Inhibiting the consumption of Cu during multiple reflows of Pb-free solder on Cu

(1)

Inhibiting the consumption of Cu during multiple reﬂows of Pb-free

solder on Cu

H.-Y. Hsiao,

a

C.-C. Hu,

a

M.-Y. Guo,

a

C. Chen

a,⇑

and K.N. Tu

b a

Department of Materials Science and Engineering, National Chiao Tung University, Hsinchu 30010, Taiwan

b

Department of Materials Science and Engineering, University of California at Los Angeles, Los Angeles, CA 90095-1595, USA

Received 25 March 2011; revised 9 August 2011; accepted 9 August 2011 Available online 25 August 2011

An effective approach to inhibiting the consumption of Cu during multiple reflows of SnAg2.3 solder on Cu is reported. By depositing a very thin layer of solder on Cu, followed by a 10-min reflow, the scallop-type morphology of interfacial Cu6Sn5

inter-metallic compounds (IMC) became ﬂat, and the channels between them closed up. When additional solder was deposited on the sample and reﬂowed again, the consumption of Cu as well as the growth the IMC was retarded.

Keywords: Intermetallic compound; Soldering; Diﬀusion; 3D-IC packaging

Molten tin wets copper at a very high rate, and they even continue to react and form Cu–Sn intermetal-lic compounds (IMC) at room temperature after solidi-fication[1–10]. Because of the high wetting rate, SnPb as well as Sn-based Pb-free solders have been widely used in electronic packaging technology for chip-to-substrate joints. For the same reason, copper serves as the most popular under-bump-metallization (UBM) in the micro-electronic packaging industry. This is because there are now thousands of solder joints on a piece of Si chip, which must all be wetted and joined simultaneously in a single reflow in which the temperature is slightly above the melting point of the solder. In device manufacturing, several reflows are required, so the solder joints are re-melted several times to form more and more IMC. How-ever, IMC induce brittleness in the solder joint, so the fewer the better. Also, since the thickness of thin film Cu UBM is limited, it can be completely consumed in a few reflows. When a thick Cu UBM is used, the pref-erential dissolution of Cu, due to anisotropic diffusivity of Sn in Cu, has caused early failure in devices. Thus, from the point of view of yield of solder joints, a quick wetting reaction is required. But from the point of view of reliability of solder joints, no more solder reaction is required after the first reflow. There is therefore a con-flict of interest. An ideal solder joint process is one which joins easily and quickly during the first reflow,

but the solder reaction should stop or become very slow in subsequent reflows. This means that it will be very beneficial to inhibit the Cu–Sn reaction after chip-join in the first reflow. Indeed, several approaches to inhibit-ing the growth of the IMC have been proposed, includ-ing alterinclud-ing the solder composition and incorporatinclud-ing Ni into the Cu UBM[11,12]. Yet, none of them is effective. Currently, the microelectronics industry is moving to-wards three-dimensional (3D) integrated circuit (IC) packaging in which the chip technology and packaging technology are merged together[13]. In 3D IC, the con-sumption and dissolution of the Cu column in through-Si via by solder reaction is a critical issue. An effective approach to retarding the consumption of Cu is reported here, and the mechanism is discussed.

SnAg2.3 solder and electroplated copper were used to demonstrate this approach. A 100-nm-thick Ti layer was deposited on a Si wafer first, followed by sputtering of a Cu seed layer 500 nm thick. Then lithography was employed to pattern cylindrical holes 100 lm in diame-ter in photo-resist for electroplating of Cu UBM and the solder. Two types of structure of solder/UBM were fabricated: the first was 19-lm-thick solder on 20-lm-thick Cu UBM, and the second was 2-lm-20-lm-thick solder on 20-lm-thick Cu UBM. These samples were reflowed at 260°C for 1 min straight after the electroplating of the solder layer.Figure 1a and b shows cross-sectional scanning electron microscopy (SEM) images for the 19-lm-thick solder and 2-lm-thick solder on 20-lm-thick Cu UBM, respectively. The solder in the

⇑Corresponding author. E-mail:chih@cc.nctu.edu.tw

Available online at www.sciencedirect.com

Scripta Materialia 65 (2011) 907–910

(2)

2-lm-thick solder sample is consumed almost com-pletely to form Cu–Sn IMC, except the central part. Yet, for the 19-lm-thick solder sample, there is plenty of unreacted solder on the Cu UBM, and the morphol-ogy of the interfacial Cu6Sn5IMC appears to be

scallop-type. Between two neighboring IMC scallops, there is a channel which connects the molten solder and the Cu during reﬂow. The channel is the fast diﬀusion and dis-solution path of Cu to the molten solder. To reduce the consumption of Cu, the channel has to be closed.

