Controlling Plating Baths in TSV Applications

When combined with non-reagent techniques, CVS analysis can ensure appropriate concentrations of inorganic components in electroplating baths, a necessary step in bringing through-silicon via (TSV) processes to production-worthy status.

Michael Pavlov, Eugene Shalyt and Peter Bratin, ECI Technology, Totowa, N.J., www.ecitechnology.com -- Semiconductor International, 3/1/2008

Consumer electronics market demand requires semiconductor manufacturers to reduce assembly size, cost and weight while increasing functionality and performance. This challenge is addressed by the search for alternative packaging methods. One of the most common approaches is the use of 3-D packages. Copper electrochemical deposition (ECD) processes are already being used for the fabrication of cost-efficient wafer-level packages (WLPs), such as bumping (studs and pillars), redistribution layers (RDLs) and some emerging through-silicon vias (TSVs) and embedded passive components (inductors).¹

While bump and RDL plating baths have already been implemented, TSV processes are still in development.^2,3 One step in TSV processing involves filling deep, blind vias completely with conductive material. The filling must be void-free to prevent reliability problems in the electrical connection. Most TSV applications use copper fill because it exhibits the second-lowest resistivity among conductive materials, as it can fill void-free vias with a 10:1 aspect ratio. Also, stable and cost-efficient ECD (plating, filling) are already used in production for advanced dual damascene interconnects and RDLs.

Part of the success of a TSV copper ECD process depends on understanding and controlling the changes in the filling bath composition, because the TSV filling process is different than the conventional damascene plating process — with significantly larger dimensions and longer process times.

Detailed analyses of copper bath compositions have been presented for various front- and back-end electroplating processes, including copper, tin, tin-lead and tin-silver baths.^4,5 The analysis method provides data about the concentration of all components in the plating solution and, in some cases, the accumulated byproducts.^6,7 This article describes new analytical techniques developed for the analysis of TSV copper-filling solutions.

TSV technology

TSV technology provides vertical connections to the 3-D package with a connection length that is close to the thickness of the chip itself. It is possible to achieve high-density and high-aspect-ratio connections that allow for the assembly of advanced multi-chip modules (MCMs).⁸

The TSV development process includes the following steps:

1. Via formation
2. Insulator, barrier and seed layer deposition
3. Copper filling
4. Removal of metal excess by chemical mechanical polishing (CMP)
5. Wafer thinning
6. Wafer bonding⁹

Each step presents challenges. During via formation, it is important to maintain shape and angle control. Via etching is performed using the Bosch process or, alternatively, laser drilling. In the next step, when the insulator and seed layer are deposited, conformity and adhesion are the main concerns. Copper filling must be performed so that the final copper plug is void-free to allow normal electrical performance of stacked device at elevated temperatures. Once the copper filling has finished, the wafer needs to be thinned with the smallest possible thickness deviation. The bonding of wafers is the final step in the stacking process.

TSV copper plating bath

Void-free, bottom-up TSV filling is critical across the wafer surface as well as the required mechanical properties of the deposit. As in copper damascene processes, a multi-component plating solution containing both organic and inorganic components is used. Copper electroplating baths are normally formulated using highly stable electrolytes containing copper sulfate and sulfuric acid. Copper concentration in these electrolytes is between 40 to 60 g/L with 5-40 g/L sulfuric acid. Other components, present in relatively small amounts include organic additives and chloride ions. The organic additives, depending on the concentration and chemical composition, affect properties of the electrodeposited copper, such as uniformity, hardness, ductility, tensile strength, etc.

The use of iron (Fe) in plating baths for TSV copper filling has recently been announced.² Such a bath has a possible advantage because a lower concentration of byproducts is generated during contact of organic additives with the anode. In this bath, the oxidizing component is iron II (Fe⁺²), which is converted to iron III (Fe⁺³). The concentration of Fe⁺³ must be monitored in solution to not exceed a certain threshold.

The types of organic additives added to copper electroplating baths typically include suppressors (polymers such as polyethylene glycols), accelerators (sulfur-contained compounds) and levelers (secondary suppressors). Organic and inorganic components are consumed at various rates during the electroplating process, thus requiring individual control. Accelerators, for example, are consumed faster than suppressors and levelers. Consumption of organic additives depends on various factors, including the condition of the electroplating cell, current density, flow rate, number of wafers plated, etc. In the TSV bath, the concentration of metal can drop during the electroplating process, and therefore needs to be tightly controlled.

To keep concentrations of plating bath constituents in the optimum range for the electroplating process, several approaches can be used. The conventional approach employs automatic dosing of individual components into the plating solution(s). A dosing system incorporates hardware for the delivery of individual components into the solution and, if needed, removal of spent solution from the bath. Replenishment of individual components is controlled via predictive software algorithms, usually in conjunction with periodic analysis of solution composition.

