CMP Becomes Gentler, More Efficient

The aim of chemical mechanical planarization (CMP) has always been global planarity. Shrinks, the coming of multi-metal architectures and techniques such as copper dual damascene are challenging this aim, forcing CMP technology — including platforms, chemistries, pads and slurries — to rapidly evolve and adapt (Fig. 1).

"CMP is an enabling technology as crucial as metal deposition or lithography," said Chris Smith, general manager of the CMP Product Business Group at Applied Materials (Santa Clara, Calif.). "We're no longer a niche application. Advanced capabilities are impossible without CMP. There's no copper interconnect without CMP — whether single- or dual-damascene."

"Clearly, copper is emerging — it's in worldwide production today," said Maria Peterson, global technical marketing director for Cabot Microelectronics (Aurora, Ill.). "But it isn't just copper anymore. Some are considering copper alloys to improve conductivity, using materials like indium and magnesium. Alternate plating technologies are being developed, using unconventional means such as electropolishing or conventional plating using standard polishing chemistries, but getting more planar surfaces. Electropolishing and standard plating are moving along those roads."

"There's no ideal slurry," said Mike Hoffman, business development manager for the Electronic Chemicals Division of Ashland Specialty Chemical (Dublin, Ohio). "Current products aren't meeting the needs for metal thinning, planarization and defectivity. The CMP engineer is not only challenged with these performance issues, but must also deal with stability and shelf life constraints, as well as the lot-to-lot variability of some slurry products. The goal remains to stop the first-step polish on the barrier layer and then remove the barrier in the second-step polish without inducing dishing, erosion and non-uniformity."

"CMP tools are achieving unparalleled technological refinements. Multi-zone carriers and in situ zone processing are just two examples," said Thomas Laursen, technical manager for copper at SpeedFam-IPEC (Chandler, Ariz.). "We're improving planarity metrics, and further improvements will be attained by using monitoring/endpoint systems to control carrier zone pressures in closed-loop process control, leading to total-wafer metrics control and increased productivity."

	1. The introduction of copper and low-k materials are forcing CMP parameters to radically change. Smaller process windows are leading to less-abrasive chemistries, new chemistries and pad developments. (Source: Applied Materials)

"Regardless of the polishing concept (rotary, orbital or linear), the reduction of the edge exclusion below 3 mm and minimizing copper and oxide loss in a double-damascene process is a challenge for CMP equipment suppliers," said Frank Gitmans, managing director at Peter Wolters CMP (Rendsburg, Germany). "The new materials' compatibility — especially low-k films — with the CMP process is questionable. These have problems with extensive erosion, defectivity and delamination. Tools will need higher levels of integration for advanced process control and CIM. The process itself will move more towards the chemical part, due to the shrinking geometry and more stringent specifications regarding defects and particles."

At present, 0.13 µm is state-of-the-art production and 0.10 µm is in the development phase. "The industry slowdown has accelerated 0.10 µm development activities, with companies currently concentrating on R&D activities to pull in the advantages of copper/low-k," said Applied's Smith. "Cycle times have dropped from 18 to 12 months or less." He added that at 0.10 µm everything — this means DSPs, microprocessors, logic — will use copper low-k material. "How to get there differs from company to company. The key question is which low-k they'll use and this, in turn, becomes CMP's challenge."

Low-k film integration manifests itself at the CMP step. "Engineers say, 'Here's a dielectric with an effective k of 2.5 or less,'" Smith said. "They put down the barrier seed layer, and it's uniform. Next, they get a nice plating step. But problems are revealed in polishing. It cannot be polished the usual way because the low-k dielectric material is more delicate and softer."

Applied sees copper low-k material as CMP's main driver, with users requiring more in situ metrology. "For years, we've used metrology for dielectrics, but now the industry needs it with copper," he added. Metrology is simple; the hurdles come with closed-loop control within the system, whether working with copper, oxide or tungsten, or doing residue detection, etc.

CMP's major technology challenge is copper low-k integration. STI and PMD are also technology drivers themselves because of their planarity requirements. Tight requirements for them are appearing both in DRAM and logic. "We can no longer group everything under oxide due to their differences in requirements," Smith said. "ILD is different from STI and PMD."

