Modulation-doped β-(Al_0.2Ga_0.8)₂O₃/Ga₂O₃ field-effect transistor

Beta-Gallium Oxide^1,2 (GO) is a promising ultra-wide band gap material with a band gap of 4.6 eV. Its large band gap and availability of bulk substrates grown using melt-based growth techniques^3–12 make it attractive for high power,¹³ high frequency, and optoelectronic^14,15 applications. Field effect transistors^16–21 and diodes with a breakdown of up to 1 kV (Ref. 22) have been demonstrated by several groups using both beta and other polymorphs of GO. Most transistor results in this material system^16–21 have been based on field-effect transistors with relatively thick channels, using Schottky or metal-oxide gate structures for charge modulation. However, as in conventional III-V semiconductors, two-dimensional electron gas (2DEG) channels provide several advantages for scaling of devices and for higher mobility.²³ In particular, in the case of GO, where the room-temperature bulk mobility is relatively low, the screening of phonon scattering in high density 2D electron gases has been predicted to lead to significantly higher mobility.²⁴ This high mobility combined with the vertical scaling enabled by 2DEGs would be very beneficial for a range of high performance device applications that exploit the large band gap of GO. Recently, such a scaled thin channel approach using delta impurity doping was used to realize a delta-doped FET²⁵ with high transconductance and current density. A modulation-doped structure would enable higher mobility than a delta-doped layer by reducing the effect of impurity scattering.

Al₂O₃ is stable in the corundum polytype, while Ga₂O₃ has multiple polytypes, with the monoclinic crystal structure (β-Ga₂O₃) being the most stable. Alloying of Al with β-Ga₂O₃ to realize β-(Al_xGa_1−x)₂O₃ (AGO) has been investigated,²⁶ and the Al₂O₃ mole fraction as high as 61% was reported while maintaining the phase purity for molecular beam epitaxy-grown films.²⁷ An Al₂O₃ mole fraction of up to 80% has been reported for AGO powders prepared by solution combustion synthesis²⁸ and pulsed laser deposition.²⁹ Evidence of modulation doping at the AGO/GO heterojunction using unintentional doping of the AGO layer has also been reported.³⁰ In this work, we report on the electrical characteristics of a delta modulation-doped epitaxial AGO/GO heterostructure and demonstrate an AGO/GO modulation-doped field effect transistor (MODFET).

An AGO/GO modulation-doped heterostructure [epitaxial stack shown in Fig. 1(a)] was grown on Fe-doped (010) β-Ga₂O₃ semi-insulating substrates.³¹ The epitaxial structure consisted of a 125 nm unintentionally-doped (UID) GO buffer layer and a 21 nm AGO barrier layer with delta doping in the AGO barrier. The AGO spacer layer with a thickness of 3 nm was used to maximize the modulation doping efficiency. Before initiating the growth, the substrate was heated up to 800 °C to remove adsorbed impurities on the substrate to avoid any parasitic channel in the buffer layer. The sample was grown³² by oxygen plasma-assisted molecular beam epitaxy using a Ga flux of 8 × 10⁻⁸Torr, an oxygen plasma power of 300 W, and a chamber pressure of 1.5 × 10⁻⁵Torr. AGO was grown³³ using an Al flux of 1.6 × 10⁻⁸Torr while keeping Ga flux at 8 × 10⁻⁸Torr. For the delta-doped layer, the Si shutter was opened for 7 s (0.25 nm), with the Si cell temperature set to 925 °C. While multiple delta-doped layers were used to achieve high charge density in our earlier report,²⁵ we used a single sheet of a delta-doped layer with a lowered target doping density of ∼9.5 × 10¹² cm⁻² in this work. The doping density was lowered to enable the complete modulation of the 2DEG charge. High resolution X-ray diffraction (HRXRD) measurements was used to determine the composition of the AGO barrier layer [Fig. 1(b)]. Assuming pseudomorphic growth, the peak separation between the AGO and the GO peak was used to determine the Al composition³⁴ of the AGO barrier layer as 20%. The full width half maximum (FWHM) value of the AGO peak was measured to be 0.24°, which is higher than the typical thickness-related broadening observed in AlGaN/GaN heterostructures for the same barrier layer thickness (0.2° FWHM for the AlGaN peak), indicating that the crystalline quality of the AGO layer can be improved further. Atomic force microscopy [Fig. 1(c)] indicates the smooth surface morphology with an rms roughness of 0.7 nm. The cross-sectional scanning transmission electron microscopy (STEM) image of a sample grown using identical growth conditions is shown in Fig. 1(d). The STEM image clearly shows the presence of the AGO/GO heterojunction with a nominal thickness in the range of 20–21 nm for the AGO layer.

Ohmic contacts were formed using a side-metal contact, where the ohmic regions were etched down to 40 nm, and a thick Ti (60 nm)/Au (50 nm)/Ni (100 nm) metal was deposited and annealed at 470 °C in N₂ ambient for 1 min in a rapid thermal anneal system. Mesa isolation was carried out using BCl₃-based inductively coupled plasma reactive ion etching (ICP-RIE) etch, and the gate metal stack of Ni/Au/Ni was deposited on the sample surface to form a Schottky barrier contact. Hall measurements were carried out using patterned van der Pauw structures, and sheet charge density of 5 × 10¹² cm⁻² and mobility of 74 cm²/V s were measured (sheet resistance of ∼16.7 kΩ/sq.).

