Breakthroughs Needed in Defect Reduction

Staff -- Semiconductor International, 2/1/1998

The newly revised National Technology Roadmap for Semiconductors features an expanded scope of detect reduction that now addresses the entire yield learning process from R&D to volume production. As explained in the Roadmap, the requirements for defect reduction technologies including defect detection, defect review and defect classification crosscut all process technologies, as well as the facility infrastructure, device design, process integration and device packaging. A variety of solutions will be needed to achieve yields of 60% (at transfer to production) and 85-95% (mature product) on leading-edge devices. To achieve targeted defect densities, yield learning must proceed at an accelerated rate, and defect isolation cycle time must be appreciably shortened.

It is for these reasons and more that the editors at Semiconductor International decided to dedicate a page in each monthly issue specifically to yield management issues. As companies continue to accelerate transitions to devices with smaller geometries, the challenges of rapidly predicting yields, isolating defects, determining the sources and eliminating the defects also increase dramatically.

In this column we will explore yield modeling, yield improvement and yield management strategies. We will cover integrated yield management, encompassing data from parametric test, functional probe, fab equipment data, in-line monitoring data from wafer inspection and classification systems and WIP data from the factory floor management systems. New technologies or methodologies that allow for yield improvement in fabs will be emphasized.

As in the Roadmap, there will be a certain crosscutting of technologies across the various news pages in SI

Please send news leads to ljpeters@Semiconductor.Net

Breakthroughs Needed in Defect Reduction

Among the many challenges identified in the 1997 Roadmap are areas of semiconductor processing requiring major breakthroughs. In the area of yield management and defect reduction, these include the need for defect detection, review and classification tools capable of delivering both high throughput and high sensitivity. For instance, the throughput of patterned wafer detection systems must increase by 300X in going from R&D to yield ramping to production levels. Tools that can accelerate the yield ramp must be characterized 18 months before production at a given technology node begins.

Defect inspection using patterned wafer inspection tools based on light scattering and optical imaging solutions will continue to be used for volume production phases for the next two to three technology nodes. Novel technologies such as holography, e-beam or X-ray may also be required. SEM-based analysis must be driven to higher throughputs. It is expected that for 300 mm manufacturing, process tools that accommodate on-wafer defect detection and characterized in situ process control sensors will also be needed.

Yield modeling and defect budgeting in the future will need to make better use of electrical characterization information and reduce emphasis on visual techniques. This will require a better understanding of systematic and parametric impacts on device yield. Yield impact at the interconnect levels is so severe that modeling of ultra-thin film integrity is needed, as well as traditional interconnect modeling.

Significant advances in test structure development are needed to determine the link between trace metallics, ions and organics on device performance, reliability and yield. Such correlations will determine whether increasingly stringent contamination limits for water, chemicals and materials are truly required, and if so, when they will be required. Fortunately, order of magnitude improvements in process critical fluids will not be needed until well into the sub-0.1 µm regime.

Timely failure analysis depends on rapid fault isolation -- essentially boosting learning rates from weeks to days and eventually hours. To source defects more quickly, there is a need for an integrated suite of data analysis software tools, using validated algorithms. According to the Roadmap, defect sourcing complexity will increase by 2.1X, 4.3X, 8.9X and 26X for the transitions from 0.25 µm to 0.18 µm, 0.18 µm to 0.15 µm, 0.15 µm to 0.13 µm and 0.13 µm to 0.10 µm, respectively.

Defect sourcing using the traditional Shewhart methodology for particle per wafer pass control will be replaced by the use of exponentially weighted moving average (EWMA) and cumulative sum techniques that are more robust to violations of the underlying assumptions of normality. Potential statistical solutions will be evaluated based on their ability to detect shifts in the process vs. the likelihood of false alarms.

Integrated yield analysis will require smart software tools that automatically access multiple databases and establish correlations between data of different types.

Defect reduction in process equipment remains paramount to achieving defect density goals. Though equipment defect targets are primarily based on horizontal scaling, vertical faults as they apply to gate stack, metallic and other non-visual contaminants need to be understood. In situ chamber monitoring will serve the dual purpose of monitoring the process and reducing the frequency of wet chamber cleans for higher equipment utilization. A more fundamental understanding of reactor contamination formation, transport and deposition on the wafer is needed to optimize the placement and interpretation of data from in situ sensors. Intelligent process control at the tool level requires a fundamental understanding of how parameters impact device performance.

Finally, the Roadmap indicates that defect prevention and elimination will require wafer environment control technology, also referred to as minienvironments, as soon as the 0.18 µm generation of devices. The ability of wafer isolation technology to facilitate factory automation will become important as targeted levels of ambient acids, bases, condensables and metals at the wafer surface become critical at certain process steps. Accurate, affordable, repeatable real-time sensors for non-particulate contamination must also be incorporated into process tools.