1. Introduction
AlGaN/GaN high-electron-mobility transistors (HEMTs) have potential applications in next-generation high-power and microwave devices. GaN exhibits high electron mobility, breakdown voltage, electron saturation speed, and thermal conductivity, because of the wide band gap. Therefore, GaN is widely used in high-frequency and high-power devices. Currently, sapphire and Si are popularly used as substrate materials for GaN. However, the large lattice mismatch between GaN buffer and Si (~17%) [
1] substrate, or the nitrogen vacancy and oxygen impurity, may result in defects and a large number of dislocations in the buffer layer, forming an n-type buffer layer [
2].
Several studies have presented doped iron (Fe) [
3,
4,
5] or carbon (C) [
2,
6,
7,
8] in the buffer layer to suppress buffer defects. If the doping position is too close to the channel layer, or the doping concentration is too strong, it may affect the reliability and electrical properties of the device. Therefore, the doped buffer devices with impurities can result in current collapse (CC), resulting in the buffer traps inducing threshold voltage (
VTH) shifts [
9].
When a high drain voltage was applied, some electrons from the two-dimensional electron gas (2DEG) were captured by dislocations or traps in the buffer layer, which caused reliability problems. Therefore, in this study, a back barrier (BB) layer was used to reduce the leakage from electron tunneling into the buffer layer and, effectively, confine electrons to the 2DEG.
2. Experiment Details
In this study, the AlGaN/GaN HEMT was grown on 6-inch Si substrates through metal organic chemical vapor deposition (MOCVD) with/without (device A/B) the Fe-doped/C-doped GaN buffer layer. TMGa, TMAl, and NH
3 were used as the sources of Ga, Al, and N, respectively. Firstly, a 50-nm initial GaN nucleation layer was grown at 550 °C, and followed by the growth of the periodically carbon-doped 200-nm GaN buffer layer. A Fe-doped GaN layer was targeted to have a Fe concentration of 8 × 10
17 atoms/cm
3 at 980 °C. Then, 1-μm intrinsically-doped GaN was grown at 1000 °C. The 50-nm Al
0.05Ga
0.95N, 300-nm GaN channel, 0.5-nm AlN, 20-nm Al
0.24Ga
0.86N, and 2-nm GaN cap were all grown under the same pressure conditions and temperature of 100 mbar and 1040 °C, respectively. To avoid the diffusion of Fe ions to the channel layer, a 50-nm Al
0.05Ga
0.95N BB was deposited between buffer layer and channel layer. A 0.5-nm-thick AlN spacer layer was grown between the GaN channel layer and the 20-nm-thick Al
0.24Ga
0.86N barrier layer. Finally, a 2-nm GaN cap was deposited through MOCVD. The schematic of a heterostructure is presented in
Figure 1a,b, which shows the 6-inch GaN on the Si HEMT wafer.
The device was fabricated through mesa isolation by using an inductively coupled plasma system with BCl3, Cl2, and a combination in the first step. Then, a Ti/Al/Ni/Au (25/130/25/80 nm) metal film was deposited through electron beam evaporation (E-gun) for the ohmic contacts (LDS = 4 µm). The device was annealed at 875 °C for 35 s in an N2 atmosphere by using a rapid thermal annealing system. The T-shaped gate (LG = 0.25 µm) was defined using electron beam lithography, and the electrode was formed using Ni/Au (50 nm/300 nm). A metal film Ti/Au (25/80 nm) was deposited as the pad for interconnection. Finally, the device was passivated with Si3N4 through plasma-enhanced chemical vapor deposition (PECVD).
To study the effect of AlGaN BB on the performance of the device, we simulated the conduction band energy diagram by using the 1-dimensional Poisson distribution, as depicted in
Figure 1c. From the energy band diagram, the peak formed by the BB can effectively suppress the penetration of electrons into the buffer layer. Er is deep level recombination centers and Et is trap energy level centers [
10,
11].
In order to understand the diffusion and incorporation of Fe into the layer structure of device A, secondary ion mass spectrometer (SIMS) profiles of Fe, C, Al, Ga, and N in the device A are shown in
Figure 2. Additionally, the slower turn-off of the Fe in the buffer layer between AlGaN back barrier and Fe-doped region may be due to memory effects of Fe diffusion associated with high growth temperatures. However, if the excess of Fe atoms diffuses into the GaN channel layer, it will work as defects, and degrade the performance of the device. Therefore, this study aims to reduce the Fe atoms diffusion by using an AlGaN back barrier.
In
Figure 2, the Fe atoms were limited to the AlGaN back barrier, because the lattice constant of AlGaN is lower than that of GaN. Because the lattice constant of AlGaN is lower than that of GaN, the distances between atoms were decreased. Therefore, the carrier is difficult to tunnel, or the tunneling speed decreases, resulting in a higher carrier concentration in the AlGaN BB region with a smaller lattice constant. Therefore, this structure can reduce the diffusion of Fe or C atoms into the channel layer, thereby improving the reliability of the device.
3. Results and Discussion
We measured
IDS–
VGS,
IDS–
VDS, and
IGS–
VGS characteristics of the two devices by using Agilent 4142B.
Figure 3a depicts the transfer characteristics (
IDS–
VGS) at
VDS = 10 V with a
VGS sweep from −6 to 2 V. The saturation current of devices A and B were 1018 and 998.2 mA/mm at
VGS = 2 V and
VDS = 10 V, respectively. The peak extrinsic transconductance values of the two devices were 318 mS/mm and 259 mS/mm. As depicted in
Figure 3a, because of the high conduction energy band of the AlGaN back barrier layer, the BB can reduce the electron distribution in the 2DEG channel of device A, causing the gate to pinch off more easily. The transfer characteristics of device A showed a higher pinch-off performance than device B. Therefore, to turn off the device, a higher negative threshold voltage was required to be applied to device B (
VTH = −3.8 V) than to device A (
VTH = −2.4 V).
