Optimized TSV Filling Processes Reduce Costs
Chemical and equipment vendors are working to reduce the cost of through-silicon via filling to ensure that it is not a cost-prohibitive roadblock to 3-D integration. Multi-step filling, in which each sub-step is optimized, could help to further reduce costs.
Arthur Keigler, Zhen Liu and Johannes Chiu, Nexx Systems, Billerica, Mass. -- Semiconductor International, 5/1/2009
M3-D integration is a rapidly advancing area of packaging that can enable high performance without the extreme the cost of scaling devices to 40 nm and below. Through-silicon vias (TSVs) are an integral element of many emerging 3-D process flows. Image sensors were the first high-volume application to use TSVs and are, at this time, using partial fill, or via lining techniques. 3-D IC stacking, in general, uses fully filled TSVs, which are the subject of this article.
Many cost models indicate that the TSV filling process is one of the most expensive steps in the process flow. Also, TSV filling steps are run conservatively today because there is not a reliable measurement technique for in-process determination of TSV voids, one of the primary sources of yield loss. We explore the differences between TSV and conventional copper damascene via filling, describe the key elements of TSV filling processes, and show results for a novel multi-step TSV filling method that ensures high-yielding bottom-up fill.
TSV vs. copper damascene
Copper damascene (Cu-D) electrodeposition (ECD) is a well known and mature process from which much has been learned for the development of TSV filling processes. Similar to Cu-D applications, TSV filling requires an anomalous bottom-up growth phenomenon to ensure filling of the via before pinch-off occurs at the mouth of the via. Contrary to conventional electroplating, wherein growth occurs faster on protrusions than in holes, for Cu-D and TSV filling it is necessary to inhibit deposition on outer surfaces and accelerate deposition in the vias and trenches. Also, Cu-D and TSV filling equipment, particularly for via-first TSV (prior to BEOL wiring), often includes post-deposition annealing and edge bevel copper removal.
ECD occurs by delivering reactants to the growth surface, first by fluid flow near the surface and, second by diffusion through the fluid boundary layer, followed by competitive adsorption and electrochemical reactions on the growth surface. The reactants for TSV filling are copper ions and several types of organic molecules. The fundamentally different scale of the TSV geometry (Fig. 1) means that reactant diffusion times in Cu-D plating are several orders of magnitude faster than those for TSV plating, which changes the ratio between characteristic diffusion times and surface reaction times. Also, because the diffusion time constant is small for Cu-D (on the order of milliseconds), the difference between diffusion rates of different species has a greater impact in TSV applications than Cu-D processes.
|
| 1. TSV filling times are >10 minutes compared with &1 minute for a copper damascene via. The relative scale of these structures indicates why a simple extension of copper damascene processes does not meet the needs of TSV engineering. |
A broad range of TSV dimensions are in development. In general, the TSV depth is determined by the downstream handling capability of the thinned wafers or die. A deeper TSV increases downstream yield but adds time to the filling process. To preserve die real estate, the TSV diameter is set as small as is feasible to provide an aspect ratio (AR=depth/diameter) that allows for economical application of the barrier/liner/seed layers (etching and oxide deposition processes are not as strongly AR-dependent). In general, there are two sizes of fully filled TSVs: via-first, which are ~20–60 μm deep and are processed at the beginning of the BEOL stage of wafer processing, and via-last and silicon interposer, which are ~60–150 μm deep and are processed after conventional wafer processing is complete. The AR is 3–10 for both types. For via-first TSVs, a blanket deposition is followed by chemical mechanical planarization (CMP) in a damascene process; for via-last, both the damascene process and pattern plating are being pursued to produce pads and lines simultaneously with TSV filling.
Mechanism of TSV filling
The TSV filling process is a combination of chemistry, fluid boundary layer control and the timing of applied current.
Economical filling requires optimizing the degree of conformal and bottom-up filling (Fig. 2 and 3). Although purely bottom-up filling is not yet feasible, it would be ideal from a yield perspective because it eliminates the chance of unwanted void formation, overburden and dimples. From a throughput perspective, purely bottom-up filling is much more expensive, theoretically requiring 2•AR fill time as the purely conformal process (i.e., 20× longer to bottom-up fill a 10:1 TSV than to conformally fill it). Organic additives are used to balance the degree of conformal and bottom-up filling.
|
| 2. Purely conformal filling (left) and bottom-up filling (right) with the same plating duration. Features to control include the overburden thickness, the dimple depth and prevention of a seam void. |
|
| 3. Example of primarily conformal filling (left), bottom-up filled (middle) and halfway through the filling process (right). |
The TSV filling chemistries use combinations of organic additives similar to copper damascene chemistries, including suppressors that inhibit copper deposition at the copper surface, mainly by covering atomic sites on the copper surface (e.g., polyethyleneglycol, PEG); accelerators that counteract the suppression and thereby allow the copper deposition to proceed faster; and levelers and/or brighteners that also inhibit deposition where the surface curvature causes high electric field, such as the protruding corners of rapidly growing grains.
|
| 4. Forced convection brings a combination of suppressor and accelerator organic species to the edge of the fluid boundary layer, from which they diffuse to the growth surface. |
The filling mechanism is depicted in Figure 4. Molecule selection and testing is quite complicated for these chemical systems because of the interactions between the different species, the base acid system and the TSV geometry. Molecule selection and chemistry development is done first by the chemical vendor and is then optimized by the tool and chemical vendor using real TSV structures.
