SPECIAL SECTION/ADVANCED CERAMICS & GLASS: A Better Bond
April 1, 2011
Lead-free glass pastes can offer multiple bonding advantages in MEMS wafer level packaging.
Due to rapid advances in consumer electronics, automotive electronics and mobile communication, the microelectromechanical systems (MEMS) market is growing rapidly.1 Applications for MEMS cover diverse fields, including automotive, aeronautics, consumer, defense, industrial, medical and life science, and telecommunications.
Various MEMS devices require a vacuum level or controlled atmosphere operation to ensure either good performance or an acceptable operation lifetime. Different wafer bonding technologies are commercially used for MEMS packaging.2-7 Among these, glass frit bonding7 offers multiple advantages, such as superior hermeticity, less stringent surface roughness requirements and tailored expansion matching to the materials being joined.
In a process known as frit bonding, the pre-glazed wafer is placed in a wafer bonder and bonded to a cap wafer in a temperature/time/pressure cycle under the desired vacuum or backfill level to achieve the required hermetically sealed assembly. The quality of the resultant seal (or bonding) is both a function of the seal glass material used and the process variables employed in both glazing and frit bonding. The interactions between these variables are important to improve the quality of the seal.
A study was undertaken to discover the effect of seven process-related factors (pre-glaze temperature, back fill type, peak temperature, time at peak temperature, ramp rate, force applied during bonding, and cool-down rate) in the frit bonding process while keeping the glazing conditions constant. Additional attention was directed at the effect of two types of pastes (lead-free pastes DL11-205 and HT753 vs. lead-based paste FX11-036) on the quality of the bonding as measured by bond strength and uniformity, the size and type of voids in the bond lines, and metal (lead or bismuth) precipitation at the interface.
The pastes were screen-printed on 6-in.-diameter (thickness, 675 æm; weight, 28.14 +/- 0.50 g), boron-doped silicon (Si) wafers. The sealing glass pattern was printed using 325 mesh screen (45° mesh angle/1.1 mil diameter/1.2 mil emulsion thickness) using a 0.3 in./min squeegee speed and 80 mil snap-off distance.
For both cases, the average dry print thickness was 27 µm; the average fired thickness was 17 µm after glazing and 9 æm following the final frit bonding step. After screen printing, the wafers were dried in an IR belt dryer with a 120°C peak temperature. The wafers were then randomized for additional processing steps.
The sealing glass patterns on the bottom wafers were fired in the glazing profile shown in Figure 1(a), 295°C (40 min) binder burnout and 410°C (15 min) glazing, in a four-hour box furnace cycle with adequate air flow. After the glazing step, the cap wafers were sealed to the bottom wafers (glazed sealing glass pattern) in an EVG520IS wafer bonder. Figure 1(b) shows the typical glass frit wafer bonding cycle that was used.
All of the experiments in Table 1 were performed in random order with one repeat for each experiment. The resulting bond lines were examined by SAM and cross-sectional SEM, as well as EDAX analysis, for adherence to the wafers, size and location of voids, cracks, metal (lead or bismuth), and other precipitates. These observations were quantified into four responses as shown in Table 1 for analysis in DOE, similar to an earlier publication for lead-based MEMS sealing glass pastes.9
A summary of the responses for each experiment is shown in Table 2. Lower numerical scores for each response indicate a better quality of the frit-bonded seal with respect to that defect. A combined minimum for all four responses signifies the best condition for glass frit bonding.
The marginal means plot and Pareto chart for the measure of metal precipitates at the glass/Si interface response is shown in Figure 3. The steeper the line in the marginal means plot and the higher the bar in the Pareto chart, the more significant the variable is for that response. Figure 3 shows that the paste type has the strongest influence on metal precipitation.
