Improved Electrical Characteristics of AlGaN/GaN High-Electron-Mobility Transistor with Al2O3/ZrO2 Stacked Gate Dielectrics

A metal-oxide-semiconductor high-electron-mobility transistor (MOS-HEMT) is proposed based on using a Al2O3/ZrO2 stacked layer on conventional AlGaN/GaN HEMT to suppress the gate leakage current, decrease flicker noise, increase high-frequency performance, improve power performance, and enhance the stability after thermal stress or time stress. The MOS-HEMT has a maximum drain current density of 847 mA/mm and peak transconductance of 181 mS/mm. The corresponding subthreshold swing and on/off ratio are 95 mV/dec and 3.3 × 107. The gate leakage current can be reduced by three orders of magnitude due to the Al2O3/ZrO2 stacked layer, which also contributes to the lower flicker noise. The temperature-dependent degradation of drain current density is 26%, which is smaller than the 47% of reference HEMT. The variation of subthreshold characteristics caused by thermal or time stress is smaller than that of the reference case, showing the proposed Al2O3/ZrO2 stacked gate dielectrics are reliable for device applications.


Introduction
In recent decades, AlGaN/GaN high-electron-mobility transistors (HEMTs) have received much attention due to their high breakdown voltage, high saturation current, and high operating temperature. These advantages make AlGaN/GaN HEMTs a potential candidate for high-frequency and high-power device applications [1][2][3]. However, conventional GaN-based HEMTs with a Schottky gate suffer from high leakage current and the current collapse effect, which degrade the device performance and reliability. Inserting dielectric layers underneath the metal gate can effectively reduce the gate leakage current [4,5], suppress current collapse phenomenon [6,7], provide better linearity of RF output power [8,9], and lower flicker noise [10,11]. As the size of the device shrinks, the thickness of the gate dielectric shrinks, resulting in leakage current or related reliability issues.
A thin oxide layer leads to large current leakage; however, high-κ materials increase the physical thickness of gate dielectrics and can thus overcome the tunneling leakage current. The high-κ materials such as HfO 2 [12], TiO 2 [13], ZrO 2 [14], HfAlO x [15], etc. are expected to increase the capacitance of the transistor without reducing the oxide thickness. Recent research has proposed that different oxide compounds have excellent dielectric properties [16]. Among these materials, ZrO 2 is a potential choice. ZrO 2 has a wide bandgap (5.16-7.8 eV) and a high dielectric constant (20)(21)(22)(23)(24)(25) [17]. However, the ZrO 2 /GaN interface has some issues, such as a poor conduction band offset with GaN of (~1.1 eV), and a very high interface state density. An ultra-thin Al 2 O 3 film can be inserted to suppress the interface charge of GaN. Al 2 O 3 has a large bandgap (7-8.8 eV) that can tolerate a high breakdown field [18]. Based on the above discussion, stacked or bilayer gate dielectrics are the effective solution for modifying the interface quality. Strong correlation was observed between different compound contents and microwave dielectric properties [19,20]. Many studies have been widely explored, and excellent results have been obtained [21][22][23][24][25] for GaN-based HEMT; however, the reliability of Al 2 O 3 /ZrO 2 on AlGaN/GaN HEMTs has rarely been evaluated [23].
Many reports have been published in the GaN-based areas and they often studied the gate leakage, dc, and C-V characteristics at room temperature or high temperature. However, flicker noise, high-frequency performance, power performance, and stability after thermal stress or time stress were less conducted. On the other hand, although stacked oxide layers are available in some studies, there are only a few on the combination of Al 2 O 3 and ZrO 2 . Authors just studied the trapped charge effects for MOS capacitor or HEMT with Al 2 O 3 or ZrO 2 (i.e., single layer, not stacked layer). In this work, a radio frequency (RF) co-sputter system was used to deposit 1 nm-thick Al 2 O 3 and 12 nm-thick ZrO 2 films as a stacked gate dielectrics on AlGaN/GaN HEMT to increase breakdown voltage and reduce leakage current. In addition to low-frequency noise and high-frequency performance, the reliabilities of thermal and time stress for proposed devices were investigated.

