arXiv Science⌕ Search

arXiv · 2611.20694

Breaking the Fixed-Substrate Limitation of Deep Learning Ellipsometry with Pseudo-Substrate Spectra

Abstract

A machine learning method, deep learning ellipsometry (DLE), enables direct and rapid estimation of the optical constants and structures of thin films from spectroscopic ellipsometry (SE) spectra. However, conventional neural-network-based DLE models are restricted to the specific substrate used during training, and applying them to a different substrate requires ~100 h of retraining. Here, we propose a framework that overcomes this fixed-substrate limitation by introducing a two-input feature-wise linear modulation DLE (FiLM-DLE). In this model, the SE spectra of a thin-film structure are input together with those of the substrate measured before film deposition, which we term the pseudo-substrate spectra. Any substrate, including a multilayer stack, is thus represented solely by its measured SE response, analogous to the pseudo-dielectric function, so that no prior knowledge of the substrate structure is required. Trained on a virtual dataset of 165 million samples representing metal, glass, and SiO2/Si substrates with different SiO2 thicknesses, a single FiLM-DLE model, trained once, extracts thin-film optical constants and structures with high accuracy and shows strong applicability to substrates not included in the training datasets. Experimental validation on 47 thin-film samples across diverse substrates yielded sample-averaged mean absolute errors, relative to conventional manual analysis, of 0.047 for refractive index, 0.030 for extinction coefficient, and 1.52 nm for thickness, with each analysis completed in under 2 s. FiLM-DLE thus eliminates the need for a separate DLE model for each substrate and enables rapid, high-accuracy optical characterization of thin films on diverse substrates.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yuki Ikarashi, Naoki Kurokawa, James N. Hilfiker, Takuji Maekawa, Hitoshi Tampo, Hiroyuki Fujiwara. 2026-11-16. Breaking the Fixed-Substrate Limitation of Deep Learning Ellipsometry with Pseudo-Substrate Spectra. https://arxiv.science/abs/2611.20694

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Controlling Mode-Locking in Liquid Light

Mode-locking is the process by which the phases of many cavity modes organize collectively into a stationary state. In lasers, mode-locking originates from the interplay of gain, loss, dispersion and nonlinearities, most commonly resulting in the emission of short pulses. However, pulse formation is prevented when the gain recovery is fast enough to suppress amplitude fluctuations. Mode-locking then produces an extended state of nearly constant intensity, whose dynamics is governed by the phase. Yet, a general framework for designing the phase of such extended photonic states remains lacking. Here we introduce a hydrodynamic approach to designing extended photonic states, in which the phase dynamics are understood and controlled as a flow of liquid-like light. The intensity, clamped by fast gain, plays the role of a constant density, the instantaneous frequency that of a flow velocity, and the programmable radio-frequency modulation applied to the laser acts as a potential landscape that controls the flow. We derive analytical solutions connecting the applied modulation to the instantaneous frequency and hence to the spectrum emitted by the laser. We further provide experimental evidence for the liquid light shaping framework by measuring the frequency comb spectra emitted by a fast-gain semiconductor laser around 1.6 micron under the control of a programmed modulation, and comparing them to the ones predicted by the analytical solution. The hydrodynamic picture of mode-locking provides a predictive design principle for frequency combs, pointing toward a programmable synthesizer of light, and establishes fast-gain lasers as an accessible, electrically driven platform for nonequilibrium fluids of light.

physics.optics↗

Backward-scattering photoelectron holography of a polar molecule driven by a sub-cycle laser pulse

We investigate photoelectron holographic interference in the polar CO molecule driven by a sub-cycle laser pulse using time-dependent Schrödinger equation simulations and a classical-trajectory-based analysis. The sub-cycle temporal confinement of the driving field enables the forward-scattering holographic (FSH) and backward-scattering holographic (BSH) interference structures in the photoelectron momentum distributions (PMDs) to be clearly distinguished. The FSH fringes exhibit a strong dependence on molecular orientation, which is associated with the orientation-dependent transverse displacement of the ionizing electron wave packet (EWP). Furthermore, the heteronuclear nature of CO results in different FSH fringe shifts for ionization occurring predominantly from the C and O sides of the molecule. In contrast, the curved interference structures in the negative parallel-momentum region are identified as BSH interference arising from the coherent superposition of direct and backward-scattered EWPs. The BSH fringe positions show weak dependence on molecular orientation and remain clearly visible after orientation averaging. We further show that BSH visibility depends on the carrier-envelope phase (CEP) and that the fringes remain discernible after CEP averaging. Moreover, the BSH interference pattern exhibits a systematic dependence on the internuclear separation, demonstrating its sensitivity to molecular geometry. These results highlight the potential of sub-cycle pulse enabled backward-scattering photoelectron holography for probing molecular structure and demonstrate its robustness against molecularorientation and CEP averaging

physics.optics↗

Active Contrast-Dissipation All-Optical Encryption

Optical encryption is meant to place secrecy in physical structure, yet adversaries can readily construct imperfect physical keys to probe the encryption system via trial-and-error iterations. Such systems inevitably leak target plaintext information and preserve invertible residual structural features, rendering the entire encryption scheme vulnerable to adversarial inversion attacks. Here, we report an active contrast-dissipation all-optical encryption architecture, which partitions wavefront modulation into M wavelength-private encryption layers and N shared decoding layers. Security rests on multi-dimensional physical keys that must match simultaneously, including illumination wavelength, precise spatial alignment and layer stacking sequence. With every key matched and authorized readout is preserved: the composite plaintext is reconstructed at a full-detector Pearson correlation coefficient (PCC) of 0.997. We further impose active contrast dissipation as a core security constraint. Under physical perturbation it suppresses the foreground-background intensity contrast, so that the slightest key mismatch collapses the target features and dissipates the plaintext into randomized, diffuse speckle within chaotic background noise, leaving no residual semantic leakage to invert. Under nanoscale wavelength detuning, a 1 degree in-plane rotation, a single-pixel lateral displacement or layer-stacking tampering, the plaintext vanishes and the PCC approaches zero, whereas conventional diffractive encryptors retain identifiable contours. This work provide a feasible and reliable solution for high-security information encryption, with great application potential in communication security, information anti-counterfeiting, and confidential data storage.

physics.optics↗