Although not shown here, similar plots for the other three responses were obtained and analyzed. In general, lower peak temperatures led to smaller voids and smaller and fewer lead precipitates at the interface for the lead paste (FX11-036), and no metal precipitates for the lead-free paste (DL11-205). Although not plotted here, interaction plots (between variables) showed that peak time must increase as the peak temperature decreases in order to produce a good glass frit-bonded microstructure.
Table 3 includes a summary of inferences that could be deduced from this study. All four responses that affect MEMS seal quality are optimized by using a lead-free glass paste with a lower preglaze temperature (410°C) and air or no gas used as back fill. A peak temperature of ≤ 450°C for 15-30 min eliminates metal precipitates and reduces debonds, voids and cracks.
Defects in SEM were also minimized with a ramp rate of 18°C/min and a bonding force of 4000 N. A cool-down rate of 20°C/min helped to eliminate metal precipitates and defects in SAM. From the DOE, the optimum processing conditions for the lead-free glass paste (DL11-205) include:
For more information, contact Ferro Corp. at 7500 E. Pleasant Valley Rd., Independence, OH 44136; call (216) 750-6947; fax (216) 750-6915; e-mail sridharans@ferro.com; or visit www.ferro.com.
2. R. Stengl, T. Tan, and U. G”sele, J. Appl. Phys., 28, 1735 (1989).
3. C. H. Tsau, S. M. Spearing, and M. A. Schmidt, J. of Microelectromech. Systems, 11, 641 (2002).
4. G. Wallis and D. Pommerantz, J. Appl. Phys., 40, 3946 (1969).
5. V. Dragoi, T. Glinsner, G. Mittendorfer, B. Wieder, P. Lindner, SPIE Proc. Series, vol. 5116, p. 160, Bellingham, WA (2003).
6. J.-W. Yang, S. Hayes, J.-K. Lin, and D. Frear, J. of Appl. Phys., 95, 6077 (2004).
7. M. Petzold, C. Dresbach, M. Ebert, J. Bagdahn, M. Wiemer, K. Glien, J. Graf, R. Muller-Fiedler, and H. Hofer, Proc. of "10th Intersociety Conf. on Thermal and Thermomech. Phenom. in Electronics Syst., ITHERM '06," IEEE Proc. Series, 1343 (2006).
8. S.R. Schmidt, and R.G. Launsby, "Understanding Industrial Designed Experiments," pp. 3-47 Air Academy Press & Associates, Colorado Springs, CO 80920 (2005); DOE software SPC XL2000, licensed from Air Academy Associates LLC was used for this study.
9. S. Sridharan et.al., "Effect of Process Variables on Glass Frit Wafer Bonding in MEMS Wafer Level Packaging," Paper 1139-GG03-35 in Mater. Res. Soc. Symp. Proc. Vol. 1139, (2009), Materials Research Society.
Editor's note: This article is based on a poster paper presented at the 11th International Symposium on Semiconductor Wafer Bonding: Science, Technology & Applications at the 218th ECS Meeting, Las Vegas, Nev., October 10-15, 2010. The information has been reproduced with the permission of The Electrochemical Society.
Table 1. DOE Taguchi L18 design matrix
Various MEMS devices require a vacuum level or controlled atmosphere operation to ensure either good performance or an acceptable operation lifetime. Different wafer bonding technologies are commercially used for MEMS packaging.2-7 Among these, glass frit bonding7 offers multiple advantages, such as superior hermeticity, less stringent surface roughness requirements and tailored expansion matching to the materials being joined.
Table 2. Responses vs. effects for experiments in Table 1. Optimum conditions for each paste type are coded in blue.
Process Basics
In glass frit wafer bonding, sealing glass paste is screen-printed into the desired geometry. This layer is dried in an infrared (IR) oven, and then fired in a furnace to burn the binder and pre-fuse the sealing glass layer. Called glazing, this process is usually done in either a belt or box furnace under atmospheric conditions.In a process known as frit bonding, the pre-glazed wafer is placed in a wafer bonder and bonded to a cap wafer in a temperature/time/pressure cycle under the desired vacuum or backfill level to achieve the required hermetically sealed assembly. The quality of the resultant seal (or bonding) is both a function of the seal glass material used and the process variables employed in both glazing and frit bonding. The interactions between these variables are important to improve the quality of the seal.