The low-k question is settling down. Two general low-k material families are emerging. One is OSG-Coral and Black Diamond. Manufacturers are qualifying their 0.13 µm processes around these low-k films replacing SiOF, the fluoride-doped oxide films, realizing some benefits from lowering the dielectric constant from 3.5 with fluorinated oxides. There are low-k films on the order of 3.0 to 2.8 with capping films. SiLK is another family that is spreading, via IBM and some of its affiliates.

Ultralow-k films with <2.3 dielectric constants should be interesting. Unfortunately, to achieve that value, porosity must be introduced into the film. There are already some materials with 20 and 40% pore volume.

	2. A low down force can eliminate the kind of CMP-induced low-k peeling shown here, at the edge of a patterned wafer. (Source: Lam Research)

Copper demands superior planarization techniques when you have an effective k<2.5 — more of the "chemical" than the "mechanical" becomes necessary. CMP providers must become planarization providers, and it is impossible to merely extend what exists today, Smith said. "In addition, by 2005 50% of all systems will be 300 mm — with integrated polish and clean technology. This means that planarization and cleaning challenges must be solved simultaneously."

Although today slurries are mainly chemical, and pads supply some abrasion, it is still a CMP process, not a purely chemical one. Abrasives are being eliminated from first-step copper polishes. Barrier polishes still require them, but in increasingly smaller amounts. Copper damascene will require abrasives for some time. Not much has changed from a defectivity standpoint; there is still corrosion, scratching potential, topography issues with damascene structures — how much is being dished in the actual copper itself and how much of the oxide field is being eroded along with the copper.

"As low-k films are integrated into the process, one key issue is film adhesion during the polishing steps," said Ashland's Hoffman. Beyond that, the problems compound with the introduction of porous low-k films. Typical down-force pressures of 5 to 6 psi used in copper-oxide processing will crush these structures. CMP must be operated with lower down force, typically <1 psi, which will cause major changes in CMP consumables and processing parameters.

The challenges are not likely to be solved by just CMP. Improvements in deposition are also going to be needed to meet the industry roadmaps. Another hurdle facing the consumable suppliers is device design variability. "Some customers design in sacrificial layers that aid in topography recovery during the second-step polishing — others don't," Hoffman said. "The predicament is to meet criteria for each different design. That means specialized formulations for each company. A starting formulation must be developed that's tunable for each different parameter. This requires a more definitive knowledge of chemistry and particle interactions vs. past CMP approaches that were mostly of the cookbook kind — empirical. Slurries must be customized to meet process needs instead of the other way around."

"With new material layers to be planarized, slurries seem to acquire additional complexities with the use of different solids such as aluminum and ceria," said Adrienne Pierce, director of business development of the chemical management division of BOC Edwards, (Wilmington, Mass.). "Whether it's controlling particle size distribution, or excessive settling associated with particular particles, new CMP processes often have different requirements."

When slurry uses a silica dioxide particle, it remains in suspension longer, providing better concentration consistency. However, these slurries tend to be shear-sensitive, and form gels and agglomerations. These are non-issues with aluminum and ceria, although they are often faster-settling.

Much is involved in delivering slurries in consistent concentrations to the production environment tool. "As we understand the slurry handling trade-offs and process optimization options, there will be more changes in slurry distribution," Pierce said. "The primary problem is that several things occur concurrently within a complex system. Once the process is 'CMP-characterized,' and the slurry chemistry is established, there's an immediate need for handling and delivery equipment. This means slurry handling issues are dealt with after the process is established. While these aren't insurmountable problems, they create delays and frustration for the user who is trying to develop a robust, high-volume production process to schedule."

It can be exasperating for the handling equipment provider as well. "A slurry producer will tell us he's supplying a specific slurry for a new CMP process to a customer," Pierce explained. "Meanwhile, they're making changes and modifications to the composition and we don't have sufficient samples to do testing. When they're generating enough slurry, and the customer wants his delivery equipment, we haven't yet had the time needed to verify performance or make modifications to provide an optimized system. Better collaboration between suppliers and customers would help mitigate this problem, but there are always proprietary issues and, because you can deal with only so much at a given point, it takes a very dedicated effort."

