Characterization of electronic states at insulator/(Al)GaN interfaces for improved insulated gate and surface passivation structures of GaN-based transistors

Gaffey et al.²⁶⁾ reported a precise characterization of interface properties of a SiO₂/SiN_x/SiO₂/n-GaN structure prepared by jet vapor deposition. From the conductance–frequency analysis, they obtained a state density of 5 × 10¹⁰ cm⁻² eV⁻¹ at E_C − 0.8 eV with a capture cross section of 2.4 × 10⁻¹⁷ cm² for the MIS sample fabricated using the optimum process condition. Matocha et al.²⁹⁾ and Kim et al.³⁰⁾ presented low state densities at the SiO₂/n-GaN interfaces prepared at high temperatures (830–900 °C). Kanechika et al.³¹⁾ fabricated a vertical-type AlGaN/GaN MOS HEMT, demonstrating good I–V characteristics with a high on/off ratio using a SiO₂ film thermally deposited at 830 °C. It was also found that high-temperature annealing process after the SiO₂ deposition was also effective in decreasing the interface state density.^32–35) Kambayashi et al.³⁴⁾ reported over 100 A operation in AlGaN/GaN hybrid HEMT using the SiO₂ gate dielectric post-deposition annealed at 700 °C. Very recently, a channel mobility of 192 cm² V⁻¹ s⁻¹ was achieved in a hybrid HEMT using a Al₂O₃/SiO₂ bilayer gate structure.³⁶⁾ Although the SiO₂ film has a low permittivity as shown in Fig. 1, it remains attractive for MIS HEMT applications because of its large bandgap and chemical stability.

The SiN_x film is widely used for surface passivation of GaN-based transistors. Hashizume et al.^20,37) demonstrated that the SiN_x deposition on N₂-plasma treated surfaces can produce a good SiN_x/n-GaN interface with a state density as low as 1 × 10¹¹ cm⁻² eV⁻¹ at E_C − 0.8 eV. The reason for the low interface state densities at the SiN_x/n-GaN interface is not yet fully understood. It is likely that nitrogen radicals generated during the SiN_x deposition process control the formation of nitrogen-vacancy-related defects at the GaN surface,³⁸⁾ leading to the prevention of interface electronic states. Green et al.³⁹⁾ were the first to apply-SiN_x passivation to the AlGaN/GaN HEMTs, and have observed a significant improvement of microwave output power as a consequence.

The in-situ deposition of SiN_x is an interesting process in the sense that it can be performed in the same growth chamber and in the same growth sequence as the rest of the layer stack.⁴⁰⁾ During the final growth stage of the AlGaN/GaN structure by metal organic chemical vapor deposition, the SiN_x layer was in-situ deposited using NH₃ and SiH₄ on the AlGaN surface at 1020 °C. Derluyn et al.⁴⁰⁾ obtained improved I–V characteristics from HEMTs with the in-situ deposited SiN_x layer. The in-situ process realized an oxide-free SiN_x/AlGaN interface.⁴¹⁾ From the high-resolution transmission electron microscopy (HR-TEM) analysis, Takizawa et al.⁴²⁾ reported that the in-situ SiN_x has a single crystalline structure. Very recently, Van Hove et al.³⁾ demonstrated that an AlGaN/GaN HEMT with a stable threshold voltage and a low current collapse can be achieved using an in-situ SiN_x/Al₂O₃ bilayer gate.

