- Email: [email protected]
New observation of nanoscale interfacial evolution in micro Cu–Al wire bonds by in-situ high resolution TEM study
Scripta Materialia 115 (2016) 1–5
Contents lists available at ScienceDirect
Scripta Materialia journal homepage: www.elsevier.com/locate/scriptamat
Regular Article
New observation of nanoscale interfacial evolution in micro Cu–Al wire bonds by in-situ high resolution TEM study Hui Xu a,⁎, Ivy Qin a, Horst Clauberg a, Bob Chylak a, Viola L. Acoff b a b
Kulicke and Soffa Industries Inc., Fort Washington, PA 19034, United States. Department of Metallurgical and Materials Engineering, The University of Alabama, Tuscaloosa, AL 35487, United States
a r t i c l e
i n f o
Article history: Received 8 December 2015 Accepted 21 December 2015 Available online 13 January 2016 Keywords: Copper–aluminum Topic: Wire bond Interfacial structure Intermetallic compounds In-situ TEM.
a b s t r a c t Three types of interfacial nanostructure are identified in as-bonded Cu–Al bonds: (1) Cu/~5 nm amorphous alumina layer/Al; (2) Cu/~20 nm CuAl2 intermetallic particle/Al; and (3) Cu/Al. During annealing, in the areas of latter two types where alumina layer is fragmented, Cu9Al4 and CuAl form as second and third intermetallic layers, and grow vertically and fast together with initial CuAl2. In the area of first type where alumina layer is present, CuAl2 grows laterally and slowly via Cu diffusion through intermetallic compounds in the neighboring area where alumina is broken to reach Al. Cu–Al interdiffusion is dominated by Cu diffusion. © 2015 Elsevier Ltd. All rights reserved.
Wire bonding is a key technique for electrical interconnection between integrated circuit chips and external circuitry in microelectronics. In recent years, transition from Au wire to Cu wire bonding on Al metallization pads has brought significant cost reduction [1,2]. A full understanding of Cu–Al interfacial structure and its evaluation under bonding and reliability test is of importance. Intermetallic formation at the Cu–Al wire bond improves strength. Five alloys (CuAl2(θ), CuAl (η2), Cu4Al3 (ζ2), Cu3Al2 (δ) and Cu9Al4 (γ2)) are possible, but excessive growth will degrade mechanical integrity. Although there have been a variety of studies on Cu–Al intermetallic compound (IMC) growth [3–13], including our previous work on intermetallic formation using Transmission Electron Microscopy (TEM) [8–10], there is no report on how the as-bonded interfacial nanostructures affect the interfacial evolution during annealing. In this letter, we take the advantage of in-situ high resolution (HR) TEM of focused ion beam (FIB) thinned specimens, to ascertain nanostructural interfacial evolution during both bonding and isothermal annealing, and especially the effect of as-bonded interfacial nanostructures on intermetallic growth, direction and phase transformation during annealing. Thermosonic wire bonding was performed on a Kulicke and Soffa (K&S) IConnPS ProCu automatic ball bonder using 20 μm Cu wire and K&S BGA devices with aluminum (Al-0.5%Cu) pads of ~1 μm thick. The bonding temperature was 175 °C and bonding time 30 ms. Both electrical flame off (EFO) and bonding parameters were optimized. Forming gas (95%N2 + 5%H2) was used to prevent Cu oxidization at ball ⁎ Corresponding author. E-mail address: [email protected] (H. Xu).
http://dx.doi.org/10.1016/j.scriptamat.2015.12.025 1359-6462/© 2015 Elsevier Ltd. All rights reserved.
