Outsourcing CMP Process Development and Materials Evaluations

The semiconductor industry's recent upturn is driving higher fab utilizations and requiring fabs to maximize the effective use of their resources. This, in turn, drives the need for additional projects to address productivity and cost issues. These types of projects are typically focused on improving process yields, reducing cycle time at a bottleneck operation, reducing consumption of consumables, or evaluating new materials for either improved performance or reduced cost. Unfortunately, the fab's resources are often constrained in upturns because of the increased demand for production time on tools. This limits their ability to execute vital engineering projects just when these projects could have the greatest impact.

One solution to increasing production output is outsourcing to foundries for additional wafer capacity. Commonly referred to as the “fab-lite” or “asset-lite” model, this option allows companies to increase wafer output beyond their internal capacity limits during periods of high demand while eliminating the risks of carrying stranded capital during periods of weak demand. While this option may be well suited for companies needing additional fab-wide capacity, a more flexible and dynamic option is preferred for addressing constraints at specific process modules.

As one of the newest and fastest growing processes in a modern fab, the chemical mechanical planarization (CMP) process module is frequently one of the processes subject to capacity constraints. It is also often targeted for cost reductions and/or productivity improvements. In order to address these issues and minimize the impact on manufacturing production, Freescale Semiconductor elected to outsource certain engineering projects for CMP materials evaluations and process improvements to Entrepix, a dedicated CMP foundry. By doing so, Freescale was able to focus internal resources on immediate production concerns while still addressing the improvement projects in a timely fashion with external resources. This article reviews the initial results of these efforts and describes the synergistic benefits that Freescale was able to achieve by leveraging Entrepix for specialized outsource CMP services.

A novel approach

Ever-advancing technologies, such as Freescale Semiconductor’s embedded microcontrollers, incorporate integration schemes that require increasing utilization of optimized CMP processes for cost and performance, which are developed and outsourced for customers at Entrepix’s dedicated CMP foundry in Tempe, Ariz.

Other factors can impact the division of hours between the fab and foundry on several of the tasks listed in the Table, depending on the level of control desired by the fab. For example, choosing variables and designing the experiment can be done effectively by the service provider's staff with a simple final review by the fab staff. However, this can also be nearly reversed with the entire experiment being planned in detail by the fab and merely reviewed and executed by the service provider's staff. The hours required for conversion of equipment away from and back to standard configuration is also highly variable, but the impact on total in-fab execution time can be dramatic depending on the type of experiment and the level of decontamination or hardware conversion required. Regardless of exact numbers, one of the strong leverage points in this approach is being able to consciously decide which tasks to keep internal and which to outsource.

Process development

Many different types of process development projects are required to sustain and evolve the CMP module in a fab. Some fraction of these projects is viable for outsourcing. The key metrics typically used for CMP evaluations are a combination of the following: removal rate and across-wafer uniformity, planarization, surface roughness, defectivity, residual surface contamination, process repeatability, and device yield. Early-stage screening trials frequently focus on just a few of these critical variables, while later-stage efforts tend to have established targets for all of these metrics (and more). Each fab imposes specific requirements based on their technology, but the general sequence of events for qualifying any new process can be described by the diagram shown in Figure 1 .

1. Qualification cycle from concept through volume ramp.

Depending on the complexity of the change and the urgency of the derived benefit, this qualification and ramp cycle typically takes anywhere from six weeks to one year or more. For fabs that do not currently have CMP in full production, the timeframe is even longer to allow for purchasing equipment, hiring and training technical staff, reconfiguring the facility, and all the other tasks associated with bringing up a new process. By utilizing outsourcing, the overall speed of implementation can be accelerated substantially, especially through the early stages. The associated time savings will yield a faster capture of increased revenue (or decreased costs) for the fab. As expected, the fab controls what fraction of the work is executed outside the fab and the success criteria for moving from phase to phase.

