Tuning composition in graded AlGaN channel HEMTs toward improved linearity for low-noise radio-frequency amplifiers Open Access

Today's rapid development of radio-frequency (RF) communication technology requires utilization of optimized devices to meet the demands of high-frequency power amplifiers.¹ AlGaN/GaN high electron mobility transistors (HEMTs) have been widely explored for RF technology applications, delivering high power and high gain. However, a weak point of such devices is their inherent non-linear behavior at high power input signal, which results in gain compression and signal distortion.² Modifying the two-dimensional electron gas (2DEG) confinement alters the 2DEG distribution and results in formation of a three-dimensional electron gas (3DEG) in the HEMT channel.³ This modification is predicted to improve the small-signal linearity rendering devices with characteristics relevant for use in low-noise amplifiers (LNAs) as well as providing improved large-signal linearity. Indeed, a spatially broader electron distribution improves the HEMT linearity,^2,4–8 rendering a constant cutoff frequency f_T and maximum oscillation frequency f_max over a wider range of input power. 3DEGs in AlGaN/GaN heterostructures can be formed through polarization doping by compositionally grading with Al the GaN HEMT channel.⁹ 

Although graded channel HEMTs are emerging as very promising devices for linear RF LNAs,^10–13 the understanding of the channel grading effects on the 3DEG properties is still limited. For example, there is no information regarding the actual Al grading profiles and their interrelation with charge carrier properties. Moreover, from a practical point of view, the development of the growth process for graded channel HEMTs is not trivial. Precise control of the compositional grading (variation of the Al-content) and the thickness (in the range of 10 nm or less) of the graded channel layer are required in order to achieve a 3DEG appropriate for improved device linearity. Additionally, achieving sharp interfaces between adjacent layers in the HEMT structure is critical for the transport properties and device performance. The aforementioned practical difficulties render the metal-organic chemical vapor deposition (MOCVD) of graded channel HEMT structures experimentally challenging. Thus, a comprehensive study combining insights on the material growth with device simulations and realization is required for getting a better understanding of the graded channel HEMT performance and structure optimization.

In this work, we present various approaches to realize graded channel AlGaN/GaN HEMT structures grown by MOCVD. We develop the growth process of a graded channel with an Al-content from 0% to 10% with three different grading profiles: (i) exponential grading, (ii) hybrid hyperbolic tangent—linear grading, and (iii) linear grading, combined with further tuning of the channel-to-barrier transition. The effect of introducing a thin AlN interlayer between the graded channel and the barrier layers on the HEMT transport properties is also investigated. The experimentally measured 2DEG/3DEG properties are discussed in comparison with simulations of the carrier density distribution related to the type of grading and the interface structure. Finally, two HEMT devices—an optimized linearly graded channel HEMT and a conventional non-graded AlGaN/GaN HEMT—are fabricated and compared in terms of device performance.

Graded channel and conventional AlGaN/GaN HEMT structures (schematics shown in Fig. 1) were grown in a horizontal hot-wall MOCVD reactor (VP508GFR, Aixtron) on semi-insulating 4H–SiC substrates with (0001) orientation. The growth process was performed under a 2 l/min constant ammonia (NH₃) flow and a mixture of N₂ and H₂ as carrier gases. NH₃, trimethylaluminum (TMAl), and trimethylgallium (TMGa) were used as the N, Al, and Ga precursors, respectively. Prior to the growth, the SiC substrates were etched under high-temperature, 1340 °C, and H₂-rich environment for step-like surface preparation. Thereafter, a 60 nm thick AlN nucleation layer was grown at 1250 °C with a V/III ratio of 1260. All the layers above the AlN nucleation layer were grown at a temperature of 1070 °C and a constant pressure of 100 mbar. The subsequent 1.1 μm thick unintentionally doped (UID) GaN buffer layer was followed by a graded channel layer in the case of samples S₁–S₅, while samples S₆–S₈ have conventional non-graded GaN channels. The barrier layer of all samples (S₁-S₈) consisted of a (11–15)-nm-thick Al_0.30Ga_0.70N layer. The growth was finished with a 2–3 nm GaN cap layer for samples S₃–S₈. An AlN interlayer was added between the channel and the barrier layer in the case of samples S₄, S₅, S₇, and S₈. During the AlN interlayer growth, the Al precursor (TMAl) flow was maintained for 15 s (S₄ and S₇) and 30 s (S₅ and S₈).

