In-situ dielectric Al₂O₃/$\beta$-Ga₂O₃ Interfaces Grown Using Metal-organic Chemical Vapor Deposition

Saurav Roy,* Adrian E. Chmielewski, Arkka Bhattacharyya, Praneeth Ranga, Rujun Sun, Michael A. Scarpulla, Nasim Alem, and Sriram Krishnamoorthy*

S. Roy, A. Bhattacharyya, P. Ranga, Dr. R. Sun, Prof. M. A. Scarpulla, Prof. S. Krishnamoorthy
Department of Electrical and Computer Engineering, University of Utah, SLC, UT 84112
Dr. A. E. Chmielewski, Prof. N. Alem
Department of Materials Science and Engineering, Pennsylvania State University, University Park, Pennsylvania 16802
Email Address: [email protected], [email protected]

The rise of $\beta$-Ga₂O₃ as a promising material for power electronic applications due to its extremely high breakdown field (8 MV/cm)^{[1, 2]} and Baliga figure of merit (BFOM) has also provided impetus for the search of an appropriate insulator material as the gate dielectric for device applications. For transistor applications, a gate dielectric should exhibit low leakage currents, have low interface trap densities to achieve a controllable threshold voltage (V_T), and should also have a higher breakdown field compared to the underlying semiconductor. Many insulators such as SiO₂,^{[3, 4, 5, 6]} Al₂O₃,^{[7, 8]} SiN,^[9] and HfO₂^{[9, 10]} etc. have been studied as a gate oxide material and passivation layers for Gallium Oxide devices. Extreme permittivity materials such as Barium Titanate (BaTiO₃) were also studied and used as dielectric material in $\beta$-Ga₂O₃ based transistors and heterojunction schottky barrier diodes.^{[11, 12]} Complex oxides such as ternary rare earth alloys (Y_0.6Sc_0.4)₂O₃ were also investigated by Masten et. al. as gate dielectric for $\beta$-Ga₂O₃ based MOS structures.^[13]

Among all the dielectric materials, Al₂O₃ is studied and used most extensively for $\beta$-Ga₂O₃ based devices due to its compatibility with $\beta$-Ga₂O₃. Chabak et. al. demonstrated various lateral MOSFETs with Al₂O₃ as gate dielectric showing excellent field strength. ^{[14, 15, 16]} Li et. al. also demonstrated $\beta$-Ga₂O₃ based vertical FinFET structures with outstanding Figure of Merit (FOM) using Al₂O₃ as gate dielectric.^{[17, 18, 19]} Jian et. al. studied the effects of post deposition annealing on the interface trap density of the ALD deposited Al₂O₃ on (001) $\beta$-Ga₂O₃.^[20] In all these reported works the deposition of Al₂O₃ is primarily performed using ex-situ technique such as Atomic layer deposition (ALD), where after its growth, the gallium oxide substrate or epilayer is taken to a different reactor for dielectric deposition. During such a transfer step, it is hard to completely remove any interface contamination. Using in-situ deposition of Al₂O₃ in the same reactor as the underlying Ga₂O₃, the surface of the Ga₂O₃ can be kept pristine without exposure to the ambient. The quality of the dielectric could also improve due to the high temperature growth by metal-organic chemical vapor deposition (MOCVD) (500 to 1000 ⁰C), compared to ALD (100 to 300 ⁰C). MOCVD also enables high deposition rates, significantly higher than that of the ALD technique. MOCVD reactor can also facilitate high temperature in-situ annealing which can reduce the dielectric/semiconductor interface state density as well as the improve the bulk dielectric quality. The in-situ growth of Al₂O₃ on both Ga-face and N-face GaN has been studied extensively using MOCVD.^{[21, 22, 23, 24, 25]} In this paper, we demonstrate for the first time in-situ growth of Al₂O₃ using MOCVD and electrical characterization of metal/in-situ Al₂O₃/$\beta$-Ga₂O₃ vertical MOSCAPs.

