Solution of pattern matching inspection problem for grainy metal layers

Abstract

In this paper, we demonstrate a new method of improving the defect controllability of grainy metal layers, for example, Hot-Al-Cu wiring, by enhancing the practical sensitivity of in-line inspectors. The problem in increasing practical sensitivity is the nuisance counts caused by grain boundaries, which do not cause electrical failures.We propose a method of decreasing the signal from the grain boundaries. On the grayscale images taken by pattern matching inspectors, grain boundaries are observed as gray on Hot-Al-Cu wiring, which is observed as white. If the illumination brightness is increased, the gray level of the grain boundaries becomes higher and saturates at the upper limit of grayscale, i.e., white. On the other hand, the gray level of the wiring stays white. Thus the signal, the grayscale difference between the grain boundaires and the wiring, can be decreased to almost zero. We call this phenomenon the "saturation effect."Our experimental results prove that the saturation effect due to illumination brightness optimization successfully cancels the nuisance counts caused by grain boundaries. The practical sensitivity limit is enhanced from 0.8 to 0.4 m. This solution will greatly improve the defect controllability of grainy metal layers.Index Terms-Grayscale, in-line inspection, pattern matching inspector, patterned inspection, semiconductor manufacturing.

Scan-Line Quality Inspection of Strip Materials Using 1-D Radial Basis Function Network

“…Assuming that the "signal" in the model mismatch signal is the defect peak and the background variation is the noise, one can conclude that the SNR relevant to the detection algorithm clearly improves. These results are also in agreement with a recent study (Sakurai et al, 2002) in the field of semiconductor inspection. Experiments have been also conducted to assess the sensitivity of the detection algorithm for different defect sizes and positions on the image signal.…”

Integrated Automatic Optical Inspection and Image Processing Procedure for Smart Sensing in Production Lines

“…In Preprocessing , several functions for gray scaling and noise cancellation will be conducted to refine the image to the proper condition for further analyses [ 13 ], which will be explained later. Similar to the actions that appear in the OM image window, “Redo” provides a chance for users to remove the existing image and re-capture the image from the camera.…”

A basic study of defect inspection methods for ASIC manufacturing line (Second report: Inspection parameter dependence on yield and cleaning cycle)

“…If the total of x and y exceeds a certain defect control limit S, it is judged as exceeding the defect control limit, and the process equipment in question is cleaned. The reason that y (≥ 0) is introduced in this paper is that y = 0 is ideal, but there are actually a significant number of cases where a false positive cannot be avoided due to the influence of instability of wafer surface roughness, insufficient sensitivity, and the like [7,15]. Also, even if no lots exceeding the defect control limit are detected, the processing equipment is periodically cleaned after processing a specified number of lots.…”

“…In other words, if the number of false positives is large, the problem occurs that the engineers who control L avr and D avr may erroneously believe that practically optimum inspection conditions do not exist. The absence of false positives is ideal, but in reality there are cases where a false positive cannot be avoided in defect inspection [7,15]. Equipment to automatically separate false positives is commercially available, but such equipment costs hundreds of thousands of dollars.…”