基于宽波段透射光谱的神经网络光学薄膜逆向设计研究

doi:10.3964/j.issn.1000-0593(2025)Z1-0373-08

Abstract
References
Related Citation (15)

Download: PDF (12337 KB)
Export: BibTeX | EndNote (RIS)

Abstract Optical thin films are widely used in fields such as optical systems optoelectronic devices and other fields, with thicknesses typically ranging from nanometers to micrometers. The design and research involve multiple disciplines. The inverse design of optical thin films is based on the target optical performance to infer structural parameters and traditional design methods face problems such as falling into local optima or high computational costs. Machine learning algorithms such as neural networks rely on multi-layer nonlinear transformation structures to quickly approximate the solution space, demonstrating breakthrough potential in inverse design. The paper combines the Transfer Matrix Method (TMM) to calculate transmission spectrum data and establishes a unified neural network framework applicable to the transmission spectra of single-layer films (SiO₂) and double-layer films (SiO₂/TiO₂), achieving end-to-end inverse design of optical thin films. The effectiveness of the model is revealed through rigorous error analysis. Ablation experiments are introduced to evaluate and optimize the neural network structure parameters using stepwise feature compression and dimensionality reduction. After the model was established, the coefficient of determination (R²) and mean absolute error (MAE) were used as core indicators to evaluate the prediction accuracy of the model. The evaluation results showed that the model achieved a high level of R², while the MAE was controlled within a relatively low range. The model exhibits strong explanatory power for the target variable, with small prediction bias, resulting in an overall satisfactory prediction effect. In order to better align with real-world application scenarios, this research introduces Gaussian random noise to simulate measurement errors in real environments, such as those encountered by detectors. The model maintains high stability even under noisy data conditions, with the decrease in accuracy remaining within an acceptable range, fully demonstrating its strong robustness. To verify the model's ability to handle transmembrane architectures, the neural network model was extended to analyze the transmission spectrum of a double-layer films (SiO₂/TiO₂). Research indicates that the model can converge normally and achieve high-precision prediction in the context of a double-layer films. The overall prediction error of the model approximately follows a gaussian normal distribution. The model possesses input dimension adaptability, enabling dynamic adjustment of the input layer dimension and its output dimension can be flexibly extended to accommodate multiple layers of film systems.

Key words：Optical thin films; Inverse design; Neural network; Transfer matrix method; Ablation experiments

Received: 2025-07-31 Accepted: 2026-01-10

ZTFLH:

O439

Corresponding Authors: DUAN Ran E-mail: rduan@niaot.ac.cn

Cite this article:

LI Ting-wei,DUAN Ran,TIAN Jie, et al. Research on Inverse Design of Neural Network Optical Thin Films Based on Wide Band Transmission Spectrum[J]. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(12zk): 373-380.

URL:

https://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2025)Z1-0373-08 OR https://www.gpxygpfx.com/EN/Y2025/V45/I12zk/373

[1] Macleod H A. Thin-Film Optical Filters. CRC Press, 2018.
[2] FAN Zheng-xiu(范正修). Optical Thin Films and Their Applications(光学薄膜及其应用). Shanghai: Shanghai Jiaotong University Press(上海: 上海交通大学出版社), 2014.
[3] TANG Jin-fa, GU Pei-fu, LIU Xu, et al(唐晋发, 顾培夫, 刘旭, 等). Modern Optical Thin Film Technology(现代光学薄膜技术). Hangzhou: Zhejiang University Press(杭州: 浙江大学出版社), 2006.
[4] Thelen A. Proceedings of SPIE-The International Society for Optical Engineering, 1993, 1782: 49.
[5] Tikhonravov A V, Trubetskov M K, Debell G W. Applied Optics, 1996, 35(28): 5493.
[6] Press W H, Teukolsky S A, Vetterling W T, et al. Numerical Recipes: the Art of Scientific Computing. 3rd ed. Cambridge: Cambridge University Press, 2007.
[7] Gen M, Cheng R. Genetic Algorithms and Engineering Optimization(遗传算法与工程优化). Translated by Xuan Guang-nan(玄光男, 译). Beijing: Tsinghua University Press(北京: 清华大学出版社), 2004.
[8] Martin S, Rivory J, Schenk M. Applied Optics, 1995, 34(13): 2247.
[9] Kennedy J, Eberhart R. Particle Swarm Optimization[C]//Proceedings of ICNN'95-International Conference on Neural Networks. Perth, Australia: IEEE, 1995: 1942.
[10] Heavens O S. Optical Properties of Thin Solid Films. London: Butterworths Scientific Publications, 1955.
[11] Joannopoulos J D, Meade R D, Winn J N. Photonic Crystals: Molding the Flow of Light (2nd ed.). Princeton University Press，2008.
[12] Yeh P. Optical Waves in Layered Media. John Wiley & Sons Inc.，2005.
[13] Yariv A, Yeh P. Photonics: Optical Electronics in Modern Communications (6th ed.). New York: Oxford University Press，2007.
[14] Fukushima K. Biological Cybernetics, 1980, 36(4): 267.
[15] Verhulst P F. Correspondance Mathématique et Physique, 1838，10：113.
[16] Zeiler M D, Fergus R. Visualizing and Understanding Convolutional Networks. In Computer Vision-ECCV，Springer, 2014. 2014, 818.
[17] QIU Xi-peng(邱锡鹏). Neural Networks and Deep Learning(神经网络与深度学习). Beijing: China Machine Press(北京: 机械工业出版社), 2020.
[18] Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. Springer New York, 2001.
[19] JIA Huai-qin(贾怀勤). Applied Statistics(应用统计). Beijng: University of International Business and Economics Press(北京: 对外经济贸易大学出版社)，2010.
[20] Papoulis A, Pillai S U. Probability, Random Variables and Stochastic Processes. McGraw-Hill, 2002.
[21] Stark H, Woods J W. Probability and Random Processes with Applications to Signal Processing. Prentice Hall, 2022.