結合類神經網絡與基因演算法於 奈米光學元件最佳化設計之探討

A Study on Optimization of Nano Photonic Devices by Combining Neural Network with Genetic Algorithm

黃崇岳(Chung-Yue Huang)

指導教授 ： 王倫

國立臺灣大學/電機資訊學院/光電工程學研究所/碩士(2007年)

https://doi.org/10.6342/NTU.2007.01010

摘要

由於傳統的光學元件的尺度無法滿足當前光電系統發展的需求，而隨著奈米製程技術的提升，促使光電元件有機會朝向高度積體化與微型化的方向發展。因此，奈微米光學元件之設計成為當前光電產業中一個極為重要的研究領域。惟傳統光學元件的設計，多數依靠設計者的經驗或試誤法的程序來找出最佳設計。在結構簡單、快速演算的模型中，設計者可以使用解析解或是全域搜索來找出最佳化模型。但是在需要大量運算或是複雜性高的光學模擬中，傳統全域搜索仍然無法有效率的搜尋最佳化參數。
本論文的目的是利用類神經網絡做出最佳化參數的預測並進而改進演化方向。藉由整合波束傳播法與嚴格耦合波兩種光學理論,配合商用電磁分析軟體Rsoft來最佳化光學元件的結構參數。並模擬幾個典型奈米光學元件之最佳化設計，比較其光學元件在效能上的提昇並討論演算法的搜索效率。

關鍵字

基因演算法 ； 類神經網路 ； 波束傳播法 ； 嚴格耦合波

並列摘要

Owing to the large size of the conventional optical element that can’t be well compatible with the high impacted and integrated micro-optical system, however, it is just the prompted nanotechnology that realizes the possibilities to manufacture such the nano scale optical elements. Therefore, nano photonic devices are now becoming more and more essential in developing electro-optical engineering. Nevertheless, traditionally optical engineers usually find the optimal parameters of the devices by their experiences or the heuristic tests. When the devices are uncomplicated, the designer could still use analytical approach or global searching methods to find the optimal parameters. When the devices become more complicated, the simulations usually need huge computations and then the analytical or conventional heuristic methods will not be practicable.
In this thesis, we utilized neural network to do the optimal design by using the database. Furthermore, the neural network affected the evolutional direction. We discussed the efficiency of combining genetic algorithm and neural network to reduce the time needed for heuristic guess. Beam propagation methods (BPM) and rigorous coupled wave analysis (RCWA) were applied to search the optimal structure parameters together with our proposed optimization method. We optimized several optical component designs such as directional coupler and diffractive grating. Neural network was used to do more flexible genetic selection. Finally, we discussed the optical characteristics of our optimized design.

並列關鍵字

Genetic algorithm ； Neural network ； Beam propagation method ； RCWA

參考文獻

[1] M. Hu, R. Scarmozzino, M. Levy, and R.M. Osgood, Jr., “A low-loss and compact waveguide y-branch using refractive index tapering,” Photon. Tech. Lett., 9, 203, 1997.

[2] Shyh-Lin Tsao, Chun-Yi Lu,” BPM Simulation and Comparison of 1 × 2 Directional Waveguide Coupling and Y-Junction Coupling Silicon-on-Insulator Optical Couplers,” Fiber & Integrated Optics, Volume 21, Number 6/November 01, 417-433 , 2002

[3] SC Cho, HH Hong, JYC Choong-reolYang, M Kang, HY,”A Novel Vertical Directional Coupler Switch with Switching Operation-Induced, Extinction Ratio-Adjusted, and Extinction Ratio-Enhanced Sections,” ETRI Journal, 2002

[5] G.R. Hadley, ” Transparent boundary condition for the beam propagation method,” J. Quantum Electron 28, 363, 1992.

[6] J.Y. Ye, X. -C. Yuan, G. Y. Zhou,” A Genetic Algorithm for Optimization Design of Diffractive Optical Elements in Laser Beam Shaping,” SPIE, 2001.