In high-volume production, plating solutions can be controlled by a continuous feed-and-bleed of virgin makeup solution, combined with automatic dosing of organic additives based on predictive algorithms and frequent analysis. This approach enables tight control of the plating solution and limits buildup of additive byproducts formed during plating. The advantage of this method, if properly performed, is that the solution is maintained at a virtual steady-state condition and can be used for a long period of time

Despite the sophistication of dosing algorithms, the composition of plating baths can change rapidly and drift out of the operational range. Such changes occur because of intermittent or unpredictable operation of the bath, or from a malfunction in the delivery of its components, causing over- or underdosing of the plating bath. These problems can be prevented and avoided if each component of the plating bath is individually monitored with an online analytical system.

The alternative approach of keeping the bath in control is based on single-use plating solutions that get dumped once a certain number of wafers passes through the plating bath. The difficulty in this approach comes from an inconsistent composition of the solutions, which must frequently be adjusted to ensure appropriate plating properties.

CVS analysis

Among numerous analytical approaches used for monitoring concentrations of organic additives in electroplating solutions, only cyclic voltammetric stripping (CVS) has demonstrated the ability to reliably monitor the activity of a wide range of components in a plating bath. In this method, the classic three-electrode cell is used where the main indicator is a platinum rotating disc electrode.² CVS involves cycling the potential of a platinum rotating disk electrode so that metal is alternately plated and stripped at the electrode surface. Organic additives are detected by the effect that they exert on the electrodeposition rate measured via the metal stripping peak area (A_r). This approach has become the most popular method for controlling damascene copper plating processes.

In the TSV process, the bath has a similar composition and, therefore, it is natural to use CVS analysis. Because of the difference in the type of organic additives, the CVS applications need to be reevaluated and tested using various solutions, including real plating solutions with different ages.

Analysis of a TSV bath

When new applications are developed for analytical control of plating solutions, there are several major requirements to consider. The performance of the analytical method should be suitable, providing the appropriate accuracy and reproducibility. It is important to evaluate the produced statistical results against the process tolerance window. If the analytical results are noisy and standard deviation is higher than 5% of the tolerance range, then the analytical process needs to be improved (Fig. 1). We used this approach to estimate the performance of analytical procedures developed for the main components in commercial TSV copper plating baths (Table 1). During this verification, each solution was analyzed 20 times for each organic additive.

1. Example of acceptable and unacceptable deviation around a process tolerance window.

The analytical method has to be selective, insensitive to the changes in the concentration of other components in the bath, and insensitive to the presence of byproduct buildup. For a TSV bath, this becomes more critical because of the longer plating times. We performed extensive testing of analytical procedures to estimate any interference between bath components and possible byproducts. Figure 2 summarizes the results of the accelerator analysis. Plating solutions prepared for this test have been composed differently and include low and high concentrations of all tested components, as well as samples that are outside of the process tolerance range of concentrations. From the results obtained, it can be concluded that accelerator concentration can be determined over a wide range, and it is not sensitive to changes in concentration of other organic additives and their byproducts.

2. Plating tests with high and low concentrations of components were performed to ensure that the measured concentrations were not influenced by the concentration of other additives or byproducts.

As noted, the concentration of copper can decrease during the TSV plating process, so we investigated copper depletion. When the concentration of copper dropped by 25%, the measured accelerator concentration was virtually unchanged (Table 2). An increase in copper concentration is unlikely, but possible because of evaporation of the electrolyte from the bath. This was also studied and reported in Table 2. As with lower concentrations, higher copper concentration did not affect the accelerator analysis. We applied the same testing to two other organic components. Copper concentration changes had no effect on their analyses.

The changes in concentration of inorganic components (chloride and acid) were not expected to be significant, but their effect was verified. It was shown that a 25% deviation in concentration of either or both inorganic components has no effect on the analysis of organic additives.

Once the procedures have been developed and tested on standard solutions with various concentrations of bath components, each procedure was tested using real plating solutions. Three types of solutions were collected: a freshly mixed bath, a mid-life bath that was used for plating, and a bath before disposal (end of life). The baths had been maintained using the best replenishment algorithms. However, the end-of-life bath was not dosed and sampled right before disposal.

The freshly measured bath was considered the baseline for the following tests. All concentrations were reported as percentages from the originally mixed concentrations (Fig. 3). The suppressor component did not change its concentration throughout the use of the bath from beginning to end. This indicated that the replenishment functions were set correctly. The concentrations of accelerator were less stable. From this we learned that the dosing algorithm needed to be tuned. The same was true for the leveler component. The adjustment in the replenishment of leveler required less attention than the accelerator, but still needed improvement.

