High-k dielectrics on (100) and (110) n-InAs: Physical and electrical characterizations

III–V compounds, such as GaAs, InGaAs or InAs, have been intensively studied to replace Si as channel material because their high electron mobilities may enable low power and high performance applications for future CMOS. InAs is a promising candidate for MOS high-electron-mobility transistor devices^1,2 because it has the highest electron mobility among in the arsenide-based III–V compounds. In addition, InAs is a binary compound that could provide a much simpler interface structure and property as compared to InGaAs. Although there are many studies on high-k/GaAs and high-k/InGaAs structures,^3–7 only a few high-k/InAs structures have been investigated.^8,9 Moreover, most of studies for high-k/InAs have focused on (100) surface orientation,^10,11 with few (110) InAs investigations published.¹² 

Besides the introduction of high electron mobility channels, the device architectures are also very important for performance enhancement. Figures 1(a) and 1(b) show the conventional planar FET and nonplanar FinFET architecture. The employment of a FinFET architecture can extend the gate scaling beyond the planar transistor limitations to obtain better off-state performance, including subthreshold swing and DIBL (drain induced barrier lowering). FinFET architecture can also offer better device matching and variability due to the low doping concentration in the channel.^13–15 However, unlike planar devices, FinFET channels consist of two surface planes, including the (100) and (110) planes, and the interface properties of both surfaces have to be considered. In this way, investigating both InAs (100) and (110) can help to understand the interface properties of InAs channel FinFET device architectures. In this article, we study the interfacial properties of Al₂O₃ and HfO₂ on (100) and (110) n-InAs epi surfaces by physical and electrical characterization, respectively.

Firstly, 400 nm InAs epilayers were grown by MOVPE (metal-organic vapor phase epitaxy) on (100) and (110) oriented Se-doped n-type InAs substrates. AFM (atomic force microscopy) scans on both the (100) and (110) n-InAs epi-layers show flat surfaces with root-mean-square roughness less than 2 Å in a 9 μm² (3μm × 3μm) area. Both as-grown (100) and (110) n-InAs epitaxial wafers were analyzed by XPS (x-ray photoelectron spectroscopy) to study the surface native oxide composition formed when the layers were removed from the deposition chamber. Prior to the high-k oxide deposition, the samples were first cleaned in acetone and isopropanol, and a standard ex-situ surface clean solution to remove native oxides followed by rinsing in de-ionized (DI) water. For each experimental condition samples of both (100) and (110) orientations were loaded together into a Picosun 100 ALD (atomic layer deposition) system. The deposition started with TMA (tri-methyl aluminum) pre-treatment before depositing films of Al₂O₃ and HfO₂. 8 nm and 2 nm of Al₂O₃ and HfO₂ sample pairs were deposited for metal-oxide-semiconductor capacitor (MOSCAP) fabrication and XPS analysis, respectively. The TMA pre-treatment prior to the high-k deposition on III–V channel materials has been widely employed and proven to eliminate native oxide on III–V surface.^16,17 MOSCAPs were fabricated by evaporation of Pd gate metal and Ti/Au backside ohmic contact, followed by annealing in forming gas (FG) ambient. TEM (transmission electron microscopy), EDX (energy dispersive X-ray spectroscopy), and XPS were used to investigate the physical and chemical structure of the high-k/III–V interface. C-V measurement was performed using an Agilent HP4294A impedance analyzer with a voltage sweep between −2 V to +2 V and frequencies varied from 1 kHz to 1 MHz at room temperature to characterize the behavior of the MOSCAP and extract interface trap density (D_it).