The engineering staff involved with outsourcing efforts has learned that a key element for successful execution of outsourced CMP development is spending a reasonable amount of time on planning. Data requirements, including wafer-level metrics and the equipment that will be used to measure them, must be understood by all participants. In most cases, an exact match in the equipment set is not required, although execution is simpler when it does match. At a minimum, the translation of all relevant data between platforms must be reasonably well characterized and any limitations clearly understood. It is important that the project plan clearly specify all required information to meet expected deliverables, including the film stack details, measurement locations, number of wafers to be measured, charts and graphs to be generated, etc.

Materials evaluations

In the complex world of semiconductor device fabrication, any material that is part of the final device or makes contact with the wafer during processing must be qualified before being released for production. For CMP, this may include incoming film composition, pads, slurries, conditioning disks, chemicals used in post-CMP cleaning, and materials of construction for critical parts of the polishing or cleaning equipment. When evaluating potential alternatives to some of these materials, even a simple engineering trial can be quite disruptive to the manufacturing area. For example, an experiment involving a slurry not currently in production requires a series of activities far beyond just polishing the wafers. There is a mandatory safety review and sign-off procedure just to get the slurry onsite. Multiple management reviews are often needed to authorize the experiment. The manufacturing group has to allocate adequate tool time and labor to perform the work. If the testing requires draining, flushing and recharging a slurry distribution loop, the impact is even more widespread and may take multiple polishing tools offline for the duration of the experiment. Finally, once the experiment is finished, all equipment must be decontaminated, returned to production-ready status, then re-qualified on the standard process. Using an outsource CMP provider avoids all of this disruption in the fab, keeps production resources focused on generating revenue, and eliminates any risk of carryover contamination.

An area of specific interest for Freescale is evaluating new CMP consumables for saving cost or improving process performance. An evaluation generally starts by defining a baseline process, which can come from either the fab or service providers' process library. A comparison is then made between the performance of the standard consumable vs. the new consumable. When defining success criteria, it should be noted that in some cases, the process targets for the new consumable may actually be different than the targets for the existing consumable. Assuming acceptable performance is achieved with the new consumable, marathon runs are performed to verify process stability over time.

A recent project illustrates the evaluation of a new developmental polishing pad in a three-way collaborative effort between the fab, service provider and a pad supplier. The initial phases of the project focused on screening and optimization of the formulation of the developmental pad. After minor iterations in the pad formulation, acceptable performance was obtained relative to baseline process expectations across all relevant metrics, though only a portion of the data will be reported in this article. Removal rate and uniformity comparisons are shown in Figure 2, while planarization curves are shown in Figure 3 .

2. Oxide CMP removal rate and uniformity for a developmental polishing pad.

3. Planarization curves comparing test wafer step height with amount removed.

For virtually any CMP consumable, it is not sufficient to simply achieve a set of process performance metrics at one point in time or for a short run typical of screening trials. Stable performance must be demonstrated across an acceptable timeframe, in a marathon run. For the developmental pad being tested, a marathon run of 1000 wafers was performed with mostly oxide filler wafers interspersed with monitor wafers for measured data points every 100 or 200 wafers throughout the run. Rate and uniformity results across the run are shown in Figure 4.

4. Oxide CMP marathon data on a developmental polishing pad.

The data presented in Figures 2-4 indicates that the new developmental polishing pad is capable of performing within reasonable expectations for removal rate, uniformity, planarization, and process stability. Referring back to Figure 1 , the status of this particular project could be described as somewhere in the process optimization stage and ready for testing in device integration. Additional data is required to move further along the path to qualification in the fab, but it is notable that the data thus far was generated entirely outside the fab.

Outsource support

Fabs running CMP in full production frequently have many other needs for process support beyond those described in this article. Start-up companies, new device development teams, or fabs just starting to adopt CMP into their process flows have even more diverse sets of needs. In any of these circumstances, it is generally possible to accelerate project execution timelines and avoid certain kinds of mistakes by properly leveraging an external provider's established capabilities and experience.

As most fabs have learned, resource requirements for CMP fluctuate over time. Using external services offers a flexible set of resources to assist customers across the CMP technology spectrum on an as-needed basis. These services can be provided either remotely or on-site. CMP outsourcing is proving to be a cost-effective option for quickly executing engineering work with minimal loading of fab resources.