Patterned wafer defect binning

04/01/1998

Patterned wafer defect binning

New tools for patterned-wafer defect inspection may enable the detection of excursions well before a process is out of control. Relying upon databases of defect categories, end-users should be able to more rapidly determine the root causes of problems.

Defects that can be detected by the scattered light from a patterned wafer can fall into many categories: particles, pits, scratches, etc. Inspection tools can register different signals for many of the different categories of defects. Each signal category can then be correlated with other metrology tools, such as SEMs or AFMs, to build confidence in a binning database.

High-speed, inline metrology systems for volume production typically only produce a wafer map showing the distribution of all defects together; defect binning requires too much detection sensitivity and too much post-detection signal processing to occur inline.

When an inline tool detects a slight rise in defect counts within control limits, the suspect wafer can be re-scanned in an off-line engineering metrology tool capable of defect binning. Engineers can then examine the defect categories to look for upward trends within one or more defect types that may point to a specific production tool subsystem problem.

For example, an increase in the number of post-CMP-clean scratches would point to a problem with the CMP buff/polish step, while an increase just in the number of particles would implicate the scrubbing tool as the defect source.

Since there is almost always a mix of defect sources producing a variety of defects on any given wafer, the ability to bin defect categories dramatically reduces the time necessary to pin down problems. This technique is particularly powerful for pilot line start-up and ramping a new process into production, though it is also useful in process sustaining.

In sustaining an established process, an engineer needs to first ensure that defects are within set control limits. Then the defect counts are monitored over time to watch for the beginning of upward trends. By catching a real increase in defects just as it begins, yield will hopefully not be degraded. Identification is easier at the beginning of an excursion in a single defect category instead of when all defects are added together.

Figure 1. Typical defect binning on monitor wafers shows that the beginning of a process excursion can be detected well before the combined defect count rises above control limits.

Figure 2. Schematic of the combined darkfield and brightfield detectors in an Applied Materials WF-736 DUO

This can be conceived of as lowering the noise-floor. If there are many semirandomly fluctuating defect counts added together to make the total, then the signal of a clear upward trend in any one category gets buried in the noise of the random variation of the others (Fig. 1). As a result, a defect source can throw the process out of control before it is noticed.

Applied Materials is promoting rapid defect binning as one of the primary advantages of its new patterned wafer defect inspection systems: the WF-736 DUO and WF-731. Both systems are based on the WF-73X platform that the company acquired with Orbot, with sufficient commonality that a WF-731 for production line-monitoring can be field retrofitted to become a WF-736 DUO for engineering analysis.

With four darkfield detectors located at cardinal points around the wafer (Fig. 2), the WF-73X systems can differentiate between defect types by comparing the amount of scattered light in different directions. For example, a small scratch will register on detectors normal to its side, while detectors inline may not register at all, and a symmetrical particle should register equally on all four detectors. The WF-736 DUO adds brightfield imaging to the four darkfield detectors.

A conceptual battle may have begun between Applied Materials and KLA-Tencor over which company`s patterned wafer inspection systems provide the best combination of throughput, resolution, and binning. Applied Materials states that its systems can do "on-the-fly" automatic defect classification as an option, and that no other commercially available system offers such capability. The company also declares that its tools provide the highest throughput.

Dave Icke, KLA-Tencor Wafer Inspection Group VP of marketing, stated, "When customers require a single-platform solution for engineering analysis and line monitoring, they have long had that capability with the KLA-Tencor 2100 series bright field inspection technology and its modular configurations." The company also claims greater resolution with broadband illumination.

- E.K.