This study examines the praseodymium-oxide- (Pr2O3-) passivated AlGaN/GaN metal-insulator-semiconductor high electron mobility transistors (MIS-HEMTs) with high dielectric constant in which the AlGaN Schottky layers are treated with P2S5/(NH4)2 + ultraviolet (UV) illumination. An electron-beam evaporated Pr2O3 insulator is used instead of traditional plasma-assisted chemical vapor deposition (PECVD), in order to prevent plasma-induced damage to the AlGaN. In this work, the HEMTs are pretreated with P2S5/(NH4)2 solution and UV illumination before the gate insulator (Pr2O3) is deposited. Since stable sulfur that is bound to the Ga species can be obtained easily and surface oxygen atoms are reduced by the P2S5/(NH4)2 pretreatment, the lowest leakage current is observed in MIS-HEMT. Additionally, a low flicker noise and a low surface roughness (0.38 nm) are also obtained using this novel process, which demonstrates its ability to reduce the surface states. Low gate leakage current Pr2O3 and high-k AlGaN/GaN MIS-HEMTs, with P2S5/(NH4)2 + UV illumination treatment, are suited to low-noise applications, because of the electron-beam-evaporated insulator and the new chemical pretreatment.

1. Introduction

Because of their inherent high breakdown voltage ( ), high two-dimensional electron gas (2-DEG) concentration, and high saturation velocity [1, 2], and AlGaN/GaN high electron mobility transistors (HEMTs) are suitable to high-power and low-noise applications. The major factors that limit the performance and reliability of GaN-based HEMTs at radio frequencies (RF) are their high gate leakage current and drain current collapse, which is associated with native oxide-induced surface states [3, 4]. Therefore, AlGaN/GaN metal-insulator-semiconductor HEMTs (MIS-HEMTs), in which SiO2 [5], Si3N4 [6], Ga2O3 [7], Al2O3 [8], and Sc2O3 [9] are used as the gate dielectrics, are studied, in order to address these problems. Related works focus on the formation of a high-k insulator, which reduces the Schottky gate leakage current at high input signal swings and improves channel modulation. However, the treatment of the interface between AlGaN and the insulator has not been studied systematically. Pretreatment before the deposition of the passivating layer between the source, the drain, and the gate terminals is dominated by the effect of surface traps, which cause flicker noise and current collapse problems. For example, (NH4)2 sulfide treatment is known to eliminate native Ga2O3 and As2O3 dangling bonds on GaAs- and InP-related semiconductors, because of the formation of stable Ga-S and As-S bonds during immersion [10, 11]. In this work, a P2S5/(NH4)2 + UV treatment that suppresses surface traps is studied. Furthermore, to increase the efficiency of the P2S5/(NH4)2 treatment, the treatment is performed in a UV chamber and both the high-k   Pr2O3 gate insulator and the passivating layer are deposited using electron-beam evaporation, which effectively prevents the plasma-induced generation of surface states. A comparison of the flicker noise determined from the pulsed I-V of Pr2O3 AlGaN/GaN MIS-HEMT with that of traditional GaN HEMTs shows that the surface traps are markedly suppressed by P2S5/(NH4)2 + UV treatment. The observed lower surface leakage current also improves the DC-RF dispersion of MIS-HEMT. X-ray photoelectron spectroscopy (XPS) measurement and secondary ion mass spectrometry (SIMS) are used to study the Ga-S energy bonds and the distributions of the depths of oxygen and sulfur atoms, following P2S5/(NH4)2 + UV treatment.