Three different approaches were implemented for the growth of the graded channel layer in samples S₁–S₃ in order to achieve distinctively different types of grading. The TMGa precursor flow was kept constant while the TMAl flow was varied: (i) TMGa flow was at 2.2 ml/min, and TMAl flow varied from 0 to 0.23 ml/min over 33 s (sample S₁), (ii) TMGa flow was at 1.4 ml/min, and TMAl flow varied from 0.05 to 0.19 ml/min over 105 s (sample S₂), and (iii) TMGa flow was at 1.4 ml/min, and TMAl flow varied from 0.02 to 0.13 ml/min over 110 s (sample S₃).

The surface root mean square (RMS) roughness of the structures was lower than 0.45 nm over a 5 × 5 μm² area according to the atomic force microscopy (AFM) measurements on samples S₁–S₃ (see supplementary material Fig. S1). The low surface roughness is indicative of the smoothness of the channel/barrier interface where the 2DEG/3DEG is formed. In order to determine the thicknesses of the AlN nucleation layer, the GaN buffer layer, and the Al_xGa_1−xN barrier layer, variable angle spectroscopic ellipsometry (VASE) data were measured and fit to an appropriate optical model. The VASE measurements were performed on a J. A. Woollam RC2-XI ellipsometer in the near-infrared to ultraviolet spectral range (⁠ $0.7 − 5.9$ eV). Note that the thickness and composition of the graded Al_xGa_1−xN layer are taken from the scanning transmission electron microscopy (STEM) and energy-dispersive x-ray spectroscopy (EDS) measurements as input for the VASE data analysis (see the supplementary for further details). Reciprocal space maps (RSMs) around the GaN asymmetric $10 1 ¯ 5$ reciprocal lattice points were recorded using a PANalytical Empyrean X-ray diffractometer in 2-axis configuration with a hybrid monochromator at the incident X-ray beam and the detector in the scanning line mode (see supplementary material Fig. S3). The Al-content in the Al_xGa_1−xN barrier layers was determined from the a and c lattice parameters obtained from the RSMs assuming validity of Vegard's law¹⁴ and using the strain-free lattice parameters of GaN and AlN.^15,16 The RSM measurements also revealed that the AlGaN epilayers were pseudomorphically grown to the thick GaN buffer layer for all Al compositions. The density of screw and edge type dislocations in the GaN buffer layers was evaluated from the respective tilt and twist angles using the lattice parameters from Ref. 15 and following the methods described by Metzger et al.¹⁷ and Srikant et al.¹⁸ The dislocation densities in the very thin (up to 10 nm) graded layers are considered similar to those in the thick GaN buffer layer and are found to be $2.5 × 10 7 cm − 2$ for screw and $4.7 × 10 8 cm − 2$ for edge type dislocations.

The 2DEG/3DEG mobility μ and sheet density $N s$ were obtained at room temperature by a contactless microwave (10 GHz) based method referred to as contactless Hall (Lehighton LEI1600). For samples S₁ and S₂, the 2DEG properties and the sheet resistance were determined by cavity-enhanced terahertz optical Hall effect (THz-OHE) measurements due to sample size limitations of the Lehighton instrument. The THz-OHE measurements were performed at room temperature and magnetic field of ±0.6 T,^19,20 and the experimental results were analyzed using the approaches described in Refs. 19 and 21. The THz-OHE and contactless Hall techniques were shown to deliver consistent values of the 2DEG properties for samples with relatively high sheet resistance, such as for samples S₁–S₂. In our case, this allows for meaningful comparisons between the samples that were measured by the two different techniques. The sheet resistance of the samples was measured by a contactless Eddy current-based method using an Eichhorn Hausmann MX604 tool with a working range of 50–3000 Ω/sq. The Al compositional profiles and quality of the interfaces were studied by STEM combined with EDS. Analysis was performed using the double corrected Linköping FEI Titan³ 60–300 microscope, operated at 300 kV. The built-in Super-X/Quantax EDS system (Bruker) was employed, and the absolute quantifications were made using the Esprit software and its built-in calibrations for TEM-EDS. The $∼ 30 %$ Al measured in the barrier, by EDS, is benchmarked to the corresponding Al-content extracted from RSM, which is also $∼ 30 %$ Al. Hence, it acts as its own reference, and the absolute values obtained from EDS are deemed reliable. Four EDS maps from four different sites per sample were acquired and integrated along the interface and subsequently quantified. The profiles were finally averaged for each sample for more reliable data. The structural and 2DEG/3DEG properties of all samples are summarized in Table I.

