A new, low pH, BTA free, noble-bond chemistry produced equivalent yield at substantially lower costs.
BY CHRISTOPHER ERIC BRANNON, Texas Instruments, Dallas, TX
The 2010 economic downturn affected many industries, semiconductor manufacturing notwithstanding. Many fabrication facilities had to layoff employees and curtail spending, all the while managing lower wafer output. This effect caused many semiconductor companies to rethink how they spend on resources. Everything was considered, from the cost of the wafers to the cost of the tool consumables and chemistries.
Texas Instruments (TI) copper chemical-mechanical planarization (Cu CMP) was no different. All spending had to be reduced and copper hillock defect had to
be eliminated. The CMP Team proposed developing a process based on the new third generation clean chemistry on the market for a number of economic and logistical reasons. The first rationale for this strategy was cost and second was time - most of the clean chemis- tries on the market were considerably cheaper than the current process of record (POR). CMP had also seen many defects due to via-to-via shorts caused by Cu hillocking (localized Cu protrusion into the above interlayer dielectric; see FIGURE 1).
There were two key reasons that TI succeeded in developing a Cu cleaning process: detailed engineering work and strong vendor support. Process development went through four generations of refinement before it was ready for high volume manufacturing. The first version focused on new clean chemistry improvements such as third generation low pH, high acid clean chemistry and an array of design of experiments (DOE) continuous improvement through optimization of the process controls and equipment modification followed in the second. The third generation attempted to adapt an existing Mirra-Desica process using a previous qualified process. A final successful attempt was made during the fourth cycle to develop a lower cost, higher throughput multi-copper platen cleaning process using a commercial chemistry from Air Products, COPPEREADY®CP72B. This paper will discuss the work that went into building TI’s successful Cu cleaning CMP process.
TI Cu CMP
Neutral pH clean chemistries using Benzotriazole (BTA) were the first generation application on most Cu CMP dual damascene back end of the line process at TI. This was dependent on using dry-in wet-out Cu CMP AMAT tools with spray acid Vertec hoods for cleaning and drying. It was also very high in cost and low in consumable life compared to most conventional CMP clean process (e.g. Tungsten, STI, Oxide). The TI POR was no different, a first generation Cu clean using three different chemistries, BTA, Electra Clean and ESC774TM. These chemistries were very expensive to use and were not very efficient at cleaning or passivating the polished copper surface. They were able to passivate the copper surface but were prone to leave many types of incompatible carbon residue defects on the wafers. Cu hillocking was very prevalent with this type of cleaning solution and via-to-via shorts in the back end of the line (BEOL) were the top defect pareto for TI.
Clean chemistry identification
To reduce the time to develop a new Cu CMP clean process, most of the development cycle focused on Cu cleans leveraging a Mirra-Desica DIDO Cu polishing process using existing pads, conditioning pucks, and heads. Early on, it was decided that to achieve maximum throughput, the wafers would need to be processed through the tool’s onboard scrubber and dry station as quickly as possible. With time running out, the Cu CMP team had contacted the major players in Cu clean chemistry to obtain their specific information and prepare a white paper screening to determine the correct path. The four candidates were evaluated on chemistry type, makeup, pH, passivation (BTA), cost, and compatibility to our current Cu and barrier slurry. Two of the chemistries fit the bill for the criteria and were selected for further testing. Chemistry 1 was a novel approach for Cu CMP and was from our current clean chemistry vendor, Chemistry 2 was similar to the current TI process of record.
The initial criteria used to judge the chemistries were blanket test wafer performance (Cu, Teos, Ta, and Nitride): etch rate, passivation, cleaning tunability via recipe parameter windowing, and defectivity. Experimental designs were run on the basic process controls with these chemistry’s with respect to the polish process: carrier speed, table speed, down-force, carrier position, carrier oscillation, and chemical flow. Both cleans performed well on the blanket experiments and were advanced to short loop, patterned wafer tests. These patterned wafer tests were used to study product behavior in the polisher and brush cleaner. A significant amount of time was spent adjusting recipe parameters to eliminate defects. The team contacted both vendors to do lifetime experiments with consumables at their facilities. The data that was collected revealed many issues with each candidate, one more so than the other (FIGURE 2).
Chemistry A was a second generation Cu clean that had high pH but had chemical additives that would aid in cleaning, still a very basic approach to wafer cleaning. The overall defectivity was sufficient on the product test wafers but would degrade after a short time window after polish. It also had to be paired with another chemistry to achieve the same Cu passivation as the POR. This chemistry was disqualified due to this reason.
Chemistry B is a third generation Cu clean that had low pH (about ~2.1) and it is BTA-free, unlike any other Cu cleans on the market at that time. This chemistry is an organic acid blend, which helps ionize Cu2O and CuO to form water and soluble Cu complex, used for passivation. This forms a strong bond with the Cu to make the surface nobel. The low pH helps to dissolve the surface defects resulting in a step function decrease in defectivity compared to baseline (see Figure 2). The chemistry was also scalable, depending on concentration making cost of ownership low. This chemistry was selected for qualification at TI Cu CMP.
