混合式多目標演化計算與幾何規劃在資通面板電路與布局之研究

Abstract

在薄膜電晶體液晶顯示(TFT-LCD)面板產業中，工程師經常需要面對許多複雜的設計問題，且這些問題幾乎都伴隨著各種實務上的考量。為了加速設計流程以及獲得較好的效果，有效率的應用優良的最佳化方法是非常重要的。在這篇論文的兩個部份中，我們嘗試使用不同的方式來解決TFT-LCD產業裡所遇到的兩個重要實務設計最佳化問題。對於這兩個問題，在傳統上設計者皆需倚靠長期的設計經驗來逐步試錯，方能得到可接受的設計。如此不僅拉長面板設計的週期，且無法保證設計的優異性。

在第一部份中，我們應用首次提出的混合式多目標演化式演算法，來對非晶矽薄膜電晶體閘極(ASG)驅動電路進行最佳化。ASG驅動電路為驅動TFT-LCD面板訊號之一新穎方案，其相關設計工作相當複雜，而且需花上大量時間。而提出的演算法由兩大方法所構成：一為多目標演化式計算法(MOEA)，屬於探索式的全域最佳化演算法；一為統計模型建構技巧。我們比較提出的演算法以及以模擬為基礎的(Simulation-based) MOEA所得到的結果。其顯示提出的演算法不只可以獲得與Simulation-based MOEA相似的解分佈，並且能夠大量節省運算時間。另外，我們亦實際下線製作出ASG驅動電路樣品並量測之，此電路之設計乃是以基因演算法進行最佳化。實驗量測結果證明所有的電特性都在要求的規格之內，且更有數項規格較模擬的結果更佳。

在第二部份中，我們首次將一個TFT-LCD繞線布局問題化為幾何規劃(Geometric Programming, GP)的標準形式，並使用全域解法器進行最佳化求解。良好的繞線結構可以有效的降低延遲時間以及避免時序錯誤，因此設計者所希望求得的，乃是由外部積體電路連接至TFT-LCD面板之最佳繞線方案。雖然目前已存在一些可進行自動繞線的商業軟體，但是這些軟體大部分乃是專為安排積體電路內部的繞線所設計，並且皆無法保證在實務要求下能獲得此問題之最佳繞線。GP為數學非線性規劃的其中一支，其已被應用在解決許多工程上的問題。如果一個問題可以被轉換為符合幾何規劃的形式，那麼已發展良好，且以內點法作為核心的全域解法器便可以在很短時間內得到最佳解。因此從原問題導出符合GP形式的問題是最重要，也是最困難的工作。我們利用所導出的TFT-LCD繞線問題的GP形式，來求得最佳繞線方案，並和商業軟體所得到的繞線相比較。結果顯示此方法確具有優越性。

總之，我們已經應用最佳化方法在兩個TFT-LCD產業中實際遇到的問題上，並且成功的獲得較已存在的方法所得更優良的解。這對TFT-LCD的製造與設計皆有正面助益。

In the industry of thin-film transistor liquid crystal display (TFT-LCD), engineers usually encounter many complicated designs with practical considerations. In order to accelerate design processes and to obtain better performances simultaneously, effective application of advanced optimization methods is an important issue. In the two parts of this thesis, we try to use different approaches to solve two practical optimization problems in TFT-LCD domain. Traditionally, the designers must rely on empirical experience to iteratively achieve some feasible solutions in these problems. Thus the design cycle may be stretched and the superiority of the designs can not be promised.

The first part is about optimization of integrated amorphous silicon TFT gate (ASG) driver circuit by the firstly proposed hybrid multi-objective evolutionary algorithm (MOEA). ASG driver circuit is a novel approach to drive TFT-LCD panel, and its design is usually a complicated and time-consuming work. The proposed algorithm is composed of MOEA which belongs to heuristic global optimization methods and statistical response model construction technique. We compare the results of proposed algorithm and full simulation-based MOEA. The results show that proposed algorithm not only achieves comparable solutions to the simulation-based MOEA but also saves considerable time cost. Besides, a sample of ASG driver circuit with an optimal design is fabricated and measured. The experimental results show that all characteristics fit in with the specifications and some terms are even better than simulation.

In the second part, a practical TFT-LCD routing problem is formulated into geometric programming (GP) standard form and then solved globally. Quality routing topology can reduce the delay time of signal and the probability of timing error. Therefore the designers want to find the optimal topology of routing wires from external integrated circuit (IC) to TFT-LCD panels. Although there exists some commercial automatic routing tools, most of them are specifically designed for the layout routing in ICs and no one can promise to obtain optimal routing with practical requirements in this problem. Geometric programming belongs to mathematical non-linear programmings and has been used to solve many engineering problems. If the problem can be transformed into the specific (GP) form, the well-developed global solver with interior point methods can be easily applied to obtain the optimal solution(s) in very short time. Thus deriving the GP formulations of the problems is the most critical and difficult work. We use the formulation to achieve the global optimal routings and compare the results with the routings by commercial tool. The results show the superiority of this approach.

In summary, we have applied optimization methods on two important design problems in TFT-LCD industry, and successfully achieved stellar results from existing methods. It may benefit design and manufacturing of TFT-LCD.