formation area and bonding area at a flow rate of 0.5 L/min. TEM specimens were prepared by dual beam focused ion beam (FEI Quanta 3D 200 FIB), and its specific location is shown in the inserted image on the top left corner of Fig. 1a. Annealing was performed inside the TEM chamber, and the evolution of nano-scale interfacial structure at the bond interface was in-situ recorded (FEI F20 system at 200 kV). Fast Fourier transformation (FFT) of lattice images calculated using ImageJ 1.42q was employed to identify the IMCs. A nanoprobe beam was used for composition analysis using energy dispersive X-ray spectrometry (EDX) in high angle annular dark field (HAADF) — scanning (S)TEM mode. Cu–Al interfaces in the as-bonded state consist of three types of morphologies, labeled as A, B and C in Fig. 1a, with details shown in Fig. 1b–d respectively. In morphology type I (Fig. 1b), a uniform amorphous layer with a thickness of approximately 5 nm is present between the crystalline copper ball and the aluminum pad. STEM–EDX suggests that such amorphous layer is alumina, which is believed to be the native aluminum oxide layer on the surface of Al pad. The alumina layer behaves as a diffusion barrier, so no intermetallic compound is formed at those areas. In morphology type II (Fig. 1c), the amorphous alumina was replaced with island-like particles of ~20 nm thick. FFT analysis of the interference lattices is consistent with CuAl2 (I4/mcm, a = 0.607 nm and c = 0.488 nm) aligned along [210] orientation forming a boundary with Al (Fm-3m, a = 0.406 nm) and Cu (Fm-3m, a = 0.361 nm) in [101] (Fig. 1e and f). The morphology types I and II are consistent with our previous TEM results [8]. However, here in this study, detailed HRTEM study reveals a third distinct morphology (Fig. 1d) which corresponds to the area where Cu is directly connected with Al.
2
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
Fig. 1. (a) Bright Field (BF) TEM showing the interface of an as-bonded Cu–Al wire bond, consisting of three types of morphologies; (b) Morphology type I: Cu/~5-nm-thick amorphous alumina/Al; (c) morphology type II: Cu/~20-nm-thick CuAl2 particle/Al; (d) morphology type III: Cu directly connects Al; (e) and (f) lattice images and Fourier reconstructed patterns of region D and E in (c) with CuAl2 [2 1 0].
Such connection is distinct as compared to the first two morphologies, because there is no uniform oxide layer or IMCs involved. The 5-nm-thick native oxide layer on Al pads has significant effect on the wire-bond quality, and the formation of IMCs during the bonding process must overcome this relatively inert thin oxide layer. For these Al surface areas where the aluminum oxide is not broken, IMC could not form, and corresponds to morphology type I. For those areas where aluminum oxide is broken during bonding, IMC will form and represents morphology type II. When aluminum oxide is fragmented at the end of wire bonding stage, there is no time for Cu and Al interdiffusion to form IMC, so Cu–Al direct contact is seen, as shown in morphology type III. The degree of fragmentation of the aluminum oxide layer is
dependent on bonding parameters, bonding wire and Al pad properties. All three types of morphologies are gap and void free, but it is believed that the bonding strength of type I (Cu/alumina/Al) is weaker than type II (Cu/CuAl2/Al) and type III (Cu/Al). In the next section, we will show that the IMC growth in the area of types II and III is much faster than in the area of type I. In mass production, the Cu–Al IMC coverage specification is usually more than 80% after baking for 4 h at 175 °C. The TEM sample of Cu–Al micro-bond was heated inside a TEM chamber, so IMC growth and phase transformation was in-situ recorded, as shown in Fig. 2. IMCs grow rapidly at the early stage of annealing at 300 °C in the area of types II and III where alumina layer is fragmented during bonding. As seen in Fig. 2b, IMC thickness increases from ~20 nm
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
3
Fig. 2. In-situ TEM observation of IMC growth in Cu–Al bonds: (a) as-bonded; (b) 5 min at 300 °C; (c) 20 min at 300 °C; (d) 40 min at 300 °C; (e) 100 min at 300 °C; (f) 150 min at 300 °C; (g) 200 min at 300 °C; (h) 360 min at 300 °C; (i) 360 min at 300 °C and 20 min at 350 °C; (j) 360 min at 300 °C and 40 min at 350 °C. TEM is under HAADF STEM mode.