2. Device Structure and Fabrication

The AlGaN/GaN HEMT heterostructures used in this study were grown using atmospheric pressure metal organic chemical vapor deposition (AP-MOCVD) on 2 inch sapphire wafers. The 4000 nm-thick undoped GaN was grown first, to form the buffer and channel layers. Then, a 35 nm-thick undoped Al0.25Ga0.75N layer was grown as the Schottky layer. The designed structure had a sheet charge density of 1.65 × 1013 cm−2 and a Hall mobility of 1060 cm2/V-s at 300 K. Figure 1 shows the cross-sections of a Pr2O3/AlGaN/GaN MIS-HEMT with P2S5/(NH4)2 + UV interface pretreatment. During fabrication of the device, the active region was protected by a photoresist and the mesa isolation region was removed using a BCl3 + Cl2 mixture gas plasma in a reactive ion etching (RIE) chamber. The ohmic contacts of the Ti/Al/Ni/Au (25 nm/125 nm/50 nm/100 nm) metal layers were deposited using electron-beam evaporation and patterned by conventional optical lithography and lift-off method, followed by 850°C RTA annealing for 30 s in an N2 environment. Before the deposition of the high-k Pr2O3 insulator and the passivating layers, the samples were immersed in a standard treatment (dilute HCl), (NH4)2 and P2S5/(NH4)2 for 15 min. Figure 2 shows the pH value of the P2S5/(NH4)2 solution versus P2S5 weight. The pH of the saturated (NH4)2 solution is 11.2 and the pH of the mixed P2S5/(NH4)2 solution was adjusted to a value of 7 by adding 10 g of P2S5 into 30 mL (NH4)2 saturated solution. Although the (NH4)2 treatment effectively removes the native AlGaN surface oxide layer, it cannot prevent the increase in surface roughness that is caused by the alkaline (NH4)2 solution. After 15 min of pretreatment immersion, the AlGaN surface roughness was 0.24 nm for the standard treatment and 0.51 nm and 0.44 nm for the (NH4)2 and P2S5/(NH4)2 solution. Therefore, the P2S5/(NH4)2 solution was used as the interface pretreatment solution for the MIS-HEMT. In an earlier study by the authors, the reduction of the number of native oxide-induced dangling bonds on the AlGaN surface, which are difficult to remove at room temperature, was achieved by increasing the temperature of the solution temperature to 60°C, but this increase makes the procedure more complicated and time-consuming. In this study, UV illumination is used during P2S5/(NH4)2 pretreatment immersion, in order to rapidly eliminate the dangling bonds and form Ga-S bonds with high-binding energy. After interface treatment, a 10 nm-thick layer of praseodymium was firstly evaporated, using an optimal oxygen flow rate of 15 sccm. During this stage, the chamber pressure was increased to around 10−4 Torr. When the chamber pressure was reduced to Torr, the conventional Ti/Au (30 nm/150 nm) gate metals were deposited. For comparison, a traditional Ni/Au Schottky gate GaN HEMT was also fabricated. Finally, the Ti/Au (30 nm/300 nm) metals were deposited to form the interconnection and probe pads, and a 200 nm-thick SiO2 layer was deposited to passivate the device. The complete process was also used to fabricate a traditional Ni/Au Schottky gate GaN HEMT, for the purpose of comparison.

The important optimization of the flow rate of the praseodymium oxide high-k layer was also considered. The electron-beam evaporation of the high-k Pr2O3 thin film is optimized by adjusting the oxygen flow rate in the chamber. Figure 3 shows the EDX measurements for Pr2O3 grown in the electron-beam evaporator using various oxygen flow rates. An analysis of the rare-earth metal atomic concentration in the oxide layer demonstrates that the optimal flow rate for the deposition of the praseodymium oxide layer is 15 standard cubic centimeters per minute (sccm), indicating that the dielectric constant is maximized, because a strong dipole is formed at the highest possible rare-earth metal concentration in the high-k oxide layer. The thermal stability of the insulator also plays an important role in high-power GaN MIS-HEMTs because the dc power is primarily dissipated near the gate contact, which causes local Joule self-heating [10]. The channel temperature can reach 100°C during high-output-power operation, which results in increased degradation and failure rates and a reduction in output power. In order to evaluate the thermal stability of Pr2O3 layer, the X-ray photoelectron spectroscopy (XPS) was used to measure the binding energy of the Pr2O3 thin films after 400°C, 600°C, and 800°C postannealing. Figure 4(a) shows the XPS 3d core-level spectra for Pr2O3 at various temperatures. The energy of the Ar-gun is 3 KeV, the operation current is 1 mA, and the analysis area is  mm2. It is seen that the binding energies of the Pr2O3 after 400°C, 600°C, and 800°C postannealing are close to the standard value of 934 eV in the 3d core level, as recorded in the XPS handbook. Figure 4(a) also shows the high signal intensities of Pr2O3. Therefore, it is concluded that a high-quality and a highly thermally stable high-k insulator is obtained by using electron-beam-evaporated Pr with a high oxygen flow rate. The results of high-resolution cross-sectional transmission electron microscopy (TEM) prove that the Pr2O3 grows on the GaN in a planar fashion, as shown in Figure 4(b). The Pr2O3 equivalent oxide thickness, measured by the TEM, is 20 nm. A (Ga2O3)Pr2O3 compound film occurs between the interface of the GaN and Pr2O3 layers, which is generated by the native GaN surface oxide layer and praseodymium materials.

