- Email: [email protected]
Effect of surface characteristic on room-temperature silicon direct bonding
Sensors and Actuators A 158 (2010) 335–341
Contents lists available at ScienceDirect
Sensors and Actuators A: Physical journal homepage: www.elsevier.com/locate/sna
Effect of surface characteristic on room-temperature silicon direct bonding Guanglan Liao a , Tielin Shi b,∗ , Xiaohui Lin a , Ziwen Ma a a b
State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074, China Wuhan National Laboratory for Optoelectronics, 1037 Luoyu Road, Wuhan, Hubei 430074, China
a r t i c l e
i n f o
Article history: Received 25 September 2009 Received in revised form 8 January 2010 Accepted 10 January 2010 Available online 25 January 2010 Keywords: Silicon direct bonding Surface characteristic Contact length Modeling
a b s t r a c t Silicon direct bonding plays an important role in micro/nano-fabrication and integration. However, silicon surface when not in good condition will result in voids or gaps in the bonding interface or even a complete failure to bond. In this paper the effect of surface characteristic on room-temperature silicon direct bonding is investigated. It is found that the occurrence of bonding is related to surface energy, micro/nano-topography and elasticity of silicon wafers. Then a dimensionless parameter, ˛, is presented in detail, and two critical values, 0.570 and 1.065 are obtained: when ˛ > 1.065, indicating that the norˆ malized combined force of both adhesion force and external force Fˆ ≤ 0 and dF/dc ≤ 0, the bonding wave will spread quickly and spontaneous bonding will occur; when 1.065 > ˛ > 0.57, Fˆ ≤ 0 and it will facilitate silicon direct bonding but not guaranteeing bonding spontaneously; when ˛ < 0.57, Fˆ > 0, the bonding resistance needs to be overcome for silicon bonding. If ˛ is very close to 0.57 and enough external pressure is provided, silicon wafer pairs will bond slowly and voids or gaps may exist in the interface, otherwise they will fail to bond. Experiments of silicon direct bonding with wafers in different surface characteristics were used to verify the model. The analysis results prove that the model describes the experiments very well. Thus, the model provides a general route for assessing the impact of surface characteristic in direct bonding, and can be employed when evaluating different processes for silicon direct bonding applications. © 2010 Elsevier B.V. All rights reserved.
1. Introduction Silicon direct bonding (SDB), a flexible process in semiconductor manufacturing, allows one to join mirror-polished silicon wafers without the addition of any glue. Now it has been widely used in a range of applications, including the fabrication of siliconon-insulator substrates, microelectronic devices, power electronic devices, micromechanic devices, optoelectronic devices, and threedimensional microelectromechanical systems [1–4]. Nevertheless, it appears that most researchers and engineers, when considering the use of SDB at room temperature, are initially confronted with a number of down-to-earth questions and problems, and the bonding mechanism remains poorly understood. It is known that SDB has been attributed to the short-range intermolecular and interatomic attraction forces, such as Van der Waals forces [5]. Consequently, the silicon surface energy and topography become the most critical parameters in the process [6–8]. According to spatial frequency range, silicon surface topography is divided into global flatness, surface waviness and surface roughness. Commercially available prime grade silicon wafers of the usual thickness exhibit a global
∗ Corresponding author. E-mail address: [email protected] (T. Shi). 0924-4247/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.sna.2010.01.025
flatness variation of 1–3 m. Variation of that order can easily be accommodated through mutual deformation of the wafers. Even a bow of up to 25 m poses no serious obstacle to wafer bonding [4]. Thus, only surface waviness and roughness are concerned in this work. Certainly, silicon wafers with obvious surface waviness and roughness will result in a small area of contact, thus yield voids or gaps in the bonding interface even failing to bond [9]. The first theory on the problem of closing gaps between contacted wafers was proposed by Stengl et al. [10]. This gap-closing theory then was further developed by Tong and Gösele [5,11,12], and used to study the energy balance between the released energy during bonding and the energy increase due to the elastic distortion of the wafer. A detailed analysis of the three-dimensional elastic field in the misfit between contacted wafers has been presented by Yu and Hu [13], leading to the same results as the gap-closing theory. Spontaneous SDB, described as a special phenomenon particularly for hydrophilic silicon wafer bonding, in which the bonding wave can spread over the entire wafer surface smoothly, has also been discussed in the literature [14–17]. However, the mechanism of spontaneous wafer bonding has not been revealed yet, and no accurate quantitative model for this process exists to date. In the present study, the silicon surface characteristic in room-temperature SDB is modeled based on the Johnson–Kendall–Roberts (JKR) theory [18]. Experiments are carried out and analyzed to verify the model.
336
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
bonding, and the dashed lines indicate the interfaces in the bonding. We need to emphasize that the combined height h in the model in Fig. 2(b) is not measurable. We can only measure the original height hi in Fig. 2(a) from the experiment, and then try to get the equivalent combined height h. This was not considered in the literature [14], thus, the available experimental data could not verify their theory reasonably. The cross-sectional surface micro/nano-profile of the upper silicon wafer in the equivalent model in Fig. 2(b) is considered as a periodically nonnegative special case examined by Johnson [22] f (x) =
Fig. 1. Cross-section of two silicon wafers in contact for direct bonding.
As depicted in Fig. 1, there are basically three factors relevant to the SDB process [19]: the material deformability, commonly expressed as its Young’s modulus E [N m−2 ]; the specific surface energy of adhesion, w [N m−1 ]; and the surface roughness of both wafers. In order to simplify the problem, the surface roughness are characterized by two variables: length L [m] and height h [m] of the gap between two wafers. Besides, the thickness of wafers also plays an important role in the bonding. Christiansen et al. investigated the bondability of wafers in different thickness conditions [4]. Turner et al. found that bonding difficulty increased with the cube of thickness [6]. Here in this work we only focus on the effect of surface characteristic on room-temperature SDB. It is clear that bonding will be easier if the surface energy of adhesion is large (w big) and the material is readily deformable (E small). So the criterion may have something to do with the ratio w/E, which has the dimension of length [m]. On the other hand, if the silicon surfaces are smooth enough (h small) and have large wavelengths L, it is also easier for wafer bonding as they do not need to be deformed much. Therefore, a criterion for bondability with a simplest dimensionless combination of w, E, h and L gives a sensible result
w E
·
L h2
x L
x ∈ (kL − x0 , kL + x0 ),
,
2x0 ≤ L
(2)
where k is an integer, x0 is the length of contacted zones, R is the mean cap radius, and L is the periodic length of the gap. Here the combined maximum height of the gaps is given by
2. Dimensional analysis
∝
L2 sin2 2R2
h=
L2 2R2
(3)
Note that for small x f (x) ≈
x2 2R
which is the usual Hertzian approximation for a cylinder of radius R. According to the contact mechanics of surfaces of periodic roughness, the pressure distribution underneath the asperities inside the contacted zones can be deduced [23] p(x) =
cos(x/L)
sin (x0 /L) − sin2 (x/L)
×
2
E∗ L 4R
sin2
x 0
L
− 2 sin2
x L
+
F L
(4)
where F is the force per asperity on the wafer surface, contributed by both the adhesion force and the external force, and E* is the
c
(1)
where c is a constant. 3. Surface characteristic modeling for room-temperature SDB It was agreed that the JKR theory is valid for solids with a large surface energy [20]. This is generally the case for hydrophilic SDB, because the silicon wafer surface roughness is in the range of micro/nano-scale where the adhesion forces between contact wafers are dominant. To evaluate the effect of surface characteristic on silicon direct bonding, a model considering the surface geometry is required. Yu and Suo presented a model in the context of direct bonding in which there is a sinusoidal varying gap at the interface, and the gap must be closed through elastic deformation during bonding [13]. Here we consider a similar way to explore the problem. In our model the asperities of silicon wafers are assumed to be periodically distributed instead of a random distribution, where the gap between silicon wafers are caused by a flatness nonuniformity with a lateral periodic extension L much larger than the gap height, h1 and h2 , as schematically shown in Fig. 2(a). According to Greenwood’s conclusions [21], the effect of a model considered roughness on two surfaces with the original height distribution hi is equivalent to the model assumed that one of the contact surfaces (the lower silicon wafer) is perfectly rigid flat not deforming in the contact, and the other (the upper silicon wafer) has a combined height distribution h, as shown in Fig. 2(b). Here the solid lines indicate the silicon wafers contact interfaces before
Fig. 2. The schematic model for SDB with periodically distributed asperities. (a) The model considered roughness on two surfaces with measurable height distribution hi ; (b) the equivalent model with one perfectly rigid flat not deforming in the contact and the other having a combined height distribution h, where the solid lines indicate the silicon wafers contact interfaces before bonding, and the dashed line the interfaces in the bonding.