in the as-bonded state to up to 500 nm after 5 min isothermal baking. A bi-layer of IMCs has formed, with CuAl2 close to Al and Cu9Al4 next to Cu. Since CuAl2 is the first and only phase which initially forms during bonding process, Cu9Al4 is believed to be the second IMC phase which forms at the early stage of annealing. CuAl2 and Cu9Al4 then grow simultaneously, and CuAl2 is the dominant phase until the Al pad is depleted. A third IMC phase CuAl appears between Cu9Al4 and CuAl2 after 20 min at 300 °C and becomes obvious after 100 min. IMC phase identification was performed by FFT analysis of the lattice images of IMCs and STEM– EDX. Taking the sample after 169 min at 300 °C for example, lattice images of three different IMCs were captured, as shown in Fig. 3. FFT analysis of the interference lattices of the IMCs was consistent with Cu9Al4 (P4-3m, a = 0.870 nm) aligned along [1 1 1] orientation (Fig. 3f), CuAl (C2/ml, a = 1.207 nm, b = 0.411 nm, c = 0.691 Å, and β = 124.96°) along [− 1 0 0] orientation (Fig. 3e), and CuAl2 (I4/mcm, a = 0.607 Å and c = 0.488 Å) along [2 1 0] orientation (Fig. 3d). EDX results (Fig. 3a and b) suggest the structure of Cu/Cu9Al4/alumina/Cu9Al4/ CuAl/CuAl2, consistent with TEM electron diffraction results. The alumina debris remains as a discontinuous line after annealing.
After the Al pad is consumed completely (Fig. 2e, 100 min @ 300 °C) in the area of types II and III, CuAl2 starts to be transformed into CuAl. As a result, the CuAl layer becomes more visible and Cu9Al4 gradually becomes the dominant phase. Due to the lack of supply of Al, CuAl also transforms into Cu9Al4 when annealing continues, and Cu9Al4 is the single finial IMC phase after 360 min @ 300 °C and 40 min @ 350 °C. 350 °C was used to fasten the reaction because the reaction becomes slower after Al is depleted. No significant delamination or void growth was observed during the whole phase transformation process. In the area where the alumina layer is present (type I, such as area A in Figs. 2f, and 3d), the alumina layer acts as a diffusion barrier, so the IMC growth is much slower. It is noted that after annealing for 200 min at 300 °C, only CuAl2 forms at the Al pad side, but little Cu9Al4 is present at the Cu side (Fig. 2g), while both CuAl2 and Cu9Al4 are present after 5 min annealing in the area where alumina is broken (Fig. 2b). The IMC growth direction is different. IMC grows vertically in the area where alumina is broken, but lateral growth of CuAl2 is seen in the area with uniform alumina layer. This is because the alumina is a diffusion barrier, and IMC formation at the area with uniform
4
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
Fig. 3. Identification of Cu9Al4, CuAl2 and CuAl in a sample after annealing for 169 min at 300 °C: (a) STEM image showing a tri-layer IMC; (b) line EDX showing Cu/Cu9Al4/alumina/Cu9Al4/ CuAl2/CuAl; (c) Bright Field TEM image of the same location as shown in (a); (d) lattice image and Fourier reconstructed pattern of region A in (c) with CuAl2 [0 0 1]/amorphous alumina/ Cu; (e) lattice image and Fourier reconstructed pattern of region B in (c) with CuAl [−1 0 0]; (f) lattice image and Fourier reconstructed pattern of region C in (c) with Cu9Al4 [1 1 1].
alumina layer has to be via Cu diffusion through IMCs in the neighboring area where alumina layer is broken to reach the Al underneath the alumina layer, as indicated by the arrows in Fig. 2d and e. It also indicates that diffusion in Cu–Al system is dominated by Cu. When Cu diffusion
is faster than Al diffusion through the IMCs, Cu atoms reach Al pad and react with Al to form CuAl2 first. In conclusion, we found three distinct interfacial structures in micro Cu–Al wire bonds in the as-bonded state: (I) ~ 5 nm thick native
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
amorphous alumina between Cu and Al and no IMC is formed at those areas; (II) in the area where the native alumina layer is removed, island-like CuAl2 particles of ~ 20 nm thick are formed between Cu and Al; and (III) Cu directly connects with Al. The as-bonded interfacial nanostructures have significant effect on IMC growth rate, direction and phase transformation during annealing. In the area of latter two types where alumina is fragmented, IMCs grow vertically and fast. Cu9Al4 and CuAl form as a second and third IMC layers and grow together with the initial CuAl2. When Al is completely consumed, both CuAl2 and CuAl transform to Cu9Al4 which is the terminal product. While in the area of type I where alumina layer is a diffusion barrier, CuAl2 grows slowly and laterally. Cu–Al interdiffusion is dominated by Cu diffusion. References [1] G.G. Harman, Wire Bonding in Microelectronics, third ed. McGraw-Hill, New York, London, 2010.
