We investigate the polarization modulation effect of a single-crystalline BeO layer on AlGaN/GaN high-electron-mobility transistors (HEMTs). The BeO layer with macroscopic polarization on top of the AlGaN barrier layer increases the 2-dimensional electron gas density in the triangular quantum well (QW) at the interface of the AlGaN/GaN heterostructure. Electronic band bending of BeO and a deeper triangular QW observed from the simulated conduction band profile indicate that the BeO layer can modify the polarization field at the AlGaN/GaN interface. A ∼20-nm-thick single-crystalline BeO thin film is grown on AlGaN/GaN HEMTs by atomic-layer deposition. Room-temperature and variable-temperature Hall-effect measurements confirm that the HEMT with BeO forms a channel with a 14% increase of the sheet carrier concentration as compared with a conventional HEMT. An improved output performance is also observed in the I-V characteristics which confirms the polarization modulation effect of the BeO layer.
Recent development in GaN-based high-electron-mobility transistor (HEMT) technology enables the realization of high-power and high-frequency devices in power and RF electronics.1 The wide-bandgap Group III-nitride (III-N) semiconductors offer many advantages in HEMTs such as a high breakdown voltage, high saturation current, and excellent carrier confinement. However, the HEMTs still suffer from various negative effects such as current collapse, gate leakage, junction heating, and trap effect for high voltage operation.2,3 Moreover, conventional AlxGa1−xN/GaN HEMTs with typically x ≤ 25% show limited sheet concentrations in the channel consisting of 2-dimensional electron gas (2DEG), e.g., ≤1 × 1013 cm−2. Increasing the AlN mole fraction (x) can further enhance the 2DEG density; however, it will also introduce additional technical challenges, such as an increasing lattice mismatch between AlGaN and GaN. The authors previously demonstrated a significant increase in 2DEG by external bending of the heterostructure, which enhances the piezoelectric polarization.4,5 However, the enhancement can only be realized in mechanically bendable thin-film structures which require a layer-transfer process followed by the removal of a brittle substrate.6,7
BeO (nonsingle-crystalline) thin films grown by atomic layer deposition (ALD) were explored and demonstrated as a gate dielectric material for Si and III–V metal/oxide/semiconductor field-effect transistors (MOS-FETs) to utilize their excellent properties such as a large binding energy and an ultrawide-bandgap energy.8 Recently, the authors demonstrated single-crystalline α-BeO (wurtzite structure) thin films on GaN by epitaxial growth using ALD.9 The single-crystalline BeO shows a low interface-trap density and superior thermal conductivity (κc = 370 W/m K at 300 K) and stability,9,10 which makes the BeO a promising material for the isolation and passivation of III-N HEMTs with efficient heat dissipation. Furthermore, single-crystalline wurtzite BeO can offer additional functionality in AlGaN/GaN HEMTs. A (0001) BeO film can be grown on (0001) III-N substrates, which means the out-of-plane direction of the BeO film is aligned in the [0001] crystallographic direction. In such a direction, the BeO film shows strong spontaneous and piezoelectric polarizations arising from its noncentrosymmetric crystal structure. The formation of a 2DEG channel in AlGaN/GaN HEMTs originates from the charge at the heterointerface induced by the polarizations.11,12 Therefore, the polarization field in the BeO film will modify the polarizations in the HEMTs when integrated with AlGaN/GaN heterostructures. In the present study, we show that a single-crystalline BeO film grown on an AlGaN/GaN HEMT can modify the total polarization fields, resulting in an increase in the 2DEG sheet density at the AlGaN/GaN interface via both numerical simulation and experimental characterization studies of the structures and devices.