A study was undertaken to discover the effect of seven process-related factors (pre-glaze temperature, back fill type, peak temperature, time at peak temperature, ramp rate, force applied during bonding, and cool-down rate) in the frit bonding process while keeping the glazing conditions constant. Additional attention was directed at the effect of two types of pastes (lead-free pastes DL11-205 and HT753 vs. lead-based paste FX11-036) on the quality of the bonding as measured by bond strength and uniformity, the size and type of voids in the bond lines, and metal (lead or bismuth) precipitation at the interface.
Table 3. Inferences that could be deduced from the study.
Experiment Details
The study used a Taguchi L18 screening design8 with eight factors-one qualitative material variable (B) and seven quantitative bonding variables (A and C to H), as well as four responses (see Table 1). The sealing glass pastes used in this study included the lead-based FX11-036 and the lead-free DL11-205 and HT753 (all manufactured by Ferro Corp.).The pastes were screen-printed on 6-in.-diameter (thickness, 675 æm; weight, 28.14 +/- 0.50 g), boron-doped silicon (Si) wafers. The sealing glass pattern was printed using 325 mesh screen (45° mesh angle/1.1 mil diameter/1.2 mil emulsion thickness) using a 0.3 in./min squeegee speed and 80 mil snap-off distance.
For both cases, the average dry print thickness was 27 µm; the average fired thickness was 17 µm after glazing and 9 æm following the final frit bonding step. After screen printing, the wafers were dried in an IR belt dryer with a 120°C peak temperature. The wafers were then randomized for additional processing steps.
The sealing glass patterns on the bottom wafers were fired in the glazing profile shown in Figure 1(a), 295°C (40 min) binder burnout and 410°C (15 min) glazing, in a four-hour box furnace cycle with adequate air flow. After the glazing step, the cap wafers were sealed to the bottom wafers (glazed sealing glass pattern) in an EVG520IS wafer bonder. Figure 1(b) shows the typical glass frit wafer bonding cycle that was used.
All of the experiments in Table 1 were performed in random order with one repeat for each experiment. The resulting bond lines were examined by SAM and cross-sectional SEM, as well as EDAX analysis, for adherence to the wafers, size and location of voids, cracks, metal (lead or bismuth), and other precipitates. These observations were quantified into four responses as shown in Table 1 for analysis in DOE, similar to an earlier publication for lead-based MEMS sealing glass pastes.9
Figure 1. Processing details for the time-temperature cycle used in the glazing step (a) and the time-temperature-piston force-backfill cycle in glass frit bonding (b).
Results
Figure 2 shows the SAM surface image and SEM cross-sectional images for optimum processing conditions for the two different types of pastes considered for this study. The SAM images depict the quality of the bond lines across the entire wafer. More sharp and uniformly dark (distinguishable) bond lines indicate a higher quality bond. The cross-sectional SEM images in Figure 2 provide information on voids, debonding tendency, cracks, metal (lead or bismuth) and other precipitates in the seal. The criteria outlined in Table 1 were used to quantify the responses from each experiment.A summary of the responses for each experiment is shown in Table 2. Lower numerical scores for each response indicate a better quality of the frit-bonded seal with respect to that defect. A combined minimum for all four responses signifies the best condition for glass frit bonding.
The marginal means plot and Pareto chart for the measure of metal precipitates at the glass/Si interface response is shown in Figure 3. The steeper the line in the marginal means plot and the higher the bar in the Pareto chart, the more significant the variable is for that response. Figure 3 shows that the paste type has the strongest influence on metal precipitation.