Band bendings and charge density distribution in the HEMT structures were obtained from solving self-consistently Poisson and Schrödinger equations, utilizing the numerical solver by Snider^22,23 The Al-profiles from the EDS measurements were used for the simulations, where a piece-wise linear approximation was adopted for samples S₁ and S₂, and single linear grading in agreement with the experimentally observed trends was employed for samples S₃–S₅. The AlN interlayers were also taken into account in the simulations accordingly, while the SiC substrate and the AlN nucleation layer were not considered.

Figure 2 shows the Al-profiles obtained from the EDS measurements across the graded channel and barrier layers for samples S₁, S₂, and S₃. If the grading in the channel starts directly from x = 0 and with the TMGa flow used for the GaN growth (S₁), the Al-content increases very slowly to ∼2% in the relatively fast growth process (33s) before reaching the intended x = 0.1. This results in an exponential channel grading for S₁. The HEMT structure can also be viewed as consisting of a ∼10-nm-thick AlGaN channel with 1%–2% of Al and a diffuse interface to the barrier layer. On the other hand, if the grading starts from a non-zero value, i.e., with a TMAl/TMGa ratio of 3.6% and a TMGa flow value employed for the growth of the barrier layer (sample S₂), there is fast on-set of the Al grading, and the Al-content increases up to 15% across the channel thickness (Fig. 2). In this case, the S₂ channel has a hybrid linear-tanh grading profile. Finally, keeping the same TMGa but reducing the initial and final TMAl/TMGa ratios to 1.4% and 9.4%, respectively, results in a linearly graded Al_xGa_1−xN channel from x = 0 to x = 0.1 for S₃.

The 2DEG/3DEG density, $N s$⁠, in the samples with the different types of channel grading (S₁–S₃) is found to be similar, in the range of $( 8 − 10.5 ) × 10 12 cm − 2$ (Table I). This can be expected since the graded channel layers have similar thickness and similar Al-content in barrier layers. The mobility parameters were found to be $964 cm 2 / V s$ for S₁, $907 cm 2 / V s$ for S₂, and $943 cm 2 / V s$ for S₃, which are significantly lower as compared to the typical mobility values in conventional non-graded HEMTs (see S₆ Table I and, e.g., Ref. 24). The latter is a result of the quasi-3D nature of the electrons and the enhanced alloy scattering²⁵ in the graded channel. In addition, the channels in S₁–S₃ are directly interfacing the barrier layers, which may result in a certain penetration of the 2DEG/3DEG electron wavefunction into the barrier layer, further enhancing the already present alloy scattering. The slightly higher 2DEG mobility for the exponentially graded channel structure (S₁) may be attributed to lesser alloy scattering within the AlGaN channel having low Al-content for a larger proportion of its thickness as compared to samples S₂ and S₃. Note that for S₁, the definition of the channel thickness is to some extent arbitrary. However, the larger error bars of the mobility parameters for samples S₁ and S₂ prevent a more solid conclusion.

The simulated electron density distribution for the graded channels using the measured Al-profiles and thicknesses is compared with the respective profile for the non-graded channel HEMT structure with no AlN interlayer, sample S₆, in Fig. 3(a). The carrier distribution in the exponentially graded channel (S₁) has an extended tail, but it is overall similar to that of the conventional non-graded channel sample (S₆). Hence, no big influence on the transconductance flatness may be expected for a device based on this sample. The carrier distribution in the sample with hybrid profile (S₂) resembles that of the linearly graded sample (S₃) with a clear 3D nature of the electron gas spanning over 15–20 nm. A comparison between the 3DEG properties of these two samples (S₂ and S₃) shows that the linearly graded channel HEMT structure (S₃) has a higher mobility compared to the hybrid graded sample (S₂) (Table I).

In addition, S₃ has a sharper channel-barrier interface compared to sample S₂ (see Fig. 2), which is expected to result in a lesser penetration of the 3DEG in the barrier layer and, consequently, in a reduced alloy scattering. The slightly higher 3DEG density of S₃ is probably related to the slightly thicker barrier layer. Therefore, the linear channel grading profile (sample S₃) was selected for further optimization.

