Effect of buffer iron doping on delta-doped β-Ga₂O₃ metal semiconductor field effect transistors

Beta-phase gallium oxide (β-Ga₂O₃) is a semiconductor with an ultra-wide bandgap of 4.5–4.9 eV,^1,2 a high breakdown field of ∼8 MV/cm,³ bulk mobility >200 cm²/V.s,³ and an estimated electron velocity of ∼2 × 10⁷ cm/s (Ref. 4) that results in large theoretical Baliga⁵ and Johnson⁶ figure-of-merits. These excellent electronic properties coupled with economical melt-based growth solutions to manufacture high-grade large-area bulk substrates^7,8 make β-Ga₂O₃ a potential candidate for next generation high-power and radio frequency (RF) applications. Schottky diodes^9–11 and field effect transistors (FETs)^12–17 exhibiting excellent device performance have already been realized. Parallel efforts towards Ga₂O₃ based RF devices include β-Ga₂O₃ metal-oxide-semiconductor FETs (MOSFETs) with a cut off frequency (f_T) of 3.3 GHz (Ref. 18) and β-(Al_xGa_1−x)₂O₃/Ga₂O₃ modulation doped FETs (MODFETs) with f_T of 3.1 GHz.¹⁹ 

Recently, a Si- delta (δ) doped metal-semiconductor FET (MESFET) utilizing a patterned regrowth process flow for low-resistance ohmic contacts was demonstrated.²⁰ The 2-dimensional electron gas (2-DEG) of the δ-doped channel, besides providing higher sheet charge density than uniformly doped channels, spatially extends to near lattice constant dimensions.²¹ Hence, δ-doped FETs can enable utmost miniaturization of the gate-channel distance, i.e., they can enable a scaled device topology while maintaining relatively high mobility and are therefore suitable for RF based device applications. Besides, a constant electric field profile between the gate and the 2-DEG²² results in higher breakdown voltage for δ-doped channels compared to uniformly doped channels where the peak electric field location is just below the gate.

For high power and RF applications, it is necessary to have semi-insulating substrates to avoid degradation in device performance due to conduction losses.²³ Fe, a deep level acceptor in β-Ga₂O₃,²⁴ is used to realize semi-insulating β-Ga₂O₃ substrates through compensation of background n-type carriers. There are prior reports of diffusion of Fe into the epitaxial layer during growth and with annealing.²⁵ In this work, we report on the effect of Fe impurities on transport properties and device performance of δ-doped MESFETs. We demonstrate excellent sheet charge (σ) with improved Hall mobility and reduced knee-walkout under pulsed conditions. This is achieved by using a thick buffer layer that ensures a negligible Fe concentration near the 2-DEG. Minimal knee-walkout/DC-RF dispersion would mean maximum power efficiency, making the device suitable for high-power and RF operations.²⁶ An increase in the buffer thickness from 100 nm to 600 nm resulted in Hall mobility improvement from 65 cm²/V.s (σ = 1.38 × 10¹³ cm⁻²) to 105 cm²/V.s (σ = 1.57 × 10¹³ cm⁻²). Moreover, δ-doped MESFETs with a 100 nm buffer thickness showed 33% DC-RF dispersion, with dispersion seen to reduce significantly for 600 nm buffer.

The following samples were investigated for this report. A thick unintentionally doped (UID) Ga₂O₃ buffer was grown on an Fe-doped substrate to investigate Fe diffusion using secondary ion mass spectroscopy (SIMS). To investigate the effect of the Fe-substrate on the 2-DEG, the separation between the 2-DEG and the substrate was varied. Figure 1(a) shows the two-dimensional schematic of the device epitaxial stack which consists of an unintentionally doped (UID) β-Ga₂O₃ buffer layer (nominally ∼100 nm, 300 nm, and 600 nm), a Si δ-doped sheet, and a Ga₂O₃ UID barrier layer (nominally ∼35 nm).

The epitaxial layers were grown by oxygen plasma-assisted molecular beam epitaxy (MBE) on commercially available Fe-doped semi-insulating (010)-oriented β-Ga₂O₃ substrates.²⁷ Growth was realized in a gallium-limited regime with a growth rate of 3.3 nm/min. Growth conditions used are as follows: chamber pressure of 1.5 × 10⁻⁵Torr, O₂ plasma power of 300 W, 8 × 10⁻⁸Torr Ga flux, 700 °C growth temperature, and Si cell temperature at 925 °C (to realize delta doping in the channel) and 1050 °C (Si cell temperature for regrowth). Ohmic source-drain contacts with low contact resistance were obtained using a patterned regrowth process flow. Source/drain regions were patterned and recessed using a SiO₂ mask and highly conductive (n⁺) β-Ga₂O₃ was regrown in these regions. The residual SiO₂ was etched using a buffered oxide etchant (BOE), which lifted off regrown n⁺ β-Ga₂O₃ outside the source-drain region. Ti/Au/Ni as ohmic metals were then deposited on the regrown region followed by a 470 °C, 1 min anneal in N₂ ambient in a rapid thermal annealer (RTA). Further details on device growth and patterned regrowth can be found in Ref. 20. Device mesa isolation was done using BCl₃/Ar based dry etch in a Plasma-Therm SLR 770 Inductively Coupled Plasma- Reactive Ion Etching (ICP-RIE) chamber. Finally, gate patterns were formed and Ni/Au/Ni was deposited as the Schottky gate metal stack.

