應用控制理論並同時考慮電路設計規格以增進光學鄰近修正之收斂

光學鄰近修正是一項用來增進解析度的技術，對於0.13 微米或是更小的技術節點而言，它已經被廣泛地應用來增進圖像的品質與解析度。光學鄰近修正技術主要可分為以規則為基礎和以模型為基礎兩類，以模型為基礎之光學鄰近修正技術的基本概念，是藉由不斷地修補光罩來使得晶圓上的圖形能夠接近事先所設計好的圖形，對於每一次的修正，光罩都須經過光學模型的模擬，以得到空間影像，進而決定修正量；因為每次的修正都需經過光學模型的模擬，而這樣的模擬是相當花時間的，因此總修正時間便會隨著修正次數的增加而大量增加，因此，我們的研究主要是以降低修正次數來減少總修正時間。我們已經提出一個控制方塊迴路圖，來達到光學鄰近修正之目的，此控制方塊圖主要可以分為四個部份，包括光學微影模型、古典控制器及經由訓練好的類神經網路系統所產生的初始光罩圖形，最後則是考慮電路設計而得到的晶圓圖形。此控制方塊圖主要設計考量是降低總修正次數。根據純量場模型和傅立葉光學理論，我們已經成功的發展出光學微影模型並和柏克萊大學所發展的線上光學模擬軟體SPLAT做了比較。其中，在控制器的部份，我們根據經驗並對於不同的光罩圖形來調整PID控制器參數，並整理出經驗法則。接下來，導入類神經網路理論再藉由給定的輸入與輸出資料訓練出一個模型，經由此模型可得到一個初始的光罩圖形，此光罩圖形是較為接近最後修正所得到的圖形，藉此初始圖形亦可降低總修正次數。而最後是考慮電路設計的部份。在給定可變化的容忍度下，並以不影響電路設計要求為前提，適度的改變設計要求，亦可使得修正次數減少。

關鍵字

光學鄰近修正； PID控制器；類神經網路；製程模型；電路設計規格

並列摘要

Optical proximity correction (OPC) is one of resolution enhancement techniques. Beyond 0.13 μm technology node, it has been broadly used to improve the image quality and resolution in lithography. OPC techniques are broadly categorized into rule-based OPC and model-based OPC. The basic concept of model-based OPC is by iteratively compensating the mask to make the wafer pattern is as closer to the desired pattern as possible. For each correction, the mask pattern will proceed to the optical model to obtain the image data to further determine the correction amount. Because the mask pattern must proceed to the optical model for each iteration to calculate the wafer pattern, it will take much processing time. Therefore, our research is focusing on how to effectively decrease the iteration number and even the total processing time. We proposed an overall model-based OPC control block diagram which can be separated from four parts, including lithography process model, classical controller, initial mask pattern generated by neural network system, and desired wafer pattern considering design intent. The main goal is to decrease the model-based OPC iteration number. Based on the theory of scalar model and Fourier optics, we have successfully built up the optical model and compared with Berkeley SPLAT simulator to verify its accuracy. In addition, we heuristically tuned the parameters of PID controller and concluded the tuning experience for different mask pattern as a rule table. In the following, we introduced the neural network theory to train a model-based OPC system to obtain a good initial mask pattern, and eventually speed up the model-based OPC correction process. Finally, with the design intent or the tolerance range of line width variation, the model-based OPC convergence requirement can be relaxed such that less iteration number just can achieve the requirement.

並列關鍵字

OPC ； PID controller ； neural network ； process model ； design intent

參考文獻

[26] W. C. Huang, C. M. Lai, B. Luo, C. K. Tsai, M. H. Chih, C. W. Lai, C. C. Kuo, R. G. Liu, and H. T. Lin, “Intelligent model-based OPC,” Proceeding of SPIE, Vol. 6154, 2006.

[2] H. J. Levinson, Principle of Lithography, SPIE, 2001

[4] A. K. K. Wong, Resolution Enhancement Techniques in Optical Lithography, SPIE, 2001

[5] J. W. Goodman, Introduction to Fourier Optics, 2nd ed., McGraw-Hill, 1996

[6] Y. T. Wang, Automated Design of Phase-Shifting Masks for Microlithography, PhD dissertation, Stanford University, 1997

國際替代計量

應用控制理論並同時考慮電路設計規格以增進光學鄰近修正之收斂

全文下載

主題瀏覽