Typically, Cu-D applications are run in a fountain cell type of plating system where the wafers are immersed face down in a vertical flow of fluid and the boundary layer thickness is set by the rotation speed of the wafer relative to the fluid. Fountain cells provide a radially dependent boundary layer with an average thickness of ~50 μm, comparable to the depth of a TSV structure. TSV filling benefits from active boundary layer thinning. For example in a ShearPlate cell configuration, the boundary layer, as measured on the wafer surface, can be thinned to 10 μm. One advantage of boundary layer thinning is that the total diffusion distance within the TSV — the sum of boundary layer on a flat surface plus the TSV depth — is similar to the fountain cell boundary layer for Cu-D applications. Boundary layer thinning can be added to fountain cell configurations and it is usually intrinsic to vertical cell configurations.
The local rate of deposition is determined by a combination of the local surface adsorbed additive concentrations and the local electric field, which can be manipulated by varying the applied current. For example, to some degree it is possible to concentrate accelerator species at the bottom of the via using a high-current step. Unfortunately, because there is no direct means of measuring these rates, especially inside the TSV, the optimization process typically involves many test runs at various current levels vs. time schedules (there are typically 5–10 steps in a fill process), and the local deposition rate is deduced by comparing growth profiles extracted through cross-section and top-down grinding and optical, SEM or FIB analysis.
Adsorbed additive concentrations at different regions in the via are determined by the species concentration in the bath, the fluid boundary layer thickness, the relative diffusion rates, and competitive adsorption kinetics of the different species. As the TSV fills with copper, the effective aspect ratio may change, particularly if the process is run with an initial primarily conformal step (which applies a large volume of copper in the shortest possible time) followed by a bottom-up step (which minimizes the risk of seam void as the via begins to be fully filled). Therefore, for some applications, it is advantageous to optimize the process by moving the wafers through several different plating baths, each with a preferred ratio of additive species, such that the filling is optimal for the via geometry as it exists at that step in the filling process. This is especially important for high-AR TSV (e.g., 10 μm diameter by 100 μm deep). Figure 5 shows bottom-up and conformal filling cross-sections.
|
| 5. Cross-sectional (left) and angled top-down (right) grinding analysis -- percentage adjacent the section view indicates the distance from top of via. The lower half of the via is solidly filled while a large opening remains in the top half of the via, so there is no risk of seam void formation. |
Multi-step TSV filling
Most of the development work on TSV filling centers on optimizing a single chemistry mixture for the complete filling process similar to the methods developed for Cu-D applications. A single chemistry approach is less complex, but also may be less optimal from a total cost perspective.
An example of a multi-step TSV filling process involves:
-
Pre-wet to ensure no air entrapped in the vias.
-
Pre-treatment to prepare the copper surface for additive adsorption.
-
Stage 1 plating with strong conformal behavior to add a large volume of copper in the shortest time.
-
Stage 2 plating with high accelerator concentration for strong bottom-up filling to minimize void risk.
-
Stage 3 plating with leveler to minimize overshoot (mound formation) and to provide a bright copper finish in the case of patterned TSV deposition.
-
Final clean and dry.
By breaking the TSV fill up into optimized sub-steps, the overall process time and, therefore, cost may be reduced. Because the same family of additives is used for the different steps, bath maintenance and control efforts are minimized. Figure 6 shows an example of a tool with 20 wafer plating positions configured for mulit-step TSV filling. The relative number of process modules for each step is adjusted to match the process times for the particular step, such that wafers proceed through the tool in a semi-continuous manner. A large number of simultaneous wafer plating positions is required, either for a single-step or multi-step filling process, because of the relatively long time of the TSV fill process.
|
| 6. An electrodeposition tool with up to 20 wafer plating positions may be configured for either a single-stage TSV fill or a multi-stage TSV filling process. |
Conclusion
Chemical and equipment vendors are pursuing a reduced cost of TSV filling to ensure that the TSV fill is not a cost-prohibitive roadblock to 3-D integration. Knowledge gained from more than 10 years of copper damascene production is being expanded to incorporate the unique aspects of TSVs. Equipment optimized for packaging processes involving TSV includes features like active boundary layer thinning and 10 or more wafer plating positions, and may include subsystems like copper annealing and edge bevel etch. Multi-step filling is being investigated as a means of further cost reduction. The rapid advancement of TSV filling engineering makes it an exciting area of process development.
-
A TSV is effectively a short, considering the large area and high-frequency applications of the capacitor.
skeptic - 5/14/2009 10:27:00 PM CDT
Addressing 0.13µm Yield Management Issues
01/01/2001
VIEW ALL PODCASTS
