A method for permanently bonding two silicon wafers so as to realize a uniform enclosure is described. This includes wafer preparation, cleaning, RT bonding, and annealing processes. The resulting bonded wafers (cells) have uniformity of enclosure ~1%1,2. The resulting geometry allows for measurements of confined liquids and gasses.
Measurements of the heat capacity and superfluid fraction of confined 4He have been performed near the lambda transition using lithographically patterned and bonded silicon wafers. Unlike confinements in porous materials often used for these types of experiments3, bonded wafers provide predesigned uniform spaces for confinement. The geometry of each cell is well known, which removes a large source of ambiguity in the interpretation of data.
Exceptionally flat, 5 cm diameter, 375 µm thick Si wafers with about 1 µm variation over the entire wafer can be obtained commercially (from Semiconductor Processing Company, for example). Thermal oxide is grown on the wafers to define the confinement dimension in the z-direction. A pattern is then etched in the oxide using lithographic techniques so as to create a desired enclosure upon bonding. A hole is drilled in one of the wafers (the top) to allow for the introduction of the liquid to be measured. The wafers are cleaned2 in RCA solutions and then put in a microclean chamber where they are rinsed with deionized water4. The wafers are bonded at RT and then annealed at ~1,100 °C. This forms a strong and permanent bond. This process can be used to make uniform enclosures for measuring thermal and hydrodynamic properties of confined liquids from the nanometer to the micrometer scale.
When clean silicon wafers are brought into intimate contact at RT, they are attracted to each other via van der Waals forces and form weak local bonds. This bonding can be made much stronger by annealing at higher temperatures5,6. Bonding can be done successfully with surfaces of either SiO2 to Si or SiO2 to SiO2. Bonding of Si wafers are most commonly used for silicon on insulator devices, silicon-based sensors and actuators, and optical devices7. The work described here takes wafer direct bonding in a different direction by using it to achieve well-defined uniformly-spaced enclosures over the entire wafer area8,9. Having a well-defined geometry where fluid can be introduced allows measurements to be performed in order to determine the effect of the confinement on the properties of the fluid. Hydrodynamic flows can be studied where the small dimension can be controlled from tens of nanometers to several micrometers.
SiO2 can be grown on Si wafers using a wet or dry thermal oxide process in a furnace. The SiO2 can then be patterned and etched as desired using lithographic techniques. Patterns which have been used in our work include a pattern of widely spaced support posts which results upon bonding in a planar or film geometry (see Figure 1). We have also patterned channels for one-dimensional characteristics, and arrays of boxes, either of (1 µm)3 or (2 µm)3 dimension1 (see Figure 2). When designing a confinement with boxes, typically 10-60 million on a wafer, there needs to be a way to fill all of the individual boxes. A separate patterning of the top wafer with a design that stands off the two wafers by 30 nm or more allows for this. Or, equivalently, shallow channels can be designed on the top wafer so that all the boxes are linked. The thickness of the oxide grown on the top wafer is different from that on the bottom wafer. This adds another degree of flexibility and complexity to the design. Being able to pattern both wafers allows for a larger variety of confinement geometries to be realized.
The size of the geometric features in these bonded wafers, or cells, can vary. Cells with planar films as small as 30 nm have been made successfully10,11. At thicknesses below this, overbonding can take place whereby the wafers bend around the support posts thus "sealing" the cell. Recently, a series of measurements on liquid 4He have been performed with an array of (2 µm)3 boxes with varying separation distance between them10,12. Features much larger in depth than 2 µm are not very practical due to the increasing length of time required to grow the oxide. However, measurements have been made with an oxide as thick as 3.9 µm9. The limits on the smallness of the lateral dimension arise from the limits of the lithography capabilities. The limit for the largeness of the lateral dimension is determined by the size of the wafer. We have successfully created planar cells where the lateral dimension spanned almost the entire wafer diameter, but one could just as easily imagine patterning several smaller structures on the order of tens of nanometers in width. However such structures would require e-beam lithography. We have not done this at this time.