The growth of in-situ Al₂O₃ layer after the epitaxial growth of $\beta$-Ga₂O₃ on Sn doped (010) $\beta$-Ga₂O₃ and the fabrication of MOSCAPs as shown in Figure 1 to analyze the in-situ Al₂O₃/$\beta$-Ga₂O₃ interface quality is discussed in the experimental section. To evaluate the abruptness of the Al₂O₃/$\beta$-Ga₂O₃ interface as well as the stoichiometry of the Al₂O₃, scanning transmission electron microscope imaging within a high angle annular dark field detector (HAADF-STEM) combined with energy dispersive x-ray spectroscopy (EDS) were used on a cross-section of the interface prepared by focused ion beam (FEI Helios 660). HAADF-STEM images were performed using a FEI Titan G2 60-300 STEM with an acceleration voltage of 300kV, a condenser lens aperture of 70 µm and a probe current of about 100 pA. Figure 2 (a) shows a HAADF-STEM image of the Al₂O₃/$\beta$-Ga₂O₃ interface in the [101] projection in which the $\beta$-Ga₂O₃ substrate is brighter then the Al₂O₃ film due to the heavier Ga atoms. The Al₂O₃/$\beta$-Ga₂O₃ interface is found to be abrupt with a transitional region of about 2 nm in which Ga atoms contrast vanishes and the high resolution is lost. The abruptness of the Al₂O₃/$\beta$-Ga₂O₃ interface was also confirmed by EDS mapping (Bruker) performed in the same region. Figure 2 (b) shows a HAADF-STEM image of the interface with its corresponding EDS maps of (c) Ga, (d) Al and (e) O. A smooth transition between the two materials is observed with a small diffusion of Al atoms at the top of the $\beta$-Ga₂O₃ epilayer. In addition, the composition of Al₂O₃ could be extracted from the EDS mapping analysis using the Cliff-Lorimer ratio technique detailed in the supporting information.^[26] As a result, the film reveals an uniform composition with 38$\pm$2 % of Al and 62$\pm$2 % of O and stays relatively homogeneous along the Al₂O₃ film as shown by the EDS linescan of Figure 3 (b).

To characterize the trap densities, the fabricated MOSCAPs with different thicknesses were subjected to a series of CV measurements using Keithley 4200, shown in Figure 3. The alternating current (AC) bias and frequency used were 30 mV and 1 MHz respectively and the MOSCAPs were swept at a rate of 200 mV/sec. The dielectric constant was extracted to be 8.2-8.5 from the accumulation capacitance on all three samples. The methodology used to extract the trap densities is same as the technique developed by Liu et. al. ^[22] and presented briefly here. The MOSCAPs were first taken from depletion (D) to accumulation (A) and kept at accumulation for 10 minutes to fill all the empty electron traps. After 10 minutes, they were again swept back to depletion (1st sweep). Another CV curve is measured by sweeping the MOSCAP from depletion to accumulation and immediately back from accumulation to depletion (2nd sweep). The total charge (Q_I) within the MOSCAPs is calculated from the equation (1),

where, V_FB is the flat band voltage, T_ox is the oxide thickness, V${}_{FB}^{Ideal}$ is the ideal flat band voltage, $q$ is the electron charge, $\epsilon$ is the relative permittivity of the Al₂O₃ layer, and $\epsilon_{0}$ is the absolute permittivity. Assuming all the charges are located at or near the oxide/semiconductor interface, Q_I can be determined from the linear fit of the V_FB vs oxide thickness plot. The change in V_FB for D to A and A to D sweeps of the hysteresis plot can be attributed to the total amount of initially empty trapped charge within the MOSCAPs as described by the equation (2),