A widely employed method for improving the 2DEG mobility in conventional non-graded channel HEMTs is the introduction of a very thin AlN interlayer between the channel and the barrier layer, which reduces the alloy scattering in the barrier layer. Such a thin AlN layer acts as sharpening of the channel-barrier interface and, at the same time, provides an effective energy barrier confining the electrons in the channel region. Due to the large bandgap of AlN, the conduction band discontinuity at the interface is increased, and the electron wavefunction is better confined in the channel layer. Moreover, it was shown that the incorporation of an AlN interlayer is beneficial for short recovery time during operation of HEMTs used in RF LNAs because it prevents hot electrons from being captured in the bulk AlGaN or at surface traps.²⁶ Indeed, for our conventional (non-graded) HEMT structures with AlN interlayers of 0.5 nm (S₇) and 1.5 nm (S₈), the mobility parameters increase noticeably as compared to the structure without an interlayer (S₆) (Fig. 4 and Table I). It is seen that the mobility can be improved more than 25% from $∼ 1860$ to $∼ 2370 cm 2 / V s$ with increasing the AlN interlayer thickness from 0 to 1.5 nm. These results are in agreement with previous reports.^24,27

However, there are no studies in the literature considering the combination of an AlN interlayer with graded channel in HEMT structures. In order to investigate the effect of AlN interlayer incorporation on the 2DEG properties of a graded channel HEMT structure, two samples with AlN interlayes with thicknesses of 0.5 nm (S₄) and 1.5 nm (S₅) were fabricated adopting the growth conditions for the 10 nm thick linearly graded channel sample (S₃). In fact, the two AlN thicknesses may be considered as a smoother (0.5 nm for S₄) or sharper (1.5 nm for S₅) onset of the Al-content and of the transition to the Al_0.30Ga_0.70N barrier layer, respectively. The simulated electron density profiles and the 2DEG/3DEG properties of the graded channel HEMT structures with different thicknesses of the AlN interlayer are presented in Figs. 3(b) and 5, respectively, together with the corresponding results for S₈. It is seen from Fig. 3(b) that a larger electron density is expected in the graded channel layer when an 1.5 nm AlN interlayer is added prior to the Al_0.30Ga_0.70N barrier. This is also reflected in the measured $N S$ data for sample S₅ (see Table I). In addition, from the simulations in Fig. 3(b), it is expected that the sample with the thicker AlN interlayer will show improved 2DEG transport properties since the largest portion of carriers is confined close to the channel-barrier interface. Consequently, an abrupt interface would reduce the alloy disorder scattering in the high Al-content AlGaN barrier. However, the mobility in the linear profile HEMT structure with the 1.5 nm thick AlN interlayer (S₅) is very similar to the mobility of its counterpart without AlN interlayer (S₃), indicating that the predominant effect on mobility is the alloy scattering in the Al-containing channel. However, the sheet resistance (⁠ $R sh$⁠) of S₅ is lower than that of S₃, implying a partial improvement on the carrier confinement in the channel. This observation is also supported by the simulated electron density profiles in Fig. 3(b), where the 2DEG penetration in the barrier is minimized for S₅ compared to S₃. Nonetheless, the improvement of the 2DEG properties as a result of incorporating an AlN interlayer is only marginal for the graded HEMT structures. The measured resistance of $686 Ω /$sq for the graded channel (S₅) HEMT structure remains significantly higher than the respective value of $290 Ω /$sq for the conventional counterpart S₈.

In order to evaluate the direct current (DC) performance, HEMT devices were fabricated on the conventional non-graded and the linearly graded channel structures (S₈ and S₅). For the HEMTs fabrication, a “passivation-first” process scheme was applied with an in situ NH₃ pretreatment, to reduce surface native oxide and minimize surface-related trapping effects,²⁸ followed by a deposition of the 15 nm SiN passivation layer with low-pressure chemical vapor deposition. The device isolation was achieved through mesa etching. Recessed and Ta-based Ohmic contacts^29,30 produced contact resistances of $0.30 − 0.35 Ω$ mm as extracted from transmission line measurements. The gate process includes two electron beam lithography (EBL) steps. The first step defines the etch mask for the recess in the SiN, which defines the gate length. The etching of the SiN passivation was performed by a combination of CF⁴ + Ar plasma etching followed by NF₃ plasma etching. The second EBL step defines the pattern for the gate metallization (Au/Pt/Ni) with an integrated source and drain field plate of 0.1 and 0.25 μm, respectively. For HEMTs with gate lengths (⁠ $L g$⁠) of 100 and $200$ nm, a gate-source distance (LGS) of 0.75 μm, gate-drain distances (LGD) of 1.5 μm, and a gate width of $2 × 50$ μm were fabricated. Figure 6 shows the DC transfer characteristics of the conventional non-graded channel (S₈—blue) and the linearly graded channel (S₅—magenta) HEMTs with $L g$ = 0.2 μm (see Fig. S5 for results on $L g$ = 0.1 μm). Both devices have a threshold voltage (⁠ $V th$⁠) $∼$−1.7 V. The maximum source-drain current density is 811 and 784 mA/mm, while the peak transconductance (⁠ $g m$⁠) at $V ds$= 4 V is 369 and 308 mS/mm for the conventional and the graded channel HEMT, respectively. As shown in Fig. 6, the $g m$ is generally flat with smaller $g m ′$ and $g m ″$ for the HEMTs with graded channel. This is a measure of the expected linearity of the HEMTs as the third-order intermodulation distortion (IM3) is proportional to $( g m ″ ) 2$⁠, and the third-order intercept point (IP3) is proportional to $( g m ) 3 / g m ″$⁠. It should be noted at this point that the measurements were performed on HEMTs with relatively long gate lengths and low $V ds$ in order to decrease short channel effects associated with the UID GaN buffer layer in these structures. Devices with Fe-doped buffer layers will be presented elsewhere.