As mentioned in Sect. 2, the Al₂O₃ has a large bandgap, a relatively high dielectric constant and a high breakdown field. In addition, there has been considerable progress of atomic layer deposition (ALD) technology, simplifying the application of Al₂O₃ to the gate and passivation structures in the GaN transistors.^43–45) Park et al.⁴⁶⁾ were the first to report AlGaN/GaN MOS HEMTs using an Al₂O₃ prepared by ALD as a gate dielectric and a surface passivation layer. The ALD process realized good Al₂O₃/n-GaN interface with a low state density of around 1 × 10¹¹ cm⁻² eV⁻¹ at E_C − 0.8 eV.^17,47)

When using Al₂O₃, special attention should be given to subsequent process temperatures. From the transmission electron microscope (TEM) observation, Hori et al.¹⁷⁾ found a flat and smooth atomic arrangement around the ALD prepared amorphous Al₂O₃/n-GaN interface. However, annealing at 800 °C for the formation of ohmic electrodes generated a large number of microcrystallized regions in the Al₂O₃ layer, causing a marked increase in the leakage current of the Al₂O₃/GaN structure. Toyoda et al.⁴⁸⁾ also reported that annealing procedures at 800 °C resulted in the phase transformation of Al₂O₃ films from amorphous to crystalline. To address this issue, an ohmic-first process, i.e., formation of ohmic electrodes including annealing at 800 °C before the Al₂O₃ deposition, was developed for Al₂O₃/n-GaN structures.¹⁷⁾ This method effectively suppresses leakage current in MOS structures. Nevertheless, even in this scheme, a surface protection layer is indispensable to avoid the chemical bond disorder at the GaN surface during the annealing process. Optimization of the ohmic-first process using a SiN_x surface protection layer then realized a good Al₂O₃/n-GaN interface with a low state density of 2 × 10¹¹ cm⁻² eV⁻¹ at E_C − 0.8 eV.¹⁷⁾

For practical applications of Al₂O₃ to the MOS HEMTs, stability and reliability characterization of the Al₂O₃ films is necessary. From a time-dependent dielectric breakdown measurement, Kikuta et al.⁴⁹⁾ reported that a time-to-breakdown (t_BD) at 3 MV/cm for the ALD-Al₂O₃ film was more than 40000 years at RT. However, the t_BD significantly decreased to 10²–10³ s at 250 °C. Esposto et al.²⁸⁾ and Son et al.⁵⁰⁾ pointed out that relatively high densities of fixed charges in Al₂O₃ films shifted flat-band voltages in the C–V curves of the Al₂O₃/n-GaN capacitors. The origin for the fixed charges is not clear yet. Choi et al.⁵¹⁾ theoretically predicted that the oxygen vacancy in Al₂O₃ introduces transition levels close to the conduction band edge of GaN, and it can act as border traps at the Al₂O₃/n-GaN interface. On the other hand, it is likely that other kinds of defects such as Al vacancy and interstitials act as fixed charges in Al₂O₃.⁵¹⁾

High-κ insulators are attractive for the MIS transistors for achieving high g_m. Recently, relatively good I–V characteristics were reported for AlGaN/GaN HEMTs with high-κ gate insulators.^52–55) Hatano et al.⁵⁵⁾ demonstrated the improved operation stability in the AlGaN/GaN HEMT with an Al₂O₃/ZrO₂ bilayer gate dielectric. However, interface properties of high-κ/AlGaN structures are not well understood. In addition, severe V_th fluctuations were reported by Hayashi et al.⁵⁶⁾ and Stoklas et al.⁵⁷⁾ in HEMTs with HfO₂- and ZrO₂-gate dielectric materials, respectively. Therefore, further investigation is necessary to gain better understanding of high-κ insulators/(Al)GaN interfaces.

To improve device performance and stability of MIS HEMTs using AlGaN/GaN and InAlN/GaN structures, evaluation of electronic state properties of the interfaces between insulators and AlGaN/InAlN is of utmost importance. However, fabrication of a simple MIS diode using a thick AlGaN or InAlN is challenging due to the difficulty of growing high-quality epitaxial layers. For the practical device application, it is desirable to characterize insulator/AlGaN or insulator/InAlN interfaces using HEMT MIS structures. In this regard, we initially attempt to interpret precisely the resulting C–V characteristics of the HEMT MIS structures.