We prepared two HEMT structures in this study: a reference HEMT and a HEMT with BeO. The reference HEMT device has a typical layer structure consisting of an Al0.25Ga0.75N Schottky barrier layer (20 nm) and a GaN layer that are grown on a Si (111) substrate with AlN and AlGaN buffer layers by metalorganic chemical vapor deposition. For the HEMTs with BeO (to be referred to as a BeO HEMT in this paper), a BeO layer (20 nm or 5 nm) was deposited on top of the reference HEMT heterostructure by ALD. The process conditions for ALD were described in an earlier report.13 The computational simulation was carried out using a commercial 1D technology computer-aided design device simulator (STR, FETIS) for the calculation of the electronic band diagram and the carrier distribution for both the reference HEMT and BeO HEMT structures. Quantum-mechanical treatment of the carrier confinement in the HEMT structure is based on a self-consistent solution of the Poisson and Schrödinger equations. The 2DEG properties were characterized by a temperature-variable Hall-effect measurement system (Ecopia, HMS-5000). For the measurement, the BeO HEMT sample was first etched in hydrofluoric acid (HF) with a Cr mask to create an opening for the ohmic contact on the AlGaN surface. Ti/Al/Ni/Au (20/80/50/50 nm) metal stacks were then deposited by electron-beam evaporation followed by rapid thermal annealing (RTA) at 850 °C in an nitrogen ambient for both HEMT structures. The HEMT structures were fabricated into devices with a Ti/Al/Ni/Au source and drain electrodes and an Ni/Au gate electrode.
Figure 1 schematically shows the spontaneous (Psp) and piezoelectric (Ppz) polarizations in an AlGaN/GaN HEMT epitaxial structure (Group-III polar) with a single-crystalline BeO layer. The Psp of BeO (−0.045 C/m2) is not significantly different from that of Al0.25Ga0.75N (∼−0.042 C/m2);14 hence, differential spontaneous polarizations ΔPsp at the BeO/AlGaN interface are negligible. Piezoelectric constants and stiffness constants of BeO are e13 = −0.02 C/m2, e33 = 0.02 C/m2, c13 = 4.70 × 1012 dynes/cm2, and c33 = 1.67 × 1012 dynes/cm2, respectively.14 The piezoelectric constants of BeO are smaller than those of GaN and AlN. However, when a large in-plane tensile strain is induced in the BeO layer, a substantial piezoelectric polarization field Ppz forms in the BeO, which changes the electronic band diagram of the heterostructure. This results in the modification of the energy levels that create the triangular quantum well (QW) at the AlGaN/GaN interface where the 2DEG channel is formed.
Simulated electronic band diagrams of the reference HEMT and BeO HEMT are compared. The previous reports on the electronic properties of BeO are limited and the electronic band structure of BeO still remains uncertain. The authors estimated the band structure of the BeO and BeO/(Al)GaN interface from the ALD-grown BeO film. A bandgap energy of 8.3 eV, a valence band maximum (VBM) of 4.96 eV, and an electron affinity (EA) of 0.36 eV were measured for the BeO film by reflection electron energy loss spectroscopy (REELS) and ultraviolet photoelectron spectroscopy (UPS).13 The authors have reported that the epitaxial growth of BeO on GaN is possible even with a significant lattice mismatch [e.g., ∼18% using a formula (as − al)/al for a BeO film (al = 2.698 Å) on a GaN substrate (as = 3.186 Å)], via domain-matching epitaxy (DME) to reduce the lattice-mismatching strain.15 A relaxation degree of 20.8% was calculated from the DME-BeO by a reciprocal space map measured by high-resolution x-ray diffraction. An equivalent lattice constant of BeO is then calculated as 2.9 Å which is to be applied in the simulation. After the parameters for BeO were defined, the simulation for both structures was performed. Figure 2 shows the simulated conduction band minimum profiles. For the reference HEMT shown in a red dot-dashed line, the negative spontaneous and piezoelectric polarization fields cause the band bending of AlGaN. As a result, a triangular QW is formed below the Fermi level (blue dashed line). The 2DEG is induced due to the negative total polarization fields and the charge density (ρ) in the QW being related to the total polarization by the equation
where Ppz is the piezoelectric polarization and Psp is the spontaneous polarization. The 2DEG sheet density for the reference HEMT is calculated as 9.2 × 1012 cm−2. For the BeO HEMT shown in a green solid line, the induced Ppz of BeO is in a negative direction due to the tensile in-plane strain in the BeO layer. Since the Psp of BeO is also in the negative direction,16–18 the total polarization fields of this three-layered structure are increased, which results in a higher 2DEG charge density. The electronic band bending of both BeO and AlGaN layers due to the negative polarization fields is observed and the triangular QW at the AlGaN/GaN interface is deeper below the Fermi level than that of the reference HEMT as shown by the dashed green and red arrows which indicate the location of QW bottoms of the BeO HEMT and reference HEMT, respectively. The 2DEG sheet density for BeO HEMT is calculated to be 1.08 × 1013/cm2. Based on the simulation results, a 17% increase in the 2DEG sheet concentration is estimated.