Where, $\Delta$V_FB is the flat band voltage difference between the D to A and A to D sweeps and $\Delta$Q_T is the trapped electron density. Figure 4 (a) and (b) shows the 1st and 2nd sweeps of the CV hysteresis plots respectively for the MOSCAPs with three different Al₂O₃ thicknesses. The CV measurements were repeatable after the 2nd sweep, overlapping with the 2nd sweep measurements. Different forward bias voltages were applied for MOSCAPs with different Al₂O₃ thicknesses to maintain approximately same electric field through the oxide ($\sim$ 2 MV/cm). Figure 4 (c) and (d) shows the corresponding V_FB and $\Delta$V_FB for the 1st and second sweeps. Flat band voltages were extracted from the deflection point of the capacitance voltage plot.^[27] V_FB1 and V_FB2 are the flat band voltages calculated from the D to A sweeps of the 1st and 2nd sweep, respectively. Fixed charges (Q_I1 $\&$ Q_I2) are determined from the slope of the flat band voltage vs oxide thickness plot. Q_I1 is the fixed charge of the MOSCAP without any bias history. So, Q_I2 is the total fixed near interface charge in the same MOSCAP, but also contains trapped electrons due to the measurement sweeps and 10 minutes of accumulation stress. Now, $\Delta$V_FB1 and $\Delta$V_FB2 are the flat band voltage differences, between each pair of D to A and A to D sweeps of the 1st and 2nd sweep respectively. During the 10 minutes of accumulation stress of the first hysteresis measurement electrons fill the empty slow near interface states and due to the high emission time of the slow traps these electrons remain trapped afterwards. However, electron will fill the fast near interface states also during the 10 min accumulation time. So, $\Delta$V_FB1 can be attributed to the initially empty fast and slow near interface states ($\Delta$Q_T1). Only the fast traps will respond during the second hysteresis and subsequent measurements since the slow traps were filled during the first sweep. So $\Delta$V_FB2 can be attributed to only the fast near interface states($\Delta$Q_T2). Density of the initially empty slow near interface traps can be evaluated from the difference between $\Delta$Q_T1 and $\Delta$Q_T2 ($\left|\Delta Q_{T1}-\Delta Q_{T2}\right|$). However, during the first D to A and A to D sweeps, there could be electron emission from some of the slow near interface states due to the applied reverse bias. So, the difference in flat band voltage for the D to A sweeps of 1st and 2nd sweep can be attributed to the more stably filled slow near interface states ($\Delta$Q_TS). All the fixed and trapped charges described here are summarized in Table 1. The positive fixed charge observed in this work is similar to the previous report of ALD Al₂O₃/ (-201) Ga₂O₃ interface.^[7]

The conduction band offset at the Al₂O₃/$\beta$-Ga₂O₃ heterostructure can be evaluated using equation (3) as V_FB can be written as,^[7]

where, E_ox is the electric field in the oxide at flat band condition due to the fixed charge at the, $\phi_{b}$ is Ni/Al₂O₃ conduction band barrier height, $\phi_{S}$ the difference between the conduction band and fermi level in $\beta$-Ga₂O₃, and $\Delta$E_C is the conduction band offset between $\beta$-Ga₂O₃ and Al₂O₃. $\phi_{b}$ was considered to be 3 eV as evaluated by Zhang et. al. using internal photoemission spectroscopy.^[28] From Figure 4 (c) and (d), ($\phi_{b}-\Delta E_{C}-\phi_{S}$) is extracted to be 1.23 eV. Using equation 3, the value of $\Delta$E_C is calculated to be 1.7 eV which matches well with the previous reports on the ALD Al₂O₃/$\beta$-Ga₂O₃ heterostructure.

Now, to find the density of all the initially filled slow near interface states and the trap distribution as a function of energy dependence near the $\beta$-Ga₂O₃ band edge, we use UV assisted CV technique developed by Swenson et. al.^[29] and later modified by Liu et. al.^[30] This technique was also used by Jian et. al.^[20] for ALD Al₂O₃/(001) $\beta$-Ga₂O₃ MOSCAPs and briefly summarized here. The MOSCAPs were first taken from D to A and kept at accumulation for 10 minutes to fill all the interface states. After 10 minutes time the MOSCAPs were again swept back to depletion without any optical excitation. This dark curve is considered as the ideal CV curve, assuming all the states are filled with electrons. Next the MOSCAPs were held at depletion and were exposed to UV light using a DUV LED with 254 nm wavelength for 5 minutes. After 5 minutes the UV light is turned off and the MOSCAPs were held at depletion for 15 minutes followed by another sweep from D to A. The dark curve is then shifted horizontally to match the post DUV CV curve and obtained as the reference CV curve. The total density of the all the interface states is calculated from the difference between the post DUV and the reference CV curve ($\Delta$V) and their energy dependence is calculated from the amount of band bending (dielectric/semiconductor surface potential) that occurs during the bias sweep.

