Mix and Match Inspection: One Size Does Not Fit All
All defect detection tools are not alike. Tools designed to detect particles on planar films can disappoint if asked to detect incomplete etching of patterned wafers. Different defects require different imaging and detection techniques. This article surveys the field and discusses the advantages and disadvantages of different approaches.
The first considerations in selecting an inspection tool are the types of defects expected and the potential sources of noise. For example, particles are the most likely defects in a blanket film, while a post-CMP (chemical mechanical planarization) film might also have microscratches and dried slurry residue. The underlying layer might degrade the signal from a transparent oxide film, or thickness variations might change its color. Grain structure or surface reflections might cause noise when imaging a metal film.
Types of defects and sources of process noise. Photo courtesy of KLA-Tencor.
As users introduce new processes and new materials, they also encounter new defect types and sources of noise. For example, CMP introduces defects due to surface mechanical damage, such as microscratches, dishing, and plug coring. The top layer thickness varies due to underlying topography, introducing color noise. On the other hand, CMP inspection tools have little need to image defects in underlying layers or high aspect ratio features.
Examples of critical CMP defects. Photo courtesy of KLA-Tencor.
Different imaging methods are more sensitive to different types of noise. For example, explained Bobby Bell, 2200 family marketing director at KLA-Tencor (Milpitas, CA), electron beam illumination offers the highest defect sensitivity. As a high-resolution, surface-sensitive technique, it minimizes most types of noise, particularly those due to underlying layers and film thickness variation. Electron beam imaging requires a vacuum chamber, however, making it both the slowest and most expensive inspection technique. Optical imaging is more cost effective for most inline inspection tools.
In brightfield optical imaging, light shines on the sample vertically, through a high resolution objective. The same objective collects and images reflected light from the wafer. The light penetrates transparent films, and the objective can see into high aspect ratio etched features. Brightfield imaging works well for planar and underlying defects, but these strengths make it susceptible to noise from underlying layers.
In darkfield imaging, a laser shines on the wafer at a low angle of incidence. The light is either imaged by an objective above the wafer (directional darkfield imaging, for instance as used in KLA-Tencor's ILM 2230 system (see previous article) or collected by a photomultiplier tube (PMT) at the same angle as the laser (double directional darkfield imaging, used in KLA-Tencor's Surfscan AIT system, for example). In either case, scattering of the laser light indicates the presence of a defect. The low angle reduces both grain noise and scatter from underlying layers. Some systems allow the user to change the incidence and detection angles to optimize the signal-to-noise ratio.
Double directional darkfield imaging is best suited to detection of surface and embedded particles. It is the fastest technique, as the illuminating laser scans the surface at high speed. Planar defects are difficult to detect by this method because they have a low scattering cross-section. The ultimate resolution of the technique is limited by the laser spot size, since any light scattered within the spot is detected at the PMT. Unfortunately, the PMT can only process a single pixel at a time. Reducing the spot size (and thus the pixel size) improves sensitivity but decreases throughput.
In directional darkfield imaging, in contrast, the objective is perpendicular to the wafer plane and a stationary laser floods the field with light. The objective images the entire field onto the detector at once. The ultimate resolution depends on the detector pixel size, which is much less than the laser spot size.
In some structures, optical filtering before the detector can reduce noise even further. With an azimuth filter, for example, vertical and horizontal lines scatter outside the objective, leaving only defects behind. Similarly, a repetitive pattern filter uses Fourier filtering to optically block repetitive diffraction orders. Both methods convert high noise regions to low noise regions, improving the defect capture rate.
Advanced optical filters remove repetitive pattern structures before image detection. Photo courtesy of KLA-Tencor.
Advanced algorithms can extract more data from the detected image, too. For example, Bell told Semiconductor Online, KLA-Tencor's local segment auto threshold algorithm can dynamically break the image into several segments based on noise characteristics. Low noise regions will have lower defect detection thresholds. This technique achieves the maximum capture rate in each region, while avoiding false defects.
CMP and other new processes place new demands on defect inspection tools. Still, with thoughtful matching between tools and applications, users should be able to find a reasonable balance between speed and accuracy.
By Katherine Derbyshire