To investigate the effect of the BB layer on the off-state of the device, the saturation current data in
Figure 3a were converted into the log-scale.
Figure 3b displays the leakage current curve of gate (
IGS−
VGS) and off-state leakage current curve of the drain (log-scale
IDS –
VGS). The gate off-state leakage current of devices A and B at
VGS = −6 V was 2.7 × 10
−3 and 6.7 × 10
−1, as depicted in
Figure 4. The drain off-state leakage current was 6.2 × 10
−3 (with BB) and 5.9 × 10
−1 (without BB) at
VGS = −6 V.
These results show that the addition of a BB layer can effectively reduce the gate leakage current. The BB layer also verifies the data in the simulation diagram of
Figure 1c and the data in
Figure 3b. Therefore, satisfactory pinch-off characteristics of device A, along with a moderate
Ion/
Ioff ratio of 4.66 × 10
5, can be calculated from the subthreshold swing (SS) of 0.119 V/dec. For the
IDS–VDS and
IGS–VGS of device B, an order of magnitude difference is found, which indicates that there is a phenomenon that the components are not tightly closed when the components without BB are in the off-state, showing that BB can have better characteristics and solve the problem that the component is not tightly closed in the off-state.
Figure 3c depicts the
IDS–VDS characteristics of devices A and B measured at
VDS ranging from 0 V to 10 V with a
VGS sweep from −6 V to 2 V and a step of 1 V. The on-resistances (
Ron) extracted from the devices’ drains to source current in the linear region in
Figure 3c is
Ron_with BB = 3.29 ohm-mm and
Ron_w/o BB = 2.96 ohm-mm. To understand the uniformity of the 6-inch GaN on the Si HEMT, the histograms of the saturated current (
IDS_max) are depicted in
Figure 3d.
For further analysis of the high-frequency characteristics of the device, the S-parameter was measured using an Agilent network analyzer.
Figure 4 displays the measurement conditions of the device with the BB for current gain (
fT) and power gain (
fmax) at
VGS = −1.2 V and
VDS = 10 V. The maximum
fT and
fmax of the device were 24.4 and 73 GHz, respectively. The current gain (
fT) and power gain (
fmax) at
VGS = −3 V and
VDS = 10 V were measured in the device without the BB. The maximum
fT and
fmax of the device were 23.1 and 61 GHz, respectively. This measurement result indicated that the use of an Al
0.05Ga
0.95N BB improved the pinch-off characteristics of the device, thereby improving the high-frequency characteristics.
When the device is operated in the cutoff region, the bias applied by the gate and drain causes defects in the device, resulting in the capture of electrons. This phenomenon prevents the device from achieving the expected operating current when turned on, thereby generating the CC effect. In 2014, Chen’s team used a short-pulse measurement method to explore the CC phenomenon [
12]. The pulse I–V was measured using an AMCAD AM241 pulsed system.
As depicted in
Figure 5, the dynamic
Ron ratio measurement conditions of devices A and B were a pulse width of 2 μs and period of 200 μs. In the measurement process, we set a static bias
VGSQ (quiescent voltage) of the gate and a static bias
VDSQ of the drain. First, (
VGSQ = 0 V,
VDSQ = 0 V) was measured to obtain a steady state current (static) without bias. Then, the static bias voltage of the drain terminal was increased from 0 to 30 V, step was set as 10 V, and the quiescent gate bias was −6 V. The electric field at the drain terminal renders electrons susceptible to defects existing on the surface of the device and below the drain terminal [
13].
Therefore, the dynamic
Ron ratio of the devices with, and without, the AlGaN BB was calculated. The dynamic
Ron ratio improved from 1.86 to 1.52 times at
VDSQ = 30 V, as indicated in
Figure 5. Because BB can effectively reduce the phenomenon of electrons being trapped during fast switching of the device, the AlGaN BB can effectively reduce electrons trapped by buffer layer defects.
In order to analyze the charger trapping and detrapping phenomena in GaN HEMTs with, and without, an AlGaN back barrier layer, the low-frequency noise (LFN) spectra were measured. The LFN spectra measurement condition of frequency was from 10 to 1 kHz, and the drain current power spectral density (
SID) was measured at low drain bias (
VDS = 0.05 V). To locate the noise in the channel layer, the normalized current spectral density
SI/I2 at 100 Hz and gate-to-source voltage of two devices at
VGS−
VTH [
14] from 0V to 1V was plotted (
Figure 6).
When
SI/I2 is proportional to
VG–1, noise will be generated in the heterostructure interface area. In the transition region,
SI/I2 is close to
VG–3, the main source of noise is from channel and buffer traps, and the drain–source resistance is mainly determined by the resistance
RU of the non-gate channel region, which is consistent with previous studies [
15,
16]. The results in
Figure 6 show that the device with AlGaN BB has lower noise, because BB effectively blocks noise from being affected by buffer layer defects. It results in the
SI/I2 being reduced at the
VGS−1 and
VGS−3 regions.
To analyze the reliability of the two devices, the two-terminal horizontal breakdown characteristics were measured by Agilent B1505 in
Figure 7. The measurement ohmic contact pattern distance was 40 μm mesa isolation region. The breakdown voltage of with and without AlGaN BB device was 540 and 710 V, respectively. The AlGaN BB device shows that the higher breakdown voltage due to this structure can improve the conduction band energy between channel layer and buffer layer to reduce the leakage current at high
VDS voltage.