In all of our work the bonded wafers formed a vacuum tight enclosure. This is achieved by retaining in the patterned oxide a solid ring of SiO2 of 3-4 mm in width at the perimeter of the wafer, see Figure 1. This, upon bonding, forms a tight seal. This design could be easily modified if one were interested in hydrodynamic studies which require an input and an output.
The bursting pressure of the bonded cells has also been tested. We found that with 375 µm thick wafers, pressure up to approximately nine atmospheres could be applied. However, we have not studied how this could be improved by bonding over larger oxide areas or, perhaps, for thicker wafers.
The procedure for interfacing the silicon cells to a filling line and the techniques for measuring the properties of the confined helium at low temperature is given in Mehta et al.2 and Gasparini et al.13 We note that changes in linear dimension for silicon are only 0.02% upon cooling the cells14. This is negligible for the patterns formed at RT.
1. Before Bonding, Wafer Preparation
This step except for 1.8 is done in the Cornell Nanoscale Facility cleanroom.
2. Bonding Preparation
3. Wafer Bonding
Properly bonded wafers will have no unbonded regions. Attempting to split the wafers after annealing will cause the cell to break into pieces due to the strength of the bond. Infrared images of properly bonded wafer are shown in Figures 5 and 6. Often annealing improves the uniformity of the cell, especially if local unbonded regions are due to lack of flatness in the wafers. In Figure 5 the light spots and border are bonded areas. The center bright spot is the hole for filling the cell. In the dark areas the wafer are at a 0.321 µm separation. The only unbonded region in Figure 5 is near the border on the top left side of the image. Since it is located beyond the edge of the oxide border, and thus could not be filled with liquid, this would not affect the use of this cell.
There are multiple symptoms of poor bonding which can manifest, however the most common is having a trapped particle between the wafers. This will cause localized lack of bonding to occur and is visible via the appearance of interference Newton rings in the infrared image, as in Figure 4A. This cell has a broad oxide ring on the outside and within this region we can see several small rings indicating unbonded regions. Also, near the center, where a square pattern of channels are formed (not visible), there is a pattern of several Newton rings. These cells would not be suitable for use. In Figure 4B we have attempted to close the unbonded region by applying pressure locally. This is partly effective, and there are fewer rings, but still there remain small inhomogeneities. These wafers were then split and the bonding process was restarted.
Another possible poor-bonding scenario is overbonding. This occurs when there are not enough support posts between the wafers to maintain uniform separation, or the posts are not large enough, thus causing the cell to collapse,16 i.e. bonding directly silicon to silicon. Bowing of the wafers occurs between posts to the point where there is no longer any gap between the wafers. This is not easily observed via the infrared imaging and is generally only discovered when the cell is unable to be filled. Overbonding is a significant concern mainly when dealing with very small gaps (tens of nanometers) where the van der Waals forces are greatest.
A third potential problem with bonding wafers is that sometimes wafers, no matter how clean, simply are not flat enough to bond. Although rare, because of the exceptionally flat wafers used, sometimes poor bonding between wafers will persist. The bonding process involves two wafers overcoming their free-state flatness and contouring to each other at a uniform separation. This necessitates a substantial stress on both wafers and may lead to lack of bonding because of excess stress. The thicker the wafer, the more difficult bonding is since the wafers lose flexibility6. When persistent lack of bonding occurs, one should use a new wafer and attempt bonding again. If bonding again is poor in the same general locations of the wafer, the reused wafer is not flat enough for bonding and must be replaced.
To achieve uniform cell structures the wafers are studied at RT both before and after bonding. Before bonding, the thickness of the oxide grown on the silicon before patterning is measured using ellipsometry. After patterning, an atomic force microscope can be used to confirm dimensions. More complicated or smaller patterns require using an electron microscope to analyze the pattern. After bonding the wafers at a desired separation, Fabry-Perot interferometry can be used to determine the local separation of the bonded structure. With multiple measurements along the face of the bonded wafers, the separation between them can be mapped as shown in Figure 7. The Fabry-Perot method uses the interference of transmitted light as it is multiply reflected by the parallel surfaces in the cell. However, this can only be used if the spacing is greater than half the cutoff absorption wavelength for Si. Thus, the lower limit for verifying bonding with Fabry-Perot interferometry is about 0.57 µm9. These methods, combined with the infrared imaging of the cell, confirm the uniformity of the cell structure.