In summary, we developed Al_xGa_1−xN graded channel HEMT structures with various graded profiles from x = 0 to x = 0.1 (exponential, hybrid tanh-linear, and linear) using hot-wall MOCVD. The Al-profiles, experimentally determined by EDS, were correlated with specific growth conditions, which can serve as guidelines for channel grading optimization in future works. It is shown that incorporating an AlN interlayer between the linearly graded channel and the barrier layer has only marginal effect on the 2DEG/3DEG properties in contrast to conventional AlGaN/GaN HEMTs where significant improvement was observed. We find that although the mobility in graded channel HEMT is limited by alloy scattering, the transconductance is generally flatter over a wider range of gate bias with smaller $g m ′$ and $g m ″$ as compared to conventional AlGaN/GaN HEMTs. These features are required for highly linear devices used in low-noise RF amplifiers.

See the supplementary material for additional details about (i) the surface morphology (AFM images) of samples S1–S3, (ii) the VASE data analysis, (iii) the XRD reciprocal space maps of a conventional (S7) and a graded channel (S3) HEMT structure, (iv) the enlarged STEM image of the top region of S3 (inset of Fig. 2), and (v) the DC transfer characteristics of the conventional non-graded channel (S8) and the linearly graded channel (S5) HEMTs with L g = 0.1 μm.

This work was performed within the framework of the competence center for III-Nitride technology, C3NiT-Janzén supported by the Swedish Governmental Agency for Innovation Systems (VINNOVA) under the Competence Center Program Grant No. 2022-03139, Lund University, Linköping University, Chalmers University of Technology, Ericsson, Epiluvac, FMV, Gotmic, Hexagem, Hitachi Energy, On Semiconductor, Region Skåne SAAB, SweGaN, Volvo Cars and UMS. We further acknowledge support from the Swedish Research Council VR under Award Nos. 2016-00889 and 2022-04812, Swedish Foundation for Strategic Research under Grant Nos. RIF14-055, EM16-0024 and STP19-0008 and the Swedish Government Strategic Research Area in Materials Science on Functional Materials at Linköping University, Faculty Grant SFO Mat LiU No. 2009-00971. The KAW Foundation is also acknowledged for the support of the Linköping Electron Microscopy Laboratory. P. O. Å. Persson acknowledges ARTEMI, the Swedish National Infrastructure in Advanced Electron Microscopy, through funding from the Swedish Research Council and the Foundation for Strategic Research (Grant Nos. 2021-00171 and RIF21-0026). The authors would like to thank Dr. Nerijus Armakavicius for general support on ellipsometry related discussions.

The authors have no conflicts to disclose.

Alexis Papamichail: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Plamen P. Paskov: Investigation (equal); Writing – review & editing (equal). Niklas Rorsman: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). Vanya Darakchieva: Conceptualization (equal); Funding acquisition (equal); Investigation (equal); Project administration (lead); Supervision (lead); Writing – review & editing (lead). Axel R. Persson: Formal analysis (equal); Investigation (equal); Writing – review & editing (equal). Steffen Richter: Formal analysis (equal); Investigation (equal); Writing – review & editing (equal). Philipp Kühne: Formal analysis (equal); Investigation (equal). Vallery Stanishev: Formal analysis (equal); Investigation (equal). Per Persson: Supervision (equal); Writing – review & editing (equal). Ragnar Ferrand-Drake Del Castillo: Investigation (equal). Mattias Thorsell: Conceptualization (equal); Methodology (equal); Writing – review & editing (equal). Hans Hjelmgren: Formal analysis (equal); Investigation (equal); Writing – review & editing (equal).

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