whole temperature range.
The mobilities of MBE-grown x ≥ 0.25 AlGaN channel HEMTs in this study are lower than the values reported previously in MOCVD-grown high Al composition AlGaN channel HEMT structures. [28][29][30] For example, Baca et al. 29 measured a mobility of 250 cm 2 /(V.s) in AlN/Al 0.85 Ga 0.15 N HEMT for a carrier density of 6 × 10 12 cm 2 , and Xue et al. 30 reported a mobility of 175 cm 2 /(V.s) for a carrier density of 8.5 × 10 12 cm 2 in the Al 0.75 Ga 0.25 N/Al 0.6 Ga 0.4 N HEMT structure. The observation of lower carrier mobilities in our films may partly be attributed to higher alloy scattering due to comparatively higher electron densities in our HEMT heterostructures. It is also possible that the MBE-grown AlGaN films in the current study suffer from relatively higher lateral alloy fluctuations, hence causing AlN/GaN energy barriers for lateral electron transport. 31 For electrons in the AlGaN QW, the mobility decreases with the increasing Al composition (decreasing by up to 90% in higher composition AlGaN compared to GaN) due to the combined effects of alloy scattering and the increase in the Γ valley conduction band edge electron effective mass. Quantitatively, under a single parabolic conduction band approximation, the alloy disorder limited mobility alloy of 2D electron carriers in an alloy channel with composition x is given by 32  are the interpolated in-plane and out-of-plane lattice constants of AlGaN. UAL is the alloy fluctuation potential and is typically estimated by fitting to experimental mobility measurements. In the absence of experimental data, UAL is approximated to be the conduction band offset EC or bandgap difference Eg between the constituents of the ternary alloy, 33,34 which in our case are GaN and AlN. For AlGaN, UAL has been estimated to be between 1.3 and 2.2 eV by various theoretical and experimental methods. [34][35][36][37][38] The series of polarization induced, undoped AlN/AlGaN/AlN 2DEG samples used in this study offer an opportunity to quantify UAL for AlGaN to understand the physical limits of 2DEG mobilities in AlGaN channel HEMTs.
Using Eq. (1), alloy is calculated for AlN/AlxGa 1−x N/AlN heterostructures as a function of Al content x = 0-0.8 for various shortrange alloy scattering potentials UAL = 1, 1.8 and 2.1 eV. For mobility calculations, ns(x) is obtained from the 1D Schrödinger-Poisson solver using a surface barrier height qϕ B of 2.2 eV, as discussed earlier in Fig. 5(b). The results of calculated alloy are plotted in Fig. 6(c) along with the measured 10 and 300 K Hall mobilities in the AlN/AlGaN/AlN samples as a function of Al composition x. Compositional dependence of the measured 10 K mobilities in the AlGaN channel fall on U-shaped curve characteristic of alloy scattering. Upon comparing experimental data with calculated alloy [gray curves in Fig. 6(c)], it is predicted that UAL 1.8 eV. It should be noted that the 2DEG mobilities in these AlN/AlGaN/AlN double heterostructures could have been adversely affected by the 2DHG formed at the bottom AlGaN/AlN heterojunction due to proximity phenomena, such as Coulomb drag effect, as has been suspected earlier for AlN/GaN/AlN heterostructures. 8,9 Thus, the low-field mobility in AlGaN channel 2DEGs is a strong function of alloy composition, and it degrades with increasing Al composition x. How do we increase the mobilities of these 2DEGs? Since the mobility in the AlGaN channel is intrinsically bound by the limits set by the statistical disorder in the alloy system, the low-field conductivity in such alloyed channels can be potentially boosted by switching to a (GaN)n/(AlN)m "digital alloy" scheme. Such a superlattice alloying method for the AlGaN-channel has been recently proposed theoretically by Pant et al. 39 to diminish the in-plane disorder scattering while still maintaining the high breakdown characteristics. The mobility is also expected to improve by moving to bulk single-crystal AlN substrates, which have less defects and dislocations compared to AlN films grown on foreign substrates, such as Si, SiC, and sapphire. Recently, high quality AlN homoepitaxy has been demonstrated both on the Al-face 40,41 and N-face 42 of native AlN substrates, which can enable further advancements in AlGaN channel HEMTs. Finally, most of the work to date has been focused on metal-polar AlGaN channel transistors. The N-polar AlGaN/AlN heterostructure, which unlike the metal-polar AlN/AlGaN/AlN heterostructure hosts a 2DEG in the AlGaN channel without an accompanying 2DHG, offers the possibility of even further mobility enhancement.
In summary, this work presents a systematic study of MBEgrown AlN/AlGaN/AlN double heterostructures with thin and strained AlxGa 1−x N quantum well channels surrounded by AlN. The material characteristics of the heterostructures and the transport properties of the 2DEGs in AlGaN channels have been examined as a function of the Al composition 0 ≤ x ≤ 0.74. Photon emission from the AlGaN QWs was observed in the photoluminescence spectra. The bandgap increased with x ranging from 3.49 to 4.9 eV. The density of the 2DEG could be manipulated from 0 to 3.7 × 10 13 cm 2 by changing x while keeping the thicknesses of the layers constant. Thus, these AlGaN channel HEMT heterostructures lay the groundwork for new generations of high-power RF and mmwave HEMTs. These structures would also be useful for pushing the breakdown voltages higher for power switching applications. More importantly, the AlN platform allows for the integration of nitride RF and mm-wave amplifiers with components, such as p-channel FETs, and RF and mm-wave filters as described by Hickman et al. 7 The authors thank Jimy Encomendero, Yongjin Cho, John Wright, and Len van Deurzen for useful discussions.

Conflict of Interest
The authors have no conflicts to disclose.

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