費洛蒙擴散控制器–結合群體智能與先進製程控制的控制器

Abstract

傳統上，統計製程管制( Statistical Process Control，簡稱SPC)被應用於半導體製程的先進製程控制(Advanced Process Control，簡稱APC)之中。然而，由於SPC視量測為一系列獨立資料，所以SPC得依不同的情況使用不同的統計圖表(chart)。本論文以嶄新的觀點設計製程控制器，將線性迴歸方程式之干擾(disturbance)視為一群「社會化的昆蟲群體」的費洛蒙，在虛擬環境(數位費洛蒙筐，digital pheromone basket)中進行擴散。故這個嶄新控制器的名稱為費洛蒙擴散控制器(Pheromone Propagation Controller，簡稱 PPC)。PPC假設不同時間的干擾具有自己的行為，並能影響其附近回合(run)的干擾，故PPC使用一維(時間軸)的數位費洛蒙框。隨著數位費洛蒙框的維度不同，本論文發展了PPC與空間-時間控制器(Space-Time Controller，STC)。在STC中，本論文使用二維(平面空間)的數位費洛蒙框，假設某一道處理程序在晶圓上對於某一個量測點的外部干擾會影響其附近的數個量測點。然後，把線性迴歸方程式的干擾項視為費洛蒙放入費洛蒙筐中，根據修改後的數位費洛蒙基礎建設(digital pheromone infrastructure) 擴散費洛蒙筐中的干擾項來獲得干擾的預估值。最後，根據由線性迴歸方程式和干擾的預估值可以算出機台下一回合的輸入參數(recipe)。更進一步地，本文分析PPC與STC的穩定度，並以模擬結果說明PPC與STC的優越性。

Traditionally, statistical process control (SPC) is employed in advanced process control (APC). However, SPC, which threats measurements as a series of isolated “statistical data”, employs different chats to deal with different problems. This paper presented a novel viewpoint of process control which treats the disturbances of the process at different runs as pheromones of “social insect colony”. Then, the digital pheromones propagate in the virtual environment, which is named digital pheromone basket. So, the novel algorithm named pheromone propagation parameter (PPC). In our research, PPC is launched by the assumption that the disturbance of the linear regression model has its own behavior and affects its nearby ones at different runs. So, PPC employs the one-dimensional (time-axis) digital pheromone basket. By the shape of pheromone propagating environment, this thesis develops Space-Time Controller (STC) subsequently. STC, which uses the two-dimensional (a plane surface) pheromone basket, assumes that the measurements have their own behavior and affect others nearby in a wafer at a run. To do so, this thesis modifies digital pheromone infrastructure [28] to remove boundary effect of pheromone basket [34]. Next, the pheromone basket filled with the disturbances initially; then the disturbances (or pheromones) in the pheromone basket propagate by the modified digital pheromone infrastructure. While accomplishing propagation, the forecasting disturbance at next run can be obtained. Consequently, the revised process recipe can be acquired by the forecasting disturbance and the linear regression model. The stability region of PPC and STC are discussed. Finally, the simulation results show the superiority of PPC and STC.