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
337
combined effective modulus given by
1 − 12
E∗ =
E1
1 − 22
+
−1
E2
where Ei (i = 1, 2) is the Young’s modulus of the wafers, i (i = 1, 2) is the Poisson ratio, and p(x), reciprocal to the square root of x0 , is a periodic function of L. Under the circumstance of k = 0 there will be −KI p(x → x0− )∼ 2(x0 − x)
(5)
where KI is the stress intensity factor KI =
E∗ L
sin2
4R
x 0
L
−
F L
·
L cot
x 0
(6)
L
If not considering the surface forces between contacted silicon wafers, the condition KI = 0 will be satisfied according to Hertz contact theory, and we obtain F=
E ∗ L2 sin2 4R
x 0
(7)
L
When two silicon wafers are fully contacted, there will be 2x0 → L, and the Hertzian contact force limit will be Flim =
E ∗ L2 4R
As the surface contact forces are assumed to act in a region of vanishing area, for surface profile given by f(x), the energy release rate of the gap outside the contact zone, , is given by =
KI2
2E ∗
=
E∗ L 4R
sin2
x0 F − L L
2
·
L cot 2E ∗
x 0
L
(8)
According to the JKR theory, the contact zone size 2x0 is determined by equating to the wafer surface energy w, and we obtain the correlation of F versus x0 E ∗ L2 F= sin2 4R
x 0
L
−
2E ∗ Lw tan
x 0
L
(9)
To study the stability of the surfaces in contact, the normalized force Fˆ and contact length c are introduced
⎧ ⎨ Fˆ = F F
lim ⎩ c = x0
Fˆ = sin2 (c) − ˛
tan(c)
(10)
where ˛ is a dimensionless parameter defined by √
˛ = 4 2
(1) ˛ = 0, a special case corresponds to the Hertzian contact with no interfacial adhesion, and Fˆ will be positive and increase with c in the range of 0 ≤ c ≤ 0.5. (2) When ˛ = 0.3, Fˆ is negative while c is very small. Then Fˆ increases gradually with c, and achieves a maximum value that is positive. When c is close to 0.5, Fˆ decreases rapidly and tends to negative infinity. (3) When ˛ = 0.57, Fˆ ≤ 0 while c increases from 0 to 0.5, which will be helpful for SDB. (4) In the case of ˛ = 0.9, it is a very similar condition to (3), where Fˆ is negative over the same interval, indicating that the wafers incline to bonding because of strong surface force. (5) When ˛ = 1.2, Fˆ is always negative for the entire interval, and decreases monotonically with c. Let Fˆ ≤ 0 favorable for SDB, according to Eq. (10) we obtain ˛≥
sin2 (c)
tan(c)
Fig. 4 displays the correlation of ˛ versus c. It can be found that the curve Fˆ = 0 increases with c while c < 0.333, then decreases with c. As
dFˆ = − tan(c) ≤ 0 d˛
L
and we obtain
Fig. 3. The relationship of the normalized contact force Fˆ vs. the normalized contact length c for different values of ˛.
wR2 E ∗ L3
(11)
Substituting Eq. (3) into Eq. (11), we obtain
√ 2 2 w L ˛= · · E ∗ h2
(12)
which is very similar to Eq. (1) with c = 0.5 presented in Section 2. 2 It can be found from Eq. (10) that the first part sin (c) is the normalized Hertz pressure, and the second part, ˛ tan(c), expresses the influence of the silicon wafer surface force on the contact zone. Fig. 3 displays the correlation of Fˆ versus c for different ˛:
Fˆ will decrease with ˛. Fˆ = 0 divides the area into two regions: Fˆ < 0 favorable for SDB and Fˆ > 0 unfavorable for SDB. There is a maximum value in the curve Fˆ = 0, ˛* ≈ 0.570. Therefore, if ˛ ≥ ˛* during the SDB procedure, Fˆ < 0, which will be very helpful for SDB. However, Fˆ is not decreasing monotonically during the whole interval when ˛ ≥ ˛* as shown in Fig. 3, not guaranteeing the contact wave spreading over entire wafers during the bonding process. For spontaneous bonding with full contact of silicon wafers (c increasing from 0 to 0.5), it must satisfy the other condition dFˆ ≤ 0, dc
0 ≤ c ≤ 0.5
which yields the other critical value ˛c ≈ 1.065. Therefore, ˛ must be big enough to meet the conditions for SDB spontaneously (˛ ≥ ˛c ). If ˛ = 0.5 during SDB, Fˆ < 0 in the beginning while c < c1 , indicating that it is in favor of bonding. When c1 < c < c2 , Fˆ > 0, meaning
338
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
Fig. 4. The correlation of ˛ vs. the normalized contact length c, showing the curve corresponding to a normalized contact force Fˆ = 0.
Fig. 5. The relationship of the combined height distribution h vs. the lateral periodic extension L for different ˛ ( 1 1 1 wafers).
that external force is needed to overcome the bonding resistance. If c > c2 , Fˆ < 0, it will be favorable for SDB again. Therefore, suitable ˛ (bigger than or around 0.57) is feasible for SDB when providing enough external forces, while too small ˛ will lead to failure in SDB even under external forces, as there are too much bonding resistance to be overcome. Also it can be found from Eq. (12) that the dimensionless parameter ˛ for the surface characteristic is determined by silicon wafer materials (the surface energy w, combined effective modulus E* ) and wafer surface characteristic (the gap length L and combined height h). Thus, the activation of wafer surface and increasing the surface energy or using wafers with smaller surface micro/nanoroughness may increase ˛ and improve silicon wafers bondability. According to Eq. (12), the condition Fˆ ≤ 0 favorable for SDB and gap closing is given by
height h is less than the critical height for SDB spontaneously. Silicon wafers will bond spontaneously, and no voids or gaps will be presented in the bonding interface due to enough surface forces. When the silicon wafers surface characteristic is located in region II, Fˆ < 0, it means that the contact wave will not be guaranteed to spread over the entire wafers spontaneously. Consequently, it may need external force to help bonding, although with voids in the interface probably. When the silicon wafer surface characteristic is located in the region I, Fˆ > 0 and h exceeds the allowed critical value for SDB. External force is needed to overcome the bonding resistance and SDB probably occurs more slowly if ˛ big enough (around 0.570), or it will fail. 4. Experimental verification
√ 2 2 w L ˛= · ≥ ˛∗ · E ∗ h2
In order to verify the model, experiments of SDB with commercially available 2 in., P type, 1 1 1 silicon wafers with thickness of 380 m were conducted. Three silicon pairs, A, B, C, with different surface roughness were used for hydrophilic bonding. The wafers were cleaned by acetone in an ultrasonic cleaner for 10 min, which contributed to removing most of the organic contaminants. After that, de-ionized (DI) water was used to flush the surface. Then the wafers were placed in a boiling mixture of 98% sulfuric acid and 40% hydrogen peroxide (H2 SO4 :H2 O2 = 2:1 by volume) for 20 min. This process eliminated metal particle and any residual organic contamination effectively and rendered the wafer surface hydrophilic. Flushing the surface with DI water was also employed after the process. Subsequently RCA1 solution consisting of H2 O:30% H2 O2 :28% NH4 OH (5:1:1 by volume) was used for 15–20 min at 70 ◦ C, in order to remove the micro particles and activate the surface layer due to the weak alkalinity of NH4 OH. The surfaces were rendered more hydrophilic. The process was also followed by DI water rinsing. Finally a spin drying process was performed for 3 min at 3500 rpm to remove the DI water. After that, silicon pairs A, B and C were brought into contact.
