Issue



New CMP process holds promise of lower costs, improved efficiencies


11/02/2012







Christopher Eric Bannon, Texas Instruments Inc., Richardson, TX


Implementing third-generation low-abrasive copper slurry


There is a certain amount of inertia involved in making a decision to deploy a new manufacturing process. However, lowering the total cost of ownership and improving the overall efficiency often provide sufficient impetus to take the first step, even when the process is so integral to semiconductor fabrication as is the chemical-mechanical polishing (CMP) of wafers. Of course, following the correct steps to discover, evaluate, qualify, and deploy the optimum materials as well as the most effective process parameters and procedures are essential to successfully migrating to a new CMP process.


Reducing costs, improving processes


The market for third-generation copper slurries has matured to the point where the cost of such slurries has reached a new low. This gets the attention of semiconductor manufacturers, but selecting, qualifying, and deploying a new consumable material is only part of the challenge. Third-generation slurries are typically less abrasive and more selective in the materials they remove than previous formulations. As a result, switching slurries would also require an array of design experiments to discover how the CMP process would be affected by changing a critical component in the process. In addition, the control parameters for the equipment performing the process would have to be modified to achieve an optimum level of production with high yields.


Still, the benefits that evolve from migrating to a new slurry would be far reaching. Specifically, the manufacturer could expect a slurry with the following characteristics: low abrasion, high copper removal rate, highly selective to the oxide, and low dishing and erosion of the wafer. At the same time, the CMP process could have increased pad/puck life, head life and tool throughput, as well as a world-class mean-time-between-failures (MTBF) of approximately 48 hours.


Evaluating slurries


Integral to choosing a new copper slurry would be evaluating its compatibility with the manufacturer's existing machine tooling. A best case scenario would be to leverage existing pads, conditioning pucks, and heads. Moreover, for maximum throughput, wafers would need to be processed through the current processing equipment's onboard scrubber and dry station as quickly as possible.


Surveying commercial slurries on the basis of slurry/solid makeup and type, chemical composition, pH level, oxidizer (H2O2) costs, and compatibility with the current barrier slurry should reduce the number of appropriate candidates to be evaluated on an experimental basis. First, the performance properties of each material would be evaluated and second, experiments would be performed by varying process control parameters to determine the optimum slurry to deploy and how it should be deployed on the machine tooling.


The initial criteria for testing wafer performance and thereby judging each slurry should involve: removal rate, nominal removal profile, removal profile turnability in relation to certain sets of recipe alternatives, and defect rates. Following these types of tests, experiments with process controls and the candidate slurries should compare: carrier speed, table speed, down force exerted on the wafer, carrier positioning, carrier oscillation, and slurry flow. Candidate slurries advancing beyond this point could also be evaluated on the basis of patterned wafer tests to discover removal rate behaviors, the ability of each slurry to detect endpoints, and the over-polishing window. In addition, slurry recipe parameters could be manipulated to eliminate residual copper, especially at the wafer's edge, and to achieve uniform and reasonable dishing and erosion performance. Lastly, slurry vendors should be asked to perform and provide the results from lifetime experiments with consumables.


Second- vs. third-generation slurries


Throughout this type of process, manufacturers will become aware of the differences in second- and third-generation slurries.


Second-generation slurries are high in abrasive solid content and have chemical additives to help stop the removal process as it reaches the tantalum nitride (TaN) layer. Overall, the results of such a slurry would be limited to a very mechanical process, one that would rely heavily on F.W. Preston's venerable equation that defined how quickly material would be removed from a surface as a result of mechanical polishing. Although experiments performed by TI have shown a second-generation slurry likely would be selective enough with regards to the materials removed, the tool's pad would probably degrade after approximately 400 wafers. This could cause increased dishing and erosion as a result of temperature spikes and reduced consumable performance.


In contrast, third-generation slurries are low in abrasives and have a different chemical composition than second-generation slurries. Third-generation slurries could be just as selective as second generation slurries, meaning they would also remove materials to the TaN layer, but they typically produce less dishing and erosion, and achieve higher copper removal rates at lower process parameters.





