Semiconductor inspection relies on imaging
A discussion with Lance Glasser of KLA-Tencor
VSD: How is KLA-Tencor using imaging and machine vision?
Glasser: KLA-Tencor's largest businesses are built on imaging technology, including, for instance, wafer inspection and reticle inspection. The wafer-inspection business is roughly three times the size of the reticle-inspection business, although this varies significantly from year to year since they are on different business cycles-one being driven by wafer starts and the other by design starts. These applications are mostly served by ultraviolet (UV) and deep-ultraviolet (DUV) wavelengths, the shorter wavelengths mandated by the requirement to detect defects in deep-submicron microelectronics technologies.
While the end of optical technology is always predicted for these applications, the use of optical inspection continues to grow because it is fast. Sub-100-nm detection is common on patterned substrates using UV and DUV technologies. KLA-Tencor also uses e-beam imaging systems for defect detection and for measuring critical dimensions on patterned wafers, in-line monitoring of electrical defects, and monitoring the composition of metal films. As the semiconductor industry drives to smaller design rules, these applications are also gaining greater importance.
VSD: What changes in the semiconductor-manufacturing market are driving these applications?
Glasser: As the critical features on semiconductor devices fall below 100 nm, specifications become increasing stringent. On a reticle used for 90-nm technology, one must achieve 100% capture of all edge-shape defects below 100 nm over the entire pixel image. The reticle is 4x the image size of the wafer and has subresolution features to guide the subwavelength lithography process used to print the wafers. In wafer inspection, our customers are inspecting up to 130 steps in the manufacturing processing using optical and e-beam inspection tools.
VSD: Within each application area, what are the most important performance criteria when qualifying a system?
Glasser: For wafer inspection, defects that lower yield must be found in the shortest amount of time for the least money and their cause understood. While dark-field systems are faster but less sensitive, bright-field systems are more sensitive but slower, and e-beam systems, which are the most expensive, are slowest but provide the ultimate in sensitivity and the ability to detect electrical faults in-line.
Our bright-field systems are now operating in the gigapixel-per-second domain. This is necessitated by the very low defect rates in modern integrated circuits. It is necessary to inspect wafers fast enough to detect enough defects for statistical control and obtain enough information to determine defect cause. This requires an inspection tool to run very small pixel sizes [on the order of tenths of microns] and inspect the entire 300-mm-wafer surface in minutes.
For reticle inspection, the most important criterion is not to miss a critical defect. Since the reticle is the master used to print the wafers, a single missed defect can cause millions of dollars of waste. Going beyond that, price, performance, and cost of ownership prevail. As reticles become more complex-they commonly modulate not only intensity, but also phase of light-the ability to sort defects is important. Thus, systems must not only analyze the wafer-lithography process and find defects but also analyze the geometry on the reticle.
Today, our reticle-inspection systems capture a DUV image on a custom TDI [time-domain integration] silicon sensor. The DUV optics in our image-acquisition system have hundreds of optical surfaces and dozens of active control loops. Decades of contamination-control knowledge help us keep the optics stable under intense DUV radiation. Components of the system error budget are in single-digit nanometers. The whole system weighs in around 5 tons.
VSD: How does KLA Tencor approach a new application?
Glasser: KLA-Tencor follows a formal product-life-cycle process for new product development. As part of that process, we spend extensive time in market validation with our customers. Like most system-oriented companies, we do extensive outsourcing of components and subsystems, and our supply-chain partners play an increasingly critical role in helping us achieve our technology roadmap and manage through the turbulent times that the semiconductor industry experiences. We have not only qualified vendor lists, but a short list of strategic partners with whom we have even closer relationships. We are active in the market, identifying potential new partners, and have disciplined programs to drive increasing competitiveness. In terms of core competencies, KLA-Tencor focuses on algorithms and on optoelectromechancial system design and integration.
VSD: Does KLA Tencor perform its own systems integration or look to outside integrators for support?
Glasser: One person's system is another's subsystem. We generally do final integration and test of our equipment before it is shipped to end users.
VSD: How do you envision the future of imaging and machine vision in semiconductor manufacturing?
Glasser: The problems of microelectronic-system manufacturing are more challenging each year. We believe this trend will only increase. Moreover, the percentage of capital that our customers spend on inspection and metrology is generally increasing, as dimensional control becomes more stringent and images larger. They do this because it makes them more productive.
VSD: How will imaging/machine-vision systems have to change to successfully meet those emerging applications?
Glasser: Semiconductor manufacturing is all about productivity. Our systems must become ever more productive and enable our customers to shrink their processes to the next generation, lowering cost per function as they do so. Because of the huge capital costs our customers experience, learning rates-and by that I mean time to results-are key, and we must constantly find new ways to provide useful information on an ever-faster pace. In a practical sense, this means increasing the speed of systems that already run at gigapixel-per-second rates, increasing resolution and sensitivity including-but not always-by moving to shorter wavelengths or using e-beam imaging systems, and, finally, increasing the sophistication of our algorithms.
Generally speaking, algorithms are moving from a world of rule-based detection to a world of modeling and simulation-based detection, incorporating more knowledge of the customer's problem domain. Algorithms are not only required to identify a defective area, they're also required to suppress process variation across the wafer and to separate defects that matter to the customer from those that don't impact the electrical yield of the device.
LANCE GLASSER is vice president and general manager for the Reticle and Photomask Inspection Division of KLA-Tencor (San Jose, CA, USA). Before joining KLA in 1996, he was director of the electronics technology office at the US Advanced Research Projects Agency. He has also been a staff member in the department of electrical engineering and computer science at Massachusetts Institute of Technology. Editor in chief Conard Holton talked to him about wafer and reticle inspection.