To close up the channels between Cu6Sn5IMC, the

2-lm-thick solder samples were reflowed at 260°C for 10 min. After the heat treatment, the solder was com-pletely consumed and transformed into Cu–Sn IMC. Then, additional SnAg solder was deposited on these samples, and another reflow of 1–3 min was needed to stabilize the deposited solder. The total amount of the deposited solder was close to that of the 19-lm-thick solder samples, so that a direct comparison can be made between these two types of samples: one is the 2-lm-thick solder with heat treatment at 260°C for 10 min, followed by the addition of a thick solder layer, and the other is the 19-lm-thick solder samples without heat treatment. Then, interfacial reactions were investigated for additional reflows at 260°C for 1–10 min of these two types of samples for comparison. The change in microstructure was examined by SEM. Focused ion

beam (FIB) was employed to observe the cross-sectional morphology of the Cu–Sn IMC.

Heat treatment at 260°C for 10 min can alter the morphology of the Cu–Sn IMC in the 2-lm-thick solder samples.Figure 1c shows the plan-view SEM image for the 2-lm-thick solder sample reﬂowed at 260°C for 10 min. The Cu6Sn5IMC in the periphery of the circular

pad merged together and became ﬂat; thus there were no obvious channels between neighboring Cu–Sn grains. In addition, the Cu6Sn5grains near the center of the pad

grew bigger, resulting in fewer channels in this sample. When additional solder was deposited onto this sample and additional reflow processes were conducted, it was found that the growth of the Cu–Sn IMC was significantly inhibited in the Cu/IMC interface of the 2-lm-thick solder sample. Figure 2a–c shows the cross-sectional SEM image for the 2-lm-thick-solder sample after the additional solder was deposited, reflowed for 5 min and reflowed for 10 min, respectively. The interfa-cial IMC were identified to be Cu6Sn5by SEM energy

dispersive X-ray. However, the morphology of the Cu– Sn IMC appears very diﬀerent from the scallop-type in the 19-lm-thick solder sample shown in Figure 1a. The shape of the Cu6Sn5IMC is not a semi-circular scallop,

but a cylinder. In the cylinder, the height of the IMC is approximately only one-third to one-quarter of its width, as shown in Figure 2b and c, which suggests that the channels closed up and the dissolution Cu ﬂux was reduced.

However, the 19-lm-thick solder sample exhibits a diﬀerent growth rate and morphology of Cu6Sn5. Figure 3a–c shows the cross-sectional SEM image on

Figure 2. Cross-sectional SEM images for the 2-lm-thick solder sample after (a) additional solder was deposited, (b) reﬂow for 5 min and (c) reﬂow for 10 min.

Figure 1. (a) Cross-sectional SEM images for the as-fabricated 19-lm-thick solder sample and (b) cross-sectional SEM images for the as-fabricated 2-lm-thick solder sample. (c) Plan-view SEM images for the as-fabricated 2-lm-thick solder sample. The remaining solder was etched away selectively.

(3)

the interfacial IMC for the 19-lm-thick solder sample after reﬂow for 0, 5 and 10 min, respectively. The mor-phology of the Cu6Sn5 grains is close to semi-circular

(scallop-type). In addition, the total IMC thickness is measured to be 4.38 lm, which is thicker than 3.11 lm for the 2-lm-thick solder sample. Figure 4 shows the total Cu–Sn IMC thickness as a function of reﬂow time. Each thickness value was obtained by averaging at least four samples. As described in the experimental section, it took 3 min at 260°C to reﬂow the additional solder on the 2-lm-thick solder sample. Therefore, the mea-sured thickness value starts from 3 min.

The inhibition mechanism was investigated. It was found that the channels between the Cu6Sn5 scallops

play a critical role in the growth of the Cu–Sn IMC. To observe the top-view morphology of the Cu6Sn5

IMC, selective etching to remove the unreacted solder was conducted using a solution of one part glycerol, one part acetic acid and one part nitric acid at 25°C.