Then the correlation between h and L for SDB is given by
√ 1 2 2 wL wL h≤ ∗ · = 1.5795 · ˛ E∗ E∗
(13)
which is very close to Tong’s conclusion about gap closing in SDB [4,24]
h < 3.6
R = 1.8 E
wL E∗
where R = L/2, E = E/(1 − 2 ), and = w. Now we take silicon wafers hydrophilic bonding as an example. The typical parameters for commercial 1 1 1 silicon wafers are the surface energy w = 0.1 J m−2 , Young’s modulus E = 190 GPa and Poisson’s ratio = 0.28. The correlation between h and L for ˛ = 0.570 and ˛ = 1.065 are shown in Fig. 5, where three regions I, II and III can be identified. When the silicon wafers surface characteristic is ˆ located in region III, that is, Fˆ < 0 and dF/dc ≤ 0, the combined gap Table 1 Experimental results of silicon direct bonding under different surface characteristics. Silicon pairs
R/m
hi = i /nm
h==
A B C
57.04 ± 21.04 22.48 ± 10.62 1.03 ± 0.22
0.13 0.77 2.22
0.18 1.09 3.14
√
2i =
√
2hi
˛
Experimental results
2.900–3.519 0.579–0.748 0.134–0.149
Spontaneously Bond slowly Not bondable
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
339
Fig. 6. Silicon wafer pairs bonding with different surface characteristic. Here in the wafers the white area meant bonded and the black unbonded. (a) SDB experiment of silicon pair A, bonding spontaneously; (b) surface morphology of silicon wafers in group A; (c) SDB experiment of silicon pair B, bonding slowly with external pressure; (d) surface morphology of silicon wafers in group B; (e) SDB experiment of silicon pair C, failing to bond; (f) surface morphology of silicon wafers in group C.
The SDB experimental results examined by infrared testing system were shown in Fig. 6. In order to investigate the effect of wafer surface characteristic on bonding results, these silicon wafers were then carefully separated and measured by atomic force microscope (AFM). The AFM scan area is 10 m × 10 m and the mean cap radius R, standard height deviation were measured and calculated from one cross-section profile as shown in Table 1. Fig. 6(a) showed the experimental result of silicon pair A bonding without external
pressure, and Fig. 6(c) and (e) bonding under 100 N. Fig. 6(b), (d) and (f) displayed the surface morphology of silicon wafers, accordingly. As can be seen, silicon wafer pair A bonds very well, silicon pair B bonds with external force although not bonding so well near the edges of the wafers, and silicon wafer pair C fails to bond, except in patches, even under external force. The asperities are assumed in the model to be distributed periodically. This differs from the actual asperities in Gaussian
340
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
The locations of these wafers in the h–L parameter space are mapped in Fig. 7. For comparison, Tong’s condition for gap closing in SDB is also shown. As can be see, the surface characteristic for silicon pairs 2# is 0.4882–0.5975 including the critical value 0.57. In other words, the surface roughness characteristic of silicon pairs 2# covers both regions I and II, Fˆ < 0 or Fˆ close to zero, meaning that the silicon wafer pair cannot bond spontaneously but slowly. For silicon pairs 1# the surface characteristic is 0.4282–0.5362 located in region I, Fˆ > 0 but close to zero. As mentioned above, when ˛ is around 0.57 (for example ˛ = 0.5 < 0.57 in Fig. 4), enough external pressure which can overcome the bonding resistance caused by Fˆ > 0 will be helpful for SDB. Therefore, silicon pairs 1# can bond but slowly under external pressure. However, if the surface characteristic of the silicon wafers is far from the region Fˆ ≤ 0, such as silicon pairs 3# (0.3309–0.3816) and 4# (0.3510–0.4338), they will fail to bond even under external pressure. For wafer pairs 5# , ˛ = 4.1825–5.7574 > 1.065, the surface characteristic is located in region III, meaning that the silicon pairs can bond spontaneously. Thus, the analysis of our model agrees with the experimental results reasonably, and is in accordance with the results from Tong’s model.
Fig. 7. The analysis results of Gui’s experiments.
distribution on the surface, and so the model parameter h and original height distribution hi cannot be obtained directly from the experiments. Usually, micro/nano-roughness is characterized by standard height deviation measured by AFM, and is similar to the gap height to represent the surface asperities. We may substitute i for hi , and then try to obtain the combined h in the equivalent model. According to Greenwood’s conclusions [20], if the height distribution on each surface is Gaussian with standard deviation i , the Gaussian distribution in √the equivalent model will possess the standard height deviation 2i . Thus,√we can √ get the combined height distribution in the model h = = 2 i = 2hi , and the silicon wafer surface characteristic ˛ can be calculated as shown in Table 1. We can find that silicon wafer pair A has good surface characteristics for bonding, where ˛ is between 2.900 and 3.519, far bigger than 1.065. That is, the roughness characteristic is located well inside region III, the silicon pair can bond spontaneously, and no voids will be observed. The surface characteristic of silicon wafer pair B with ˛ between 0.579 and 0.748 is located in region II, and the silicon wafer pair can bond, though probably not perfectly. For silicon wafer pair C with surface characteristic 0.134 < ˛ < 0.149, which is much less than 0.57 and all located in region I, wafer bonding cannot occur even under external pressure. Therefore, the analysis deduced from the model is in reasonable agreement with the experimental results. Then we take Gui’s SDB experiments with wafers having different surface roughness [19] for further verification. Five pairs of 1 0 0 silicon wafers with Young’s modulus 130 GPa and Poisson ratio 0.28 were used. The primary grade polished surfaces of six pieces of wafers were kept as original, while the surfaces of the other four pieces of wafers were modified by means of chemical mechanical polishing and chemical etching. Wet chemical cleaning and activation processes were employed, and the silicon wafers possessed hydrophilic surfaces, which would provide specific adhesion energy of about 0.1 J m−2 . The room-temperature bonding was carried out, where the 1# bonded slowly with pressure, 2# bonded slowly, 3# and 4# failed to bond, and 5# bonded spontaneously. Then 1# , 2# and 5# silicon wafers were debonded, and all wafers were measured using AFM, with R 10.9 ± 4.6, 13.3 ± 5.1, 1.8 ± 0.5, 2.0 ± 0.8, 76.2 ± 43 m and 1.2, 1.1, 1.0, 0.9, <0.1 nm, respectively. More details can be found in the literature [19].
