1. Introduction

Conventional AlGaN/GaN high electron mobility transistors (HEMTs) have shown excellent performances in high-frequency applications, for the advantages associated with GaN-based wide band gap semiconductors which include the high breakdown fields, high peak electron and saturation drift velocities, and very high sheet charge densities at the interface resulting from large conduction band offset and strong polarization effects^[1-5]. Many new and advanced device structures are being explored, and material quality and device processing are ameliorated^{[6, 7]}. For the larger conduction band offset and higher polarization charge density at the interface, high Al content AlGaN/GaN heterostructures have been demonstrated, and Palacios et al. demonstrated the AlGaN/GaN/InGaN/GaN HEMTs structure with improved buffer isolation^[8], and an improved structure with higher 2DEG mobility was reported^{[9, 10]}. In our earlier research, the AlGaN/AlN/GaN structure with the step-graded AlGaN barrier layer was demonstrated, which improved the crystal quality of high Al content AlGaN layer^[11]. However, high Al content AlGaN barrier results in a larger conduction-band discontinuity and a higher barrier that help confine the 2DEG. But the increased lattice mismatch and strain associated with high Al content seriously deteriorate the crystal quality of the AlGaN layer. Furthermore, the 2DEG mobility will decrease due to the poor interface quality at the AlGaN and GaN layers^[11]. In order to combine the advantages of InGaN back barrier and step-graded AlGaN barrier, the novel AlGaN/AlN/GaN/InGaN/GaN HEMTs structure with a compositionally step-graded AlGaN barrier layer has been investigated.

In this study, we focus on the growth and characterization of the InGaN back barrier HEMTs structure with a compositionally step-graded AlGaN layer. Structural and electrical properties of two structures are investigated and compared. Double crystal X-ray diffraction (DCXRD) and atomic force microscopy (AFM) were performed to study the structural properties and crystalline quality, while electrical properties were obtained by the Hall measurements. Benefitting from the optimized structure and improved growth processes, the crystal quality of the AlGaN layer and the properties of the device are better than the results of Ref.[11] in our earlier works.

2. Device fabrication and experimental setup

The AlGaN/AlN/GaN/InGaN/GaN HEMTs structure with a compositionally step-graded AlGaN barrier layer (sample A) was grown by MOCVD on 2-in $c$-plane (0001) sapphire substrate, and the schematic cross section of the structure is shown in Fig. 1(a). As a comparison, the Al$_{0.5}$Ga$_{0.5}$N/AlN/GaN/InGaN/GaN HEMTs structure with a 25 nm Al$_{0.5}$Ga$_{0.5}$N barrier (sample B) shown in Fig. 1(b) was grown under identical growth conditions. After initial desorption in ambient H$_{2}$ at 1100 ℃, a low temperature (LT) GaN nucleation layer was grown at 530 ℃, subsequently a 3 $\mu$m undoped high resistance (HR) GaN buffer layer was grown at 1050 ℃. An InGaN interlayer was grown at 800 ℃ under N$_{2}$ atmosphere, with 10% indium composition. Ammonia (NH$_{3})$, triethyl-gallium (TEG) and trimethyl-indium (TMI) were used as source materials. Then the low temperature (LT) GaN spacer layer was grown at 800 ℃, with the thickness of 2-5 nm. High mobility (HM) GaN channel layer was grown at 1000-1100 ℃ under H$_{2}$ atmosphere, and ammonia (NH$_{3})$, trimethyl-gallium (TMG) were used as source materials. Subsequently an ultrathin AlN interlayer of about 1 nm was grown at 1000-1100 ℃, followed by a 25 nm undoped compositionally step-graded AlGaN barrier layer, grown at 1000-1100 ℃. The step-graded AlGaN barrier contained a 5 nm Al$_{0.5}$Ga$_{0.5}$N layer, a 10 nm Al$_{0.35}$Ga$_{0.65}$N layer, and a 10 nm Al$_{0.2}$Ga$_{0.8}$N layer along the direction of epilayer growth in sequence. Finally a thin GaN cap layer was grown at 1000-1100 ℃. The indium content in the InGaN interlayer was about 10%, which was determined by both Rutherford backscattering (RBS) and X-ray diffraction measurements. The electrical properties of the heterostructure were studied by Hall and eddy current sheet resistance measurements. The surface morphology of the wafer was characterized by atomic force microscopy (AFM).