3. Experimental Results for the Device

Atomic force microscopy (AFM) or scanning force microscopy (SFM) has a very high resolution, with a demonstrated resolution of the order of fractions of a nanometer, more than 1000 times better than the optical diffraction limit. It is also one of the most widely used tools for imaging, measuring, and manipulating matter at the nanoscale. The information is gathered by “feeling” the surface with a mechanical probe. Piezoelectric elements that facilitate tiny but accurate and precise movements on (electronic) command enable very precise scanning [11]. Figure 5(a) and Figure 5(b) show the 2D and 3D images of the surface roughness on the AlGaN/GaN surface with different treatments, as measured using a Park Systems XE-70. The P2S5/(NH4)2 + UV-treated surface exhibits a better roughness than the other sulfurization treatments and forms a superior interface between the AlGaN Schottky layer and the Pr2O3 high-k gate insulator layer.

Table 1 shows the mobility, sheet charge density, and surface roughness for variously treated devices, characterized by Hall measurement at 300 K. The P2S5/(NH4)2 + UV-treated devices have a sheet charge density of 1.403 × 1013 cm−2 and a Hall mobility of 1150 cm2/V-s at 300 K; these values are, respectively, 1.648 × 1013 cm−2, 1060 cm2/V-s and 1.512 × 1013, 1087 cm2/V-s for the standard treatment and (NH4)2 + UV treatment for the devices. These results clearly demonstrate that P2S5/(NH4)2 + UV treatment improves the channel mobility by reducing the number of surface traps.

The surface of composite semiconductor substrates was subject to heat treatment, using an ultraviolet (UV) light, in order to enhance the surface reaction [12]. The photoluminescence (PL) measurements in Figure 6 apply to the complete structure of AlGaN/GaN after sulfurization and demonstrate that P2S5/(NH4)2 + UV treatment yields a greater PL intensity for the AlGaN Schottky layer than either P2S5/(NH4)2 treatment or standard treatment. The increase in PL intensity is due to the elimination of surface states, which generates nonradiative recombination centers on the AlGaN surface [13]. No obvious Al0.25Ga0.75N signal is evident, because the laser used is He-Cd laser and the PL signal intensity of Al0.25Ga0.75N is much smaller than that for GaN. Because only 35 nm AlGaN was grown on GaN for the PL evaluation structure after sulfurization, the Al0.25Ga0.75N signals are not obvious in the PL measurements. Although the P2S5/(NH4)2 treatment can suppress the surface state density, as demonstrated by the PL results, a more concentrated P2S5/(NH4)2 solution with UV illumination treatment produces a stable phosphorus oxide layer and more Ga-S bonds, because the phosphorus and sulfur concentrations are higher [14].

Figure 7 shows the and XPS spectra of the variously treated devices. The sulfur 2p core level and the binding energy of pure sulfur are all 163.8 eV. The binding energy of gallium is 160 eV at the signal peak in the spectrum of the standard treated GaN sample while the signal peak of the P2S5/(NH4)2 -treated device is shifted to 160.3 eV, because the AlGaN surface contains Ga-S bonds (Ga-S = 163.2 eV) after this process. However, P2S5/(NH4)2 + UV treatment shifts the signal peak in the spectrum to 160.4 eV, with a linear distribution of intensity from 160 eV to 161 eV. The peaks are shifted to a higher binding energy after P2S5/(NH4)2 + UV treatment. This phenomenon is related to the Ga sulfurized states [15]. Therefore, more Ga-S bonds are generated on the AlGaN surface by P2S5/(NH4)2 + UV treatment than by P2S5/(NH4)2 treatment and these high-energy bonds are more stable than the Ga-O bonds that are formed in a moist environment.

