Automation Comes to Litho Inspection

Alexander E. Braun, Associate Editor -- Semiconductor International, 2/1/1999

Automated defect inspection and control strategies are commonplace across fabs, except for the litho cell. Because many litho process defects are relatively large, applicable solutions are limited. For example, the COO of using highly sensitive inspection systems utilized in other process steps often cannot be justified. Anyway, litho error detection requires a system that can be tuned to detect defect types with a wide range of characteristics.

Since over 75% of yield-relevant defect types are fairly large and visible to the eye, macro after-develop inspection (MADI) traditionally has been a manual, operator-intensive process. Fabs rely mostly on manual visual inspection to determine whether a wafer passes (to further processing), is reworked (stripped and rerun through the litho cell) or scrapped. Current MADI systems are manual or semi-automated, and illumination may range from common green lights or spotlights to special point sources and flat, monochromatic panels. Always, the detector is human.

This is limited; defect detection and classification are inconsistent and unreliable, with results varying due to wafer complexity, background patterning noise and human boredom and fatigue. Up to 80% of all MADI defects may go undetected until after etch or final test, when it is too late, resulting in higher scrap rates and lower sort yields. Depending on fab size, ~$3.6 million a year are wasted due to defects undetected by human operators.

Increases in wafer complexity are adding to problems likely to go undetected by manual inspection. While other process steps are improved by baseline defect density reduction through automated detection and control of defects, the litho cell ­ limited to manual inspection techniques that provide little visibility or control and collect very little usable data ­ is a source of yield loss and scrap issues.

Fig. 1. Macro inspection is relatively inexpensive ( ~ $0.35 per wafer). If done properly, it ensures against expensive problems.

KLA-Tencor (San Jose, Calif.) will introduce an inspection suite designed to overcome these limitations, the 2401 Automated Macro Defect Inspection System. It is expected to replace the bright-light MADI performed by human operators. It provides automated detection, classification and reporting of all yield-critical MADI defect types, including hot spots, scratches, large particles, extra and missing resist, unexposed fields, striations and developer spots and splash-back (see Figure).

The system's sensitivity and inspection consistency, repeatability and accuracy far exceed a human operator's, allowing disposition decisions to be made quickly and accurately, reducing scrap and averting investment in low-yielding wafers. Integrated with other yield analysis systems, it provides information that can be used to correct defect mechanisms.

The system integrates brightfield and darkfield inspection technologies, necessary for detection of all MADI defect types. These technologies, combined with detection algorithms, simultaneous ADC and 80-wafers-per-hour throughput at 50 µm sensitivity, permit detection of individual or continued excursions on any wafer at every layer. Lacking the human equation, operator inattention or inadequate sampling are no longer issues.

The system's analysis capabilities provide recommended go/no-go disposition decisions, summary statistics, defect maps and defect images. This allows more effective dispositioning for higher productivity, lower scrap and more accurate rework decisions.

Integrated as part of a yield analysis system, the system can provide production and engineering analysis capabilities impossible with manual MADI. For example, it can create a composite map of wafers in a lot, revealing defect mechanisms invisible when looking at one wafer. It also produces defect images that are transferable to the analysis database for a better understanding of defect characteristics. Other analysis capabilities include control charts, summary statistics, defect source analysis, layer subtraction and zone analysis. Pass/fail settings can be configured for each recipe, enabling automated disposition decision-making. Information supporting the recommended decision ­ such as wafer maps and images ­ is also automatically provided.