Quantifying inhomogeneities in silicon-rich oxide thin films

Nonstoichiometric silicon oxide (SiO_x with x<2) thin films are technologically relevant materials because they serve as thin boundary layers between crystalline silicon and thermally grown oxides in many electronics and optoelectronics. The presence of these boundary oxides strongly affects the electronic properties of silicon.¹

After detecting visible luminescence in porous silicon, researchers began using an annealing process to produce nanocrystalline silicon from silicon-rich oxides.² Producing nanocrystalline silicon from SiO_x thin films offers an advantage because SiO_x thin films can be created using CMOS-compatible methods. The performance and properties of electronic devices and light-emitting materials are dictated by the quality of the silicon-rich oxide films, which is influenced by inhomogeneities in the films. While the real microscopic structure of SiO_x is still a subject of debate, studies have been conducted to establish the composition and structure of SiO_x phases.³

Particular attention has been paid to three models that describe the global structure of an SiO_x network: the mixture model,⁴ the random bonding model,⁵ and the intermediate model.⁶ However, the structure of SiO_x and the effective use of one model appears to depend on the specific fabrication technique of the films.⁷ For our work, we prepared SiO_x thin films using low-pressure chemical vapor deposition (LPCVD), quantified inhomogeneities in the deposited films using a variety of spectroscopic techniques, drew conclusions about the relationship between the silicon-rich structures and the oxygen content in the films, and evaluated which model best characterized our films.⁸

We prepared SiO_x thin films using LPCVD at 570°C with silane gas diluted with argon and oxygen as the reactant gases. We produced five samples (S1–S5) with varied oxygen contents and confirmed their amorphous structures using Raman spectroscopy. Next, we estimated the oxygen content in the films using time-of-flight elastic recoil detection analysis (TOF-ERDA) and energy-dispersive x-ray spectroscopy (EDX). We also used ellipsometry (for films less than 100nm thick) and m-line measurements (for films greater than 100nm thick) to determine the complex refractive index of the deposited films. From the complex refractive indices, we estimated the oxygen content in the films using Bruggeman's effective medium approximation (EMA). Next, we used IR spectroscopy, which estimates oxygen content from the position ω (in cm⁻¹) of the Si-O stretching mode⁹ by:

We observed a shift in the position of the Si-O stretching mode peak with increasing oxygen content in our samples: see Figure 1(a). We obtained consistent estimations of oxygen content using TOF-ERDA, EDX, and EMA, but the estimations we acquired using IR spectroscopy for S1–S4 corresponded to systematically higher oxygen contents: see Figure1(b). This effect can be explained by the inhomogeneous distribution of oxygen atoms in the films. In IR spectroscopy, the position of the Si-O stretching peak does not depend on the real oxygen content, x, but on the microscopic amount of oxygen atoms around a given silicon atom. Therefore, IR spectroscopy estimates can suffer from non-negligible errors if the oxygen is non-homogeneously distributed in the sample. In particular, if the oxygen atoms in an SiO_x matrix form SiO_y clusters (y>x), then the oxygen content shown by the IR spectra would be equal to y rather than x.

Figure 1. (a) IR spectra of silicon-rich oxide thin films (samples S1–S5) deposited on silicon substrates using low-pressure chemical vapor deposition. (b) Oxygen content, X, obtained from time-of-flight elastic recoil detection analysis (TOF-ERDA), energy dispersive x-ray spectroscopy (EDX), Bruggeman's effective medium approximation using ellipsometry and m-line measurements, and IR spectroscopy. (c) X-ray photoelectron spectroscopy (XPS) spectra of S1–S5. The left column shows the experimental XPS spectra (orange circles), the individual components of the Si 2p core level corresponding to different oxidation states of silicon (dashed lines), and their sum (blue line). The right column shows the relative intensities of the individual components to the Si 2p core level (blue circles), and the calculated components according to the random bonding model (red squares) and mixture model (green triangles). The x, y, and z parameters used in the calculations are given inside the boxes.

Due to the disparate estimation of oxygen content using IR spectroscopy, we concluded that the structure of our films did not correspond to the random bonding model, which describes the SiO_x structure as a single-phase system consisting of a number of homogenously distributed random bonded Si-(Si_nO_4−n) tetrahedra (0<n<4). In the mixture model, a phase separation of the SiO_x structure into two components—one richer in silicon and the other richer in oxygen—is assumed as:

where the oxygen content in the two separated phases, y and z, are used as fit parameters. The intermediate model assumes smooth variation of the chemical composition at the boundaries between silicon clusters and SiO₂ matrix, leading to a continuous distribution of the oxygen content.

Next, we further quantified the structural inhomogeneties we detected with IR spectroscopy using x-ray photoelectron spectroscopy (XPS), which is sensitive to the specific chemical bond formed by oxygen with silicon in its different degree of oxidation. We determined x from the experimental intensities of the different Si 2p core level components using the formula:⁴

where the overall intensity of the Si 2p core level was normalized to unity. We fit the spectrum of the Si⁰ peak to two Gaussians corresponding to the two spin-orbit Si⁰ levels. The other peaks were fit using Gaussian components and by assuming a constant energy shift of the Si¹⁺, Si²⁺, and Si³⁺ peaks from the midpoint between the two spin-orbit split components of Si⁰ to the Si⁴⁺ peak. We used the intensities of the peaks as fitting parameters and varied the position of the Si⁴⁺ component and the midpoint between the Si⁰ components to optimize the fit. The widths of the peaks and their energy separation were kept constant.

The XPS method yielded a slightly higher value for oxygen content (0.4–0.5 higher) than the TOF-ERDA, EDX, and ellipsometry methods. These results can be explained by surface oxide. The XPS probe depth was approximately 6nm, while the native oxide depth could have been 1–2nm. XPS analysis of the Si 2p core levels confirmed inhomogeneous distribution of oxygen atoms in the films and an almost-complete separation of the silicon-rich oxides into amorphous silicon (z=0) and silicon dioxide (y=2): see Figure1(c). The fit provided values of y>1.7 and z<0.22. These results indicated that the mixture model was the most appropriate model for our films.

The quality of silicon-rich oxide films plays a crucial role in electronics and optoelectronics. In this work, we measured inhomogeneities in nonstoichiometric silicon oxide thin films using a variety of analytical techniques. IR absorption spectroscopy estimated higher levels of oxygen than TOF-ERDA, EDX, and ellipsometry and m-line measurements using Bruggeman's EMA. This finding can be explained by inhomogeneities due to phase separation of the oxygen-rich SiO_y and silicon-rich SiO_z (y>x>z) structures. This phase separation was further quantified using the XPS technique. The XPS spectra of the Si 2p core levels showed an almost-complete phase separation of the SiO_x films into amorphous silicon and silica. The results indicated that the structure of the samples was best described by the mixture model. Future work will include using the model to develop detailed structural analysis procedures.

This research was performed in the framework of the NSBMO research project (2010–2013) of the Provincia Autonoma di Trento and of project 098-0982904-2898 of the Ministry of Science, Education and Sports of the Republic of Croatia.