HEMT based on the step-graded structure was fabricated. Ti/Al/Ni/Au metals were deposited in sequence by electron beam evaporation, followed by rapid thermal annealing at 870 ℃ for 30 s in ambient N$_{2}$. Subsequently the device isolation was realized by using multiple-energy boron (B) ion implantation with an isolation resistance of more than 10$^{11}$ $\Omega$/square. A silicon nitride (SiN) film was then deposited using plasma-enhanced chemical vapor deposition (PECVD). The recessed gate etching was performed by dry etching technology, and the Schottky gates with a field plate were formed by electron beam lithography. The gate metallization was realized by using electron-beam evaporated Ni/Au. The drain side edge of the gate metal overlapped the SiN passivation layer, thus forming the field plate. The direct current (DC) measurements of the device were carried out using HP4142 semiconductor parameter analyzers.

3. Results and discussion

The two samples are characterized by DCXRD measurements. Figure 2 shows $\omega$/2$\theta$ scans around (0002) reflection of the two samples. The main strong peaks are the (0002) 0th-order Bragg reflection of thick GaN, and the weak and broad peaks at the right side of the main peaks correspond to the (0002) 0th-order Bragg reflection of thin AlGaN barrier layers. In order to figure out the Al compositions of the AlGaN barriers, we assume that the thin AlGaN barriers are fully strained to the thick GaN layers. Using the method in Ref.[12], the Al compositions of AlGaN barriers in samples A and B are determined to be 34% and 45%. For the step-graded layer, the Al quasi composition is lower than the high Al structure.

To characterize the crystalline quality of the two samples, we have performed XRD $\omega$-scan on different positions across the 2-inch wafers. The full width at half maximum (FWHM) values as well as standard deviations, determined from the $\omega$-scan XRD rocking curves (RCs) of the (0002) planes of barriers in samples A and B, are shown in Table 1. The density of screw-type threading dislocations (TDs) correlates with the FWHM of $\omega$-scan RC of (0002) planes^[13]. From Table 1, one can see that when the Al composition of the barrier increases, FWHM values of (0002) planes increase, indicating more TDs are formed in the barrier layer with higher Al composition. This is consistent with the results published earlier^[11]. Using the method developed by Ref.[14], by means of calculating FWHM, the densities of screw-type TDs are 8.34 $\times$ 10$^{8}$ cm$^{-2}$ (sample A), 11.44 $\times$ 10$^{8}$ cm$^{-2}$ (sample B). So it can be concluded that the step-graded structure, which decreases the lattice strain, improves the crystal quality of AlGaN layer.

In order to investigate the surface morphology of the structures, AFM scan was performed and 2 $\times$ 2 $\mu$m$^{2}$ AFM images for the two samples are shown in Fig. 3. Benefitting from the compositionally step-graded AlGaN barrier layer which decreases the lattice strain, the sample exhibits a smooth surface with clear atomic steps and good surface morphology with a low dislocation density. As listed in Table 1, the measured RMS surface roughness is 0.172 nm and 0.232 nm for the samples A and B, respectively. Furthermore, we use the AFM technique for the confirmation of threading dislocation density (TDD) differences among the samples. As demonstrated recently^[15], the most adapted technique to determine TDD is AFM which images the pits associated with TDs at their termination on the surface. From Fig. 3 many small pits are visible at the surfaces of the samples. The pit numbers are counted and the pit densities are estimated to be 7.25 $\times$ 10$^{8}$ cm$^{-2}$ and 10.75 $\times$ 10$^{8}$ cm$^{-2}$ for sample A and B, respectively. This implies that TDD increases with the increase of Al composition of the AlGaN barrier, which is consistent with the result of XRD. However, the calculated TDD by AFM is lower than that obtained by XRD. The underestimate of TDD is probably because the size of the pits for TD is very small ($\approx$ 16 nm) and they are only observed in scans smaller than 2 $\times$ 2 $\mu$m$^{2}$.

The electrical properties of the 2DEG were studied by performing Hall measurement. The 2DEG mobilities and the sheet electron concentrations of the step-graded structure and the high Al structure as a function of temperature are shown in Fig. 4. As listed in Table 1, by using the step-graded barrier structure, the room-temperature 2DEG mobility increases from 1300 to 1820 cm$^{2}$/(V$\cdot$s). The low temperature mobilities (80 K) of 8770 and 6030 cm$^{2}$/(V$\cdot$s) are also observed on step-graded and high Al structures, respectively. The decreased strain in the step-graded AlGaN barrier can reduce mismatch dislocation, which actively affects the quality of AlN/GaN interface. Thus, we speculate that the significant improvement in the 2DEG mobility for the step-graded structure is mainly attributed to the high AlN/GaN interface quality. Furthermore, the AlN interfacial layer will confine the electrons nearly exclusively to the binary compounds and almost completely eliminate scattering due to alloy disorder, as reported in Refs.[16, 17]. The high mobility also benefits from the divided GaN layers' growth which consist of the GaN channel layer that provides high mobility 2DEG and GaN spacer layer that was grown at low temperature to prevent indium cluster formation and indium diffusion. Compared to the high Al structure, the step-graded structure with the same AlGaN barrier thickness has less total Al content in the barrier, and therefore has lower 2DEG sheet density due to the reduction of polarization. Another attractive advantage of the step-graded structure with the InGaN back barrier, i.e. the 2DEG sheet density is independent of the measurement temperature, which indicates that the 2DEG is very well confined in the GaN channel layer.