Although not shown here, similar plots for the other three responses were obtained and analyzed. In general, lower peak temperatures led to smaller voids and smaller and fewer lead precipitates at the interface for the lead paste (FX11-036), and no metal precipitates for the lead-free paste (DL11-205). Although not plotted here, interaction plots (between variables) showed that peak time must increase as the peak temperature decreases in order to produce a good glass frit-bonded microstructure.
Table 3 includes a summary of inferences that could be deduced from this study. All four responses that affect MEMS seal quality are optimized by using a lead-free glass paste with a lower preglaze temperature (410°C) and air or no gas used as back fill. A peak temperature of ≤ 450°C for 15-30 min eliminates metal precipitates and reduces debonds, voids and cracks.
Defects in SEM were also minimized with a ramp rate of 18°C/min and a bonding force of 4000 N. A cool-down rate of 20°C/min helped to eliminate metal precipitates and defects in SAM. From the DOE, the optimum processing conditions for the lead-free glass paste (DL11-205) include:
- Glazing: 410°C (15 min)
- Backfill: Vacuum or partial air
- Bonding: 450 ± 15°C (15-30 min)
- Ramp and cool-down rates of ~ 20°C/min
- Bonding force: 4000 ± 800 N for 6-in. wafer
Figure 2. SAM and SEM cross-section images of the lead-free paste (a) and lead paste (b).
Conclusions
Both lead-free and lead-based glass pastes can be used in the glass frit bonding of MEMS wafers for MEMS wafer level packaging. A lead-free paste was shown to offer a wide processing window. In addition, metal precipitates at the Si/glass interface are absent with the lead-free paste, while they are prevalent with the lead-based paste.For more information, contact Ferro Corp. at 7500 E. Pleasant Valley Rd., Independence, OH 44136; call (216) 750-6947; fax (216) 750-6915; e-mail sridharans@ferro.com; or visit www.ferro.com.
Figure 3. Marginal means plot and Pareto chart for metal precipitates response.
References
1. Market Forecast, Yole Development, France, www.yole.fr.2. R. Stengl, T. Tan, and U. G”sele, J. Appl. Phys., 28, 1735 (1989).
3. C. H. Tsau, S. M. Spearing, and M. A. Schmidt, J. of Microelectromech. Systems, 11, 641 (2002).
4. G. Wallis and D. Pommerantz, J. Appl. Phys., 40, 3946 (1969).
5. V. Dragoi, T. Glinsner, G. Mittendorfer, B. Wieder, P. Lindner, SPIE Proc. Series, vol. 5116, p. 160, Bellingham, WA (2003).
6. J.-W. Yang, S. Hayes, J.-K. Lin, and D. Frear, J. of Appl. Phys., 95, 6077 (2004).
7. M. Petzold, C. Dresbach, M. Ebert, J. Bagdahn, M. Wiemer, K. Glien, J. Graf, R. Muller-Fiedler, and H. Hofer, Proc. of "10th Intersociety Conf. on Thermal and Thermomech. Phenom. in Electronics Syst., ITHERM '06," IEEE Proc. Series, 1343 (2006).
8. S.R. Schmidt, and R.G. Launsby, "Understanding Industrial Designed Experiments," pp. 3-47 Air Academy Press & Associates, Colorado Springs, CO 80920 (2005); DOE software SPC XL2000, licensed from Air Academy Associates LLC was used for this study.
9. S. Sridharan et.al., "Effect of Process Variables on Glass Frit Wafer Bonding in MEMS Wafer Level Packaging," Paper 1139-GG03-35 in Mater. Res. Soc. Symp. Proc. Vol. 1139, (2009), Materials Research Society.
Editor's note: This article is based on a poster paper presented at the 11th International Symposium on Semiconductor Wafer Bonding: Science, Technology & Applications at the 218th ECS Meeting, Las Vegas, Nev., October 10-15, 2010. The information has been reproduced with the permission of The Electrochemical Society.