本論文有兩個主要研究目標,一是發展出人工神經網路進行多熱源晶片封裝之溫度分佈研究,另一目標是應用灰色理論在多熱源晶片封裝溫度之最高溫度、平均溫度的預測。 研究結果得知,比較倒傳遞神經網路(Back-Propagation Neural Network, BPN)與Icepak套裝軟體兩種方法顯示,BPN準確度達97﹪,由此顯示應用BPN在多熱源晶片封裝之溫度分佈預測上,BPN是可行、可用的。 灰色建模運算只需利用少量數據(至少4筆資料)做預測使得計算效率加快。在本研究中,經由Icepak的分析數據進行灰預測建模,經由殘差分析與後驗差檢驗來驗證灰預測模型的準確性,結果顯示本研究所建立的灰預測模型屬於“好”(Good)的等級,這意味著本研究所建立的灰預測模型是精確與準確的。 本研究結果的主要貢獻在於依此方法在晶片封裝的熱設計與分析上提供設計者更為快速簡便的方法。 An objective of this thesis is to develop an ANN model to obtain the temperature distribution of a IC chip in the packaging processes. Another objective is to predict the highest temperature and mean temperature of it by using the Grey system theory. The calculated results of BPN(Back-Propagation Neural Network) are compared with the simulated results of the software package(Icepack). The accuracy of the results between these two methods is 97%. It shows that the BPN technique is useful to predict the temperature distribution of the IC packaging. The advantage of the Grey prediction model is the higher calculating efficiency for a few original data (at least 4 data in the series). In the research, the highest and mean temperatures of eight chips from Icepack simulation are set to be the original data in the GM(1,1) model separately. Then, the predicted results are applied to make the post residual error analysis. The analyzed results show that this Grey prediction model is belong to the “Good” grade. It means this developed GM(1,1) model is precise and accurate. The contribution of this research results can be applied to the thermal analysis and design for an IC chip packaging processes.