The typical sheet resistance and resistance uniformity measurement results for the step-graded structure are shown in Fig. 5, which was conducted by Lehighton contactless sheet resistivity mapping across the 2-inch wafer. The wafer displays an average sheet resistance of 380 $\Omega$/$\square$, with a sheet resistance non-uniformity of 3.68%. The relation of sheet resistance with 2DEG sheet density ns and mobility can be given by

$\begin{equation} \label{eq1} R_{\rm s} =\frac{\rho }{t}=\frac{1}{\sigma t}=\frac{1}{(nq\mu ) t}=\frac{1}{nt} \frac{1}{\mu q}=\frac{1}{qn_{\rm s} \mu }, \end{equation}$

(1)

where $\rho$ is the resistivity, $\sigma$ is the conductivity, $q$ is the electric charge, and $t$ is the 2DEG width along the growth direction in the triangular quantum well. From the results of Hall measurement, the calculated value of sheet resistance is about 404 $\Omega$/$\square$, which is consistent with the result of Lehighton contactless sheet resistivity mapping.

According to the results of AFM, XRD, Hall measurements and sheet resistance mapping, the grown step-graded structure with InGaN back barrier is very suitable for the device fabrication. For the high Al content HEMTs structure, the large lattice mismatch between AlGaN and GaN ($\Delta$$a/a$ $=$ 1.21%) results in high tensile stress in the AlGaN layer. Thus, more strain-induced defects are formed, leading to the degradation of the surface morphology and crystal quality. However, for the step-graded structure, the lattice mismatch $\Delta$$a/a$ decreases gradually with the step-like decrease of the Al composition in the AlGaN layer, from 1.21% and 0.85% to 0.16%, as shown in Table 2. Thus the lattice strain decreases correspondingly step by step, therefore the total strain in the AlGaN layer decreases compared with the high Al structure and realizes high surface and crystal quality.

HEMT based on the InGaN back barrier HEMTs structure with a compositionally step-graded AlGaN layer was fabricated. Typical current-voltage ($I_{\rm ds}$-$V_{\rm ds})$ and transfer characteristics of the device with 0.35 $\mu$m $\times$ 1 mm gate periphery are illustrated in Figs. 6(a) and 6(b). The maximum drain current density is obtained to be 713.8 mA/mm at a gate bias of 1 V. The maximum extrinsic transconductance is 376.3 mS/mm at a gate bias of -1 V. The device pinched off completely at a gate bias of -2.8 V. Furthermore, the transconductance ($g_{\rm m})$ is quite flat after onset and remains close to its peak value at high current levels, which shows a good linearity.

The off-state $I_{\rm ds}$-$V_{\rm ds}$ characteristic of the step-graded structure at the gate voltage of -3 V is shown in Fig. 7. As seen, the leakage current is only 28 $\mu$A when the drain voltage is 10 V, which indicates the structure has a good isolation of GaN buffer. The low leakage current for the device is mainly due to the improved 2DEG confinement by the InGaN back-barrier, which is caused by the opposite piezoelectric polarization in the InGaN layer compared to the AlGaN layer^{[9, 10]}. Therefore electrons spilling over into the buffer layer is effectively avoided and the buffer leakage current is greatly reduced in the device. The optimized thickness of InGaN layer in the structures has been investigated and reported in our earlier works^[9].

These results are better than those of the InGaN channel HEMTs^[18] and the step-graded AlGaN/AlN/GaN HEMTs^[11] in our earlier works.

4. Conclusion

The structural and electrical properties of AlGaN/AlN/GaN/InGaN/GaN HEMTs structure with a compositionally step-graded AlGaN barrier layer were studied. The densities of screw-type threading dislocations are 8.34 $\times$ 10$^{8}$ cm$^{-2}$ and 11.44 $\times$ 10$^{8}$ cm$^{-2}$ for the step-graded structure and the high Al structure, respectively, which are consistent with the results of AFM. The AFM result of this structure shows a good surface morphology. The measured 2DEG mobility was 1820 cm$^{2}$/(V$\cdot$s) for the step-graded structure, 1300 cm$^{2}$/ (V$\cdot$s) for the high Al structure, respectively. It can be concluded that the step-graded structure, which decreases the lattice strain, improves the crystal quality of AlGaN layer. The HEMTs device using the material was fabricated, and a maximum drain current density of 713.8 mA/mm and an extrinsic transconductance of 376.3 mS/mm are achieved, which are better than our earlier work.