Figure 3(a) shows a schematic illustration of the HEMT MIS structure used in this study. An undoped AlGaN/GaN heterostructure grown on a sapphire substrate by metal organic chemical vapor deposition (MOCVD) was used as the starting wafer (provided by NTT-AT). The sheet resistance and mobility of the AlGaN/GaN heterostructure were 500 Ω/sq. and 1750 cm² V⁻¹ s⁻¹, respectively. We prepared Al₂O₃/AlGaN/GaN samples with and without dry etching of the AlGaN surface. For the etching of AlGaN, we used a Cl₂-based dry etching process assisted by inductively coupled plasma (ICP) at RT. The ICP and bias powers were 300 and 5 W, respectively. The resulting etching depth was 7 nm. We then deposited a 10-nm-thick SiN_x film as a surface protection layer to avoid damage to the AlGaN surface during ohmic annealing.¹⁷⁾ As an ohmic electrode, a ring-shaped Ti/Al/Ti/Au multilayer structure was deposited on the AlGaN surface, followed by annealing at 800 °C for 1 min in N₂ ambient. After removing the SiN_x film, an Al₂O₃ layer with a nominal thickness of 20 nm was deposited at a deposition rate of 0.11 nm/cycle on the AlGaN surface using an ALD system (SUGA-SAL1500) at 350 °C for 170 cycles. During the deposition process, water vapor and TMA were introduced into an ALD reactor in alternate pulse forms. Finally, a circular Ni/Au (20/50 nm) gate electrode concentric with the ohmic electrode was deposited on the Al₂O₃ layer.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 3(b) shows a typical C–V curve of the Al₂O₃/AlGaN/GaN structure without the etching of the AlGaN surface. C–V measurements were carried out using an impedance analyzer (HP4192 LF) with a measurement frequency of 1 MHz at RT. A characteristic C–V curve with two steps was observed, which is peculiar to the MIS structure fabricated on the heterostructure including a two-dimensional electron gas (2DEG).^10,44,58,59) The constant capacitance at the forward bias corresponds to the Al₂O₃ capacitance, whereas that at the reverse bias is determined by the total capacitance (C_TOTAL) of the Al₂O₃ and AlGaN layers. At a deep reverse bias, the capacitance steeply decreases to nearly zero, indicating the 2DEG depletion at the AlGaN/GaN interface.

To understand this peculiar C–V behavior, we carried out calculations using a numerical solver of the Poisson equation based on the one-dimensional Gummel algorithm, taking into account the fixed charge at the AlGaN/GaN interface originating from spontaneous and piezoelectric polarization as well as the charge in the electronic states at the insulator/AlGaN interface.⁶⁰⁾ We assumed the 2DEG density of 1 × 10¹³ cm⁻² and an insulator thickness of 20 nm. Other parameters used in calculation are listed in Table I, where Φ_B is the Schottky barrier height at the metal/insulator interface, ΔE_C1 and ΔE_C2 are the conduction-band offsets between insulator/AlGaN and AlGaN/GaN interfaces, respectively. Figure 4(a) shows the calculated MIS C–V curves neglecting interface state density (ideal curve) for various insulators. The two-step behavior is reproduced in all HEMT MIS structures. Insulator capacitance, total capacitance (C_TOTAL) and threshold voltage at the step 2 systematically change along with the changing value of permittivity as expected. The AlGaN-thickness dependence of the C–V curves is shown in Fig. 4(b). Similarly, a systematic change is seen in C_TOTAL and the threshold voltage.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The ideal curves show almost the same slope at both steps 1 and 2. This is completely different from the experimental result shown in Fig. 3(b). To shed a light on this, we carried out calculations for the Al₂O₃/AlGaN/GaN structure, taking into account the effect of interface electronic states.⁶⁰⁾ Figure 5(a) shows the interface state density distributions D_it(E) used in calculation, illustrating the acceptor- and donor-like states divided by the charge neutrality level E_CNL.^60,61) We adapted the E_CNL values calculated by Mönch in Ref. 62. In addition, we used electron emission time constant τ from interface states to the conduction band calculated using Shockley–Read–Hall (SRH) statistics:

$$\begin{equation} \tau = \frac{1}{v_{\text{TH}}\sigma_{\text{TH}}N_{\text{C}}}\exp\left(\frac{E_{\text{T}}}{kT}\right), \end{equation} \tag{ 1 }$$

where $v$_TH, σ_TH, N_C, and E_T are electron thermal velocity, capture cross section, density of state at the conduction band, and interface state energy, respectively.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The calculated C–V curves are shown in Fig. 5(b). In the forward bias region, the C–V slope drastically decreases with increasing interface state density, similar to the experimental results. At forward bias, the nearly flat potential of the AlGaN layer shown in Fig. 6(a) can lead to the electron transfer with ease from the AlGaN/GaN interface to the Al₂O₃/AlGaN interface, resulting in the steep C–V slope relative to the case without interface states. However, in the actual case, electron trapping at the Al₂O₃/AlGaN interface states is highly probable. The acceptor-type states result in excess negative charges when they trap electrons. Such negatively charged interface states can screen the gate electric field, suppressing the potential modulation of the AlGaN layer. This causes the stretch out of the C–V curve, as shown in Fig. 5(b). In an extreme case, if a HEMT MIS sample includes very high interface state densities, then the step 1 will not be observed even at high forward bias. Indeed, several papers showed no C–V step even at high forward bias.^11,12) On the other hand, in the reverse bias region, the stretch-out behavior was not observed in the C–V curve. Figure 6(b) shows the potential distribution at V_G = −8 V. The Fermi level (E_F) is located far below the valence band maximum of AlGaN at the Al₂O₃/AlGaN interface, which makes electron occupation of interface states no longer a function of the gate bias, leading to absence of stretch-out behavior in the C–V curve. In addition, Eq. (1) gives very long time constants of electron emission from deep states to conduction band. Therefore, electrons captured at deep states remain unaffected even when a large negative bias is applied to the gate electrode. This leads to the fact that the interface states act as "fixed and frozen" charges.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 7 schematically shows the charge condition of the interface states at RT. When a value of 1 × 10⁻¹⁶ cm² is used for the capture cross section of the interface states, only the upper half of the acceptor-like traps can change their charge state (N_A⁰/N_A⁻) accordingly with the gate voltage sweep. The acceptor-type states in this energy region produce excess negative charges when electrons are trapped, whereas they are in the neutral charge state when electrons are detrapped, mainly causing the stretch out of the C–V curve at the forward bias. Due to the associated long emission time constants, electrons in the lower half of acceptor-like traps remain trapped and act as negatively fixed charges, even when the E_F is located far below the valence band maximum of AlGaN [Fig. 6(b)]. This leads to the parallel C–V shift (V_th shift) at step 2 toward the positive bias direction with respect to the ideal C–V curve, as shown in Fig. 5(b). The V_th shift is enhanced with increasing interface state density. Similarly, the donor-like traps remain on their neutral charge state (filled with electrons), independent of the bias sweeping. Since their neutral condition remains unchanged at RT, it is rather difficult to evaluate the effect of the donor-like traps by a standard C–V measurement.

Zoom In Zoom Out Reset image size

The conductance method is often applied in HEMT MIS structures to analyze interface states.^63–65) Using this technique, careful attention should be given to the E_F position at the insulator/AlGaN interface. At reverse bias regime, the interface states could not be responsible for the conductance peak in the conductance–frequency (G–f) curve,^63,64) because of the associated long emission time constants of interface states. In addition, Shih et al.⁶⁵⁾ pointed out that the conventional conductance method is viable only for a narrow range of gate bias voltages, and invalid for the analysis of deeper interface states. They proposed a unique characterization method using G–f characteristics at forward bias and high temperatures. From an analysis based on the G–f–T (temperature) mapping, they estimated an energy range of interface states responding to the G–f measurement and a gate control efficiency related to an interface state density.⁶⁵⁾