TI’s internal polishing engineering staff was augmented with exceptional support from several consumable vendors during development. Together TI engineers developed proprietary and patent-pending technologies to enhance the Mirra Desica cleaner performance on Cu BEOL CMP. TI also benefited from strong relationships with its contact clean brush suppliers. Rippy was instrumental in brush evaluations and consul- tation on process developments. To improve the tool’s performance, DOW was pivotal in adding additional functionality to the process through end of life evaluations. Perhaps most important of all relationships that developed was with Air Products, who provided an invaluable education into Cu cleaning process development.
Solving defect issues
During process development, TI engineers encountered several defect related issues. Some issues like photo-induced corrosion were resolved quickly after some technical research. There were two others that took more troubleshooting: carbon residue defects and Cu hillock formation.
The presence of gross surface defects, like carbon residue is an obvious yield killer. The Cu CMP Engineers come to the conclusion through EDX (Energy-dispersive X-ray spectroscopy) and much lab analysis that the current Cu slurry still had traces of BTA in it and were causing this residue defect to form on the wafers after polish. Many DOE later determined that extending the clean chemistry buff polish would eliminate this defect.
With residue defects effectively eliminated, the next major technical challenge was Cu hillock formation. TI had been experiencing higher defectivity due to back end of the line via to via shorts on the previous Cu CMP clean chemistry process. It was understood that the formation of Cu hillocks were the cause for this signature. To solve this problem, a completely different wafer cleaning chemistry was needed to passivate the copper surface. TI Cu CMP Engineers looked for one that did not use BTA or other high pH chemistries, but, would coat the wafer surface and not allow native oxides to grow on the Cu. The new chemistry (CoppeReady®CP72B) proved to form a nobel bond with the Cu (CuO2) and eliminated hillock growth formation, thereby reducing via-to-via shorts (see FIGURE 3).
Further process development
One of the last stages of development on the new process was a project to develop a faster through-put process. Although this work was successful, it highlights some of the challenges in pursuing this type of strategy. The motivation for this work was to dramatically boost the throughput and to further cut process expense. The POR process was limited by the cleaner and was much slower causing higher cost and higher wafer-per-hour rates. To maximize throughput, the new process would have two components: speed up the on board cleaner, brush box 1&2 throughput, as well decrease the platen 2&3 process times but include a clean chemistry buff. Because of the high down forces employed to achieve a flat removal profile, the Cu polishing component of this work, platen 1, was surprisingly fast but was the intended bottle neck. These changes allowed for a 10 percent increase in overall wafer through put compared to the baseline process. This had an alternate effect on the Cu polish process. TI’s current Cu slurry is thermally driven, with making platen 1 the bottle neck it kept that platen at one constant temperature throughout the lot, causing the overall end point times (EPD) to be reduced and streamlined. This further increased the tools throughput by 2 percent and reduced wafer to wafer EPD variation down to 2 to 3 seconds; previous was 10 to 12 sec between wafers (see FIGURE 4).
For initial qualification and benchmarking, TI installed and setup the best known method (BKM) Cu polishing process on an Applied Materials Mirra-Desica. To
bring the new clean process into production, Cu Polish engineers needed to demonstrate equivalent or better yield between the two competing process. The new clean chemistry needed to be tested for EM (electro migration), which is a stress test of Cu interconnects between two metal lines. This test had to be outsourced to a third party company that specializes in oven-baking stress tests (FIGURE 5). After extensive electrical and yield testing, the new clean process was fully released. Sample yield comparisons consistently demonstrated that the performance is equivalent to slightly better and the new process has higher through-put (~12 percent). The chemical costs (dilute 60 to 1 CP72B®) are 68 percent less per wafer pass than the competing process. The pad/ conditioner life had increased by 13 percent from the previous process due to thermal driven Cu slurry through put modification (FIGURES 6 AND 7).
TI engineers developed a Cu CMP cleaning process using new third generation low pH Cu chemistry. Despite the tool’s many limitations, the engineering staff successfully delivered an integrated process capable of producing equivalent yield at substantially lower costs over the best alternative method. There were undoubtedly challenges along the way, only a fraction of which have been described in this paper. By leveraging an existing deep reservoir of engineering, maintenance, and operational talent, an existing and efficient supply chain, and the outstanding support of numerous vendors, TI Polish module was able to realize its goal of making efficient use of its assets to achieve a competitive advantage.
1. Tsung-Kuei Kanga, and Wei-Yang Choub Author. ‘Avoiding Cu Hillocks during the Plasma Process’
Journal of The Electrochemical Society, 151
CHRISTOPHER ERIC BRANNON is a TI Cu CMP Manufacturing Engineering, Texas Instruments, Dallas, TX.