provided by: Semiconductor International Automated edge inspection has revealed several immersion-specific defectivity modes. Defects related to the activity of the immersion fluid, interactions with resists and topcoats, and edge bead removal processes can have a significant effect on process yield.
The availability of high-speed automated edge inspection has created a growing awareness of the impact of the wafer-edge defects on process yield. Some manufacturers estimate that 30% or more of killer defects originate from wafer-edge regions. Increased focus on edge defectivity during the development of advanced processes and materials for 300 mm manufacturing has revealed a number of edge-specific mechanisms that will require careful monitoring and control as new processes move into production. Developing and qualifying new processes require sensitive detection and detailed characterization capabilities.
Controlling these processes at high volume demands production-worthy inspection capability - adding such requirements as high throughput and robust automatic classification of a broad range of physical- and process-related defects. The imminent introduction of immersion lithography is an excellent case in point - interactions of the wafer and scanner through the immersion medium introduces a number of novel defectivity modes, several of which are edge-specific. Although edge defects typically do not cause yield loss directly, they are often precursors to killer defects in the wafer's active area. Real-time edge inspection ensures that edge defects can be detected, and their causes can be corrected, thereby minimizing yield impact.
Automated edge inspection
Available edge-inspection technologies use one of two basic approaches: laser scanning or image-based. Scanning techniques use a monochromatic source focused to a small spot and scanned over the edge region. They typically detect light scattered by the defect, although the signal may be supplemented by detectors intended to spot phase effects and surface topography. They may return to detected defects to acquire images, or construct images from the scanned signals. In contrast, image-based techniques acquire a continuous series of 2-D images covering the region of interest with one or more cameras using broad-spectrum illumination.
The fundamental requirements for any production-worthy defect inspection technology are speed, useful sensitivity and reliable data reporting. For both scanned and image-based technologies, defect inspection in the edge region is complicated by the edge's complex, multiple curved shapes. Scanner techniques may address this challenge with a scanning head that rotates around the edge to acquire signals from the full region: top, bevel and apex. Large deviations from the expected edge profile or inaccurate centering can cause tracking problems, and the mechanical scanning process, coupled with sequential data acquisition, is inherently slow. Increased sensitivity requires a reduced spot size and increased scan density, resulting in longer acquisition times. Scanned images are monochromatic, and the interpretation of multiple signals is not intuitive, particularly when the signals are contradictory.
Defect sizing based on scanned darkfield scattering is also problematic. First, the analysis must resolve and interpolate the multiple peaks generated by adjacent, overlapping scans. Even then, the derived signal intensity is affected by independent factors that include illumination wavelength, incidence angle, collection angle, defect refractive index, underlying film refractive index, and defect orientation. For these reasons, comparisons of inspection sensitivities between systems based on their ability to detect darkfield or high-scattering defects can be misleading. In a production environment, it is more relevant to compare inspection sensitivities based on the ability of the systems to detect brightfield or non-scattering defects - defects that can be correlated to measurements made by other techniques available in a typical production line.
Image-based systems address the edge curvature issue with multiple cameras and optics optimized to provide sufficient resolution over the required depth of field. Image-based inspection can completely cover the edge region in a single pass with a 300 mm production-capable throughput of more than 80 wph. Defects retain their familiar visual characteristics, readily recognized by an operator or automatically classified by well-established image-based classification engines. Both brightfield and darkfield images are available, and the color information provided by the broad-spectrum illumination and imaging is particularly insensitive to film variation along the wafer edge (Fig. 1). Image-based edge inspection has been qualified for production at multiple 300 mm fabs.
Edge inspection cannot rely on die-to-die comparisons used for inspecting the active regions of the wafer. A fab-proven alternative approach builds a "defect-free" surface reference model for each acquired image, compares the raw image with the "defect-free" surface reference model, and extracts the defects according to user-controlled parameters. A second algorithm detects very large defects, such as edge chipping. Finally, the system automatically classifies defects according to edge zones, color, size, area, morphology, etc. The classification results, including defect locations and corresponding color images, can be exported to offline defect review software, a data analysis system or a SEM review station (Fig. 2).
Immersion lithography defectivity
In immersion lithography, interactions between the scanner head and wafer surface through the immersion fluid create new defect mechanisms. On the wafer edge, the fluid dynamics are strongly affected by the abrupt wafer edge as well as the more subtle, but no less significant, edges of the various film layers. Immersion-specific defects include bubbles, film delamination, droplets, stains and residues.
Unintended wafer-edge conditions contribute to the formation of bubbles in the immersion fluid as the wafer edge slides through the fluid at high speed. Bubbles can act as concave lenses that distort the ultraviolet (UV) light from the scanner, resulting in tell-tale circular patterning defects on the wafer. Bubbles have many causes; they can come from dissolved gas, photoresist outgassing, surface roughness, and from gas trapped in a surface cavity. The bubbles are not usually transported far from their origins, but can move with the flow of fluid. Since surface conditions near the wafer edge are the most unpredictable, it is common for bubbles to appear there first (Fig. 3).
