Introduction
The structures of interest consist of various shapes with different dimensions as depicted in the Fig 1 (a) and (b). The bonding yield can be extracted by comparing the theoretical open-areas derived from the reticle design, to the measured open-area which are extracted from the scanning acoustic microscopic (SAM) analysis. Image recognition software is used to calculate the bonding yield, defined as the measured open-area divided by the theoretical open-area, of each structure separately. By using this method, the effect of the structure geometry, surrounding channel width and depth on the bonding yield can be quantified.
Method
The donor wafers used for this experiment are prepared by depositing 1.2 µm of PECVD SiO2 followed by planarization using CMP. The carrier wafers are prepared by growing 1 µm of thermal oxide, then patterning structures into the silicon. The oxide is then removed completely by wet etching. Finally, a new thermal oxide layer is grown.
The wafers were then activated with N2 plasma, cleaned with DIW and bonded. Subsequently, the SAM images of all the wafers were analyzed and the bonding yield for the structures indicated in figure 2.b was extracted. In figure 2.a the inner area of that structure and the width of the surrounding channel is listed.
Results and discussion
The histogram plot of the bonding yield is shown in Fig.3, indicating a disparity that most of the structures are either bonded with high yield or failed to be bonded.
On top of that, no significant difference in mean bonding yield between circle and square structures could be found. (p = 0.49 at alpha 0.05) The bonding yield for the serpentines is much lower and the difference between its mean and the circles and squares mean is significant. (p < 0.0001). According to [1] a bond wave that encounters a trench in its path will come to a complete stop, subsequently it will initiate on the other side of the trench and continue bonding. The velocity of a bond wave traveling next to a channel is proportional to the channel volume. It is observed that the proportion of bonded structures is higher for smaller and shallower channels. A possible explanation is that as the bond wave at the surrounding area of the structures will travel faster around the deep and wide channels due to the improved lateral airflow[2], it results in insufficient time for the bond wave inside the channels to initiate. In figure 4 the proportion of bonded and unbonded structures is plotted grouped by channel width and channel depth.
On the other hand, the size of the structures has less impact on the mean bonding yield for the serpentines and the squares but a larger effect for the circles. Overall the larger structures have a slightly lower mean bonding yield. In figure 5 the mean bonding yield vs. the inner area grouped by structure type is shown.
When oxidizing a sharp silicon corner, it is well known that corner-rounding will occur. [3] In figure 6 an example of corner rounding on the edge of a channel is shown. To investigate the effect of this phenomenon on the bonding yield, the thickness of the final bonding oxide layer was varied between 200 and 1000 nm. The thickness of the layer had a significant impact on the bonding yield. (p < 0.0001) In figure 7.a the proportion of bonded and unbonded structures are shown grouped by re-oxidation thickness. In figure 7.b and 7.c a SAM image of a wafer with 200 nm and 1 um of thermal oxide is visible. Thinner oxides and less corner rounding resulted in a much higher bonding yield. This could mean that initiation of the bond wave on the inside of the channels occurs faster when less corner rounding is present.
References
[1] D. Radisson, “Direct bonding of patterned surfaces,” p. 151.
[2] D. Radisson, F. Fournel, and E. Charlaix, “Modelling of the direct bonding wave,” Microsyst. Technol., vol. 21, no. 5, pp. 969–971, May 2015.
[3] R. B. Marcus and T. T. Sheng, “The Oxidation of Shaped Silicon Surfaces,” J. Electrochem. Soc., vol. 129, no. 6, pp. 1278–1282, Jun. 1982.