The As3d and In3d XPS spectra were collected to reveal the composition of native oxide for both surface planes, as shown in Figures 2(a) and 2(b). Figure 2(a) shows that there is more As-oxide on (100) than on (110), which is consistent with the As-rich condition at the (100) surface.¹⁸ Figure 2(b) shows slightly stronger In-oxide intensity on (110) surface indicating that the (110) surface may be In-rich.¹⁹ The intrinsic difference of the native oxides on the two different surface orientations was eliminated or minimized by TMA pre-treatment before the high-k (Al₂O₃ and HfO₂) deposition. Figures 3(a)–3(d) show the In3d and As3d XPS spectra of Al₂O₃ on (100) and (110) n-InAs epi, respectively. The Al₂O₃ samples for XPS were pretreated with TMA. Figures 3(a) and 3(b) show that the XPS spectra of In-oxide on (100) and (110) after Al₂O₃ deposition are comparable and the peak location of the indium-oxide bond energy is similar on both surface orientations. Figures 3(c) and 3(d) show that the As-oxide bonds were suppressed on both (100) and (110) samples after TMA pre-treatment. As-As dimers were detected with comparable intensities and binding energy on (100) and (110) surface after Al₂O₃ deposition. Samples with HfO₂ high-k were processed for XPS using the same TMA pre-treatment before HfO₂ deposition. Figures 4(a) and 4(b) show that HfO₂ samples on (100) and (110) n-InAs have similar In-oxide bond intensities on the surface and the same bond energy peak location in the energy spectrum. Similarly, As-oxide bonds were suppressed on both surface orientations. A similar intensity of As-As dimers and the binding energies were found, as shown in Figures 4(c) and 4(d). The intensity and binding energy of Hf5p in the HfO₂ film on (100) and (110) were also found to be similar, as shown in Figures 4(c) and 4(d). This implies comparable HfO₂ film quality on both surface orientations.

TEM micrographs for the (100) and (110) oriented structures consisting of Pd/Al₂O₃/n-InAs epi/n-InAs substrate/Ti/Au MOSCAP are shown in Figures 5(a) and 5(b). The physical thickness for Al₂O₃ is 8.9 nm on both crystal planes was measured by TEM. Figures 6(a)–6(d) show the LAADF (low-angle annular dark field) TEM micrographs and EDX analysis for the HfO₂ MOSCAP on both (100) and (110) InAs epi. The physical thickness extracted from TEM for HfO₂ (Figures 6(a) and 6(b)) is 7.1 nm with 7 Å from the TMA (Al₂O₃) pre-treatment layer at the interface between HfO₂ and InAs for both (100) and (110) planes. EDX analysis (Figures 6(c) and 6(d)) shows strong Al and O signals due to the pre-treatment at the interface of HfO₂ and InAs. A sharp transition from InAs to the HfO₂ layer is indicated by the fast decay of the In and As signals. The TEM analysis shows that the Al₂O₃ and HfO₂ MOSCAPs on (100) and (110) have good thickness uniformity over the examined area. The Al₂O₃ and HfO₂ dielectric layers are amorphous without any crystallite formation. In addition, there are no visible defects and no surface roughening at the Al₂O₃ interface with InAs and the Al2O3/HfO2 interface, indicating good interface quality on both surface planes. No interfacial native oxide is observed in the Al₂O₃ layer, while a thin layer is observed in HfO₂ samples, which can be attributed to the TMA pre-treatment step.

Figures 7(a) and 7(b) show the multi-frequency C-V responses of Al₂O₃ and HfO₂ MOSCAPs on (100) and (110) oriented n-InAs epitaxial layers, respectively. In the accumulation regime, the multi-frequency responses of Al₂O₃ on (100) and (110) n-InAs epi show similar behavior in frequency dispersion and C_acc (accumulation capacitance), with only a slight C_acc difference due to variation in physical oxide thickness. The frequency dispersion for Al₂O₃ on (100) and (110) n-InAs are 1 %/dec and 1.1 %/dec, respectively. The HfO₂ samples also show similar multi-frequency responses for both orientations. The values of frequency dispersion for HfO₂ on (100) and (110) n-InAs epi are 2.9%/dec and 2.6%/dec, respectively. The larger frequency dispersion in HfO₂ samples is attributed to the presence of more border traps close the InAs/HfO₂ interface. For Al₂O₃ MOSCAPs on (100) and (110) InAs in the depletion regime, the minimum capacitance C_dep, and the corresponding V_g are similar, ranging from -0.45 V (1 kHz) to −0.75 V (1 MHz). For HfO₂ MOSCAPs on (100) and (110) InAs C_dep and the corresponding V_g are also similar, ranging from 0 V (1 kHz) to −0.5 V (1 MHz). As shown in the Figures 7(a) and 7(b), the capacitance at negative bias and C-V stretch-out of Al₂O₃ and HfO₂ MOSCAPs on (100) and (110) InAs shows similar behavior in depletion and inversion. However, the visible difference in frequency dispersion in negative bias between the Al₂O₃ and HfO₂ MOSCAPs implies that the high-k traps on the (100) and (110) surfaces have a different frequency response. Larger frequency dispersion was observed in the (110) InAs samples with both Al₂O₃ and HfO₂ deposition.