In order to investigate the material atomic composition of the AlGaN surface after various sulfide treatments, the samples were subject to secondary ion mass spectroscopy (SIMS). Figure 8 shows the sulfur and oxygen atom concentration profiles of the samples treated using the three aforementioned methods. Based on the measurement results shown in Figure 8(a), the concentration of S atoms in the P2S5/(NH4)2 and P2S5/(NH4)2 + UV-treated samples is relatively high near the AlGaN surface, dropping by two to three orders of magnitude in the channel. However, few S atoms are observed after the standard treatment. The S atoms produced from the sulfide solution are incorporated into the sample surface, which is similar to the effect of the immersion of CF4 in plasma, used to enable a GaN HEMT to operate in enhancement mode [16]. In order to further investigate the mechanism for the removal of Ga-O bonds by sulfur treatments, an experiment was conducted to determine the oxygen concentration distribution; Figure 8(b) shows the results. The lowest curve in the figure shows that the P2S5/(NH4)2 + UV treatment removes a huge amount of AlGaN/GaN native oxide and reduces the number of oxygen atoms more effectively than the other treatments, because UV illumination provides enough energy to the sulfur atoms to replace the Ga-O bonds and form stable Ga-S bonds. Therefore, P2S5/(NH4)2 + UV treatment not only produces a smoother Schottky interface, reducing the gate leakage current, but also reduces the number of oxygen atoms more effectively. Both effects reduce the density of the surface states.

Figure 9 plots the gate-to-drain I-V curves for a standard GaN HEMT, a Pr2O3 MIS-HEMT, and a P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMT. A reduction in the number surface states and leakage current in the sulfide-treated sample means that P2S5/(NH4)2 + UV-treated MIS-HEMTs have a higher value than the others. The improved value of 1.71 V for the P2S5/(NH4)2 + UV-treated Pr2O3-gate device results in a significant reduction in the gate leakage current at a high pumped gate voltage, which improves the linearity and results in a reduction in the signal dispersion of the device. The reversed gate-to-drain breakdown voltages ( ), defined as the voltage at which the gate leakage current is −1 mA/mm, are −131.3 V for the P2S5/(NH4)2 + UV-treated samples, which allows operation at high drain voltage, and only −128.2 V and −107.5 V for the Pr2O3 MIS-HEMT and standard-treated samples, respectively.

In order to study the dc characteristics, the drain-to-source current ( ) versus drain-to-source voltage ( ) curves for three devices are shown in Figure 10. The drain current does not vary significantly between the treated devices. The turn-on resistance ( ) is 3.958 Ω for the HEMTs subject to standard treatment, at of 0 V, and 4.675 Ω and 4.210 Ω for the Pr2O3 MIS-HEMTs and the P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMTs, respectively. This is primarily because the sulfide solution influences only the surface states, rather than the intrinsic parameters. However, since the traps within the interface between the gate and channel are suppressed by P2S5/(NH4)2 + UV treatment, the sample with this treatment has the highest output resistance, which results in an increase in device linearity and power gain cut-off frequency ( ). The output conductance ( ) of the P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMTs is 0.46 mS/mm; the corresponding values for the Pr2O3 MIS-HEMTs and the standard device are 0.6 mS/mm and 1 mS/mm, respectively.

Figure 11 plots the transistor transconductance ( ) versus curves for the three devices of interest at a of 8 V. The maximum drain-to-source currents ( ) at are 923 mA/mm, 864 mA/mm, and 920 mA/mm, for the standard HEMTs, the Pr2O3 MIS-HEMTs, and the P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMTs, respectively. The maximum transconductance values ( ), biased at , are 144 mS/mm, 121 mS/mm, and 132 mS/mm, respectively. All of these values are reasonably favorable. The standard HEMTs exhibit a higher peak , because a high-k insulator is inserted in the MIS-HEMTs structure. The gate-to-channel improves channel modulation and effectively modulates any increase in the depletion region between the metal gate and the channel. Obviously, high-k GaN MIS-HEMTs demonstrate large swing voltage and low gate leakage current.

The measurement of on-wafer microwave S-parameters for μm2 devices was performed in a common-source configuration using an Agilent E8364C PNA network analyzer, from 0.1 GHz to 20.1 GHz. S-parameter measurements demonstrate a maximum current gain cut-off frequency ( ) of 9.2 GHz and a maximum oscillation frequency of 16.8 GHz for standard HEMTs. These values are 7.9 GHz and 10.7 GHz for Pr2O3 MIS-HEMTs and 8.5 GHz and 16.2 GHz for P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMTs, at and , respectively. The superior RF characteristics for P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMTs prove their greater power and linearity. Table 2 presents dc and radio RF characteristics for various devices.