Room-temperature Hall-effect measurement results of the reference and BeO (20 nm) HEMT structures are shown in Table I. The 2DEG sheet concentration of the reference HEMT was measured to be 9.22 × 1012 cm−2 which is very close to the simulated result of 9.2 × 1012 cm−2. A high electron mobility of 1430 cm2/V s was also observed. For the BeO HEMT, a 2DEG sheet density of 1.05 × 1013 cm−2 was measured, which corresponds to an ∼14% increase in the electron concentration in the channel, while the mobility remains at a similar value. Hall coefficients along two diagonal directions show a good consistency due to the square shape. The coefficients at room temperature are −1.69 cm3/C and −1.69 cm3/C for the directions of the reference HEMT and −1.5 cm3/C and −1.49 cm3/C for the BeO HEMT. The conductivity value of the BeO HEMT shows an increase of 11% compared to that of the reference HEMT. Temperature-variable Hall-effect measurement was also performed and the results are shown in Fig. 3. The inset shows a schematic cross section of the Hall-effect measurement sample of a BeO HEMT. The reference HEMT and BeO HEMT show the same trend in both the electron sheet concentration and mobility when plotted vs temperature. However, the BeO HEMT demonstrates a higher level of electron sheet concentration which represents an increase of the 2DEG sheet density. The decrease in the carrier density after 250 K can be attributed to the thermal escape of carriers out of the triangular QW.19,20 The BeO HEMT also shows a smaller decrease in the carrier density than the reference HEMT at a higher temperature, which suggests a deeper QW of the BeO HEMT. The 2DEG mobility values for both HEMT structures are also shown in the secondary y-axis of Fig. 3. A small difference of 2% is found between the HEMTs. The electron mobility decreases with a higher temperature, i.e., from 4300 cm2/V s at 135 K to around 1000 cm2/V s at 350 K. The temperature dependence of mobility here mostly arises from the temperature dependence of scattering due to acoustic phonons.21 Both room-temperature and variable-temperature Hall-effect measurements show a significant increase of the 2DEG sheet concentration for the BeO HEMT and the increased value (14%) is also close to the simulated result (17%).
. | Sheet conc. (cm−2) . | Mobility (cm2/V s) . | Conductivity (Ω−1 cm−1) . | Average Hall coef. (cm3/C) . |
---|---|---|---|---|
Ref. HEMT | −9.22 × 1012 | 1430 | 844 | −1.69 |
HEMT w/BeO | −1.05 × 1013 | 1400 | 938 | −1.50 |
. | Sheet conc. (cm−2) . | Mobility (cm2/V s) . | Conductivity (Ω−1 cm−1) . | Average Hall coef. (cm3/C) . |
---|---|---|---|---|
Ref. HEMT | −9.22 × 1012 | 1430 | 844 | −1.69 |
HEMT w/BeO | −1.05 × 1013 | 1400 | 938 | −1.50 |
The output characteristics were measured for both HEMT devices with gate length Lg = 6 μm, gate-drain distance Lgd = 6 μm, and gate-source distance Lgs = 3 μm. For the comparison of devices, the BeO HEMT with a 5-nm-thick BeO layer was fabricated to compare with the reference HEMT with similar threshold voltages. By decreasing the BeO thickness from 20 nm to 5 nm, Vth changes from −4.8 V to −4.2 V which is closer to −3.8 V of the reference HEMT (data not shown here). Figure 4 shows the DC output characteristics with the Vds ranging from 0 V to 10 V. The saturation drain currents of the BeO HEMT at Vg = −1 to −3 V are higher than those of the reference HEMT by 10% or more than the current change with the threshold voltage difference, which reflects a higher 2DEG density in the BeO HEMT. In addition, the off-state drain current, which is mostly caused by the gate leakage current below the threshold voltage, of the BeO HEMT (∼6 × 10−4 mA/mm at Vg = −10 V) is 3 orders of magnitude lower than that of the reference HEMT (∼5 × 10−1 mA/mm) (data not shown here). However, the saturation current of the BeO HEMT at Vg = 0 V is slightly higher than that of the reference HEMT. Furthermore, both HEMTs show a slight current collapse behavior at Vg = 0 V, which requires further investigation whether self-heating takes place, overshadowing the effect of the 2DEG density difference.