Figure 1. Results of removal rate tests on a third-generation slurry.

Figure 1. Results of removal rate tests on a third-generation slurry.



Experiments performed by Texas Instruments (TI) with one third-generation slurry compared to a currently deployed first-generation slurry showed that there could be a problem with premature balding of the machine's pad (Fig. 1). The removal rate for the polishing process decreased as wafers were processed and as the grooves on the machine's pad became shallower due to excessive conditioning required because of the third-generation slurry's nonabrasive nature.





Figure 2. A third-generation slurry's polish time as a function of short-loop dishing.

Figure 2. A third-generation slurry's polish time as a function of short-loop dishing.



Further experimentation revealed that this could be overcome by reducing the conditioner's down-force and time. Since the third-generation slurry was low-abrasive, it would not wear down the conditioner diamonds, which caused the degrading of the cut rate of the pad. Figure 2 shows the performance of a third-generation slurry with regard to the removal of dishing over several polishing times and compares this to the same results of a first-generation slurry. This clearly demonstrates the superior performance of the third-generation slurry.


Process development challenges


Once the benefits of a new slurry have been identified, the actual processes in the manufacturing line where the slurry will be deployed must be re-developed to take into account the new material. With a CMP process, these challenges often involve identifying and correcting any defects the new process may introduce. Issues such as photo-induced corrosion can usually be resolved quickly, while others like residual copper and micro-scratching, may require adjustments in the process.


Residual copper can have a devastating effect on yields, but polish and plate engineers can usually produce custom film depositions based on the requirements of the process. In addition, wafer-level inspections on every lot processed will also help. Micro-scratching is particularly problematic with first-generation slurries which incorporate fumed silica. These processes typically lack a final table conditioner and, as a result, the machine's soft pad will become embedded with slurry until it effectively becomes a fine-grain sandpaper. Nonabrasive, third-generation slurries which employ colloidal silica avoid this problem entirely.


Replacement costs


Even though experiments demonstrated the benefits of a third-generation slurry, these advantages must outweigh the cost of replacing a process that is currently in place. This involves benchmarking consumable costs and the wafer yield rate with a new CMP process versus the old CMP process. The benchmark results performed at TI were quite positive in favor of a third-generation slurry. The cost of consumables, such as pads, conditioners, the slurry itself and other materials, revealed a 17 percent cost savings with the new process over the old (Fig. 3.) The useful life of pads and conditioner was 50 percent greater with the third-generation process because of low abrasiveness and lower conditioning down-force. Moreover, the life cycle of equipment heads nearly doubled, increasing by 98 percent. This was caused by the lower process parameters required by the third-generation slurry, which reduced wear on the retaining ring. Longer head life cycles reduced the number of failures resulting from loading and unloading heads, lengthening the tool's MTBF to a range of 30 to 40 hours. In addition, wafer yield comparisons showed that the new process achieved slightly better yields than the old process.





Figure 3. Consumables cost savings of third-generation slurry CMP process over first-generation slurry CMP process.

Figure 3. Consumables cost savings of third-generation slurry CMP process over first-generation slurry CMP process.



Figure 4 benchmarks the availability performance of a new third-generation slurry CMP process versus an older first-generation process. The point of switch-over from the old process to the new is indicated by a vertical green line in each of the four plots. In every case, the performance of the new process significantly surpassed that of the old.





Figure 4. Sample copper tool availability with the new third-generation slurry process.

Figure 4. Sample copper tool availability with the new third-generation slurry process.



Conclusion


Qualifying, developing, and deploying a new copper polishing process with a third-generation low-abrasive copper slurry presents certain challenges, but the opportunity to lower costs and achieve an equivalent or better wafer yield outweighs the challenges. In the end, leveraging the manufacturer's engineering, maintenance, and operational talent ,as well as that of supply chain vendors will be essential for a successful transition to a new process.


Christopher Eric Brannon (cbrannon@ti.com) is a member of technical staff at Texas Instruments (www.ti.com) and currently holds a position at TI's DMOS5 factory as a copper CMP manufacturing engineer.


Solid State Technology | Volume 55 | Issue 9 | November 2012