5. Conclusions SDB is the surface contact deformation of two silicon wafers in essence, in which surface characteristic is a local, microscopic parameter of wafer surface quality crucial in wafer bonding. In this paper the surface characteristic is modeled to describe the bondability for SDB, and a dimensionless analysis was carried out in detail. It can be found from the model that the SDB results can be predicted through the analysis on the surface energy and micro/nano-topography of wafers. There are two ways to improve the silicon wafer bonding capacity: increasing the surface energy, or reducing the surface roughness. Based on dimensional analysis, the silicon wafer bonding possibilities can be identified: ˆ (1) when ˛ ≥ 1.065, Fˆ ≤ 0 and dF/dc ≤ 0, the silicon wafers can direct bond spontaneously with full contact and no voids or gaps; (2) when ˛ < 0.57 and Fˆ not close to zero, the silicon wafers will fail to bond even under external force; (3) when 0.57 ≤ ˛ < 1.065 or ˛ < 0.57 but Fˆ close to zero, external pressure can help to overcome the bonding resistance, and SDB can occur slowly although some voids will probably exist. The SDB experimental results were consistent with the model, indicating that the model can provide a general route for assessing the impact of surface characteristic in direct bonding, and can be employed when evaluating different processes for silicon direct bonding applications. As reported in the literature [8], the standard deviation or RMS roughness does not necessarily provide sufficient information about the surface topography for SDB, and bearing ratio is proved experimentally to be a useful characterization. However, there is no quantitative model about bearing ratio for SDB yet. In this paper, we model the surface characteristic for SDB, not taking the bearing ratio into consideration. Next, we will try to combine the bearing ratio into a new model for room-temperature SDB. Acknowledgements This research is sponsored by the National Key Basic Research Special Fund of China (Grant no. 2009CB724204) and National Natural Science Foundation of China (Grant nos. 50775091 and 50975106).
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
References [1] N. Miki, X. Zhang, R. Khannaa, A.A. Ayón, D. Ward, S.M. Spearing, Multi-stack silicon-direct wafer bonding for 3D MEMS manufacturing, Sens. Actuat. A: Phys. 103 (2003) 194–201. [2] M. Alexe, U. Gösele, Wafer Bonding Applications and Technology, SpringerVerlag, Berlin, 2004. [3] A.A. Ayón, X. Zhang, K.T. Turner, et al., Characterization of silicon wafer bonding for power MEMS application, Sens. Actuat. A 103 (2003) 1–8. [4] S.H. Christiansen, R. Singh, U. Gösele, Wafer direct bonding: from advanced substrate engineering to future applications in micro/nanoelectronics, Proc. IEEE 94 (2006) 2060–2106. [5] Q.Y. Tong, U. Gösele, Semiconductor wafer bonding: recent developments, Mater. Chem. Phys. 37 (1994) 101–127. [6] K.T. Turner, S.M. Spearing, W.A. Baylies, et al., Effect of nanotopography in direct wafer bonding: modeling and measurements, IEEE Trans. Semiconductor Manuf. 18 (2005) 289–296. [7] N. Mikia, S.M. Spearing, Effect of nanoscale surface roughness on the bonding energy of direct-bonded silicon wafers, J. Appl. Phys. 94 (2003) 6800–6806. [8] Z. Tang, P. Peng, T. Shi, G. Liao, L. Nie, S. Liu, Effect of nanoscale surface topography on low temperature direct wafer bonding process with UV activation, Sens. Actuat. A 151 (2009) 81–86. [9] U. Gösele, H. Stenzel, T. Martini, J. Steinkirchner, D. Conrad, K. Scheerschmidt, Self-propagating room-temperature silicon wafer bonding in ultrahigh vacuum, Appl. Phys. Lett. 67 (1995) 3614–3616. [10] R. Stengl, K. Mitani, V. Lehman, U. Gösele, Silicon wafer bonding: chemistry, elasto-mechanics and manufacturing, in: Proceedings of the 1989 IEEE SOS/SOI Technology Conference, Stateline, Nevada, 1989, pp. 123–124. [11] U. Gösele, Q.U. Tong, Silicon layer transfer by wafer bonding, in: M.A. Schmidt, T. Abe, C.E. Hunt, H. Baumgart (Eds.), Proceedings of the 2nd International Symposium on Semiconductor Wafer Bonding: Science, Technology, and Applications, Pennington, NJ, America, 1993, pp. 395–409. [12] Q.Y. Tong, U. Gösele, Thickness considerations in direct silicon wafer bonding, J. Electrochem. Soc. 142 (1995) 3975–3979. [13] H.H. Yu, Z. Suo, A model of wafer bonding by elastic accommodation, J. Mech. Phys. Solids 46 (1998) 829–844. [14] Z. Tang, T. Shi, G. Liao, S. Liu, Modeling the formation of spontaneous wafer direct bonding under low temperature, Microelectron. Eng. 85 (2008) 1754–1757. [15] K. Ljungberg, A. Söderbärg, Y. Bäcklund, Spontaneous bonding of hydrophobic silicon surfaces, Appl. Phys. Lett. 62 (1993) 1362–1364. [16] F. Rieutord, B. Bataillou, H. Moriceau, Dynamics of a bonding front, Phys. Rev. Lett. 94 (2005), 236101-1–236101-4.
341
[17] D.V. Kubair, S.M. Spearing, Cohesive zone model for direct silicon wafer bonding, J. Phys. D: Appl. Phys. 40 (2007) 3070–3076. [18] K.L. Johnson, K. Kendall, A.D. Roberts, Surface energy and the contact of elastic solids, Proc. R. Soc. Lond. A 324 (1971) 301–313. [19] C. Gui, M. Elwenspoek, N. Tas, J.G.E. Gardeniers, The effect of surface roughness on direct wafer bonding, J. Appl. Phys. 85 (1999) 7448–7454. [20] D. Maugis, Adhesion of spheres: the JKR-DMT transition using Dugdale model, J. Colloid Interface Sci. 150 (1992) 243–269. [21] J.A. Greenwood, J.H. Tripp, The contact of two nominally flat rough surfaces, Proc. Inst. Mech. Eng. 185 (1970–1971) 625–633. [22] K.L. Johnson, The adhesion of two elastic bodies with slightly wavy surfaces, Int. J. Solids Struct. 32 (1995) 423–430. [23] C.Y. Hui, Y.Y. Lin, J.M. Baney, E.J. Kramer, The mechanics of contact and adhesion of periodically rough surfaces, J. Polym. Sci., Part B: Polym. Phys. 39 (2001) 1195–1214. [24] Q.Y. Tong, U. Gösele, Semiconductor Wafer Bonding: Science and Technology, Wiley, New York, 1999, p. 45.
Biographies Guanglan Liao received the PhD degree from Huazhong University of Science and Technology, Wuhan, China, in 2003. He is currently an associate professor in the State Key Laboratory of Digital Manufacturing Equipment and Technology at Huazhong University of Science & Technology. His research focuses on developing gas sensors and micro/nano-fabrication technologies. Tielin Shi received the PhD degree from Huazhong University of Science and Technology, Wuhan, China, in 1991. He is currently the director of Division of Optoelectronic Materials and Micro/Nano-Manufacturing in Wuhan National Laboratory for Optoelectronics, Wuhan, China. His research focuses on micro/nano-fabrication technologies and developing advanced fabrication systems for microelectronic industry. Xiaohui Lin received the PhD degree from Huazhong University of Science & Technology, Wuhan, China, in 2008. His PhD work is focusing on the silicon wafer direct bonding under low temperature. Ziwen Ma received the MS degree from Huazhong University of Science & Technology, Wuhan, China, in 2008. His research focuses on micromachining and wafer-scale packaging.