3. The dosing algorithm for the accelerator and leveler needs to be tuned to adjust for changes with use of the baths.

Additional tests were run on the end-of-life bath. This solution has a higher chance of being loaded with byproducts left after electroplating. Considering this, we dosed the aged solution with each component and then analyzed it (Fig. 4). The analyzed concentration for each component correlated well with expected values (within 5%).

4. The end-of-life bath was spiked with each component and analyzed. Each measured value correlated well with the expected concentration.

We also diluted the same plating bath by 50% with an inorganic matrix and retested it (Fig. 5). The results were accurate for each component. The accuracy was difficult to estimate for the accelerator because of the low concentration in the solution before dilution, but an approximate estimate indicated that the accelerator result after dilution of the aged bath is accurate.

5. The end-of-life bath was diluted by 50% with an inorganic matrix and then retested.

The tests described are just a small portion of the verifications we performed during the development of the TSV processes. All results indicated that the CVS analytical process can deliver very accurate, reproducible and interference-free results.^4,5 The CVS technique can also provide information about the accumulation of certain byproducts generated during plating. Figure 6 shows the effect of possible accelerator byproduct (MPS) on the slope of the argon (Ar) signal.

6. Effect of accelerator byproduct on the slope of the argon signal.

Aside from accuracy, reproducibility and other statistical parameters, there are a few additional aspects that need to be considered during the analysis of TSV plating solutions. They include time of analysis and cost, both of replaceable parts and other consumables. A low-cost analysis is achieved when non-reagent techniques are used. For example, the inorganic components can easily be analyzed using a combination of spectroscopic and electrochemical techniques.⁷ These techniques allow for the monitoring of components in plating solutions quickly and without reagents. It is also important that these methods do not require a chemometric approach in resolving the interference between components. The analytical signals should be strong, not requiring significant mathematical transformations.

7. Linear relationship between spectroscopic signal and the acid concentration in the plating solution.

Figure 7 shows a calibration graph for acid measured by a non-reagent technique. The response is linear and dependent only on the concentration of acid in the plating solution. The non-reagent technique has been compared with a standard acid/base titration analysis. Table 2 compares the statistical results obtained using the spectroscopic non-reagent technique to titration results. The performance of the non-reagent technique is similar to titration, which allows for reliable monitoring of acid in the plating solution. The same approach is taken for copper and chloride. These species can be also monitored using non-reagent techniques.

When a CVS technique is used, the consumption of reagents is negligible and no larger than 1 mL per analysis. The reagents used for CVS analysis are the same organic additives used for the electroplating process and, therefore, do not affect the cost of the analysis. However, consumption of the plating solution may be a concern. The plating bath volume may vary depending on plating tool type and process throughput. In this case, minimal volume should be withdrawn from the bath for analysis. This is possible using newly developed sampling algorithms and modified electrochemical cells.

Conclusions

With the TSV process gaining full acceptance in high-volume manufacturing of the next generation of semiconductors, reliable metrology is of utmost importance. CVS has been shown to be a highly accurate and precise method for the control of organic additives used for the copper filling of TSVs. When combined with non-reagent techniques used for the analysis of inorganic components, the metrology tool becomes fast and effective with a low cost of ownership.

References

1. B. Kim, “Back-End Copper Plating for 3D Chip Stacking Applications,” Device Packaging 2007.
2. R. Preisser, “Copper Electroplating Techniques for the Interconnect Formation in Advanced Semiconductor Devices using an Inert Anode/Redox-Mediator Concept,” 211th ECS Meeting, 2007.
3. T. Ritzdorf, “Challenges and Opportunities for Electrochemical Processing in Microelectronics,” 211th ECS Meeting, 2007.
4. M. Pavlov, E. Shalyt and P. Bratin
5. M. Pavlov, E. Shalyt and P. Bratin, “The Advanced Monitoring of Constituents in Copper Damascene Electroplating Baths,”
6. M. Pavlov, P. Bratin, E. Shalyt and M. Tench, “Detection of Accelerator Breakdown Products in Copper Plating Baths,” 204th ECS Mtg., 2003.
7. P. Bratin et al., “Detection of Suppressor Breakdown Contaminants in Copper Plating Baths,” 203rd Mtg., 2003.
8. P. Garrou, “3-D IC Integration: Rumors and Ruminations,” Semiconductor International, Feb. 8, 2008
9. S. Spiesshoefer et al. “ Z-axis Interconnects Using Fine Pitch, Nanoscale Through-Silicon Vias: Process Development,” Electronic Components and Technology Conference, 2004.