以穿透式電子顯微鏡觀察在高電流密度下覆晶封裝中 Ni(V)金屬墊層的消耗現象

Abstract

依據過去本實驗室的研究，我們提出了一個新的失效機制，在此失效機制中，電流叢聚(current crowding)效應與引發的局部焦耳熱效應(local Joule heating)首先導致位於晶片端電子流入口處的局部鎳釩金屬墊層[Ni(V) under bump metallization; Ni(V)UBM]被快速消耗，此區域稱之為所謂的”鎳消耗區”[“consumed Ni(V)”]。隨著反應的時間拉長，此鎳消耗區將轉變為一所謂的”多孔性結構”(“porous structure”)。生成此多孔性結構將會造成不導電的結果，電子流因而轉向流入鄰近區域，因此鄰近區域將成為電子流入口處，此一區域的Ni(V) UBM又將被消耗。在反應的過程中，上述現象將不斷地重複，使得非導電性的區域逐漸擴大。一旦非導電性的區域幾乎橫跨了整個銲點與導線的界面時，整個電子流線路將成為斷路而失效。 過去乃是以場發射掃描式電子顯微鏡(field emission scanning electron microscopy; FE-SEM)來觀察上述現象的整個過程。此多孔性結構在經聚焦離子束(focused ion beam)拋光後，呈現的外部型態乃一兩相結構(two-phase structure)。然而，在如此倍率之下，以及掃描式電子顯微鏡的儀器限制，此多孔性結構的細微結構仍然無法被鑑定出來。因此本實驗更進一步地運用掃描穿透式電子顯微鏡(scanning transmission electron microscopy; STEM)來觀察此Ni消耗後的結構以及其後形成的多孔性結構。在明視野影像上，我們發現在Ni消耗區的外部型態乃一均勻的結構，與Ni(V) UBM在經過一次迴銲之後所擁有的柱狀結晶明顯不同。另外，consumed Ni(V)與(Cu,Ni)6Sn5介金屬的界面變的很模糊，此結果與之前FE-SEM的背向散射影像(back-scattered electron image; BSE)是相呼應的。而多孔性結構的外部型態中會有一整排孔洞散佈在原來Ni(V) UBM的基地中，而這一排孔洞尤其分布於Ni(V)與(Cu,Ni)6Sn5介金屬的界面附近。此排孔洞的形成我們認為與通電與否有密切的關係。此外也進行了掃描穿透式電子顯微鏡的線掃描的成分分析，其結果說明了在經歷Ni消耗之後，代表有許多的Ni原子已經擴散出Ni(V) UBM。Ni(V) UBM中的Ni原子百分率濃度(at.%)降低了許多，我們知道原先Ni(V) UBM中的原子百分濃度Ni約為93at.%而V約為7at.%。另外雖然在此過程中亦有Sn、Cu原子擴散入Ni(V)基地。然而此時V相對含量已由原先的7at.%提高到30-40at.%，但Ni(V)層體積卻沒有變化，這說明了Ni(V)層基地的空洞化。多孔性結構也進行了穿透式電子顯微鏡的線掃描成份分析， Ni原子訊號更進一步減弱，只剩下在Ni(V)與Al的界面上還有些許Ni的訊號被偵測到。之後進行的是選區繞射(selected area diffraction pattern; SADP)的檢驗，我們發現Ni消耗區及多孔性結構的SADP結果均為寬大的入射電子束伴隨著一些模糊的繞射環以及繞射點，這說明了這兩者的結構有可能為非晶質的基地上分布了一些奈米微晶。其中多孔性結構的繞射點比較起來稍微明銳一點，這有可能是因為多孔性結構經歷的反應時間較久因而有較高程度的結晶性。更進一步地我們利用了高解析穿透式電子顯微鏡(high resolution transmission electron microscopy; HRTEM)來觀察多孔性結構，發現其基地確實為非晶質，上面佈滿了一些奈米微晶，此結果呼應了先前的SADP觀察。我們利用影像處理得到這些微晶的快速傅立葉轉換(FFT)的模擬繞射點，經過匹配進一步說明了這些微晶乃Cu6Sn5與VSn2(V2Sn3)微晶。 在電性方面，在前人的實驗中推論N(V)金屬墊層與Al導線中應該會生成一層NiAl3介金屬，並評估此介金屬生成後伴隨形成的氧化層與不導電的結果有密切關聯。我們利用TEM明視影像與暗視影像，STEM線掃描分析及HRTEM影像，在N(V)金屬墊層與Al導線中間發現了一層相當薄的連續層，經過鑑定此連續層為NiAl3介金屬。另外經由STEM-HAAEF影像輔助說明NiAl3介金屬與多孔性結構之間存在一層氧化層。而電子流主要流經鎳消耗區時造成的些微電阻上升，歸因於其基地非晶化所造成。而電子流主要流經多孔性結構時造成的電阻快速上升，歸因於及基地非晶化、奈米晶的生成以及連續氧化層生成的綜合效果。