Contents lists available at ScienceDirect
Sensors and Actuators A: Physical journal homepage: www.elsevier.com/locate/sna
Effect of surface characteristic on room-temperature silicon direct bonding Guanglan Liao a , Tielin Shi b,∗ , Xiaohui Lin a , Ziwen Ma a a b
State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074, China Wuhan National Laboratory for Optoelectronics, 1037 Luoyu Road, Wuhan, Hubei 430074, China
a r t i c l e
i n f o
Article history: Received 25 September 2009 Received in revised form 8 January 2010 Accepted 10 January 2010 Available online 25 January 2010 Keywords: Silicon direct bonding Surface characteristic Contact length Modeling
a b s t r a c t Silicon direct bonding plays an important role in micro/nano-fabrication and integration. However, silicon surface when not in good condition will result in voids or gaps in the bonding interface or even a complete failure to bond. In this paper the effect of surface characteristic on room-temperature silicon direct bonding is investigated. It is found that the occurrence of bonding is related to surface energy, micro/nano-topography and elasticity of silicon wafers. Then a dimensionless parameter, ˛, is presented in detail, and two critical values, 0.570 and 1.065 are obtained: when ˛ > 1.065, indicating that the norˆ malized combined force of both adhesion force and external force Fˆ ≤ 0 and dF/dc ≤ 0, the bonding wave will spread quickly and spontaneous bonding will occur; when 1.065 > ˛ > 0.57, Fˆ ≤ 0 and it will facilitate silicon direct bonding but not guaranteeing bonding spontaneously; when ˛ < 0.57, Fˆ > 0, the bonding resistance needs to be overcome for silicon bonding. If ˛ is very close to 0.57 and enough external pressure is provided, silicon wafer pairs will bond slowly and voids or gaps may exist in the interface, otherwise they will fail to bond. Experiments of silicon direct bonding with wafers in different surface characteristics were used to verify the model. The analysis results prove that the model describes the experiments very well. Thus, the model provides a general route for assessing the impact of surface characteristic in direct bonding, and can be employed when evaluating different processes for silicon direct bonding applications. © 2010 Elsevier B.V. All rights reserved.
1. Introduction Silicon direct bonding (SDB), a flexible process in semiconductor manufacturing, allows one to join mirror-polished silicon wafers without the addition of any glue. Now it has been widely used in a range of applications, including the fabrication of siliconon-insulator substrates, microelectronic devices, power electronic devices, micromechanic devices, optoelectronic devices, and threedimensional microelectromechanical systems [1–4]. Nevertheless, it appears that most researchers and engineers, when considering the use of SDB at room temperature, are initially confronted with a number of down-to-earth questions and problems, and the bonding mechanism remains poorly understood. It is known that SDB has been attributed to the short-range intermolecular and interatomic attraction forces, such as Van der Waals forces [5]. Consequently, the silicon surface energy and topography become the most critical parameters in the process [6–8]. According to spatial frequency range, silicon surface topography is divided into global flatness, surface waviness and surface roughness. Commercially available prime grade silicon wafers of the usual thickness exhibit a global
∗ Corresponding author. E-mail address: [email protected] (T. Shi). 0924-4247/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.sna.2010.01.025
flatness variation of 1–3 m. Variation of that order can easily be accommodated through mutual deformation of the wafers. Even a bow of up to 25 m poses no serious obstacle to wafer bonding [4]. Thus, only surface waviness and roughness are concerned in this work. Certainly, silicon wafers with obvious surface waviness and roughness will result in a small area of contact, thus yield voids or gaps in the bonding interface even failing to bond [9]. The first theory on the problem of closing gaps between contacted wafers was proposed by Stengl et al. [10]. This gap-closing theory then was further developed by Tong and Gösele [5,11,12], and used to study the energy balance between the released energy during bonding and the energy increase due to the elastic distortion of the wafer. A detailed analysis of the three-dimensional elastic field in the misfit between contacted wafers has been presented by Yu and Hu [13], leading to the same results as the gap-closing theory. Spontaneous SDB, described as a special phenomenon particularly for hydrophilic silicon wafer bonding, in which the bonding wave can spread over the entire wafer surface smoothly, has also been discussed in the literature [14–17]. However, the mechanism of spontaneous wafer bonding has not been revealed yet, and no accurate quantitative model for this process exists to date. In the present study, the silicon surface characteristic in room-temperature SDB is modeled based on the Johnson–Kendall–Roberts (JKR) theory [18]. Experiments are carried out and analyzed to verify the model.
336
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
bonding, and the dashed lines indicate the interfaces in the bonding. We need to emphasize that the combined height h in the model in Fig. 2(b) is not measurable. We can only measure the original height hi in Fig. 2(a) from the experiment, and then try to get the equivalent combined height h. This was not considered in the literature [14], thus, the available experimental data could not verify their theory reasonably. The cross-sectional surface micro/nano-profile of the upper silicon wafer in the equivalent model in Fig. 2(b) is considered as a periodically nonnegative special case examined by Johnson [22] f (x) =
Fig. 1. Cross-section of two silicon wafers in contact for direct bonding.
As depicted in Fig. 1, there are basically three factors relevant to the SDB process [19]: the material deformability, commonly expressed as its Young’s modulus E [N m−2 ]; the specific surface energy of adhesion, w [N m−1 ]; and the surface roughness of both wafers. In order to simplify the problem, the surface roughness are characterized by two variables: length L [m] and height h [m] of the gap between two wafers. Besides, the thickness of wafers also plays an important role in the bonding. Christiansen et al. investigated the bondability of wafers in different thickness conditions [4]. Turner et al. found that bonding difficulty increased with the cube of thickness [6]. Here in this work we only focus on the effect of surface characteristic on room-temperature SDB. It is clear that bonding will be easier if the surface energy of adhesion is large (w big) and the material is readily deformable (E small). So the criterion may have something to do with the ratio w/E, which has the dimension of length [m]. On the other hand, if the silicon surfaces are smooth enough (h small) and have large wavelengths L, it is also easier for wafer bonding as they do not need to be deformed much. Therefore, a criterion for bondability with a simplest dimensionless combination of w, E, h and L gives a sensible result
w E
·
L h2
x L
x ∈ (kL − x0 , kL + x0 ),
,
2x0 ≤ L
(2)
where k is an integer, x0 is the length of contacted zones, R is the mean cap radius, and L is the periodic length of the gap. Here the combined maximum height of the gaps is given by
2. Dimensional analysis
∝
L2 sin2 2R2
h=
L2 2R2
(3)
Note that for small x f (x) ≈
x2 2R
which is the usual Hertzian approximation for a cylinder of radius R. According to the contact mechanics of surfaces of periodic roughness, the pressure distribution underneath the asperities inside the contacted zones can be deduced [23] p(x) =
cos(x/L)
sin (x0 /L) − sin2 (x/L)
×
2
E∗ L 4R
sin2
x 0
L
− 2 sin2
x L
+
F L
(4)
where F is the force per asperity on the wafer surface, contributed by both the adhesion force and the external force, and E* is the
c
(1)
where c is a constant. 3. Surface characteristic modeling for room-temperature SDB It was agreed that the JKR theory is valid for solids with a large surface energy [20]. This is generally the case for hydrophilic SDB, because the silicon wafer surface roughness is in the range of micro/nano-scale where the adhesion forces between contact wafers are dominant. To evaluate the effect of surface characteristic on silicon direct bonding, a model considering the surface geometry is required. Yu and Suo presented a model in the context of direct bonding in which there is a sinusoidal varying gap at the interface, and the gap must be closed through elastic deformation during bonding [13]. Here we consider a similar way to explore the problem. In our model the asperities of silicon wafers are assumed to be periodically distributed instead of a random distribution, where the gap between silicon wafers are caused by a flatness nonuniformity with a lateral periodic extension L much larger than the gap height, h1 and h2 , as schematically shown in Fig. 2(a). According to Greenwood’s conclusions [21], the effect of a model considered roughness on two surfaces with the original height distribution hi is equivalent to the model assumed that one of the contact surfaces (the lower silicon wafer) is perfectly rigid flat not deforming in the contact, and the other (the upper silicon wafer) has a combined height distribution h, as shown in Fig. 2(b). Here the solid lines indicate the silicon wafers contact interfaces before
Fig. 2. The schematic model for SDB with periodically distributed asperities. (a) The model considered roughness on two surfaces with measurable height distribution hi ; (b) the equivalent model with one perfectly rigid flat not deforming in the contact and the other having a combined height distribution h, where the solid lines indicate the silicon wafers contact interfaces before bonding, and the dashed line the interfaces in the bonding.