Pulse measurements were made to characterize the carrier trapping phenomenon and heating effect in the device. With respect to the trapping carriers, the response time dominates the pulse measurement. The response time for these trapping carriers is typically of the order of μs longer than the ns of carrier transportation, especially for high-speed and high-power devices. In order to more easily observe the output signals from the drains of each device, a 50 Ω resistor was added between the drain electrode and the power supply, in order to determine the total drain-to-source current and to reduce the damping effect in the output square waveform. Pulse I-V measurements for the three devices were also made, in order to confirm their surface trapping effects. Figure 12 plots the dc to 1 μs pulse I-V measurement for μm2 devices at a of 0 V and a of 8 V, for the three devices. The gate width of the measured devices is 100 μm, so the heating effect could therefore be neglected. Since the surface states determine the dispersion effect, the current density at high current decays as the pulse width decreases. As shown in the figure, the standard GaN HEMTs demonstrate a larger slope with respect to the pulse period than do the Pr2O3 MIS-HEMTs and P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMTs. It is evident that P2S5/(NH4)2 + UV treatment produces reliable and stable surface performance and less load-line hysteresis and better linearity for the device in high-power applications are which both expected.

To investigate the relationship between the flicker noise and the quantity of the variously treated devices, a low-frequency noise measurement was performed, because, at low frequency, this is sensitive to the semiconductor surface [17]. The bias point for low-frequency noise measurement was selected as V, which gives an of 100 mA/mm for all devices. Since the measurement is dominated by the series resistance of each device, identical bias point confirms the flicker noise characteristics. As shown in Figure 13, the P2S5/(NH4)2 + UV-treated MIS-HEMT demonstrates a lower 1/f spectral noise than both the Pr2O3 MIS-HEMT and the GaN HEMT, which proves the reduction in the number of surface dangling bonds due to P2S5/(NH4)2 + UV treatment.

The fabricated 1 μm long gate GaN HEMT, Pr2O3 MIS-HEMT, and P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMT were tested on-wafer and the microwave power characteristics were evaluated using a load-pull system with automatic tuners, which simultaneously provides conjugate-matched input and load impedances for the maximum output power. The microwave load-pull power performance was conducted at 2.4 GHz, with a drain bias of 8 V, using various devices. The bias points for class AB operation of the standard HEMT, the Pr2O3 MIS-HEMT, and the P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMT must be biased at −3.1 V, −3.7 V, and −3.4 V (at 1/4 ), respectively. Figure 14 shows the output power ( ), power gain ( ), and PAE as a function of the input power ( ), for various devices with gate dimensions of μm2. The P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMT has better dc current and lower gate leakage current than that of the standard HEMT or the Pr2O3 MIS-HEMT, at high-input-power swing. The PAE values are 22.4%, 23.6%, and 24.2%, for the standard HEMT, the Pr2O3 MIS-HEMT and the P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMT, respectively. As a result, the microwave-power performance is improved by the MIS-gate structure, and the power-gain degradation is also improved in the high-input-power regime. The reduction in gate leakage current in the P2S5/(NH4)2 + UV-treated Pr2O3 MIS-HEMT allows a significant improvement in device linearity [18].

4. Conclusion

In summary, Pr2O3 MIS-HEMTs with low gate leakage current and low flicker noise, as a result of P2S5/(NH4)2 + UV treatment, were developed and characterized. An electron-beam-evaporated high-k insulator and a passivating layer prevent the formation of surface states by plasma. P2S5/(NH4)2 + UV treatment represents a simple and efficient means of reducing the number of surface dangling bonds. Based on the results of Hall, XPS, and SIMS measurements, this P2S5/(NH4)2 + UV treatment prevents the formation of strong Ga-S bonds on the AlGaN surface. This reduction in surface states improves carrier mobility and simultaneously suppresses unstable native Ga-O bonds. This novel pretreatment is therefore proven to be eminently suitable to low-noise GaN MIS-HEMT applications.


The authors would like to thank the Nano Device Labs (NDL) for providing the low frequency noise measurements, and the National Central University (NCU) for providing the thermal image measurements. This work was financially supported by the National Science Council, ROC [NSC-97-2221-E-182-048-MY3 and NSC-100-2221-E-182-009], and High Speed Intelligent Communication (HSIC) Research Center of Chang Gung University, Taoyuan, Taiwan.