多點振鏡掃描式全域彩色共焦表面顯微量測系統之研發

Abstract

隨著半導體產業的日漸發展，目前產業製程已經邁入了晶圓級封裝，這無疑會使得關鍵尺寸量測上面臨前所未有的挑戰。目前產業主流的量測方法為雷射三角法，但因雷射三角法其本身的遮蔽效應，使其在量測緊密排列的微凸結構甚至是高深寬比的深孔時會顯得相對劣勢，然而在如今的產業發展下，這些微結構已然成為現今先進封裝製程中常見的結構，因此，具備同軸量測的檢測技術勢必將成為半導體量測之必要元素。目前具備同軸架構且高精度之彩色共焦顯微術將具有相當高之發展潛能。 為達到半導體製成所需之高速且高解析能力的量測需求，本研究將以多點振鏡掃描式的彩色共焦量測方法為核心主軸。本研究的關鍵技術主要可列成下列三點，首先，有鑑於先前提出以數位微鏡裝置(DMD)或是液晶面板(LCoS)來做數位空間濾波器的彩色共焦顯微術方法其光效率低落，而其主要原因在於為保持共焦之特性，在使用以上空間濾波器時，無法同時讓所有光線進入系統中，因此為了提升整體系統的光效率，本團隊將會以微透鏡陣列(Microlens Array)來聚焦成多點的架構來進行整體系統的設計，以確保所有聚焦光能同時進入系統中。二，為了達到全域的量測，本研究使用光學掃描透鏡搭配二維振鏡，以達到精準且快速量測。此方法主要的優勢在於，可完全避免系統與樣品間的相對移動，移除平台移動時所產生的震動且減少位移導致的體積誤差，且振鏡本身在小角度旋轉的響應時間也較為快速，可進一步提升系統量測速度。三，本研究在偵測端特別開發出二維轉換成線性架構的光纖陣列，其主要用途為改進先前技術中，光譜解析能力受限於兩點間的距離之問題，進而增加量測解析能力。相較於現有的彩色共焦顯微術，本研究具備了高速量測的能力同時也維持住高解析能力，若與線型的彩色共焦技術相比，本研究同時還避免了橫向解析能力下降的問題。 目前本研究在量測平面鏡時，在經過影像重建後，其最大誤差為400 nm，而造成此誤差的原因為光纖轉換陣列在編排上並非為理想的線性架構。為進一步探討系統的量測精度，之後便對50 µm的階高進行量測，系統量測到的結果為53.72 µm與商用雷射共焦儀的結果相比，其偏差量為2.33 µm。之後再量測微凸塊陣列，其量測高度為28.37 µm，與Keyence量測結果相比，其偏差量為6.07 µm。造成上述偏差量較大的原因為，樣品載台與系統間有傾斜角，導致量測精度因阿貝誤差而降低。

With the continuous development of the semiconductor industry, the manufacturing process has entered the wafer-level packaging, which undoubtedly poses unprecedented challenges for critical dimension measurement. The current mainstream measurement method in the industry is the laser triangulation method, but due to the occlusion effect of the laser triangulation method, it is relatively disadvantaged in measuring closely arranged micro-convex structures or deep holes with high aspect ratios. However, in today's industry development, these microstructures have become common structures in advanced packaging processes, and therefore, detection technology with coaxial measurement will inevitably become a necessary element of semiconductor measurement. Currently, the high-precision color confocal microscopy with coaxial structure will have considerable potential for development. To meet the high-speed and high-resolution measurement requirements for semiconductor manufacturing, this study will focus on the multi-point scanning color confocal measurement method as the core. The key technologies of this study can be listed as follows. Firstly, in view of the previously proposed color confocal microscopy methods using digital microscopes (DMD) or liquid crystal panels (LCoS) to make digital spatial filters, their optical efficiency is low. The main reason is that in order to maintain the characteristic of confocal, not all rays can enter the system when using the above spatial filters. Therefore, to improve the overall system's optical efficiency, the team will design the system using a microlens array to focus into a multi-point structure to ensure that all focused light can enter the system simultaneously. Secondly, in order to achieve global measurement, this study uses a pair of the optical scan lens with a two-dimensional galvanometer to achieve precise and rapid measurements. The main advantage of this method is that it completely avoids relative movement between the system and the sample, removes the vibration generated by platform movement, reduces volume errors caused by displacement, and the response time of the mirror itself to small angle rotation is also faster, further improving the system measurement speed. In addition, this study has developed a two-dimensional fiber array that can be converted into a linear structure at the detection end, which is mainly used to improve the spectral analysis ability that was previously limited by the distance between two points, thereby increasing the measurement resolution. Compared to existing confocal microscopy, this study not only has the ability to perform high-speed measurements but also maintains high resolution. In comparison with linear confocal technology, this research also avoids the issue of decreased lateral resolution capability simultaneously. Currently, in this study, when measuring flat mirrors, the maximum error after image reconstruction is 400 nm. The cause of this error is that the arrangement of the fiber conversion array is not an ideal linear structure. To further investigate the measurement accuracy of the system, subsequent measurements were conducted on step heights of 50 µm. When comparing the results obtained by the system to those of a commercial confocal laser microscope, the measurement results is53.72 µm and the deviation was found to be 2.33 µm. Subsequently, measurements were taken on a micro-bump array with a height of 28.37 µm. When compared to Keyence measurement results, the deviation amounted to 6.07 µm. The primary reason for the larger discrepancies mentioned above is the presence of an inclination angle between the sample stage and the system, leading to a reduction in measurement accuracy due to Abbé error.