Capacitance-voltage characterization [Fig. 2(a)] (10 kHz) showed characteristic accumulation behavior, indicating the presence of a 2D electron gas. A total sheet charge density of 8.5 × 10¹² cm⁻² was extracted from the CV curve. The anti-clockwise hysteresis of approximately 0.1 V was observed in double-sweep CV measurements, suggesting the motion of charged ions or defects within the structure. Further characterization is required to understand the observed hysteresis behavior. No frequency dispersion was observed between CV measurements at 1 kHz and 10 kHz. CV measurements could not be performed at higher frequencies due to the high series resistance in this structure.

The discrepancy between the charge density measured using Hall measurements and the CV charge is attributed to the incomplete ionization of donors in AGO at equilibrium (V_GS = 0 V). Delta doping in the AGO barrier layer results in the reduction of separation between the AGO conduction band edge and the fermi level. This results in a reduced energy difference between the donor energy level and the fermi level, leading to incomplete ionization of donors at equilibrium. When a negative gate bias is applied, the conduction band in the delta-doped AGO layer and the donor energy level are pulled up relative to the fermi level, resulting in the increased ionization of donors. The donor ionization process continues with increased negative bias until all the donors are completely ionized. When the donors are completely ionized, a further increase in negative gate bias voltage depletes the 2DEG at the heterojunction. Hence, CV characterization is expected to overestimate the 2DEG equilibrium charge density. Furthermore, the modulation of the ionized donors with gate bias results in a parasitic capacitance parallel to the 2DEG capacitance, affecting the extraction of 2DEG characteristics (charge location) exclusively from CV measurements.

The apparent charge concentration profile extracted from CV characteristics is shown in Fig. 2(b). We have taken into account the spread of the dielectric constant [10 (Refs. 35 and 36)–13 (Ref. 37)] reported in the literature. It should be noted that the dielectric constant of GO/AGO along the (010) orientation has not been characterized yet. The charge profile extracted from the CV measurements clearly shows charge confinement. The location of the apparent charge profile peak (assuming ε_r = 13) at the dopant site is due to the effect of the parasitic capacitance resulting from donor ionization, which is confirmed from CV simulation explained below.

The equilibrium energy band diagram [Fig. 3(a)] was calculated using self-consistent Schrodinger-Poisson simulations (bandeng),³⁸ for a nominal Al₂O₃ mole fraction of 20% for the AGO barrier. Material parameters (ε—relative permittivity, m*_e—effective electron mass, Φ_B—AGO surface barrier, and ΔE_c—AGO/GO conduction band offset) were based on previous experimental and theoretical estimates and are listed in Table I. The Schottky barrier height was extracted to be 1.4 eV using internal photoemission measurements [Fig. S1 see the supplementary material]. The calculated equilibrium sheet charge (5 × 10¹² cm⁻²) matches well with the experiment (Hall measurement) if an AGO/GO conduction band offset of 0.6 eV and a donor energy level of 135 meV (E_C-E_D) are assumed in AGO. This would indicate that the band gap difference²⁷ appears completely as the conduction band offset at the AGO/GO heterojunction. The Si donor sheet density in the delta-doped layer was set to be 9.5 $×$ 10¹² cm⁻². The simulated CV curve matched well with the experiments when the dielectric constant was assumed to be 13.³⁷ Further detailed experiments are required to measure band offsets, donor energy level in AGO and dielectric constant of AGO.

Electrical characteristics of the AGO/GO MODFET is shown in Fig. 4. Due to the non-ohmic source/drain contacts (Fig. S2 in the supplementary material) and high contact resistance, the transistor characteristics were measured on I-shaped structures consisting of wide source/drain contact/access regions (W_contact = 100 μm) and a narrow channel region (W_channel = 2 μm) [Fig. 4(a)]. The maximum current [Fig. 4(a)] normalized with respect to the channel width was measured to be 5.5 mA/mm. Measurements on devices with different channel width and length dimensions clearly indicate that the measured current density is rather limited by the contacts (Fig. S3 and Table S1). The pinch-off voltage of −3 V measured in the transfer characteristics [Fig. 4(b)] was similar to the pinch-off voltage in the CV characteristics, and an on-off ratio of 2.5 × 10⁵ was obtained. A peak transconductance of 1.75 mS/mm was measured. While this report of modulation doping in such a transistor structure shows the feasibility of this approach, further growth optimization would be needed to achieve the optimal mobility.

In summary, we report modulation-doped (Al_0.2Ga_0.8)₂O₃/Ga₂O₃ heterostructures with the evidence of carrier transfer from the (Al_0.2Ga_0.8)₂O₃ layer to the Ga₂O₃ layer. Further work on improving the transport and resistance in such devices could enable the realization of high performance heterostructure devices based on the AGO/GO material system.

See supplementary material for internal photoemission characterization of the Ni/(Al_0.2 Ga_0.8)₂O₃ barrier height, source/drain contact characterization, and the effect of the device width and length on maximum drain current of AGO/GO MODFET.

We acknowledge funding from the Office of Naval Research under Grant No. N00014-12-1-0976 (EXEDE MURI). The project or effort depicted was or is sponsored by the Department of the Defense, Defense Threat Reduction Agency (Grant HDTRA11710034). The content of the information does not necessarily reflect the position or the policy of the federal government, and no official endorsement should be inferred. We acknowledge funding from The Ohio State University Institute of Materials Research (IMR) Multidisciplinary Team Building Grant. We thank Air Force Research Laboratory, WPAFB, Dayton Ohio for support.