A smooth surface morphology was observed in all the samples using atomic force microscopy (AFM), with an RMS roughness of ∼0.7 nm [Figs. 1(b)–1(d)]. The SIMS profile for a thick UID buffer grown on an Fe-doped substrate is plotted in Fig. 2(a). The diffusion of Fe in the buffer is evident, and its doping concentration falls to the detection limit above 200 nm. The electron Hall mobility as a function of buffer thickness is plotted in Fig. 2(b). Hall charge/mobility values of 1.38× 10¹³ cm⁻²/65 cm²/V.s for 100 nm, 1.46 × 10¹³ cm⁻²/79 cm²/V.s for 300 nm, and 1.57 × 10¹³ cm⁻²/105 cm²/V.s for the 600 nm buffer are seen. Hall mobility is seen to increase with increase in the buffer thickness. While the residual Fe concentration at the 2-DEG is different between the three samples, it is not sufficient to explain the change in mobility, since the Fe concentration is much lower than the electron density. The electron mobility in delta-doped electron gases is higher than in bulk layers with the same effective doping²⁸ because the electrons have less overlap with the sheet donor layer than in a bulk-doped layer with the same donor concentration. However, the wavefunction of the 2-DEG formed by delta doping is strongly dependent on the electric field due to the gate/surface potential, as well as the buffer potential. While the top electric field is the same in all the samples investigated, the electric field due to the buffer is significantly different. As the buffer thickness decreases, the electric field can make the electron wavefunction more localized, thereby increasing the overlap of the electron wavefunction with the donor sheet and thereby the scattering rate. This qualitatively explains the increasing mobility trend with buffer thickness observed here. The increase in charge density as a function of buffer thickness is attributed to a combination of variation in growth and back-depletion due to the Fe doped substrate, as seen in Fig. 2(c). The figure shows the impact of Fe back-depletion to conduction band energy for the 100 nm, 300 nm, and 600 nm buffers simulated using a self-consistent Schrodinger Poisson solver.²⁹ 

δ-doped transistors fabricated using 100 nm, 300 nm, and 600 nm buffer thicknesses were analyzed. The leakage current between the isolation patterns with 7 μm separation between the ohmic pads was measured up to 200 V. Lower leakage is seen for the 100 nm and 300 nm buffers. The 600 nm buffer shows the highest leakage which is one-two order higher than the 100 nm one. The lower leakage for the 100 nm and 300 nm buffers is attributed to the presence of Fe in between the isolation patterns which yields more resistive IVs. The isolation leakage for 600 nm buffer is still very low (40 nA/mm at 200 V) which shows that the undoped layer in the thick buffer does not create significant additional parasitic leakage paths between source and drain. Using regrown contacts, sheet resistances of 3.5 kΩ/□ for the 600 nm buffer, 5.4 kΩ/□ for the 300 nm buffer, and 6.7 kΩ/□ for the 100 nm buffer were measured, in agreement with van der Pauw structures. A low contact resistance of ∼0.35–0.44 Ω . mm was extracted from transfer length measurements (TLMs).

