Page 45 - Chip Scale Review_January-February_2024-digital

Figure 7 shows the bonding energy
measured by NI at different locations in a
wafer (two edges and the center) for two sets
of bonded wafers, whose bonding energy
was measured by DCB. Edge 1 is the edge
of the silicon wafer in the <100> crystal
orientation, and Edge 2 is the edge in the
<110> crystal orientation.
The first sample set has weaker plasma
2
before bonding (bond strength 3.08J/m )
and the second sample has stronger plasma
2
before bonding (bond strength 4.1J/m ). In
total, 30 points were measured per each
location. Then, some obvious outliers
and/or failed measurements were removed
for analysis We did this because we
considered it necessary to allow for a
certain range of measurement error—
considering that the NI test is a destructive
test—to obtain some certain range
of measurement.
For the case of low plasma, the
bonding energy in the DCB method was
2
3.08J/m . For the case using measurement
by NI, the average bonding energy of Figure 7: Bonding energy for each wafer using NI.
2
Edge 1 was 3.23J/m , the average bonding
energy of Edge 2 was 3.51J/m , and the
2
average bonding energy of the Center
2
was 3.14J/m . The bonding energy in the
DCB test fell within the quartile range
for all samples. Furthermore, the average
bonding energy at Edge 1 and the Center
differed from the DCB test by 4.8%, and
2.0%, respectively.
The bonding energy in the DCB
method was 4.1J/m for the case using
2
high plasma before bonding. When
the measurement was done by NI, the
average bonding energy of Edge 1 was
2
3.96J/m , the average bonding energy
2
of Edge 2 was 4.3J/m , and the average
bonding energy of the Center was
2
3.98J/m . The bonding energy in the DCB Figure 8: SAM image of indentation: a) with a thermal SiO 2 film; and b) with an LT-SiO 2 film.
test also fell within the quartile range for
all samples. Furthermore, the average It should be noted that in this study, the bonded samples. Figure 8 shows SAM
bonding energy at Edge 1, Edge 2, and the bonding energy did not change between the images with D2W. The interface dielectric
Center differed from the DCB test by 3.3%, edge and the center in W2W. This indicates material for bonding is thermal SiO 2 film
4.9%, and 2.0%, respectively. that the plasma radiation and annealing (thickness=100nm) and LT-SiO 2 film
As expected, the measurement error of during W2W bonding are uniform. (thickness=100nm).
the NI test is large, therefore it is difficult Additionally, it is worth noting that the bond In Figure 8, one can see the SAM image
to measure a valid bonding strength by strength is equivalent for the case with the of D2W bonded samples after annealing
measuring only a single point, so it is obtained value in “GB.” This indicates that at 250°C. There are some minor voids at
necessary to statistically estimate the the bonding interface is not exposed to air the die edge. The root cause of the void
bonding strength from a large number of by the indentation. Therefore, the impact formation is currently being studied.
measurement points. This method, however, of water stress corrosion in this case can However, we are quite sure that it is not due
can be considered as effective as a wafer be negligible, which enables precise bond to the dicing or bonding processes because
bonding strength measurement method strength measurement by NI. the bonding voids obviously appeared after
because the statistical measurement results Bond strength measurement by NI annealing. Therefore, it might be due to a
lead to a similar range as the bonding energy (D2W bonding). Based on the result non-optimized film condition that can outgas
obtained by the DCB test. obtained in W2W, we utilized the bond water after annealing. However, we used
strength measurement method for D2W this sample for the measurement because the

43
Chip Scale Review January • February • 2024 [ChipScaleReview.com] 43