Figure 5 (a) to (c) shows energy band diagrams of the MOSCAPs under applied bias and UV illumination. During the UV exposure, electron hole pairs are generated and the photo-generated holes either recombine with the electrons trapped in the interface or gets self-trapped, which results in the ledge observed in the post DUV curve. Due to the recombination of photo generated holes with the trapped electrons at the Al₂O₃/$\beta$-Ga₂O₃ interface, the trapped states become positively charged, which changes the width of the depletion region. Some of the extra holes also get trapped in the dielectric bulk hole traps inside the Al₂O₃ layer. Figure 6 (a) to (c) shows CV plots for the MOSCAPs with three different oxide thicknesses. The black curve (Dark) is the A to D sweep after the 10 minutes of accumulation stress. The purple curve is the D to A sweep after DUV exposure and the dashed black curve is the reference curve obtained after shifting the dark curve to match the post DUV plot. $\Delta$V in Figure 6 (a) to (c) includes the effects of both interface traps at the Al₂O₃/$\beta$-Ga₂O₃ interface and bulk hole traps inside the Al₂O₃ layer. Density of all the initially filled interface and bulk dielectric traps is calculated using the equation (4),

where, $\psi_{S}$ is the surface potential at the Al₂O₃/$\beta$-Ga₂O₃ interface, given by $\psi_{S}$ = $\frac{q\epsilon\epsilon_{0}N_{D}A^{2}}{2C_{dep}^{2}}$. N_D is the doping density, $A$ is the area of the contacts, C_dep is the depletion capacitance and C_ox is the oxide capacitance.

Figure 6 (d) to (f) shows the D_t vs the trap energy distribution for the MOSCAPs with three different oxide thicknesses. Due to the existence of a positive valence band offset, the extra holes (which did not recombine with the interface traps) get accumulated at the Al₂O₃/$\beta$-Ga₂O₃ interface (some holes also get self-trapped). The rise in the D_t near the conduction band edge can be attributed to the accumulated holes, either at the Al₂O₃/$\beta$-Ga₂O₃ interface or possibly also due to the self-trapped holes in the depletion region.^[31] The average D_t for each of the MOSCAPs are obtained after eliminating the contribution from holes by neglecting the rise in D_t.

We can also see from Figure 6 (d) to (f) that the average D_t reduces as the thickness of the Al₂O₃ layer decreases. There would not be any dielectric thickness dependence if all the traps were located at the Al₂O₃/$\beta$-Ga₂O₃ interface. Instead, due to the presence of bulk hole traps inside the Al₂O₃ layer, some of the holes get captured inside the Al₂O₃ layer. So, the density of the interface states (D_it) without the contribution from the bulk hole traps in the Al₂O₃ is calculated from the equation (5).^[30]

Where, n_ht is the amount of bulk hole traps inside the Al₂O₃. So, average D_it is obtained to be 7.8 $\times$ 10¹¹ cm^-2 eV^-1 by extrapolating the linear fit of average D_t vs oxide thickness for T_ox = 0 as shown in Figure 7. The amount of bulk hole traps (n_ht) is estimated from the equation 5 to be 2.1 $\times$ 10¹⁷ cm^-3 eV^-1. The fact that we see a linear relationship of D_t vs oxide thickness clearly indicates that not all the holes are self-trapped in the $\beta$-Ga₂O₃ and in-fact can move and get inside the Al₂O₃ layer and get trapped. The mobility of holes could be significant enough to cause bulk hole trapping in the dielectric as also extracted by Akyol ^[32] for $\beta$-Ga₂O₃ schottky barrier diode based photodetector. The value of D_it obtained for the in-situ Al₂O₃ MOSCAPs are significantly lower than the previous reports of ALD Al₂O₃/ (001) $\beta$-Ga₂O₃ interface and may have resulted from the O vacancies and Al dangling bonds at the interface.