Figure 1. Schematic drawing of two wafers ready to be bonded together (upper). The blue represents the Si while red represents SiO2. The left wafer has been patterned lithographically with support posts. The right wafer has not been patterned in this example, although often it will be patterned. Combining the two wafers as indicated creates a planar geometry of uniform separation interrupted by the support posts. The wafers are bonded together at RT (lower). This bond is weak, and the wafers will need to be annealed to strengthen the bond. Click here to view larger image.
Figure 2. A cross-sectional drawing of two patterned wafers bonded together. The bottom wafer has boxes which have been etched in the oxide using ion beam lithography (these are the dark purple squares). The top wafer has support posts, shown by the red squares, which keep the top waver 33 nm above the bottom wafer. These features are not to scale in this drawing. Click here to view larger image.
Figure 3. Schematic diagram of the RT rinsing and drying process in the micro-clean chamber. A) shows the two wafers. B) the wafers have been placed on the spinner and are separated a distance of approximately 1 mm by three spacer tabs. A jet of deionized water is sprayed between the wafers as they spin slowly. C) the wafers have been covered and are spun at 3,000 rpm to dry them under an infrared heat lamp. After this process, the separating tabs are moved out of the way by rotating the cover before exposure to the laboratory environment. Click here to view larger image.
Figure 4. A) Infrared images of a cell after initial RT bonding. There are some clearly unbonded areas (light rings) in the border which are not large enough to compromise the use of the cell. However, near the center the multiple rings indicate that there is an unbonded area where the separation is ~3 µm. B) After attempting to force bonding in this region by applying a pressure locally, it is clear that there is a particle trapped between the wafers near the center causing the lack of bonding. These wafers will have to be split and the process restarted. Note that throughout the images there is a faint waviness seen most clearly along the bonded dark broad border. This is due to the variations of thickness of the silicon wafers themselves and not their separation. Click here to view larger image.
Figure 5. A close up infrared image of a section of a cell. Because of the thickness of the oxide grown for this cell, 0.321 µm, the support posts can clearly be seen in this image as the regular light spots throughout the cell. The bright spot in the center is the filling hole. A slight lack of bonding can be seen at the edges of the image on the left side. Click here to view larger image.
Figure 6. Infrared image of a cell immediately before (A) and after (B) annealing. There are two places where there is a lack of bonding, as evidenced by the light rings. Annealing caused the location and size of the unbonded areas to change. The "squarish" patter covering most of the wafer is the active area for experimental use. This is completely uniform. The dark area around the bright center hole is likely a chemical reaction due to backstreaming from the mechanical pump. Click here to view larger image.
Figure 7. Typical uniformity of spacing for well bonded wafers. This plot was obtained using Fabry-Perot interferometry in a series of measurements over an area ~20 mm x 20 mm on the bonded wafers. The cell was designed for a separation of 0.989 µm. As measured, the bonded wafer agrees well with this to better than one percent. Click here to view larger image.
Figure 8. A cross-sectional drawing of wafers patterned with a Corbino17 ring geometry. Two regions are isolated from each other by a ring. A thin film of 30 nm will be formed on top of this ring by the pattern on the top wafer. The resulting geometry will have two relatively large chambers separated by a nanofilm. Click here to view larger image.