In our previous studies, a two-stage failure mechanism of flip-chip solder joints during electromigration was pointed out. When the test vehicles were under current stressing and solid-state aging, the combination of current crowding and resulting Joule heating would make the original Ni(V) UBM be replaced by the so-called “consumed Ni(V)” at the edge of the passivation where electron flows entered into the solder joints. With increasing the reaction time, the “consumed Ni(V)” would transform to the so-called “porous structure” by comparing the morphology change in scanning electron microscopy (SEM) back-scattered electron (BSE) images. The formation of porous structure would block electron flows. Hence, the electron flows would be diverted to the neighboring Ni(V) and subsequently this area would also be attacked to form consumed Ni(V). The above processes would proceed and the range of porous structure would expand. Once porous structure propagated all the Ni(V) layer, the device would fail. All the above processes were observed by SEM. In this study, transmission electron microscopy (TEM) was used to observe the microstructures of “consumed Ni(V)” and “porous structure” in detail. The sputtered Ni(V) UBM after one reflow was with columnar structures in TEM bright field (BF) images. However, consumed Ni(V) was a solid layer without crystalline grains. Besides, the original Ni(V)/(Cu,Ni)6Sn5 interface became blurred. These results were due to the Ni-Cu-Sn interdiffusion at the original Ni(V)/(Cu,Ni)6Sn5 interface. Confirmation was done by scanning TEM (STEM) compo- sitional analyses. The consumed Ni(V) became a V-Cu-Sn-Ni region. The selected area diffraction patterns (SADP) of consumed Ni(V) showed a broad incident beam with faint diffraction rings and spots, which indicated that it was composed of an amorphous phase and crystalline phases with ultra-fine grains. The BF images of porous structure showed that there were many voids near the original Ni(V)/(Cu,Ni)6Sn5 interface. This result corresponded to that porous structure was shown as a two-phase structure by SEM in our previous studies. For the original Ni(V) layer, Cu and Sn in-fluxes could not balance the serious Ni out-flux, which resulted in the void formation. It was confirmed by the fact that almost no Ni signal was detected in porous structure by STEM-EDX analyses. V atoms were immobile and trapped in Ni(V) layer without agglomeration, which was beneficial to form an amorphous matrix. The SADP of porous structure implied that it was composed of an amorphous matrix and fine-crystalline Cu6Sn5 and VSn2 intermetallic compounds (IMC). The following HRTEM observation supported this implication. Compared to consumed Ni(V), the contrast of porous structure was lighter in TEM BF images. This implied that porous structure had lower atomic weight. The V content could be a critical value since V was immobile during electromigration. From consumed Ni(V) to porous structure, the related V content increased from 30-40 at.% to over 40 at.%. Since the volume of Ni(V) layer was unchanged, it implied that porous structure had higher degree of porosity. In addition, by comparing the SADP types of consumed Ni(V) and porous structure, porous structure was convinced to have higher degree of crystallization. Finally, the current applying was considered to play an important role on the void formation. The electron flows enhanced the Ni out-flux and retarded the Cu and Sn in-fluxes, which resulted in more serious unbalanced fluxes. Porous structure with higher degree of crystallization also provided more fast diffusion paths for vacancies. These phenomenons would accelerate the void formation. It was pointed out in the past literature that a NiAl3 phase was presumably formed at the original Al/Ni(V) interface. This phase formation followed by the oxide formation should be correlated to the abrupt increase of resistance of the interconnect after Ni(V) consumption. Based on our TEM BF/DF images, STEM line scan analyses and HRTEM observation, a thin continuous layer was exactly observed between porous structure and Al. After identification, this continuous layer was convinced to be NiAl3. In addition, according to STEM high angle angular dark field (HAADF) images, an oxide layer was likely to form between NiAl3 and porous structure. The slight increase of electrical resistance when electron flows mainly passed through consumed Ni(V) was due to the amorphization of Ni(V) matrix. The fast increase of resistance when electron flows mainly passed through porous structure was due to the combination of amorphization of Ni(V) matrix, formation of crystalline fine grains and continuous oxide layer.