Experimental Study
The wafer structure was grown with metal-organic chemical vapor deposition (MOCVD) on a 6-inch p-type (≤0.03 Ω·cm) Si-(111) substrate. The epi-structure from the top to bottom consisted of a 2-nm-thick GaN cap layer, a 20-nm-thick Al 0.25 Ga 0.75 N barrier layer, a 300-nmthick GaN channel layer, and a 3.9-µm-thick buffer/transition layer. The sheet resistivity, sheet carrier concentration, and Hall mobility were 448.6 Ω/cm, 2338 cm 2 /V·s, and 6.137 × 10 12 cm −3 , respectively. Samples were cleaned in an ultrasonic machine with acetone, methanol, and deionized water for 5 min. For device fabrication, the mesa isolation was conducted using an inductive coupled plasma reactive ion etching (ICP-RIE) system under a Cl 2 and BCl 3 mixed-gas environment. Ohmic contact of source/drain regions were deposited with Ti/Al/Ni/Au (25 nm/150 nm/30 nm/120 nm) by electron beam evaporation and followed by rapid thermal annealing (RTA) at 875 • C for 45 s in N 2 ambient. After that, the gate region was defined by photolithography. For reference Schottky-gate HEMT, gate electrode was deposited directly. For MOS-HEMT, the gate dielectrics of 1-nm Al 2 O 3 and 12-nm ZrO 2 stacked layer was then deposited by RF co-sputter system. The co-sputtering system ULVAC ACS-4000-C3 includes electrodes of two DC power supplies and one RF power source. The argon gas flow rate was 30 sccm and the RF power was 100 W for Al 2 O 3 of 2 min and ZrO 2 of 11 min, respectively. Finally, a gate electrode of Ni/Au (80 nm/100 nm) was deposited using the electron beam evaporation. The gate length (L G ), gate width (W G ), and source-to-drain distance (L SD ) were 1 µm, 100 µm, and 5 µm, respectively. The dc characteristics of the proposed devices were measured using an Agilent B1500A semiconductor analyzer. The lowfrequency noise spectrum was measured using a ProPlus system. NoisePro Plus software was used to extract low-frequency noise model parameters. S-parameter measurements of the high-frequency devices were performed by a Keysight N5245A PNA-X microwave network analyzer.

Results and Discussion
A cross-sectional schematic diagram of the device is shown in Figure 1a. Figure 1b shows the atomic force microscopy (AFM) image (5 µm × 5 µm) of the Al 2 O 3 and ZrO 2 . The root mean square (RMS) value of Al 2 O 3 was 0.36 nm and that of ZrO 2 was 0.79 nm, indicating that the surface roughness of oxide films is still good. Figure 1c-e show the X-ray diffraction (XRD) patterns of Al 2 O 3 , ZrO 2 , and Al 2 O 3 /ZrO 2 stacked layer, respectively. The XRD patterns did not reveal any peaks corresponding to the single crystal or polycrystalline phases, indicating that the RF co-sputtered oxide films were amorphous. The diffraction peaks of AlGaN, GaN, and substrate are not measured due to the use of grazing incidence at a small angle. The insets of Figure 1c,d show the related energy-dispersive X-ray spectroscopy (EDS) spectra for Al 2 O 3 and ZrO 2, respectively. The elemental composition ratio of the Al 2 O 3 was nearly 2:3, and that of the ZrO 2 was around 1:2. The inset of Figure 1e shows the transmission electron microscopy (TEM) image of the Al 2 O 3 /ZrO 2 stacked layer between the metal and semiconductor. The image reveals that the Al 2 O 3 and ZrO 2 layers are not crystalline, which is consistent with the XRD pattern.   The lower subthreshold swing can ensure excellent pinch-off and low power dissipation in digital applications, and good power-added efficiency in analog applications. Due to the Al2O3/ZrO2 between the gate electrode and GaN, the off-state IDS is lower than that of the reference case. Figure 3b shows the two-terminal gate leakage current for the Schottkygate HEMT and stacked-layer MOS-HEMT. The reverse gate leakage currents are 6.91 × 10 −4 mA/mm and 1.75 × 10 −6 mA/mm at VGS = −8 V, respectively. The gate leakage current can be reduced by around three orders of magnitude for the MOS-HEMT due to the higher energy barrier between the gate metal and GaN. The related turn-on voltages are 1.6 V  The lower subthreshold swing can ensure excellent pinch-off and low power dissipation in digital applications, and good power-added efficiency in analog applications. Due to the Al 2 O 3 /ZrO 2 between the gate electrode and GaN, the off-state I DS is lower than that of the reference case. Figure 3b shows the two-terminal gate leakage current for the Schottky-gate HEMT and