5
[2] P. Chauhan, Z.W. Zhong, M. Pecht, J. Electron. Mater. 42 (2013) 2415–2434. [3] H.J. Kim, J.Y. Lee, K.W. Paik, K.W. Koh, J.H. Won, S.H. Choe, J. Lee, J.T. Moon, Y.J. Park, IEEE Trans. Comp. Pack. Technol. 26 (2003) 367–374. [4] J. Li, L. Han, L.J. Duan, J. Zhong, Appl. Phys. Lett. 90 (2007) 242902. [5] M. Drozdov, G. Gur, Z. Atzmon, W. Kaplan, J. Mater. Sci. 43 (2008) 6029–6037. [6] M. Drozdov, G. Gur, Z. Atzmon, W. Kaplan, J. Mater. Sci. 43 (2008) 6038–6048. [7] C.J. Hang, C.Q. Wang, M. Mayer, Y. Zhou, H.H. Wang, Microelectron. Reliab. 48 (2008) 416–424. [8] H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, Scr. Mater. 61 (2009) 165–168. [9] H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, V.L. Acoff, Acta Mater. 59 (2011) 5661–5673. [10] H. Xu, I. Qin, H. Clauberg, B. Chylak, V.L. Acoff, Acta Mater. 61 (2013) 79–88. [11] I.T. Bae, D. Young Jung, W.T. Chen, Y. Du, J. Appl. Phys. 112 (2012) 123501. [12] H.G. Kim, S.M. Kim, J.Y. Lee, M.R. Choi, S.H. Choe, K.H. Kim, J.S. Ryu, S. Kim, S.Z. Han, W.Y. Kim, S.H. Lim, Acta Mater. 64 (2014) 356–366. [13] C.L. Gan, F.C. Classe, B.L. Chan, U. Hashim, J. Electron. Mater. 43 (2014) 1017–1025.
Contents lists available at ScienceDirect
Scripta Materialia journal homepage: www.elsevier.com/locate/scriptamat
Regular Article
New observation of nanoscale interfacial evolution in micro Cu–Al wire bonds by in-situ high resolution TEM study Hui Xu a,⁎, Ivy Qin a, Horst Clauberg a, Bob Chylak a, Viola L. Acoff b a b
Kulicke and Soffa Industries Inc., Fort Washington, PA 19034, United States. Department of Metallurgical and Materials Engineering, The University of Alabama, Tuscaloosa, AL 35487, United States
a r t i c l e
i n f o
Article history: Received 8 December 2015 Accepted 21 December 2015 Available online 13 January 2016 Keywords: Copper–aluminum Topic: Wire bond Interfacial structure Intermetallic compounds In-situ TEM.
a b s t r a c t Three types of interfacial nanostructure are identified in as-bonded Cu–Al bonds: (1) Cu/~5 nm amorphous alumina layer/Al; (2) Cu/~20 nm CuAl2 intermetallic particle/Al; and (3) Cu/Al. During annealing, in the areas of latter two types where alumina layer is fragmented, Cu9Al4 and CuAl form as second and third intermetallic layers, and grow vertically and fast together with initial CuAl2. In the area of first type where alumina layer is present, CuAl2 grows laterally and slowly via Cu diffusion through intermetallic compounds in the neighboring area where alumina is broken to reach Al. Cu–Al interdiffusion is dominated by Cu diffusion. © 2015 Elsevier Ltd. All rights reserved.