At sufficiently low frequency all charges (inversion and accumulation carriers and traps) are assumed to respond to the ac signal. To extract D_it from the measured low-frequency C-V curve we can solve Poisson's equation:

where V(y) is the position-dependent electrostatic potential, q the electronic charge, N_d and N_a the ionized donor and acceptor concentrations, n(y) and p(y) the electron and hole concentrations and ε_r the relative permittivity of the semiconductor. Following,²⁰ we account for non-parabolicity of the conduction band Γ-valley by writing the conduction band density of states as:

where |$m_\Gamma $|$mΓ$ is the gamma-valley electron effective mass, E_c is the conduction band minimum and α = (1/E_g) · (1 − m_Γ/m₀) is the non-parabolicity factor where E_g is the semiconductor band gap energy; we take α∼2.69 eV⁻¹ for InAs. Fermi-Dirac statistics are applied for both CB (conduction band) and VB (valence band).

Subsequently, by setting V(y) = 0 at y = infinity and applying a Cauchy boundary condition at the semiconductor/dielectric interface, we can calculate the intrinsic (without interface traps) MOSCAP V_gi(E_f) and Q_sc(E_f), where V_gi is the intrinsic gate voltage, Q_sc the integrated semiconductor charge and E_f the Fermi energy at the semiconductor/dielectric interface. To calculate V_gi the work function difference between gate and semiconductor ϕ_ms is arbitrarily chosen to be equal to zero.

Now we introduce a fitting function Q_it(E_f) which represents the total trapped charge at the semiconductor/dielectric interface at a given interface Fermi energy, such that C_gate(V_g) = −d(Q_sc + Q_it)/dV_g matches the experimental data. Here,

is the applied gate voltage with C_ox = ɛ₀κ/T_ox the gate dielectric capacitance. The effective interface trap density is then given by D_it(E) = −1/q dQ_it(E)/dE. In this procedure, uncertainty in the metal gate effective work function ϕ_m,eff appears as a constant offset in the fitted Q_it-function, and is eliminated when the derivative is taken to calculate the effective D_it.

In the above analysis, input parameters N_a and N_d are well-known from layer growth calibration. Despite the good layer thickness control of the ALD process, this is not the case for C_ox since the dielectric constant κ is a priori unknown for low-temperature ALD. Therefore, we chose to treat C_ox as fitting parameter. In our analysis, it was found that the condition of a monotonously decreasing Q_it(E_f) function (i.e. positive D_it) in combination with a good fit with experimental data could only be met for a small Cox window - in practice we can estimate C_ox with an uncertainly of about 10%.

By using a low-frequency fitting routine we discussed above on the 1 kHz samples, the D_it profiles for (100) and (110) samples has been extracted and shown in Figure 8. Nearly identical D_it profiles (Figure 8) were obtained for Al₂O₃ on (100) and (110) n-InAs epi, with D_it minimum of 6.1 × 10¹² and 6.5 × 10¹² cm⁻² eV⁻¹, respectively. HfO₂ on (100) and (110) n-InAs epi-layers have slightly higher D_it values with D_it minimum of 6.6 × 10¹² and 7.3 × 10¹² cm⁻² eV⁻¹, respectively. The TMA pre-treatment before high-k deposition can efficiently passivate the surface on (100) and (110) n-InAs and provide similar D_it profiles and D_{it_min} on both surface orientations, which can be detected and quantified by LAADF TEM and EDX analysis as shown in Figures 6(a)–6(d). The difference of D_{it_min} location between Al₂O₃ and HfO₂ were attributed to more border traps rather than the difference between D_it in the HfO₂ film compared to the Al₂O₃ film since the interface of both types of samples are almost identical. The process conditions and parameters of all samples (Al₂O₃ or HfO₂ on (100) and (110) InAs) are listed in Table I, including ex-situ pre-treatment, post-treatment (PMA: Post Metalization Anneal), high-k thickness, and D_{it_min}.

We have reported the physical and electrical characterization of Al₂O₃ and HfO₂ on (100) and (110) n-InAs epitaxial layers. Both TEM and XPS results show nearly identical interface properties between Al2O₃ and HfO₂ for both (100) and (110) InAs surface orienations. Similar C-V characteristics in accumulation and depletion regimes as well as D_it profiles were also observed for both orientations. The similar interface trap density (D_it) on (100) and (110) surface orientation was observed, which can be extended to InAs FinFET device architecture with both (100) and (110) surfaces.

The authors would like to thank Y. C. Sun of TSMC and L. Samuelson of Lund University for their support.