萃取接觸阻抗係數方法之比較研究──CBKR結構與改良式TLM結構

Abstract

為提升積體電路之性能，金氧半場效電晶體的尺寸不斷微縮，金屬與源極和汲極之接觸面積也隨之縮小，在相同接觸電阻係數下會導致較大的接觸電阻，甚至可能限制藉微縮達成的性能提升。因此，如何降低接觸電阻係數成為首要之務。而低接觸電阻係數的萃取亦是研究接觸電阻的一大課題。目前廣為使用的Cross-Bridge Kelvin Resistor (CBKR)結構其萃取誤差小，但仍會受製程限制造成的寄生效應影響。Transmission line method (TLM)結構的製作方式簡單，但易受製程變異影響也將造成萃取誤差。本論文提出一改良之TLM結構(Modified TLM, mTLM)並與CBKR結構相較，從模擬及實作探討此兩方法的準確度與極限。 過去已有文獻發表接觸電阻係數測量結構的二維模擬結果，近年亦有三維模擬結構提出。文獻中藉由模擬提出接觸電阻係數萃取之數值分析解，藉以修正萃取時測試結構造成的誤差。然分析解採用之模型通常經過簡化，且仍為一相當複雜之關係式。因其需進行大量計算且只適用於特定狀況，對於真實情況下的接觸電阻係數萃取僅能提供參考。因此，本論文使用三維數值模擬元件之電性以進行接觸電阻係數之萃取。對CBKR結構而言，討論可能影響萃取結果之參數，其中包含符合當今製程技術的測試結構尺寸。首先，縮小接觸窗可提升萃取之精確度；其次，製程容忍度(δ)對萃取的相關性變低，此結論不同於二維模擬之結果。另外，本模擬也將元件製作時可能遇到的問題列入考量。光學近接效應(optical proximity effect)導致接觸窗的邊角圓化(corner rounding)，造成萃取誤差增加。下沉式接觸結構使等效接觸面積增加，並改變電流分布，因而低估接觸電阻係數。至於mTLM結構，利用數值模擬設計結構尺寸後，同樣討論各參數對於萃取結果的影響。因mTLM結構對製程變異甚為敏感，本論文從統計觀點研究其與載子濃度分布變異及元件區側壁傾斜角度變異之相關性。結果發現mTLM結構對載子濃度分布變異有高度相關，而元件區側壁傾斜角度變異之相關性則較輕微。另外，同樣對mTLM考慮下沉式接觸結構，發現在接觸電阻係數較高時，接觸窗下沉會造成接觸電阻係數的高估；而在低接觸電阻係數時則有低估的情形。 第二部分以實驗數據對照模擬結果。為能明確定義出元件區，元件之間採用淺溝渠隔離技術(STI)，在同一製作流程下完成CBKR和mTLM兩種測試結構。CBKR萃取方式簡單，但寄生電阻的影響將導致無法避免之萃取誤差；mTLM萃取過程相對複雜，但透過大量數據平均值趨近期望值之概念，可將其對製程變異的敏感度降低，進而提升萃取的精確度。 本論文從模擬和實驗兩方面討論CBKR和mTLM兩種萃取接觸電阻係數之方法。當CBKR隨著元件微縮而有較小之接觸窗，理論上可得較精確的萃取結果，但寄生項的存在限制了萃取之精確度；而本論文提出之mTLM結構，使用STI技術完成元件製作，利用平均化將其對製程變異敏感度降低之後，對於精確萃取接觸電阻係數應具潛力。但若考慮下沉式接觸結構，其造成的影響複雜，對CBKR和mTLM兩種結構都將產生無法預測之誤差。因此，透過本研究可看出接觸電阻係數之萃取無論CBKR或mTLM兩種方法都遭遇困難，新穎結構與萃取方法仍為迫切需求。

To enhance the performance of Integrated circuits (IC), the MOSFETs have been scaled down continuously, and contact areas between metal and source/drain have shrunk consequently. Then, with a fixed specific contact resistivity (ρc), the contact resistance increases and would suppress the performance improvement caused by scaling. Therefore, how to reduce the ρc is the first task, and how to extract such a low ρc is also challenging. The Cross-Bridge Kelvin Resistor (CBKR) is a common used test structure, which has a lower extracted error according to two-dimensional (2-D) analysis while the error still exists due to the parasitic effect caused by process limitation. The transmission line method (TLM) is easy to be fabricated but sensitive to the process variation. In this thesis, a modified TLM (mTLM) procedure is proposed and compared with the CBKR method by both simulation and experiment. 2-D simulations on the CBKR method have been widely studied, and three-dimensional (3-D) simulations have also been reported in recent years. Some analytic solutions were proposed to correct the extracted error. Nevertheless, as analytic methods are used, the models are usually developed after some simplifications, though still in a complex form. Thus, for a real case, analytic methods can only provide a rough estimation, since they have to perform complex calculation but only are suitable for particular conditions. In this thesis, 3-D simulation of device characteristics is performed to evaluate the accuracy of the ρc extraction by the CBKR and mTLM methods. For the CBKR method, several parameters are considered. The dimensions of the test structures are close to the design rule of the current IC technology. First, reducing the contact area will enhance the extraction accuracy. Second, the process tolerance (δ) has less influence on the extraction, which is different from the fact presented in 2-D simulation. Moreover, issues occurring in fabrication are also taken into account. The corner-rounding contact resulting from the optical proximity effect increases the extraction error. The recessed contact structure underestimates the ρc because of the increase of the effective contact area and the change of the current distribution. As for the mTLM method, the dimensions of the mTLM structure are designed and optimized by simulation at first. Because the mTLM structure is sensitive to the process variation, this thesis studies the dependence on the variation of the dopant concentration and the variation of tapered sidewall angles of the active region according to the statistic. It is observed that there is a strong dependence on the variation of the dopant concentration, while a relative slight dependence on the variation of tapered sidewall angles of the active region. In addition, as the recessed contact is considered, the ρc is overestimated with higher ρc but underestimated with lower ρc. The second part shows the experimental data compared with the simulation. In order to define the active region explicitly, shallow trench isolation (STI) is utilized. The CBKR and the mTLM structures are realized in an identical process flow. The CBKR method is easier to extract the ρc, but the parasitic resistance is about to result in inevitable extracted error. The mTLM method is complicated; however, its sensitivity to the process variation could reduce and the extracted accuracy could be enhanced, by means of averaging sufficient data according to the fact that the average of sufficient data is closed to the expected value which is the true ρc. This thesis discusses the CBKR and mTLM methods to extract the ρc by simulation and experiment. The CBKR structure with a smaller contact area due to the devices scale down would obtain more accurate results in theory, while the parasitic resistance would still limit the accuracy. On the other hand, the mTLM structure proposed in this thesis is realized by using the STI process. Its sensitivity to the process variation could be diminished by averaging data. Therefore, the TLM method could be more accurate and is promising for ρc extraction. However, if the recessed contact structure is considered, due to its complex influence on the ρc extraction, an incalculable error would be caused for the CBKR and mTLM methods. Therefore, according to this thesis, it would be observed that the ρc extraction encounters great challenge for both CBKR and mTLM methods. Novel test structure and extraction procedure are still critical issue.