Wire bonding is a key technique for electrical interconnection between integrated circuit chips and external circuitry in microelectronics. In recent years, transition from Au wire to Cu wire bonding on Al metallization pads has brought significant cost reduction [1,2]. A full understanding of Cu–Al interfacial structure and its evaluation under bonding and reliability test is of importance. Intermetallic formation at the Cu–Al wire bond improves strength. Five alloys (CuAl2(θ), CuAl (η2), Cu4Al3 (ζ2), Cu3Al2 (δ) and Cu9Al4 (γ2)) are possible, but excessive growth will degrade mechanical integrity. Although there have been a variety of studies on Cu–Al intermetallic compound (IMC) growth [3–13], including our previous work on intermetallic formation using Transmission Electron Microscopy (TEM) [8–10], there is no report on how the as-bonded interfacial nanostructures affect the interfacial evolution during annealing. In this letter, we take the advantage of in-situ high resolution (HR) TEM of focused ion beam (FIB) thinned specimens, to ascertain nanostructural interfacial evolution during both bonding and isothermal annealing, and especially the effect of as-bonded interfacial nanostructures on intermetallic growth, direction and phase transformation during annealing. Thermosonic wire bonding was performed on a Kulicke and Soffa (K&S) IConnPS ProCu automatic ball bonder using 20 μm Cu wire and K&S BGA devices with aluminum (Al-0.5%Cu) pads of ~1 μm thick. The bonding temperature was 175 °C and bonding time 30 ms. Both electrical flame off (EFO) and bonding parameters were optimized. Forming gas (95%N2 + 5%H2) was used to prevent Cu oxidization at ball ⁎ Corresponding author. E-mail address: [email protected] (H. Xu).
http://dx.doi.org/10.1016/j.scriptamat.2015.12.025 1359-6462/© 2015 Elsevier Ltd. All rights reserved.
formation area and bonding area at a flow rate of 0.5 L/min. TEM specimens were prepared by dual beam focused ion beam (FEI Quanta 3D 200 FIB), and its specific location is shown in the inserted image on the top left corner of Fig. 1a. Annealing was performed inside the TEM chamber, and the evolution of nano-scale interfacial structure at the bond interface was in-situ recorded (FEI F20 system at 200 kV). Fast Fourier transformation (FFT) of lattice images calculated using ImageJ 1.42q was employed to identify the IMCs. A nanoprobe beam was used for composition analysis using energy dispersive X-ray spectrometry (EDX) in high angle annular dark field (HAADF) — scanning (S)TEM mode. Cu–Al interfaces in the as-bonded state consist of three types of morphologies, labeled as A, B and C in Fig. 1a, with details shown in Fig. 1b–d respectively. In morphology type I (Fig. 1b), a uniform amorphous layer with a thickness of approximately 5 nm is present between the crystalline copper ball and the aluminum pad. STEM–EDX suggests that such amorphous layer is alumina, which is believed to be the native aluminum oxide layer on the surface of Al pad. The alumina layer behaves as a diffusion barrier, so no intermetallic compound is formed at those areas. In morphology type II (Fig. 1c), the amorphous alumina was replaced with island-like particles of ~20 nm thick. FFT analysis of the interference lattices is consistent with CuAl2 (I4/mcm, a = 0.607 nm and c = 0.488 nm) aligned along [210] orientation forming a boundary with Al (Fm-3m, a = 0.406 nm) and Cu (Fm-3m, a = 0.361 nm) in [101] (Fig. 1e and f). The morphology types I and II are consistent with our previous TEM results [8]. However, here in this study, detailed HRTEM study reveals a third distinct morphology (Fig. 1d) which corresponds to the area where Cu is directly connected with Al.
2
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
Fig. 1. (a) Bright Field (BF) TEM showing the interface of an as-bonded Cu–Al wire bond, consisting of three types of morphologies; (b) Morphology type I: Cu/~5-nm-thick amorphous alumina/Al; (c) morphology type II: Cu/~20-nm-thick CuAl2 particle/Al; (d) morphology type III: Cu directly connects Al; (e) and (f) lattice images and Fourier reconstructed patterns of region D and E in (c) with CuAl2 [2 1 0].
