Conventional equipment is not capable of detecting fine defects in new materials. Dr Kenji Watanabe explains how Hitachi and Tokyo Siesmitsu have developed the DUV optical wafer inspection system for the 65nm technology node.
As the trend towards miniaturisation of advanced semiconductor devices shows no signs of abating, some advanced device manufacturers are proceeding from the 90nm development process node to 65nm technology; the mass production process node, too, is shifting from 100nm to 90nm and below. For these leading-edge devices, fine processing technology with ArF exposure that makes free use of Optical Proximity Correction (OPC) and the use of new materials that include high-k materials such as the gate insulation layer, low-k materials used between wiring layers, Cu, strained Si, and SOI (Si On Insulator) is gaining ground.
Wafer inspection systems are tackling problems, such as fine defects, that cannot be detected by conventional equipment. Market demands on optical wafer inspection systems include high sensitivity for detecting fine defects, pattern noise reduction and Signal-to-Noise (SN), ratio improvement, and an automatic defect classification function. To meet such needs, Hitachi High-Technologies Corporation and Tokyo Seimitsu Co Ltd have developed the high-performance DUV optical wafer inspection system.
Highly sensitive defect detection is achieved by using an optics that combines a DUV laser light source for high sensitivity with Super Resolution (SR) technology together with a high-performance die comparison algorithm. The detection of pattern wafer defects involves the problem of detecting nuisance defects produced by colour variations or grain. The system is capable of reducing such pattern noise by using a die comparison algorithm that tolerates differences in brightness.
To cope with the burgeoning amount of image processing as inspection pixel size becomes smaller and wafer size increases, Hitachi has achieved throughput several times that of conventional systems by developing one of the world's fastest image processing devices and a high-speed comparison algorithm.
Generally, the limit on optics resolution is proportional to the wavelength of the illuminating light, L, and is expressed approximately using the numerica aperture of the lens as L/2NA. A 266nm wavelength DUV laser is employed as a high-output laser that produces a stable, continuous oscillation. Using the laser eliminates the effects of chromatic aberration, realising extremely low aberration and wide field of view in this high-performance DUV objective lens.
Furthermore, because there is abundant optical power, even for fast image scanning, this system can be applied under various detection conditions, ensuring superior process adaptability. Speckle noise from use of the laser light source is eliminated by development of a low-coherence optics
There is a growing need for the detection of short defects in the Cu damascene process, which is becoming the mainstream wiring process. Light that has a wavelength of 266nm has minimal reflectivity with regard to copper, and so has the advantage of allowing contrast with the background. A high-sensitivity image sensor and die comparison algorithm also allow detection of fine differences between dies, thus achieving sensitivity that is sufficient to detect small defects of a fraction of the wavelength.
The aim is to improve sensitivity by using proprietary SR optics to sharply increase pattern contrast to near the limit of resolution. The optimum SR conditions can be selected for the process, providing high process adaptability.
To take advantage of the resolution obtained by the optics, a fast and accurate linear motor stage was developed with laser interferometers on the X and Y axes as well as a new highly sensitive broadband image sensor that features very fast and accurate real-time automatic focusing. A new high-speed, programmable image processing system was also developed for flexibility and to strengthen compatibility with various processes.
This system is equipped with a die-to-die algorithm that smooths out process-induced colour variations and is capable of distinguishing fine differences in shape. Furthermore, a hybrid cell / die comparison algorithm allows simultaneous comparison of repeated pattern cell regions and random-pattern peripheral circuits to increase sensitivity for cells in die comparison.
A new Real-time Defect Classification (RDC) function has also been added. This function uses differences in attributes to classify nuisance defects, which are detected as defects but do not directly affect device yield, separately from actual defects. It removes them and then reports the final detection results in real time.
The system is equipped with this RDC as a standard function, which lightens the load on the operator and review system by automatically classifying nuisance defects and other defects. This greatly reduces the number of defects that must be dealt with.
The use of short-wavelength illuminating light achieves a striking improvement in resolution compared to conventional systems that employ white light or UV light. This greatly improves the capability for fine defect detection in fine patterns. For 90nm node devices, the Hitachi system detected two to four times as many defects as did the other system.
The favourable results compared well with the other DUV system come from the system's use of intense laser illumination, which allows a smaller detection pixel size than does lamp illumination. Furthermore, this system was able to detect ten times more defects than the other system for the even finer 70nm node devices. This system is expected to be even more effective in the development and production of subsequent 65nm node devices.
APPLICATION TO RETICULE EVALUATION
For the exposure of 65nm node devices and other fine patterns, the desired pattern processing cannot be done without using an ArF light source and a reticule that has been corrected for various proximity effects. As this kind of correction is done, the exposure margin is smaller than with the conventional method, thus making it difficult to maintain a stable, high yield in mass production.
The exposure margin is determined with a pragmatic approach whereby the actual exposure is done with amount of exposure and the focus as parameters, the pattern is transferred, and the transferred pattern is examined. Evaluation of all of the patterns over the entire mask would take a very long time, so various efficient methods are being sought.
The use of a wafer inspection system to detect and evaluate defects that arise according to changes in exposure conditions is a particularly effective method.
And because this system uses an illumination light that has a wavelength close to the ArF exposure wavelength, greater contrast between the photoresist pattern and the background anti-reflection film can be obtained than is possible with conventional wafer inspections systems, making high-sensitivity inspection. The OPC reticule pattern is transferred to the wafer under various exposure conditions. At that time, the wafer inspection system is in the die-to-die mode, so both of the dies that are being inspected must have patterns that were exposed under standard conditions.
Looking to the 90nm node era and beyond, inspection technology will play an important role in facilitating the development of advanced devices and production factory start-up. In the future, the company intends to strive for higher performance to respond to the requirements that come with finer patterns, new materials and higher speeds.