Mask inspection with locally variable sensitivity
The cost of masks for photolithographic patterning is increasing dramatically along with continued semiconductor scaling, in part because of growing inspection costs. Leading-edge masks have line-edge roughness virtually equal to defect dimensions, and conventional defect-detection techniques result in excessive and redundant detection of pseudo- and nuisance defects. In addition, advanced techniques that selectively reduce the influence of mask errors require a substantial change in the defect-detection concept, because it is no longer adequate to judge the presence of an imperfection solely on whether the pattern figures of the mask are correct.
In 2006, the Association of Super-Advanced Electronics Technologies (ASET) consortium, with support from the New Energy and Industrial Technology Development Organization (NEDO), started a 4-year project to reduce mask manufacturing cost and turnaround time by optimizing mask data preparation, writing, and inspection. We are developing inspection equipment and a data-processing system to enable practical inspection based on pattern prioritization entered at the device-design stage, as shown in Figure 1.1–3 The pattern-priority information will be used to dynamically change the defect-judgment level or algorithm, or both. In detecting defects, this system will take into account pattern priority, the pattern-repeating description in the design data, and printability. It will feature low cost and rapid turnaround, primarily by avoiding excessive detection of flaws.
Traditionally, only design-pattern data was provided to the inspection equipment, and the mask was inspected using fixed, preset, defect-judgment algorithms. As part of OASIS (open artwork system interchange standard, SEMI-P39), a common data format was defined to effectively express a repetition pattern and pattern prioritization. A procedure was developed to generate mask data rank (MDR) information from device designers, to be provided to the mask-inspection system in the same format as that used for conventional pattern data and in parallel.
The MDR information expresses multiple levels of pattern-priority data. For example, some designations, such as ‘power supply line’ or ‘dummy patterns,’ can be assigned relaxed defect-judgement threshold levels, while others, such as ‘clock line,’ should be assigned a tightened threshold. MDR data having several pattern-priority levels is converted in a pattern memory by the pattern-generating circuit. If different priority patterns exist in a single area, the tightest judgment level among them is applied.
The converted MDR information is then provided to a defect-judgment unit, which decides based on the pattern priority that was determined at the device-design stage. Based on the MDR pattern, detecting pseudo- or nuisance defects can be reduced by relaxing the defect-judgment threshold, while for other, critical problems the threshold is tightened up and the necessary algorithm applied.
This process dramatically shortens review time. Figure 2 and Table 2 show simulated inspection and review times for leading-edge masks. As shown in Table 1, we adopted estimated review times of 30s per detected defect for obvious flaws and 1min for uncertain cases. While the inspection time itself, namely, the sensor's image acquisition time, is constant regardless of the number of defects, review time depends on how many there are. The effect of employing the MDR information is expected to be high because the number of uncertain defects is huge for complex patterns like those on the metal-1, metal-4 or poly layer.
In summary, we have proposed new mask-inspection methods that change defect-judgement threshold on the fly for every local area, depending on the pattern importance determined at the device design stage. The new method is shown to shorten mask inspection time. In addition to pattern importance, the input information for the mask-inspection system can include layout-analysis information obtained using pattern-analyzing software. For example, pre-analyzing the post-OPC (optical proximity correction) pattern allows us to adjust the defect-judgment threshold or algorithm to assure accurate post-OPC pattern-edge positioning. Furthermore, extracting a mask-error enhancement factor expansion area and including its index information in the input information will permit extra functions such as printability simulation. ASET is committed to continue developing such expanding functionality for the mask-inspection system.
Industrial Science and Technology, the University of Kitakyushu Semiconductor Technology Academic Research Center, and Advanced Mask Inspection Technology. My thanks also to all the members of ASET Mask D2I.