Thus, slurry process characterization work and the OEM's tool design directly affect how slurry is delivered and handled. As Pierce puts it, "If you have a slurry with a time-sensitive chemical like an oxidizer, which rapidly deteriorates, and you have a point-of-use blending system on your CMP tool, the issue of handling that particular chemical and maintaining its concentration in the slurry vanishes. However, you must consider the process, slurry chemistry issues, the CMP equipment, and bulk slurry delivery to achieve the best system possible — all impact CMP at the production level."

The move toward abrasive-free chemistries is founded upon using the pad's abrasive capabilities. A fluid containing complexing and corrosion-inhibiting agents is brought into the wafer. "The current generation of pad materials has lent itself well to demonstrate and validate process capabilities with multi-level structures," Kistler said. "We've partnered with IMEC, looking at porous SiLK, which has a 2.3 k value. We've also looked at XLK, which has a 40% pore volume and a 2.0 k value." He added that Lam is in the final selection process to qualify 0.10 µm and has encountered no fundamental limitations in terms of existing consumable sets on the linear platform.

	3. CMP users are focusing on defectivity at every step of the copper process, not just after the barrier slurry step. Copper surfaces that were acceptable last year (left) are being improved (right) to achieve the ultrasmooth surfaces required for new technology nodes. (Source: Cabot Microelectronics)

"At 0.07 µm, further improvements are required regarding planarization and total metal loss," Kistler said. "We'll have to develop a set of pad materials with better engineered surfaces, in terms of their asperity profile." Primarily, urethanes are being considered — closed-cell as well as open-cell architectures. Closed-cell has been the leading technology for hard-pad choices. It has limitations in terms of slurry transport, while some of the newer materials with a more porous open-cell structure show some advantages in how they wet the pad and facilitate slurry transport to the substrate's surface while accommodating transport of abraded materials from the wafer surface.

Cabot views integration as a major opportunity to design a slurry that addresses this issue. "If I must integrate copper with an organic or inorganic low-k film and some kind of cap, commercializing different slurries to meet all integration requirements is an industry we've identified and that is currently devolving in R&D," said Peterson.

"Currently we have copper R&D groups, one focused on 0.13 µm, another on the 0.10 µm node, and we're on the early phases of 0.07 µm. But, for example, at 0.10 µm, we have customers integrating these different low-k films, with or without caps."

David Watts, director of technology for Ebara Technologies (Sacramento, Calif.), believes CMP has matured. "It must improve productivity and capability. In addition to upgrading capabilities to handle 0.10 µm technology, we must improve the tool's cost of ownership and productivity. Before, the focus had been on improving capability and handling new materials such as copper and low-k. Now things like robust endpoint capability and in situ monitoring capability become critical."

Nothing fundamentally limits CMP in situ process control. It is just a matter of getting it proven in production. "We can monitor in situ thickness across the wafer," Watts said. "With each pass we do a thickness scan, and get a thickness profile while the wafer is processing in copper CMP. The next step is process control that not only manages average thickness, but also provides center-to-edge control across the wafer." This is critical for copper CMP. The ideal situation is to simultaneously clear the copper at all areas in the wafer. This would be the ultimate in process control.

"We're seeing a shift. The first-tier people who were first into tungsten have now come back through and are addressing CoO issues," said Bob Small, CMP technical director at EKC (Hayward, Calif.). "Simultaneously, they're looking at new parameters and new metrics that are now required. One or two years back, you talked about 1000 cumulative dishing oxide erosion, etc. Then it went down to 600, and now we're hearing 300."

Metrology is another challenge. "A year ago, a 6420 for particle analysis was OK," Small said. "Today, if you aren't reporting SP1 data, the process engineer won't be impressed. Obviously ATI equipment for pattern recognition and other metrology techniques are coming. Anybody in this field better prepare to offer good pattern recognition, whatever it might be, because even blanket SP1 data must extend beyond what we've been doing."

A hurdle is that the new metrology, particularly SP1, is so recent that some do not fully understand how to run it or interpret data. Consumables producers at times gnash their teeth over it, because sometimes the data coming back reflects the company's learning curve, and they are chasing ghost problems.

"Last year we were focused on working out copper migration issues for pilot line manufacturing," Kistler said. "These are mostly resolved and we're now in volume production at 0.13 µm, with up to eight levels of copper dual damascene."