To evaluate near-midgap electronic states at insulator/AlGaN interfaces at RT, we have developed a photo-assisted C–V method.¹⁰⁾ Figure 8 shows schematic illustrations of photo-ionization effects of interface states under a monochromatic light with energy less than the bandgap of GaN. First, under dark condition, a forward gate voltage high enough to observe the insulator capacitance is applied. At this state, almost all of interface traps are filled with electrons under a nearly flat band condition. Then, the gate bias is swept toward a value more negative than the threshold voltage. After reaching a sufficiently negative gate bias, a monochromatic light with photon energy of hν₁ is illuminated to the sample surface. Consequently, we are effectively inducing photo-assisted electron emission from the interface states within the energy range corresponding to the photon energy range, as schematically shown in Fig. 8(a). After switching the light off, we restarted the voltage sweep toward 0 V under dark condition. As shown Fig. 8(b), we can observe the C–V curve shift toward the reverse bias direction, corresponding to the change in the interface state charge Q_it (hν₁).

Zoom In Zoom Out Reset image size

Using different photon energies, we can induce a systematic C–V shift according to the photon energy, as shown in Fig. 8(b). The voltage shift difference (ΔV) between two photon energies corresponds to the interface charge difference, ΔQ_it, in the energy range Δhν shown in Fig. 8(c). The state density can be simply estimated using the observed ΔV in the following equation:

$$\begin{equation} D_{\text{it}}(E = E_{\text{AV}}) = \frac{C_{\text{TOTAL}}\cdot\Delta V}{q\cdot\Delta h\nu}, \end{equation} \tag{ 2 }$$

where C_TOTAL is the total capacitance of Al₂O₃ and AlGaN and E_AV is the average interface energy schematically shown in Fig. 8(c).

It should be emphasized that the present method uses incident photons with energies less than the bandgap of GaN. On the other hand, some groups have reported photo-assisted C–V characterization of MIS structures using a light source with photon energies larger than the GaN bandgap.^66,67) In this case, holes are generated and accumulated at insulator/AlGaN and AlGaN/GaN interfaces. In addition, interaction of holes with the interface states especially at reverse bias has to be considered. This makes the charge dynamics in HEMT MIS structures complicated and the evaluation of interface state states very difficult. Very recently, Matys et al.^68,69) proposed a simulation method for the photo-assisted C–V characteristics, taking into account the electron–hole generation and the recombination processes as well as the hole capture process at the interface states. Comparing their calculation results to the experimental data, they were able to estimate the state density distribution at the Al₂O₃/AlGaN interface near the valence-band maximum of AlGaN.

Using the combination of the numerical fitting of C–V curves and the photo-assisted C–V method, we have attempted to estimate the state density distribution of the Al₂O₃/AlGaN interfaces. Figure 9(a) shows the comparison between the calculated and experimental C–V curves of the Al₂O₃/AlGaN/GaN structure with and without the ICP dry etching of the AlGaN surface. As shown in Fig. 3(a), the initial HEMT structure has a 34-nm-thick AlGaN layer. After the ICP etching, the AlGaN thickness was reduced to 27 nm,⁷⁰⁾ causing an increase in C_TOTAL and a pronounced V_th shift in the C–V curve. By assuming interface state density distributions indicated by the solid lines in Fig. 9(b), the calculation reasonably reproduced the experimental data. From the fitting results, we can determine the state density distribution near E_C at the Al₂O₃/AlGaN interface. Note that it is difficult to evaluate the state density distribution at the energies above E_C − 0.2 eV from C–V measurements at the frequency of 1 MHz, because electrons captured at such shallow interface states can respond to the measurement AC signal and accordingly give an inaccurate estimate of capacitance.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 10 shows the measured photo-assisted C–V curves of the sample with the ICP etching of AlGaN. We observed systematic C–V shifts toward the reverse bias direction with increasing photon energy, as explained in Sect. 4.2. The completely parallel C–V shift toward the negative direction indicates that the interface states near midgap or at deeper energies act as fixed charges in this bias range. Ozaki et al.⁷¹⁾ recently reported a similar parallel C–V shift in an ALD-Al₂O₃/n-GaN/AlGaN/GaN structure. Using Eq. (2), we estimated the interface state densities in the samples with and without the ICP etching of AlGaN. Figure 11 shows the state density distributions at the Al₂O₃/AlGaN interface determined by the photo-assisted C–V and the fitting methods. The sample without the ICP etching of AlGaN showed state densities of around 1 × 10¹² cm⁻² eV⁻¹ or less near the midgap. On the other hand, much higher state densities were obtained at the Al₂O₃/ICP-etched AlGaN interface.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