stacked-layer MOS-HEMT. The reverse gate leakage currents are 6.91 × 10 −4 mA/mm and 1.75 × 10 −6 mA/mm at V GS = −8 V, respectively. The gate leakage current can be reduced by around three orders of magnitude for the MOS-HEMT due to the higher energy barrier between the gate metal and GaN. The related turn-on voltages are 1.6 V and more than 5 V, respectively. Similar results can be observed in the three-terminal off-state breakdown voltage for both devices as shown in the inset. The off-state breakdown voltages of the reference HEMT and Al 2 O 3 /ZrO 2 MOS-HEMT are 104 and 170 V, respectively. The off-state breakdown voltage is defined as the drain voltage at the drain current density of 1 mA/mm. Therefore, the gate leakage current and breakdown can be suppressed through the MOS structure instead of the Schottky-gate structure.     Figure 4c shows the normalized dynamic R on ratio and the related static R on at V GS = 0 V for both devices. Under the condition of high V DS , the device with a stacked oxide layer has less drain current degradation than that of the Schottky-gate HEMT, so its normalized R on is close to 1. Surface traps on the GaN are suppressed and passivated by the Al 2 O 3 /ZrO 2 , leading to the lower drain current degradation than with the reference HEMT. The degradation may be also caused by the interaction between the self-heating and trapping effect [26].  Figure 4c shows the normalized dynamic Ron ratio and the related static Ron at VGS = 0 V for both devices. Under the condition of high VDS, the device with a stacked oxide layer has less drain current degradation than that of the Schottky-gate HEMT, so its normalized Ron is close to 1. Surface traps on the GaN are suppressed and passivated by the Al2O3/ZrO2, leading to the lower drain current degradation than with the reference HEMT. The degradation may be also caused by the interaction between the self-heating and trapping effect [26].   In order to have better insight into the impact of interface traps, high-frequency capacitance-voltage (C-V) and low-frequency noise (i.e., flicker noise) for both devices were measured, as shown in Figure 5a,b, respectively. When the up and down sweep lines do not coincide, the presence of hysteresis (ΔV) is mainly result from the interface traps. The interface trap density (Dit) was calculated via the capacitance and ΔV from the high-frequency capacitance measurements [27]. The ΔV (Dit) were 220 mV (1.92 × 10 12 cm −2 eV −1 ) for the Schottky-gate HEMT and 40 mV (2.44 × 10 11 cm −2 eV −1 ) for the stacked-layer MOS-HEMT. Lower flicker noise indicates fewer defects or traps at the interface between the metal gate and semiconductor. Defects or traps capture/emission charges via biases, and the total charges within the channel change, resulting in drain current degradation as shown in Figure 4. In addition, the frequency exponent (γ) was fitted as 1.63 and 1.04 for the reference HEMT and MOS-HEMT, respectively. A larger γ closely corresponded to generation-recombination center. Due to the stacked Al2O3/ZrO2 layer, the electric field strength around the gate-drain region is relatively small, resulting in less carrier scattering in the channel. Thus, the Al2O3/ZrO2 stacked layer reduces the gate leakage current at the interface and can also satisfy the dangling bonds on the GaN to suppress defects or traps, leading to reduced carrier scattering and flicker noise. In order to have better insight into the impact of interface traps, high-frequency capacitance-voltage (C-V) and low-frequency noise (i.e., flicker noise) for both devices were measured, as shown in Figure 5a,b, respectively. When the up and down sweep lines do not coincide, the presence of hysteresis (∆V) is mainly result from the interface traps. The interface trap density (D it ) was calculated via the capacitance and ∆V from the high-frequency capacitance measurements [27]. The ∆V (D it ) were 220 mV (1.92 × 10 12 cm −2 eV −1 ) for the Schottky-gate HEMT and 40 mV (2.44 × 10 11 cm −2 eV −1 ) for the stackedlayer MOS-HEMT. Lower flicker noise indicates fewer defects or traps at the interface between the metal gate and semiconductor. Defects or traps capture/emission charges via biases, and the total charges within the channel change, resulting in drain current degradation as shown in Figure 4. In addition, the frequency exponent (γ) was fitted as 1.63 and 1.04 for the reference HEMT and