A high premium is placed on a CMP system that provides more within-wafer nonuniformity control and flexibility. "Here, a linear polishing system can rapidly tune for an incoming within-wafer nonuniformity profile without compromising planarization," he said. "Particularly in the foundry environment — where several products are being run and different deposition systems are being screened and considered — having a flexible tool affords an advantage."

Endpoint technology is a main concern. In the ASIC environment, manufacturers are having endpointing difficulties with upper-level structures. As they reach metal 4 and above, they are attaining densities ranging from 50% to as much as 70%. "With some single-wavelength endpoint techniques, you get reflections from the underlying levels, so you have multiple slope changes as you process that data," Kistler said. "This makes it difficult to determine where to call endpoint. This can be solved with a broadband endpoint technology that looks at wavelengths from 300 to 700 nm. This provides additional signal-to-noise capabilities that allow a more accurate endpointing on the upper levels."

"We've had good experience with optical broadband endpoint technology," said SpeedFam's Laursen. "It is reliable and the endpoint system monitors spectral shape changes in the 400 to 800 nm range rather than intensity changes. It enables users to call endpoint on all metalization levels, including the upper levels."

The hurdle being dealt with is that consumables are designed for what are essentially new tools, whether from Applied, Ebara, SpeedFam or Lam. "There's a need to work with OEMs to develop tool-specific platforms," said Rodel's Rhoades. "That's a major thrust for us presently — platform-specific solutions."

Like others, Rodel faces uncertainty with copper in terms of future low-k and barrier materials. "We're working now on solutions for the next generation, but we really don't know what the materials will be, particularly with low-k," said Todd Buley, integration engineering for Rodel. "We're planning on materials that'll require low polishing pressure and low friction, but must work on someone else's guesses."

Since low-k selection seems dependent on the manufacturer's back-end integration scheme, there may not be a single industry choice. Instead, there could be a variety of integration schemes, placing even more demand on consumables because of varied requirements.

Another issue with 300 mm and continuing device shrinks is defectivity performance demands. Defectivity constraints are very challenging as technology progresses because materials, processes, etc., that have been satisfactory are suddenly bumping into barriers that did not exist before.

Laursen believes that meeting the planarity metrics for 0.13 and 0.10 µm technologies will be complicated. "However, using abrasive-free slurry with our orbital technology, we've demonstrated the capability to meet those planarity targets. The ITRS roadmap has a copper thinning target — the combination of dishing and erosion, and oxide loss as well. When this is combined with abrasive-free polish, there is little oxide loss. We are getting average planarity levels of 300 Å, which meets the 0.10 µm technology target (Fig. 4). The values are even better for the finer structures. The same is true of electrical measurements, which reflect the copper thinning values. We've also seen a very tight distribution — little variation in copper thinning across the wafer and also in the electrical measurements."

4. Work done by the Damascus Alliance, a collaborative group of equipment vendors aiming to accelerate the transition of customers to copper/low-k technologies, demonstrated a 3× improved capability for relative copper loss over conventional CMP, using abrasive-free slurry. (Source: SpeedFam-IPEC)

There are many good CMP products and equipment on the market, but what the industry and manufacturers want, especially when it comes to 300 mm, is a more cost-effective CMP solution. "They expect state-of-the-art, performance-integrated metrology and cleaning — but at a lower price and, of course, it must be smaller and have a lower CoO compared to what is out there," said Mike Kirkpatrick, vice president, worldwide sales, at Strasbaugh (San Luis Obispo, Calif.).

Kirkpatrick added that tools must not be overdesigned or overengineered. "In the early days, cleaning systems were tacked on to the side of a machine. Then they were integrated, but they were still large, bulky systems, whereas if you used design expertise and thought carefully about footprint, size, complexity and cost, it could all be reduced. Smaller is better, but you also need to make a more robust and reliable tool by simplifying it while building in the components necessary for improving yield."

For the present, we probably will not see revolutions in chemistries or abrasives. We are using pretty much the same inventory of products; however, the required parameters have radically changed over the last two years — not just the ones for material loss, but also those for CoO. So while on the surface everything appears the same, we are still doing tungsten, using familiar chemistries, and doing copper. Underneath this seemingly placid surface, parameters are getting increasingly tighter. A reason for hope is that CMP is becoming more systematic. Many universities are beginning to put together and come out with equations that in a year or so may result in a "unified CMP field theory."

That will give us a wider horizon to work on.