Such connection is distinct as compared to the first two morphologies, because there is no uniform oxide layer or IMCs involved. The 5-nm-thick native oxide layer on Al pads has significant effect on the wire-bond quality, and the formation of IMCs during the bonding process must overcome this relatively inert thin oxide layer. For these Al surface areas where the aluminum oxide is not broken, IMC could not form, and corresponds to morphology type I. For those areas where aluminum oxide is broken during bonding, IMC will form and represents morphology type II. When aluminum oxide is fragmented at the end of wire bonding stage, there is no time for Cu and Al interdiffusion to form IMC, so Cu–Al direct contact is seen, as shown in morphology type III. The degree of fragmentation of the aluminum oxide layer is
dependent on bonding parameters, bonding wire and Al pad properties. All three types of morphologies are gap and void free, but it is believed that the bonding strength of type I (Cu/alumina/Al) is weaker than type II (Cu/CuAl2/Al) and type III (Cu/Al). In the next section, we will show that the IMC growth in the area of types II and III is much faster than in the area of type I. In mass production, the Cu–Al IMC coverage specification is usually more than 80% after baking for 4 h at 175 °C. The TEM sample of Cu–Al micro-bond was heated inside a TEM chamber, so IMC growth and phase transformation was in-situ recorded, as shown in Fig. 2. IMCs grow rapidly at the early stage of annealing at 300 °C in the area of types II and III where alumina layer is fragmented during bonding. As seen in Fig. 2b, IMC thickness increases from ~20 nm
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
3
Fig. 2. In-situ TEM observation of IMC growth in Cu–Al bonds: (a) as-bonded; (b) 5 min at 300 °C; (c) 20 min at 300 °C; (d) 40 min at 300 °C; (e) 100 min at 300 °C; (f) 150 min at 300 °C; (g) 200 min at 300 °C; (h) 360 min at 300 °C; (i) 360 min at 300 °C and 20 min at 350 °C; (j) 360 min at 300 °C and 40 min at 350 °C. TEM is under HAADF STEM mode.
in the as-bonded state to up to 500 nm after 5 min isothermal baking. A bi-layer of IMCs has formed, with CuAl2 close to Al and Cu9Al4 next to Cu. Since CuAl2 is the first and only phase which initially forms during bonding process, Cu9Al4 is believed to be the second IMC phase which forms at the early stage of annealing. CuAl2 and Cu9Al4 then grow simultaneously, and CuAl2 is the dominant phase until the Al pad is depleted. A third IMC phase CuAl appears between Cu9Al4 and CuAl2 after 20 min at 300 °C and becomes obvious after 100 min. IMC phase identification was performed by FFT analysis of the lattice images of IMCs and STEM– EDX. Taking the sample after 169 min at 300 °C for example, lattice images of three different IMCs were captured, as shown in Fig. 3. FFT analysis of the interference lattices of the IMCs was consistent with Cu9Al4 (P4-3m, a = 0.870 nm) aligned along [1 1 1] orientation (Fig. 3f), CuAl (C2/ml, a = 1.207 nm, b = 0.411 nm, c = 0.691 Å, and β = 124.96°) along [− 1 0 0] orientation (Fig. 3e), and CuAl2 (I4/mcm, a = 0.607 Å and c = 0.488 Å) along [2 1 0] orientation (Fig. 3d). EDX results (Fig. 3a and b) suggest the structure of Cu/Cu9Al4/alumina/Cu9Al4/ CuAl/CuAl2, consistent with TEM electron diffraction results. The alumina debris remains as a discontinuous line after annealing.
After the Al pad is consumed completely (Fig. 2e, 100 min @ 300 °C) in the area of types II and III, CuAl2 starts to be transformed into CuAl. As a result, the CuAl layer becomes more visible and Cu9Al4 gradually becomes the dominant phase. Due to the lack of supply of Al, CuAl also transforms into Cu9Al4 when annealing continues, and Cu9Al4 is the single finial IMC phase after 360 min @ 300 °C and 40 min @ 350 °C. 350 °C was used to fasten the reaction because the reaction becomes slower after Al is depleted. No significant delamination or void growth was observed during the whole phase transformation process. In the area where the alumina layer is present (type I, such as area A in Figs. 2f, and 3d), the alumina layer acts as a diffusion barrier, so the IMC growth is much slower. It is noted that after annealing for 200 min at 300 °C, only CuAl2 forms at the Al pad side, but little Cu9Al4 is present at the Cu side (Fig. 2g), while both CuAl2 and Cu9Al4 are present after 5 min annealing in the area where alumina is broken (Fig. 2b). The IMC growth direction is different. IMC grows vertically in the area where alumina is broken, but lateral growth of CuAl2 is seen in the area with uniform alumina layer. This is because the alumina is a diffusion barrier, and IMC formation at the area with uniform
4
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
Fig. 3. Identification of Cu9Al4, CuAl2 and CuAl in a sample after annealing for 169 min at 300 °C: (a) STEM image showing a tri-layer IMC; (b) line EDX showing Cu/Cu9Al4/alumina/Cu9Al4/ CuAl2/CuAl; (c) Bright Field TEM image of the same location as shown in (a); (d) lattice image and Fourier reconstructed pattern of region A in (c) with CuAl2 [0 0 1]/amorphous alumina/ Cu; (e) lattice image and Fourier reconstructed pattern of region B in (c) with CuAl [−1 0 0]; (f) lattice image and Fourier reconstructed pattern of region C in (c) with Cu9Al4 [1 1 1].