Bubbles <2 ?m in size are thermodynamically unstable, and quickly dissolve back into the fluid. However, some bubbles grow in size when exposed to enough UV radiation. Bubbles >2 ?m have longer lifetimes, and will distort the pattern before dissolving back into the fluid. The bubbles closest to the wafer surface will cause the most damage because, due to boundary fluid conditions, they move slower and their shadows are in focus. New immersion fluid delivery system designs and inline deionized (DI) water degassers have reduced bubble generation from dissolved gas. However, bubbles caused by wafer-surface conditions remain a problem. Inspection for edge defects and control of edge conditions will remain ongoing requirements for immersion lithography.
Delamination
Hydrodynamic forces generated between the immersion fluid and wafer can cause delamination at film edges (Fig. 4). For example, current topcoat materials are known to have poor adhesion with silicon. If they spill over the underlying photoresist layer and come into contact with bare silicon near the wafer edge, the fast-moving stage can create enough force to delaminate the weakened topcoat. Flakes from these defects may then be transported by the immersion fluid and redeposited in the wafer's active area. Dislodged topcoat flakes can also contaminate the fluid delivery system, causing expensive shutdown of the lithography cell.
Delaminated flakes can affect subsequent wafers. The immersion fluid can carry and deposit them on the wafer stage, only to dislodge the flakes at a later time and scatter them onto subsequent wafers. The residue can also slowly build up from the constant wetting and drying of the stage until it acquires critical mass and is transported onto the wafer.
It is possible to reduce delamination by ensuring that the topcoat fully covers the photoresist while resting on top of the bottom antireflective coating (BARC). This solution requires using chemical edge bead removal (EBR) to eliminate the BARC, an optical EBR to strip off the photoresist and another chemical EBR to remove the topcoat. Additional constraints are placed on these processes to ensure that the EBR regions do not extend too far from the wafer-edge apex, so as to maximize the surface area available for devices. This solution requires tight control of the EBR process to ensure that the EBR cuts are clean and the individual layers in the film stack are centered with each other and wafer edge.
Conventional EBR metrology is typically a manual process, using an optical microscope to measure the distance from the EBR line to the wafer edge at four opposite "corners" of the wafer. Manual EBR metrology has shortcomings, such as the wafer-bevel transition is easily mistaken for the wafer-edge apex; the EBR line is difficult to distinguish on patterned wafers; the EBR line may be discontinuous around the wafer; and four measurement points per wafer are statistically insufficient.
Image-based edge inspection can perform automatic high-resolution EBR metrology on both patterned and unpatterned wafers. One novel approach stitches together images acquired by the edge-top camera and compresses them to form an EBR Fingerprint map (Fig. 5). The process filters non-circumferential features from the images and allows the EBR algorithm to measure EBR features even when patterning extends to the wafer edge. Circular EBR features appear as vertical lines in the "EBR Fingerprint" map, while oblong or off-center EBR features appear as sinusoidal lines. Since many edge defects, especially immersion-related ones, are caused by poor EBR, measuring EBR and edge-bevel width can help eliminate an entire class of EBR-induced defects, including delamination and flakes.
Droplets, residue and stains
Defects do not have to interfere with the exposure step to cause yield problems. They can cause problems indirectly by altering surface conditions. Stage speed, fluid delivery system design, topcoat materials and surface conditions all affect the immersion-fluid contact angle. For example, contamination on the surface can alter its hydrophobic properties and change the receding contact angle. If the contact angle is too small, it can cause the fluid to leave behind water droplets. Water droplets are especially common near the wafer edge because of the various surface conditions. Resulting defects can differ greatly depending on the chemical interactions between the droplet and topcoat, and may include residues, stains and bridging defects.
Conclusions
Edge defectivity has a significant impact on process yields. Edge-inspection systems have been applied to immersion lithography, revealing a number of edge-defectivity modes that will require close control as these advanced processes move into production. Image-based edge-inspection technologies appear to have the high throughput, useful brightfield sensitivity, and accurate defect reporting capability required for high-volume production.
Tuan Le started in the semiconductor equipment industry with Tencor Instruments (now KLA-Tencor). Before joining Rudolph Technologies as an all-surface product manager, he held marketing and management positions at Cymer and Ushio. Le received his B.S. in mechanical engineering from the University of the Pacific and his MBA from Cornell University.
author: Tuan Le, Rudolph Technologies, Flanders, N.J., www.rudolphtech.com
Semiconductor International. Copyright © 2007 Reed Business Information, a division of Reed Elsevier Inc. All rights reserved.