MOS-HEMT, respectively. A larger γ closely corresponded to generation-recombination center. Due to the stacked Al 2 O 3 /ZrO 2 layer, the electric field strength around the gate-drain region is relatively small, resulting in less carrier scattering in the channel. Thus, the Al 2 O 3 /ZrO 2 stacked layer reduces the gate leakage current at the interface and can also satisfy the dangling bonds on the GaN to suppress defects or traps, leading to reduced carrier scattering and flicker noise. Based on the S-parameter measurement in Figure 6a, the unity-current-gain cutoff frequency (fT) and the maximum oscillation frequency (fmax) were 3 (6.4) GHz and 4.1 (9.1) GHz at maximum gm for the Schottky-gate HEMT (MOS-HEMT). The increased microwave performances of the Al2O3/ZrO2 MOS-HEMT may be attributed to the increase in the ratio of gm to gate-source capacitance or the addition of high-κ material [26]. Similar results were also observed in [28,29]. Figure 6b shows the comparison of output power, power gain, and power-added efficiency (PAE) versus input power at 2.4 GHz for both Based on the S-parameter measurement in Figure 6a, the unity-current-gain cutoff frequency (f T ) and the maximum oscillation frequency (f max ) were 3 (6.4) GHz and 4.1 (9.1) GHz at maximum g m for the Schottky-gate HEMT (MOS-HEMT). The increased microwave performances of the Al 2 O 3 /ZrO 2 MOS-HEMT may be attributed to the increase in the ratio of g m to gate-source capacitance or the addition of high-κ material [26]. Similar results were also observed in [28,29]. Figure 6b shows the comparison of output power, power gain, and power-added efficiency (PAE) versus input power at 2.4 GHz for both devices. The saturated output power, power gain, and maximum PAE are 13.4 (15.2) dBm, 8.6 (11.6) dB, and 16.7 (27.1)% for the reference HEMT (MOS-HEMT), respectively. The PAE can be supposed that the rate of input DC power is transformed into output AC power. Improved current drive, g m , and gate leakage current obtained in Al 2 O 3 /ZrO 2 MOS-HEMT are beneficial to the PAE. Therefore, Al 2 O 3 /ZrO 2 MOS-HEMT demonstrated better saturated output power and maximum PAE than those of the reference case.
devices. The saturated output power, power gain, and maximum PAE are 13.4 (15.2) dBm, 8.6 (11.6) dB, and 16.7 (27.1)% for the reference HEMT (MOS-HEMT), respectively. The PAE can be supposed that the rate of input DC power is transformed into output AC power. Improved current drive, gm, and gate leakage current obtained in Al2O3/ZrO2 MOS-HEMT are beneficial to the PAE. Therefore, Al2O3/ZrO2 MOS-HEMT demonstrated better saturated output power and maximum PAE than those of the reference case.    Figure 7a,b show the temperature-dependent I DS -V DS degradation characteristics with various temperatures ranged from room temperature (i.e., 25 • C) to 100 • C for the Schottkygate HEMT and stacked layer MOS-HEMT, respectively. The temperature program ranged from 25 • C to 100 • C as the devices were measured with an Agilent B1500A semiconductor parameter analyzer. Degradation of the Schottky-gate HEMT and the stacked-layer MOS-HEMT is 47% and 26%, respectively. The R on increased and the I DS decreased at 0 V of V GS for both devices as the temperature increased. Because of the good thermal conductivity of Al 2 O 3 , the heat dissipation is better and the degradation of the I DS is smaller than those of the reference HEMT. Figure 8a,b show the subthreshold characteristics with various temperatures for the Schottky-gate HEMT and stacked-layer MOS-HEMT, respectively. The degraded on-state I DS is due to the decreased saturation velocity or mobility [30] results from phonon scattering [31] that dominates the temperature effect at high drain current region, which is consistent with the result in Figure 7. On the contrary, the increased off-state I DS with increasing temperature is owing to the ionized traps that dominate the temperature effect results in increasing the carrier concentration in the low drain current region. The threshold voltage shift to the positive bias direction with increasing temperature can be observed for both devices. Although the electrons can jump to the conduction band easily with increasing temperature, the Al 2 O 3 /ZrO 2 stacked layer can block the tunneling current, leading to smaller variation in I DS and threshold voltage shift compared with those of the reference HEMT in Figure 8a.