Figure 5a shows the top-view or plan-view SEM images of the Cu6Sn5IMC after etching away the unreacted

sol-der in the 19-lm-thick solsol-der sample after 10 min reﬂow at 260°C. The shape of the Cu6Sn5 IMC appears

scallop-like, and there are channels between the scal-lop-like IMC. It is reported that the channels serve as

Figure 5. Plan-view SEM images of the Cu6Sn5 IMC after etching

away of the unreacted solder after 10 min reﬂow at 260°C: (a) for the 19-lm-thick solder sample; (b) for the 2-lm-thick solder sample. Figure 4. Plot of the total Cu–Sn IMC thickness as a function of reﬂow time for the two sets of samples.

Figure 3. Cross-sectional SEM images for the 19-lm-thick-solder sample. (a) as-fabricated sample; (b) reﬂow for 5 min; and (c) reﬂow for 10 min.

(4)

rapid diﬀusion and dissolution paths of Cu into molten solder to facilitate the growth of the Cu–Sn IMC[5].

However, the morphology for the 2-lm-thick solder sample appears diﬀerent. Figure 5b presents the plan-view SEM image for the Cu6Sn5 IMC after selective

etching. The average grain size of the IMC appears much bigger than that in Figure 5a. In addition, the channel area appears much less than that inFigure 5a. ComparingFigure 1c andFigure 5b, it is observed that some of the Cu–Sn IMC with closed channels on the periphery of the pad became open again after reﬂow for 10 min. However, the grain size of the Cu6Sn5

re-mained very large. The total channel area did not in-crease much. Therefore, the 19-lm-thick solder sample still has more channels between the Cu6Sn5IMC than

the 2-lm-thick solder sample after reﬂow for an addi-tional 10 min. Therefore, the 2-lm-thick solder sample possesses a slower growth rate of the Cu–Sn IMC.

An effective way to slow down the growth of the Cu– Sn IMC in reflow was demonstrated. By depositing a 2-lm-thick solder on Cu and reflowing at 260°C for 10 min, the channels between the Cu6Sn5 scallops can

be closed up. When the sample is jointed to a thick sol-der later, this layer-type Cu6Sn5IMC becomes a

diﬀu-sion barrier for Cu/solder reaction during additional reﬂows. Although some of the channels may reopen, the total channel area is much less than the sample with-out the heat treatment.

Financial support from the National Science Council, Taiwan, under contract NSC 98-2221-E-009-036-MY3 is acknowledged.

[1] B.F. Dyson, T.R. Anthony, D. Turnbull, J. Appl. Phys. 38 (1967) 3408.

[2] K.N. Tu, Acta Met. 21 (1973) 347.

[3] K.N. Tu, R.D. Thompson, Acta Met. 30 (1982) 947. [4] S. Bader, W. Gust, H. Hieber, Acta Met. Mater. 43 (1995)

329.

[5] H.K. Kim, K.N. Tu, Phys. Rev. B53 (1996) 16027. [6] J.Y. Song, J. Yu, T.Y. Lee, Scripta Mater. 51 (2004) 167. [7] H.F. Zou, H.J. Yang, Z.F. Zhang, Acta Mater. 56 (2008)

2649.

[8] T. Laurila, V. Vuorinen, J.K. Kivilahti, Mater. Sci. Eng., R 49 (2005) 1.

[9] J. Gong, C. Liu, P.P. Conway, V.V. Silberschmidt, Scripta Mater. 60 (2009) 333.

[10] L. Jiang, N. Chawla, Scripta Mater. 63 (5) (2010) 480. [11] S.-H. Kim, J.-Y. Kim, J. Yu, And T.Y. Lee, J. Elec.

Materi. 33 (9) (2004) 948.

[12] W.T. Chen, C.E. Ho, C.R. Kao, J. Mater. Res. 17 (2002) 263.

[13] J.C. Lin, W.C. Chiou, K.F. Yang, H.B. Chang, Y.C. Lin, E.B. Liao, J.P. Hung, Y.L. Lin, P.H. Tsai, Y.C. Shih, T.J. Wu, W.J. Wu, F.W. Tsai, Y.H. Huang, T.Y. Wang, C.L. Yu, C.H. Chang, M.F. Chen, S.Y. Hou, C.H. Tung, S.O. Jeng, D.C.H. Yu, IEEE Int. Elect. Dev. Meeting (2010) 25.