To determine leakage currents and breakdown voltages of the MOSCAPs, reverse and forward current voltage (IV) characteristics is measured and shown in Figure 8 (a) and (b) respectively. The measurement is performed at a sweep rate of 200 mV/sec. Breakdown voltages with positive bias on the gate metal (accumulation) for MOSCAPs with 14, 24 and 55 nm Al₂O₃ thicknesses are 9, 14, and 32 V, respectively. The breakdown field for all the MOSCAPs were determined using Sentaurus device simulator^[33] and the field profile is shown in Figure 9 (a). For the accumulation case with positive potential on the gate, the entire voltage is dropped across the Al₂O₃ layer and so the breakdown can be attributed to the breakdown of the dielectric. The average breakdown field at the center of the gate metal region and peak field at the edges of the metal for the Al₂O₃ dielectric are estimated to be $\sim$ 5.8 MV/cm and 15 MV/cm respectively from electric field simulations as shown in Figure 9 (a). Electric fields were shown only in the Al₂O₃ region during forward bias as the entire bias voltage is dropped across the oxide layer. The barrier to tunneling for the MOSCAPs were also estimated using trap assisted tunneling model and extracted to be 1.1 eV and presented in the supporting information. The reverse breakdown voltage from the IV characteristics in Figure 8 (a) for MOSCAPs with 14, 24 and 55 nm Al₂O₃ thicknesses are approximately 40, 80, and 200 V, respectively. From the electric field distribution in Figure 9 (b), a peak field of $\sim$ 15 MV/cm is observed at the metal edges but the center of the gate metal, the electric field is less than the observed field in case of forward bias ($\sim$ 3.5 MV/cm as compared to 5.8 MV/cm). This is due to the fact that, at reverse bias, the total applied voltage is dropped across both Al₂O₃ and Ga₂O₃ region and a significant amount of electric field is present in the Ga₂O₃ ($\sim$ 2.6 MV/cm) as also shown by the dashed curves in Figure 9 (b). At reverse bias, all the three MOSCAPs (T_ox = 14, 24 and 55 nm) have a peak electric field of around $\sim$ 15 MV/cm, indicating that the breakdown is probably happening at the metal edges. These high edge electric fields can be reduced by creating a field plate structure using the same in-situ Al₂O₃ material showing the versatility of this technique.

The in-situ Al₂O₃ grown using MOCVD, in this work, is found to have reduced fixed charges and interface traps (D_it) compared to the more commonly used ALD technique. A lower fixed charge concentration of +2 × 10¹² cm^-2 is found when compared to ALD Al₂O₃ on (-201) $\beta$-Ga₂O₃ (+3.6 $\times$ 10¹² cm^-2) reported by Hung et. al.^[7] A lower interface trap density (D_it) of 7.8 $\times$ 10¹¹ cm^-2eV^-1 is also found in comparison to ALD Al₂O₃/ (001) $\beta$-Ga₂O₃ interface reported by Jian et. al. (1.34 $\times$ 10¹² cm^-2eV^-1).^[20] It is to be noted that no high temperature post deposition annealing (PDA) or post metallization annealing (PMA) is performed on these MOSCAPs, which is expected to further improve the interface quality.^[34] Further experiments to understand the effect of growth parameters such as growth temperature, chamber pressure and precursor ratios and also higher temperature PDA will be necessary to improve the dielectric constant, fixed charge, interface traps and critical breakdown fields of the Al₂O₃ layers in these fully MOCVD-grown Al₂O₃/$\beta$-Ga₂O₃ MOSCAPs. Also, detailed work is necessary to investigate the origin of the interface traps which may have resulted from native defects and impurities at the interface and can be performed using structural and spectroscopic techniques.

In conclusion, we report in-situ Al₂O₃ growth using MOCVD as a potentially better alternative to ALD and the interface quality is investigated using capacitance voltage characterization. The presence of near interface fixed charge in these MOSCAPs were identified and quantified using the linear relationship between flat band voltage and oxide thicknesses and a stable positive fixed charge of +2 $\times$ 10¹² cm^-2 is calculated. The sheet density of trap states with fast and slow emission times were also calculated from linear relationship of the CV hysteresis versus Al₂O₃ thickness plots. The density of all the near interface traps and their energy dependence were calculated using UV-assisted CV technique as low as +7.8 $\times$ 10¹¹ cm^-2. Conduction band offset using the linear relationship between flat band voltage and oxide thicknesses is calculated to be 1.7 eV. Furthermore, the average breakdown fields for the MOCVD grown Al₂O₃ material is found to be approximately 5.8 MV/cm. This approach of in-situ dielectric deposition on $\beta$-Ga₂O₃ can pave the way as gate dielectrics for future $\beta$-Ga₂O₃ based high performance MOSFETs due to its promise of improved interface and bulk quality compared to other conventional dielectric deposition techniques.