From transmission electron microscopy (TEM) observations, it was found that the ICP etching process introduced a monolayer-level crystalline roughness to the AlGaN surface.⁷⁰⁾ As a result, lattice disorder and/or a high density of dangling bonds near step edges remained at the AlGaN surface, leading to the generation of high-density electronic states with a continuous energy distribution. During ICP etching using the Cl₂/BCl gas mixture, active Cl-based radicals dominantly react with the AlGaN surfaces to form volatile products, such as GaCl₃ and AlCl₃. However, there are other possible chemical reactions. For instance, a highly volatile NCl₃ can be formed by a reaction between the Cl-based radical and N atoms, causing a preferential loss of N atoms at the AlGaN surface.³⁸⁾ In fact, Fang et al.⁷²⁾ reported that several kinds of deep levels originating from defect complexes related to nitrogen vacancy were increased at the GaN surface after the Cl₂-based ICP etching. It is thus likely that ICP etching causes the monolayer-level interface roughness, disorder of the chemical bonds and formation of various types of defect complexes at the AlGaN surface, leading to poor C–V characteristics due to high-density interface states at the Al₂O₃/AlGaN interface.

Although various kinds of insulator materials have been applied to improve the performance of AlGaN/GaN HEMTs, several problems remain unsolved. Among the most serious issues is the V_th fluctuation. Several papers reported that different gate bias conditions cause the V_th shifts in MIS HEMTs.^13–15) Fixed charges in insulator films,^28,50) border traps⁵¹⁾ and interface states at the insulators/AlGaN interfaces are involved in the V_th shift mechanism. As shown in Fig. 5(b), a higher interface state density leads to a shallower V_th. In addition, the charging state of the interface traps varies with the gate bias (E_F position), as shown in Fig. 7, and the deeper states with long time constants for electron emission contribute to a slow V_th fluctuation.¹⁴⁾ Lu et al.¹⁵⁾ and Johnson et al.¹⁴⁾ reported that the higher positive gate biasing of the MIS HEMTs induces larger V_th shift toward the forward bias direction. This is mainly related to the change in the density of ionized acceptor-type states, as shown in Fig. 7. Ťapajna and Kuzmík¹³⁾ also pointed out the effect of interface states on the V_th shift in MIS HEMTs. To realize normally-off operation, a combination of a recessed and an insulated gate structure is often implemented in AlGaN/GaN HEMTs. However, dry etching process step for the formation of the recessed structure introduces some kinds of surface defects, resulting in high densities of electronic states at insulator/(Al)GaN interfaces, as shown in Fig. 11 and as reported in Refs. 47 and 70.

Yang et al.⁴⁴⁾ demonstrated that surface nitridation of GaN/AlGaN/GaN structure using remote N₂ plasma is effective in improving a sub-threshold gate control and V_th stability of the Al₂O₃-gate HEMT. Hori et al.⁷³⁾ reported that N₂O radical treatment can decrease interface states at the Al₂O₃/AlGaN interface, resulting in the improved stability of V_th even after applying an off-stress gate bias. Van Hove et al.³⁾ utilized in-situ SiN_x as an interfacial control layer, and reported improved operation stability with a small V_th fluctuation in the SiN_x/Al₂O₃ gated AlGaN/GaN HEMT. On the basis of a systematic characterization of interface states, further investigation is absolutely necessary to control electronic states at insulator/(Al)GaN interfaces for improved stability of AlGaN/GaN MIS HEMTs.