AVS 71 Session EL-ThP: Spectroscopic Ellipsometry Poster Session

A custom multi-bounce, prism-based liquid cell has been developed for use with infrared ellipsometry to enable quantitative analysis of liquids and the species dissolved within them. Traditional optical methods for liquid-phase analysis often rely on attenuated total reflection (ATR), where changes in signal intensity are caused by absorption via interaction between the evanescent wave and the liquid sample. In contrast, this approach leverages ellipsometric polarization measurements, capturing the reflected light’s amplitude ratio (Ψ) and phase difference (∆). These measurements probe molecular vibrations in the fingerprint region, enabling analysis of the chemical composition of liquids.

Conventional optical measurement schemes are limited in the IR spectrum for liquid applications due to strong absorption, particularly in water-based systems. The ATR measurement technique addresses this limitation by coupling light into a prism and probing the sample with the evanescent wave produced by total internal reflections. Our new method combinesthe traditional ATR concept with ellipsometric detection. Multiple internal reflections within a high-index prism produce evanescent waves that repeatedly interact with the liquid sample. The repeated interactions with the liquid enhance sensitivity to small constituent fractions of species within the liquid. By capturing both the intensity and the polarization changes, this configuration extends the sensitivity of ellipsometric measurements to liquid environments.

The prism is housed in a custom-built sealed liquid cell and cut to provide a 45° angle of incidence. Submerged length wise in the liquid, the prism supports multiple internal reflections, with each reflection generating an evanescent field that probes the sample. Our results demonstrate the potential of ellipsometric ATR for concentration-based analysis of complex liquids, with future applications in quality control for the food, beverage, and water industries.

The objective of any noise-reduction filter is to preserve information and eliminate noise, both to the maximum extent possible.Up to now filters have been assessed by trial-and-error.Here, we report a cost function that quantifies the action of a filter on information and noise in a spectrum, that is, on distortion and leakage, respectively.For linear filters, which act by direct-space (DS) convolution or (equivalently) by selective attenuation of reciprocal-space (RS) Fourier coefficients, the expression is exact.Consequently, optimal parameters for any linear filter operating on any spectrum can now be determined unambiguously.We find that the best practical linear filter is the Gauss-Hermite filter introduced by Hoffman and co-workers in 2002 [1].

Nonlinear filters operate differently, retaining low-order Fourier coefficients exactly up to the white-noise cutoff and replacing those in the white-noise region with model-independent most-probable analytic extrapolations.With distortion nominally eliminated and noise input limited to retained coefficients, these should outperform any linear filter.However, the only known example, the corrected-maximum-entropy(CME) approach [2], can be used only with positive-definite spectra consisting of superpositions of Lorentzian absorption lines.Here, we report a forward-prediction approach with performance equivalent to the CME, but one that is based on physical principles and is completely general.Examples dealing with spectroscopic-ellipsometric and other types of data, for instance X-ray photoelectron and Auger electron spectra, will be presented.

[1] D. K. Hoffman, D. J. Kouri, and E. Pollak, Computer Phys. Commun. 147 (2002) 759-769.

[2] L. V. Le, Y. D. Kim, and D. E. Aspnes, Thin Solid Films 761 (2022) 139515.

Decades of intensive research on transparent conducting oxides (TCOs) have enabled advancements across a wide range of technologies, including flat-panel displays, photovoltaic cells, LEDs, transparent electronics, smart windows, and wearable devices. TCOs are unique in that they combine high electrical conductivity with optical transparency, a property made possible by their wide bandgaps. Among them, indium oxide (In₂O₃) and tin-doped indium oxide (ITO) are widely used due to their excellent optical transmission and tunable electrical conductivity, which can be adjusted by varying the tin content and deposition conditions. More recently, manganese-doped ITO (Mn-ITO) has attracted attention for applications in spintronic devices. In this study, Mn-ITO thin films with Mn concentrations ranging from 0 to 12.8% were deposited on glass substrates using DC magnetron sputtering. The structural, electrical, and optical properties of the films were systematically investigated. Optical characterization using spectroscopic ellipsometry revealed a monotonic decrease in the energy bandgap with increasing Mn content. X-ray diffraction analysis of the undoped ITO films, using the (440) peak, yielded a lattice constant of approximately 10.15 Å, which is smaller than the standard value of ~11.2 Å. A possible explanation for this deviation is discussed.

Ellipsometry can be used to determine the optical properties and thickness of a thin film depositing on a known substrate based on light reflecting from the surface.This approach has the advantage of being able to be used in situ during the growth of the film with commercially available equipment to pass the light in and out of the deposition chamber. Nevertheless, a serious challenge in practice arises if the material structure underlying the growing film is composed of multiple layers. In this case, very accurate knowledge of all of the underlying structure is required in order to obtain accurate results by conventional means. Another challenge is that the computation normally takes significant time using conventional iterative solution methods such as Levenberg Marquardt. The work here demonstrates the use of an Artificial Intelligence (AI) method suitable for real-time growth in which the underlying structure is unknown and complicated. This method is based upon previous development using five separate reflections to solve for the underlying reflection and the film parameters simultaneously. The method is sufficiently fast that multiple groups of five measurements can be analyzed during the growth to confirm results and to examine the vertical homogeneity of the film being deposited. Examples will be given for absorbing films (up to 45 nm) using a multilayer perceptron configuration consisting of 10 input neurons and 10 output neurons with two hidden layers of 80 neurons each. Solutions are performed at each wavelength independently and do not rely on fitting functions. The design, training and use of a number of neural networks will be presented.