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
337
combined effective modulus given by
1 − 12
E∗ =
E1
1 − 22
+
−1
E2
where Ei (i = 1, 2) is the Young’s modulus of the wafers, i (i = 1, 2) is the Poisson ratio, and p(x), reciprocal to the square root of x0 , is a periodic function of L. Under the circumstance of k = 0 there will be −KI p(x → x0− )∼ 2(x0 − x)
(5)
where KI is the stress intensity factor KI =
E∗ L
sin2
4R
x 0
L
−
F L
·
L cot
x 0
(6)
L
If not considering the surface forces between contacted silicon wafers, the condition KI = 0 will be satisfied according to Hertz contact theory, and we obtain F=
E ∗ L2 sin2 4R
x 0
(7)
L
When two silicon wafers are fully contacted, there will be 2x0 → L, and the Hertzian contact force limit will be Flim =
E ∗ L2 4R
As the surface contact forces are assumed to act in a region of vanishing area, for surface profile given by f(x), the energy release rate of the gap outside the contact zone, , is given by =
KI2
2E ∗
=
E∗ L 4R
sin2
x0 F − L L
2
·
L cot 2E ∗
x 0
L
(8)
According to the JKR theory, the contact zone size 2x0 is determined by equating to the wafer surface energy w, and we obtain the correlation of F versus x0 E ∗ L2 F= sin2 4R
x 0
L
−
2E ∗ Lw tan
x 0
L
(9)
To study the stability of the surfaces in contact, the normalized force Fˆ and contact length c are introduced
⎧ ⎨ Fˆ = F F
lim ⎩ c = x0
Fˆ = sin2 (c) − ˛
tan(c)
(10)
where ˛ is a dimensionless parameter defined by √
˛ = 4 2
(1) ˛ = 0, a special case corresponds to the Hertzian contact with no interfacial adhesion, and Fˆ will be positive and increase with c in the range of 0 ≤ c ≤ 0.5. (2) When ˛ = 0.3, Fˆ is negative while c is very small. Then Fˆ increases gradually with c, and achieves a maximum value that is positive. When c is close to 0.5, Fˆ decreases rapidly and tends to negative infinity. (3) When ˛ = 0.57, Fˆ ≤ 0 while c increases from 0 to 0.5, which will be helpful for SDB. (4) In the case of ˛ = 0.9, it is a very similar condition to (3), where Fˆ is negative over the same interval, indicating that the wafers incline to bonding because of strong surface force. (5) When ˛ = 1.2, Fˆ is always negative for the entire interval, and decreases monotonically with c. Let Fˆ ≤ 0 favorable for SDB, according to Eq. (10) we obtain ˛≥
sin2 (c)
tan(c)
Fig. 4 displays the correlation of ˛ versus c. It can be found that the curve Fˆ = 0 increases with c while c < 0.333, then decreases with c. As
dFˆ = − tan(c) ≤ 0 d˛
L
and we obtain
Fig. 3. The relationship of the normalized contact force Fˆ vs. the normalized contact length c for different values of ˛.
wR2 E ∗ L3
(11)
Substituting Eq. (3) into Eq. (11), we obtain
√ 2 2 w L ˛= · · E ∗ h2
(12)
which is very similar to Eq. (1) with c = 0.5 presented in Section 2. 2 It can be found from Eq. (10) that the first part sin (c) is the normalized Hertz pressure, and the second part, ˛ tan(c), expresses the influence of the silicon wafer surface force on the contact zone. Fig. 3 displays the correlation of Fˆ versus c for different ˛:
Fˆ will decrease with ˛. Fˆ = 0 divides the area into two regions: Fˆ < 0 favorable for SDB and Fˆ > 0 unfavorable for SDB. There is a maximum value in the curve Fˆ = 0, ˛* ≈ 0.570. Therefore, if ˛ ≥ ˛* during the SDB procedure, Fˆ < 0, which will be very helpful for SDB. However, Fˆ is not decreasing monotonically during the whole interval when ˛ ≥ ˛* as shown in Fig. 3, not guaranteeing the contact wave spreading over entire wafers during the bonding process. For spontaneous bonding with full contact of silicon wafers (c increasing from 0 to 0.5), it must satisfy the other condition dFˆ ≤ 0, dc
0 ≤ c ≤ 0.5
which yields the other critical value ˛c ≈ 1.065. Therefore, ˛ must be big enough to meet the conditions for SDB spontaneously (˛ ≥ ˛c ). If ˛ = 0.5 during SDB, Fˆ < 0 in the beginning while c < c1 , indicating that it is in favor of bonding. When c1 < c < c2 , Fˆ > 0, meaning
338
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
Fig. 4. The correlation of ˛ vs. the normalized contact length c, showing the curve corresponding to a normalized contact force Fˆ = 0.
Fig. 5. The relationship of the combined height distribution h vs. the lateral periodic extension L for different ˛ ( 1 1 1 wafers).
that external force is needed to overcome the bonding resistance. If c > c2 , Fˆ < 0, it will be favorable for SDB again. Therefore, suitable ˛ (bigger than or around 0.57) is feasible for SDB when providing enough external forces, while too small ˛ will lead to failure in SDB even under external forces, as there are too much bonding resistance to be overcome. Also it can be found from Eq. (12) that the dimensionless parameter ˛ for the surface characteristic is determined by silicon wafer materials (the surface energy w, combined effective modulus E* ) and wafer surface characteristic (the gap length L and combined height h). Thus, the activation of wafer surface and increasing the surface energy or using wafers with smaller surface micro/nanoroughness may increase ˛ and improve silicon wafers bondability. According to Eq. (12), the condition Fˆ ≤ 0 favorable for SDB and gap closing is given by
height h is less than the critical height for SDB spontaneously. Silicon wafers will bond spontaneously, and no voids or gaps will be presented in the bonding interface due to enough surface forces. When the silicon wafers surface characteristic is located in region II, Fˆ < 0, it means that the contact wave will not be guaranteed to spread over the entire wafers spontaneously. Consequently, it may need external force to help bonding, although with voids in the interface probably. When the silicon wafer surface characteristic is located in the region I, Fˆ > 0 and h exceeds the allowed critical value for SDB. External force is needed to overcome the bonding resistance and SDB probably occurs more slowly if ˛ big enough (around 0.570), or it will fail. 4. Experimental verification
√ 2 2 w L ˛= · ≥ ˛∗ · E ∗ h2
In order to verify the model, experiments of SDB with commercially available 2 in., P type, 1 1 1 silicon wafers with thickness of 380 m were conducted. Three silicon pairs, A, B, C, with different surface roughness were used for hydrophilic bonding. The wafers were cleaned by acetone in an ultrasonic cleaner for 10 min, which contributed to removing most of the organic contaminants. After that, de-ionized (DI) water was used to flush the surface. Then the wafers were placed in a boiling mixture of 98% sulfuric acid and 40% hydrogen peroxide (H2 SO4 :H2 O2 = 2:1 by volume) for 20 min. This process eliminated metal particle and any residual organic contamination effectively and rendered the wafer surface hydrophilic. Flushing the surface with DI water was also employed after the process. Subsequently RCA1 solution consisting of H2 O:30% H2 O2 :28% NH4 OH (5:1:1 by volume) was used for 15–20 min at 70 ◦ C, in order to remove the micro particles and activate the surface layer due to the weak alkalinity of NH4 OH. The surfaces were rendered more hydrophilic. The process was also followed by DI water rinsing. Finally a spin drying process was performed for 3 min at 3500 rpm to remove the DI water. After that, silicon pairs A, B and C were brought into contact.