Near flat C-V curves (not shown here) confirmed the presence of 2-DEG in the channel. The apparent doping profiles extracted from C-V plots for the devices are shown in Fig. 3(a). 2-DEG peak concentrations of 2.7 × 10¹⁹ cm⁻³ at ∼35 nm, 2.8 × 10¹⁹ cm⁻³ at ∼37 nm, and 3.1 × 10¹⁹ cm⁻³ at ∼38 nm from the gate/UID cap layer interface were seen for the 100 nm, 300 nm, and 600 nm devices, respectively. Differences are attributed to growth rate variation. The transfer characteristics (I_DS-V_GS) are shown in Fig. 3(b). The device dimensions L_G (gate length), L_GS (gate to source regrowth edge), and L_GD (gate to drain regrowth edge) are 1 μm, 0.8 μm, and 3 μm for the 100 nm buffer; 1.2 μm, 1.4 μm, and 4 μm for the 300 nm buffer; and 1.1 μm, 1.3 μm, and 4 μm for the 600 nm buffer. All dimensions were confirmed using a scanning electron microscope (SEM). An on/off ratio, I_ON/I_OFF ∼10⁵–10⁶, was measured in the devices. The off-state current leakage is ∼2 μA/mm (at −16 V for 600 nm and −12 V for 100 nm) and ∼0.4 μA/mm (at −12 V for 300 nm), which is much higher compared to the isolation leakage currents shown in Fig. 2(d). The off-state leakage is hence limited by the Ni/Au/Ni gate. Pinch-off voltages of −8 V (100 nm), −10 V (300 nm), and −12.5 V (600 nm) were estimated from the transfer curves. The sub-threshold slope, as extracted from the transfer curves, was found to be ∼650 mV/decade, ∼145 mV/decade, and ∼300 mV/decade for the 600 nm, 300 nm, and 100 nm buffers, respectively. The sub-threshold slope showed a non-monotonic dependence on the buffer thickness, with the lowest slope being observed for the 300 nm buffer layer. To understand this, 2D simulations were carried out using Silvaco ATLAS, assuming that Fe was present with a uniform concentration (10¹⁷ cm⁻³) up to 200 nm from the regrowth interface. Figure 3(c) plots measured and simulated sub-threshold slopes as a function of the buffer thickness. Fe as a deep level acceptor was seen to increase the sub-threshold slope when present in the vicinity of the 2-DEG. The large sub-threshold slope for the 600 nm buffer is attributed to the higher integrated charge density of background (N_D-N_A) doping of the buffer as compared to the 100 and 300 buffer. While the simulations do not match quantitatively with the measured results, they do predict the non-monotonic trend observed in the experiment. Other trap levels in the buffer layer were not considered here and could be the reason for this discrepancy. For example, since growth was realized in a gallium limited regime, gallium vacancies³⁰ may be present and could therefore have an impact on the subthreshold slope. Double-pulsed current-voltage and constant drain current DLTS measurements also predict a trap level at E_C–0.7 eV.³¹ 

Pulsed IV measurements were performed on a Keithley 4200-SCS parameter analyzer using a pulse width of 5 μs and 0.1% duty cycle. Pulsed IV data for 100 nm, 300 nm, and 600 nm buffer devices are shown in Figs. 4(a)–4(c), respectively. The devices were biased at off-state quiescent bias points below pinch-off and V_DSQ = 15 V (Q is used in the subscript for pulsed voltage values). Substantial knee-walkout was seen for the 100 nm buffer sample. The saturated current density was reduced by 33% going from DC to pulsed conditions. Reduced DC-RF dispersion was seen for the 300 nm buffer compared to the 100 nm buffer, with further reduction for the 600 nm buffer. This is attributed to the presence of Fe in the 2-DEG for the 100 nm buffer, 2-DEG being close to the region with appreciable Fe doping for the 300 nm buffer, and 2-DEG being far away from the Fe doped region for the 600 nm buffer. The saturated current density for the pulsed condition (600 nm) was found to be higher than the DC value, due to the absence of self-heating. However, some knee walkout or dynamic on-resistance (R_ON) degradation was still evident in the IV curves. R_ON degradation (ΔR_ON) normalized to R_{ON, (0, 0)} [R_ON extracted from the I_DS-V_DS curve with devices biased at an off-state (V_GSQ, V_DSQ) = (0, 0)] at different off-state V_DSQ using a pulse width of 5 μs and 0.1% duty cycle is plotted in Fig. 5 keeping off-state V_GSQ below pinch-off for the devices. R_ON was extracted from the triode region of I_DS-V_DS curve measured at V_GS = 0 V. ΔR_ON is extracted using the equation

where R_{ON, (x, y)} stands for R_ON extracted at the quiescent bias condition (V_GSQ, V_DSQ) = (x, y).

Since R_{ON, (0, 0)} is similar to R_ON extracted from the DC I_DS-V_DS curve with no self-heating effects, Fig. 5 indicates the degree of knee-walkout in both the devices as a function of off-state V_DSQ. A large increase in normalized ΔR_ON is seen for the 100 nm buffer. Reduced knee-walkout for the 300 nm buffer compared to the 100 nm buffer with V_DSQ is observed, with further reduction for the 600 nm buffer. A similar phenomenon is observed for R_ON plotted at different V_DSQ for I_DS-V_DS curves with different V_GS. This highlights the robustness of δ-doped MESFET with a thick buffer towards obtaining higher output power at higher drain voltages. However, the knee-walkout is non-negligible and suitable surface passivation is needed to further reduce it.

In summary, thick UID buffer which essentially results in a negligible Fe concentration near the 2-DEG is shown to have better device performance in terms of improved mobility and minimal DC-RF dispersion compared to a thin buffer layer where the 2-DEG is subjected to Fe doping compensation. The reported results put forward β-Ga₂O₃ δ-doped MESFETs with thick buffers as promising devices for scaled β-Ga₂O₃ high-power and RF FETs. Future enhancements in field management and passivation could facilitate high performance transistors that would benefit from the high breakdown field of β-Ga₂O₃.

We acknowledge funding from the Department of Defense, Defense Threat Reduction Agency (Grant No. HDTRA11710034), Office of Naval Research EXEDE MURI program (Grant No. N00014-12-1-0976), and OSU Institute for Materials Research Seed Program. We thank the Air Force Research Laboratory (D. Dorsey, G. Jessen, and K. Chabak) for support. 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.