alumina layer has to be via Cu diffusion through IMCs in the neighboring area where alumina layer is broken to reach the Al underneath the alumina layer, as indicated by the arrows in Fig. 2d and e. It also indicates that diffusion in Cu–Al system is dominated by Cu. When Cu diffusion
is faster than Al diffusion through the IMCs, Cu atoms reach Al pad and react with Al to form CuAl2 first. In conclusion, we found three distinct interfacial structures in micro Cu–Al wire bonds in the as-bonded state: (I) ~ 5 nm thick native
H. Xu et al. / Scripta Materialia 115 (2016) 1–5
amorphous alumina between Cu and Al and no IMC is formed at those areas; (II) in the area where the native alumina layer is removed, island-like CuAl2 particles of ~ 20 nm thick are formed between Cu and Al; and (III) Cu directly connects with Al. The as-bonded interfacial nanostructures have significant effect on IMC growth rate, direction and phase transformation during annealing. In the area of latter two types where alumina is fragmented, IMCs grow vertically and fast. Cu9Al4 and CuAl form as a second and third IMC layers and grow together with the initial CuAl2. When Al is completely consumed, both CuAl2 and CuAl transform to Cu9Al4 which is the terminal product. While in the area of type I where alumina layer is a diffusion barrier, CuAl2 grows slowly and laterally. Cu–Al interdiffusion is dominated by Cu diffusion. References [1] G.G. Harman, Wire Bonding in Microelectronics, third ed. McGraw-Hill, New York, London, 2010.
5
[2] P. Chauhan, Z.W. Zhong, M. Pecht, J. Electron. Mater. 42 (2013) 2415–2434. [3] H.J. Kim, J.Y. Lee, K.W. Paik, K.W. Koh, J.H. Won, S.H. Choe, J. Lee, J.T. Moon, Y.J. Park, IEEE Trans. Comp. Pack. Technol. 26 (2003) 367–374. [4] J. Li, L. Han, L.J. Duan, J. Zhong, Appl. Phys. Lett. 90 (2007) 242902. [5] M. Drozdov, G. Gur, Z. Atzmon, W. Kaplan, J. Mater. Sci. 43 (2008) 6029–6037. [6] M. Drozdov, G. Gur, Z. Atzmon, W. Kaplan, J. Mater. Sci. 43 (2008) 6038–6048. [7] C.J. Hang, C.Q. Wang, M. Mayer, Y. Zhou, H.H. Wang, Microelectron. Reliab. 48 (2008) 416–424. [8] H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, Scr. Mater. 61 (2009) 165–168. [9] H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, V.L. Acoff, Acta Mater. 59 (2011) 5661–5673. [10] H. Xu, I. Qin, H. Clauberg, B. Chylak, V.L. Acoff, Acta Mater. 61 (2013) 79–88. [11] I.T. Bae, D. Young Jung, W.T. Chen, Y. Du, J. Appl. Phys. 112 (2012) 123501. [12] H.G. Kim, S.M. Kim, J.Y. Lee, M.R. Choi, S.H. Choe, K.H. Kim, J.S. Ryu, S. Kim, S.Z. Han, W.Y. Kim, S.H. Lim, Acta Mater. 64 (2014) 356–366. [13] C.L. Gan, F.C. Classe, B.L. Chan, U. Hashim, J. Electron. Mater. 43 (2014) 1017–1025.