Then the correlation between h and L for SDB is given by
√ 1 2 2 wL wL h≤ ∗ · = 1.5795 · ˛ E∗ E∗
(13)
which is very close to Tong’s conclusion about gap closing in SDB [4,24]
h < 3.6
R = 1.8 E
wL E∗
where R = L/2, E = E/(1 − 2 ), and = w. Now we take silicon wafers hydrophilic bonding as an example. The typical parameters for commercial 1 1 1 silicon wafers are the surface energy w = 0.1 J m−2 , Young’s modulus E = 190 GPa and Poisson’s ratio = 0.28. The correlation between h and L for ˛ = 0.570 and ˛ = 1.065 are shown in Fig. 5, where three regions I, II and III can be identified. When the silicon wafers surface characteristic is ˆ located in region III, that is, Fˆ < 0 and dF/dc ≤ 0, the combined gap Table 1 Experimental results of silicon direct bonding under different surface characteristics. Silicon pairs
R/m
hi = i /nm
h==
A B C
57.04 ± 21.04 22.48 ± 10.62 1.03 ± 0.22
0.13 0.77 2.22
0.18 1.09 3.14
√
2i =
√
2hi
˛
Experimental results
2.900–3.519 0.579–0.748 0.134–0.149
Spontaneously Bond slowly Not bondable
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
339
Fig. 6. Silicon wafer pairs bonding with different surface characteristic. Here in the wafers the white area meant bonded and the black unbonded. (a) SDB experiment of silicon pair A, bonding spontaneously; (b) surface morphology of silicon wafers in group A; (c) SDB experiment of silicon pair B, bonding slowly with external pressure; (d) surface morphology of silicon wafers in group B; (e) SDB experiment of silicon pair C, failing to bond; (f) surface morphology of silicon wafers in group C.
The SDB experimental results examined by infrared testing system were shown in Fig. 6. In order to investigate the effect of wafer surface characteristic on bonding results, these silicon wafers were then carefully separated and measured by atomic force microscope (AFM). The AFM scan area is 10 m × 10 m and the mean cap radius R, standard height deviation were measured and calculated from one cross-section profile as shown in Table 1. Fig. 6(a) showed the experimental result of silicon pair A bonding without external
pressure, and Fig. 6(c) and (e) bonding under 100 N. Fig. 6(b), (d) and (f) displayed the surface morphology of silicon wafers, accordingly. As can be seen, silicon wafer pair A bonds very well, silicon pair B bonds with external force although not bonding so well near the edges of the wafers, and silicon wafer pair C fails to bond, except in patches, even under external force. The asperities are assumed in the model to be distributed periodically. This differs from the actual asperities in Gaussian
340
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
The locations of these wafers in the h–L parameter space are mapped in Fig. 7. For comparison, Tong’s condition for gap closing in SDB is also shown. As can be see, the surface characteristic for silicon pairs 2# is 0.4882–0.5975 including the critical value 0.57. In other words, the surface roughness characteristic of silicon pairs 2# covers both regions I and II, Fˆ < 0 or Fˆ close to zero, meaning that the silicon wafer pair cannot bond spontaneously but slowly. For silicon pairs 1# the surface characteristic is 0.4282–0.5362 located in region I, Fˆ > 0 but close to zero. As mentioned above, when ˛ is around 0.57 (for example ˛ = 0.5 < 0.57 in Fig. 4), enough external pressure which can overcome the bonding resistance caused by Fˆ > 0 will be helpful for SDB. Therefore, silicon pairs 1# can bond but slowly under external pressure. However, if the surface characteristic of the silicon wafers is far from the region Fˆ ≤ 0, such as silicon pairs 3# (0.3309–0.3816) and 4# (0.3510–0.4338), they will fail to bond even under external pressure. For wafer pairs 5# , ˛ = 4.1825–5.7574 > 1.065, the surface characteristic is located in region III, meaning that the silicon pairs can bond spontaneously. Thus, the analysis of our model agrees with the experimental results reasonably, and is in accordance with the results from Tong’s model.
Fig. 7. The analysis results of Gui’s experiments.
distribution on the surface, and so the model parameter h and original height distribution hi cannot be obtained directly from the experiments. Usually, micro/nano-roughness is characterized by standard height deviation measured by AFM, and is similar to the gap height to represent the surface asperities. We may substitute i for hi , and then try to obtain the combined h in the equivalent model. According to Greenwood’s conclusions [20], if the height distribution on each surface is Gaussian with standard deviation i , the Gaussian distribution in √the equivalent model will possess the standard height deviation 2i . Thus,√we can √ get the combined height distribution in the model h = = 2 i = 2hi , and the silicon wafer surface characteristic ˛ can be calculated as shown in Table 1. We can find that silicon wafer pair A has good surface characteristics for bonding, where ˛ is between 2.900 and 3.519, far bigger than 1.065. That is, the roughness characteristic is located well inside region III, the silicon pair can bond spontaneously, and no voids will be observed. The surface characteristic of silicon wafer pair B with ˛ between 0.579 and 0.748 is located in region II, and the silicon wafer pair can bond, though probably not perfectly. For silicon wafer pair C with surface characteristic 0.134 < ˛ < 0.149, which is much less than 0.57 and all located in region I, wafer bonding cannot occur even under external pressure. Therefore, the analysis deduced from the model is in reasonable agreement with the experimental results. Then we take Gui’s SDB experiments with wafers having different surface roughness [19] for further verification. Five pairs of 1 0 0 silicon wafers with Young’s modulus 130 GPa and Poisson ratio 0.28 were used. The primary grade polished surfaces of six pieces of wafers were kept as original, while the surfaces of the other four pieces of wafers were modified by means of chemical mechanical polishing and chemical etching. Wet chemical cleaning and activation processes were employed, and the silicon wafers possessed hydrophilic surfaces, which would provide specific adhesion energy of about 0.1 J m−2 . The room-temperature bonding was carried out, where the 1# bonded slowly with pressure, 2# bonded slowly, 3# and 4# failed to bond, and 5# bonded spontaneously. Then 1# , 2# and 5# silicon wafers were debonded, and all wafers were measured using AFM, with R 10.9 ± 4.6, 13.3 ± 5.1, 1.8 ± 0.5, 2.0 ± 0.8, 76.2 ± 43 m and 1.2, 1.1, 1.0, 0.9, <0.1 nm, respectively. More details can be found in the literature [19].
