電子束微影技術（Electron Beam Lithography, EBL）因具備亞十奈米解析度與高圖樣自由度，廣泛應用於次奈米元件製作與光罩原型開發。然而，電子束於材料內部的散射效應導致能量外溢與邊界模糊，即鄰近效應（Proximity Effect），造成圖樣失真，為實務製程中亟需解決的關鍵問題。我們提出一套基於非負最小平方法（Non-Negative Least Squares, NNLS）之劑量反解模型，實現在保證物理可行性的前提下，提升修正精度與邊界控制能力, 並比較三種傳統鄰近效應修正策略：空間域下的最佳化圖樣修正、頻域反卷積法與最大熵誤差最小化法。
模擬實驗部分採用MATLAB建立雙高斯點擴散函數模型，並選用多種代表性圖樣（正方形、L形、U字形與柵欄圖樣）進行交叉比較。結果顯示，NNLS 在小尺寸圖樣下修正精度明顯優於其他方法，特別在對比度與邊界細節還原上表現最穩定。實體實驗部分則利用Raith VOYAGER電子束微影系統與CSAR 62光阻材料進行圖樣直寫與顯影，透過SEM觀察驗證各方法在物理實作上的圖樣還原差異。整體而言，NNLS在模擬與實驗中皆展現出兼具物理合理性與數值穩定性的優勢，為後續實用微影流程提供一個具體且可擴展的修正框架。

Electron Beam Lithography (EBL) is widely used in single-digit nanometer device and mask fabrications due to its sub-10 nm resolution and high patterning flexibility. However, the scattering effects of electrons within materials cause blurred edges and pattern distortion when patterns are closely packed. This is commonly referred to as the proximity effect, which poses a critical challenge in fabrication of nanometer devices and circuits. We propose a novel dose inversion model based on the Non-Negative Least Squares (NNLS) method, aiming to enhance correction accuracy and edge control while maintaining physical feasibility. This method is shown to be better than three conventional proximity effect correction (PEC) schemes; namely, optimal shape correction, Fourier-Based Deconvolution, and Local Area dosage Correction using Maximum Entropy Method.
In the simulation experiments, a double-Gaussian point spread function model was implemented in MATLAB, and several representative patterns, such as squares, L-shaped and U-shaped patterns, as well as gratings, were used for cross-comparison. Results indicate that NNLS significantly outperforms other methods in correcting small patterns, particularly in terms of contrast and edge fidelity. For physical verification, actual exposures were carried out using the Raith VOYAGER EBL system with CSAR 62 resist, followed by SEM imaging to evaluate the pattern fidelity of each correction method. Overall, NNLS demonstrated both physical validity and numerical stability in simulations and experiments, offering a concrete and scalable correction framework for future advanced lithography applications.