Interface trap density and total polarization charge change can both modify the 2DEG density. The basic expression for sheet charge concentration in the HEMT device is given by22
where σpol is the total spontaneous and piezoelectric polarization induced charge concentration, εAlGaN and dAlGaN are the permittivity and thickness of AlGaN, q is the charge of the electron, EF(ns) is the difference between the Fermi level and the bottom of the conduction band in the GaN layer, φBn is the Schottky barrier height, and ΔEC is the conduction band offset at the AlGaN/GaN interface. When a single crystalline BeO layer is deposited on AlGaN by ALD, it can reduce the interface trap density resulting from the efficient passivation of the GaN dangling bonds with crystalline BeO and increase the Schottky barrier height φBn.23 Based on Eq. (2), the reduced interface trap density and increased Schottky barrier height φBn will lead to a decrement of the 2DEG density. However, the measured results show the opposite which confirms that the increased 2DEG is actually generated from the increased polarization charge change of σpol.
In addition, the high κc of BeO offers a potential thermal benefit by enhancing the heat dissipation away from the device active region. A coupled electrothermal simulation was performed to demonstrate the effectiveness of BeO films as heat spreaders for an AlGaN/GaN HEMT operating under 5 W/mm. As shown in Fig. 5, a 2-μm-thick BeO film can reduce the device peak temperature by 16% as compared to a device employing a standard Si3N4 passivation layer.
In summary, we have studied the effect of the BeO layer on a conventional HEMT via both numerical simulation and experimental demonstration. The simulated conduction band of the BeO HEMT showed a deeper triangular QW and a higher polarization induced charge density than that of the conventional HEMT. BeO HEMT was fabricated by growing an ∼20-nm or 5-nm BeO layer on top of the AlGaN barrier layer by ALD. Both room-temperature and variable-temperature van der Pauw Hall-effect measurements demonstrated a higher sheet carrier concentration of about 14%. The output characteristics showed an improved performance of the BeO HEMT which confirms a higher 2DEG density. The origin of 2DEG increase was discussed and we believe the surface passivation effect of BeO can reduce the 2DEG density while the additional polarization charge from the BeO layer can offset this effect and lead to the final increment of the 2DEG density. Besides the polarization modulation effect, other functions such as passivation and fast heat dissipation make BeO even more promising for AlGaN/GaN HEMTs at a high voltage and high temperature operation.
This material is based upon work at the University of Houston supported by the National Science Foundation under Grant No. 1842299 [Electrical, Communications and Cyber Systems (ECCS)]. J.-H.R. also acknowledges partial support from the Texas Center for Superconductivity at the University of Houston (TcSUH) and the Advanced Manufacturing Institute (AMI). The study at Yonsei University was supported by the MIST (Ministry of Science and ICT), Korea, under the “ICT Consilience Creative Program” (IITP-2019-2017-0-01015) supervised by the IITP (Institute for Information & Communications Technology Promotion) and by the Korea Electric Power Corporation (under Grant No. R18XA06-03). Funding for efforts by the Pennsylvania State University was provided by the AFOSR Young Investigator Program (Grant No. FA9550-17-1-0141, Program Officers: Dr. Brett Pokines and Dr. Michael Kendra. This work was also monitored by Dr. Kenneth Goretta).