The development of suitable silicon lithography in combination with direct wafer bonding has allowed us to make vacuum tight enclosures with highly uniform small dimensions over all the full area of a 5 cm diameter silicon wafer. These enclosures have allowed us to study the behavior of liquid 4He in the neighborhood of its phase transitions from a normal liquid to a superfluid. These studies have verified predictions of finite-size scaling, as well as pointed out failures which remain to be explored. The work has also identified, for the first time, a very strong coupling that exists between two regions of liquid when separated by a very thin, ~30 nm film. Studies along these lines are continuing with cells designed in the Corbino geometry, as shown in Figure 8. This geometry has two region isolated from each other by a ring and connected only by a film 30 nm thick.
Our method of cell construction is limited because SiO2 thickness much greater than 2 µm is difficult to achieve. This is because of the long furnace growth time. In the other limit, large planar structures with separation smaller than ~30 nm are difficult to achieve while avoiding overbonding. Overbonding happens when the two wafers bend over the support posts and touch. One way to avoid this is to use thicker wafers and/or space the support posts closer together. We have not explored all of these variables fully. A thicker wafer in particular might prevent overbonding, however it may also be too stiff and not bond to give a uniform separation. We have achieved separation as small as 10 nm in a structure where studies were made in channel of widths ranging from 2-20 µm18. In this limit one has to worry about the short range variations in the surface of the silicon which can be mapped out with an Atomic Force Microscope18.
There are other bonding methods which can be considered. Electrostatic bonding can be used for glass bonding to silicon. This process is more suitable for bonding over a small area since one initiates bonding with an electrode at high voltage and the bonding wave starts wherever the surfaces are closest together. Thus the bonding wave is not symmetric over the surface of the wafers. Another bonding technique with which we experimented had a similar problem. In our earlier bonding procedures we initiated bonding by using tweezers to pick up the wafers from the microclean chamber. This was not satisfactory. Thus, as described, we went to the use of a holder and the initiation of bonding using a ball press. This step could also be improved since we have not explored the parameters for optimum ball stiffness and press arrangement.
Overall successful bonding of silicon must start with exceptionally flat wafers. Ours are specified to be flat with 1 µm over the full 5 cm size. Since we have spaced two wafers as close as 30 nm, one can see that there must be substantial deformation of the wafers as they bend to achieve this separation. This suggests that wafers cannot be too thick. We have not explored variations in wafer thickness since we have been successful with 375 µm.
Small cavities can also be achieved using a process of anodic bonding, using either glass on glass19 or glass on silicon20. These techniques have yielded planar cavities in the 30 nm to 11 µm range. These structures have a smaller cross section than the cells we make by more than an order of magnitude, 0.2-0.7 cm2 vs 12 cm2 for our cells. They can also be made without support posts because much thicker glass and silicon are used. Thus, while their techniques represent another viable way of achieving micro- to nanofluidic chambers, it would seem to us that direct wafer bonding with the possibility of patterning both wafers is a more variable technique which has allowed the formation of two dimensional, one dimensional, and zero dimensional structures. The cells from Dimov et al.19 and Duh et al.20 would not be suitable for our own measurements.
The authors have nothing to disclose.
This work was funded by NSF grants DMR-0605716 and DMR-1101189. Also, the Cornell NanoScale Science and Technology Center was used to grow and pattern the oxides. We thank them for their assistance. One of us FMG is grateful for the support of the Moti Lal Rustgi Professorship.
SmartCut | North American Tool | FL 130 | Not much is needed per cell. Smaller sizes are available. |
Silicon Wafers | Semiconductor Processing Co | There are many suppliers. Pay attention to thickness and thickness variation when ordering. | |
Deionized Water | General Availability | ||
Peroxide | General Availability | ||
Hydrochloric Acid | General Availability | ||
Ammonium Hydroxide | General Availability | ||
Nitrogen Gas | General Availability | ||
Helium Gas | General Availability | ||
Diamond Paste | Beuler Metadi II | e.g. 406533032 | |
Diamond Drills | Starlite | e.g. 115010 | |
Pyrex Dishes | General Availability | ||
Filter Paper | Whatman | 1001-110 | |
Acetone | General Availability | ||
Methanol | General Availability | ||
Quartz tubes for flushing furnace | General Availability | ||
Rubber vacuum hose | General Availability |