5. Conclusions SDB is the surface contact deformation of two silicon wafers in essence, in which surface characteristic is a local, microscopic parameter of wafer surface quality crucial in wafer bonding. In this paper the surface characteristic is modeled to describe the bondability for SDB, and a dimensionless analysis was carried out in detail. It can be found from the model that the SDB results can be predicted through the analysis on the surface energy and micro/nano-topography of wafers. There are two ways to improve the silicon wafer bonding capacity: increasing the surface energy, or reducing the surface roughness. Based on dimensional analysis, the silicon wafer bonding possibilities can be identified: ˆ (1) when ˛ ≥ 1.065, Fˆ ≤ 0 and dF/dc ≤ 0, the silicon wafers can direct bond spontaneously with full contact and no voids or gaps; (2) when ˛ < 0.57 and Fˆ not close to zero, the silicon wafers will fail to bond even under external force; (3) when 0.57 ≤ ˛ < 1.065 or ˛ < 0.57 but Fˆ close to zero, external pressure can help to overcome the bonding resistance, and SDB can occur slowly although some voids will probably exist. The SDB experimental results were consistent with the model, indicating that the model can provide a general route for assessing the impact of surface characteristic in direct bonding, and can be employed when evaluating different processes for silicon direct bonding applications. As reported in the literature [8], the standard deviation or RMS roughness does not necessarily provide sufficient information about the surface topography for SDB, and bearing ratio is proved experimentally to be a useful characterization. However, there is no quantitative model about bearing ratio for SDB yet. In this paper, we model the surface characteristic for SDB, not taking the bearing ratio into consideration. Next, we will try to combine the bearing ratio into a new model for room-temperature SDB. Acknowledgements This research is sponsored by the National Key Basic Research Special Fund of China (Grant no. 2009CB724204) and National Natural Science Foundation of China (Grant nos. 50775091 and 50975106).
G. Liao et al. / Sensors and Actuators A 158 (2010) 335–341
References [1] N. Miki, X. Zhang, R. Khannaa, A.A. Ayón, D. Ward, S.M. Spearing, Multi-stack silicon-direct wafer bonding for 3D MEMS manufacturing, Sens. Actuat. A: Phys. 103 (2003) 194–201. [2] M. Alexe, U. Gösele, Wafer Bonding Applications and Technology, SpringerVerlag, Berlin, 2004. [3] A.A. Ayón, X. Zhang, K.T. Turner, et al., Characterization of silicon wafer bonding for power MEMS application, Sens. Actuat. A 103 (2003) 1–8. [4] S.H. Christiansen, R. Singh, U. Gösele, Wafer direct bonding: from advanced substrate engineering to future applications in micro/nanoelectronics, Proc. IEEE 94 (2006) 2060–2106. [5] Q.Y. Tong, U. Gösele, Semiconductor wafer bonding: recent developments, Mater. Chem. Phys. 37 (1994) 101–127. [6] K.T. Turner, S.M. Spearing, W.A. Baylies, et al., Effect of nanotopography in direct wafer bonding: modeling and measurements, IEEE Trans. Semiconductor Manuf. 18 (2005) 289–296. [7] N. Mikia, S.M. Spearing, Effect of nanoscale surface roughness on the bonding energy of direct-bonded silicon wafers, J. Appl. Phys. 94 (2003) 6800–6806. [8] Z. Tang, P. Peng, T. Shi, G. Liao, L. Nie, S. Liu, Effect of nanoscale surface topography on low temperature direct wafer bonding process with UV activation, Sens. Actuat. A 151 (2009) 81–86. [9] U. Gösele, H. Stenzel, T. Martini, J. Steinkirchner, D. Conrad, K. Scheerschmidt, Self-propagating room-temperature silicon wafer bonding in ultrahigh vacuum, Appl. Phys. Lett. 67 (1995) 3614–3616. [10] R. Stengl, K. Mitani, V. Lehman, U. Gösele, Silicon wafer bonding: chemistry, elasto-mechanics and manufacturing, in: Proceedings of the 1989 IEEE SOS/SOI Technology Conference, Stateline, Nevada, 1989, pp. 123–124. [11] U. Gösele, Q.U. Tong, Silicon layer transfer by wafer bonding, in: M.A. Schmidt, T. Abe, C.E. Hunt, H. Baumgart (Eds.), Proceedings of the 2nd International Symposium on Semiconductor Wafer Bonding: Science, Technology, and Applications, Pennington, NJ, America, 1993, pp. 395–409. [12] Q.Y. Tong, U. Gösele, Thickness considerations in direct silicon wafer bonding, J. Electrochem. Soc. 142 (1995) 3975–3979. [13] H.H. Yu, Z. Suo, A model of wafer bonding by elastic accommodation, J. Mech. Phys. Solids 46 (1998) 829–844. [14] Z. Tang, T. Shi, G. Liao, S. Liu, Modeling the formation of spontaneous wafer direct bonding under low temperature, Microelectron. Eng. 85 (2008) 1754–1757. [15] K. Ljungberg, A. Söderbärg, Y. Bäcklund, Spontaneous bonding of hydrophobic silicon surfaces, Appl. Phys. Lett. 62 (1993) 1362–1364. [16] F. Rieutord, B. Bataillou, H. Moriceau, Dynamics of a bonding front, Phys. Rev. Lett. 94 (2005), 236101-1–236101-4.
341
[17] D.V. Kubair, S.M. Spearing, Cohesive zone model for direct silicon wafer bonding, J. Phys. D: Appl. Phys. 40 (2007) 3070–3076. [18] K.L. Johnson, K. Kendall, A.D. Roberts, Surface energy and the contact of elastic solids, Proc. R. Soc. Lond. A 324 (1971) 301–313. [19] C. Gui, M. Elwenspoek, N. Tas, J.G.E. Gardeniers, The effect of surface roughness on direct wafer bonding, J. Appl. Phys. 85 (1999) 7448–7454. [20] D. Maugis, Adhesion of spheres: the JKR-DMT transition using Dugdale model, J. Colloid Interface Sci. 150 (1992) 243–269. [21] J.A. Greenwood, J.H. Tripp, The contact of two nominally flat rough surfaces, Proc. Inst. Mech. Eng. 185 (1970–1971) 625–633. [22] K.L. Johnson, The adhesion of two elastic bodies with slightly wavy surfaces, Int. J. Solids Struct. 32 (1995) 423–430. [23] C.Y. Hui, Y.Y. Lin, J.M. Baney, E.J. Kramer, The mechanics of contact and adhesion of periodically rough surfaces, J. Polym. Sci., Part B: Polym. Phys. 39 (2001) 1195–1214. [24] Q.Y. Tong, U. Gösele, Semiconductor Wafer Bonding: Science and Technology, Wiley, New York, 1999, p. 45.
Biographies Guanglan Liao received the PhD degree from Huazhong University of Science and Technology, Wuhan, China, in 2003. He is currently an associate professor in the State Key Laboratory of Digital Manufacturing Equipment and Technology at Huazhong University of Science & Technology. His research focuses on developing gas sensors and micro/nano-fabrication technologies. Tielin Shi received the PhD degree from Huazhong University of Science and Technology, Wuhan, China, in 1991. He is currently the director of Division of Optoelectronic Materials and Micro/Nano-Manufacturing in Wuhan National Laboratory for Optoelectronics, Wuhan, China. His research focuses on micro/nano-fabrication technologies and developing advanced fabrication systems for microelectronic industry. Xiaohui Lin received the PhD degree from Huazhong University of Science & Technology, Wuhan, China, in 2008. His PhD work is focusing on the silicon wafer direct bonding under low temperature. Ziwen Ma received the MS degree from Huazhong University of Science & Technology, Wuhan, China, in 2008. His research focuses on micromachining and wafer-scale packaging.