is smaller than those of the reference HEMT. Figure 8a,b show the subthreshold characteristics with various temperatures for the Schottky-gate HEMT and stacked-layer MOS-HEMT, respectively. The degraded on-state IDS is due to the decreased saturation velocity or mobility [30] results from phonon scattering [31] that dominates the temperature effect at high drain current region, which is consistent with the result in Figure 7. On the contrary, the increased off-state IDS with increasing temperature is owing to the ionized traps that dominate the temperature effect results in increasing the carrier concentration in the low drain current region. The threshold voltage shift to the positive bias direction with increasing temperature can be observed for both devices. Although the electrons can jump to the conduction band easily with increasing temperature, the Al2O3/ZrO2 stacked layer can block the tunneling current, leading to smaller variation in IDS and threshold voltage shift compared with those of the reference HEMT in Figure 8a.  The time stress was biased at VGS of 1 V and VDS of 4 V, with times ranging from 1 s to 1000 s at room temperature. The monitoring of gm after time stress was performed at VDS of 4 V and VGS sweeping from −6 to 3 V for the Schottky-gate HEMT and stacked-layer MOS-HEMT, respectively, as shown in Figure 9a,b. The degradation of peak gm (16%) for both devices are almost the same; however, the gm becomes somewhat unstable as the gate bias increases in Figure 9a. The subthreshold characteristics were measured after stress at VGS of 1 V and VDS of 4 V with different time ranged from 1 s to 1000 s at room temperature. The monitoring was conducted at VDS of 4 V and VGS sweeping from −6 to 0 V for the Schottky-gate HEMT and stacked-layer MOS-HEMT, respectively, as shown in Figure  10a,b. The aforementioned results in Figure 5 describing the defects or traps behavior implies that fewer carriers were easily trapped versus detrapped due to the high energy gap of Al2O3/ZrO2, which means the smaller degradation of IDS or subthreshold current suggests that there are less interface states between the metal gate and GaN. In other words, the subthreshold current decreases with increasing time for both devices; however, a larger variation in threshold voltage shift is attributed to the capture/emission process of The time stress was biased at V GS of 1 V and V DS of 4 V, with times ranging from 1 s to 1000 s at room temperature. The monitoring of g m after time stress was performed at V DS of 4 V and V GS sweeping from −6 to 3 V for the Schottky-gate HEMT and stackedlayer MOS-HEMT, respectively, as shown in Figure 9a,b. The degradation of peak g m (16%) for both devices are almost the same; however, the g m becomes somewhat unstable as the gate bias increases in Figure 9a. The subthreshold characteristics were measured after stress at V GS of 1 V and V DS of 4 V with different time ranged from 1 s to 1000 s at room temperature. The monitoring was conducted at V DS of 4 V and V GS sweeping from −6 to 0 V for the Schottky-gate HEMT and stacked-layer MOS-HEMT, respectively, as shown in Figure 10a,b. The aforementioned results in Figure 5 describing the defects or traps behavior implies that fewer carriers were easily trapped versus detrapped due to the high energy gap of Al 2 O 3 /ZrO 2 , which means the smaller degradation of I DS or subthreshold current suggests that there are less interface states between the metal gate and GaN. In other words, the subthreshold current decreases with increasing time for both devices; however, a larger variation in threshold voltage shift is attributed to the capture/emission process of traps happening at the interface between the metal gate and GaN as shown in Figure 10a. Table 1 summarizes the dc characteristics of MOS-HEMTs with different stacked gate dielectrics in this work and by other researchers [5,12,21,23,25]. The dc characteristics presented in this work are comparable with those of other groups. The proposed Al 2 O 3 /ZrO 2 stacked-gate dielectrics are reliable for device applications.

Conclusions
This study demonstrated the improved electrical characteristics of AlGaN/GaN MOS-HEMT with Al2O3/ZrO2 stacked gate dielectrics. Improved electrical characteristics included suppressed gate leakage current, decreased flicker noise, increased high-frequency performance, better power performance, and enhanced stability after thermal stress or time stress. The gate leakage current can be reduced by three orders of magnitude due to the Al2O3/ZrO2 stacked layer, which also contributed to the lower flicker noise. To further understand the stability of the proposed device, thermal and time stresses were conducted. The thermally induced degradation of IDS was smaller than that of reference HEMT. The variation of subthreshold characteristics caused by thermal or time stress was

Conclusions
This study demonstrated the improved electrical characteristics of AlGaN/GaN MOS-HEMT with Al 2 O 3 /ZrO 2 stacked gate dielectrics. Improved electrical characteristics included suppressed gate leakage current, decreased flicker noise, increased high-frequency performance, better power performance, and enhanced stability after thermal stress or time stress. The gate leakage current can be reduced by three orders of magnitude due to the Al 2 O 3 /ZrO 2 stacked layer, which also contributed to the lower flicker noise. To further understand the stability of the proposed device, thermal and time stresses were conducted. The thermally induced degradation of I DS was smaller than that of reference HEMT. The variation of subthreshold characteristics caused by thermal or time stress was smaller than that of the reference case, showing the proposed Al 2 O 3 /ZrO 2 stacked gate dielectrics are reliable for device applications.