| dc.description.abstract | Testing wafers requires manpower and time. Finding a way to quickly detect errors is inevitable. The purpose of this study is to explore another way of wafer homogeneity, and use the verification of homogeneity to screen out wafers with inconsistent randomness or inconsistent yield to achieve the effect of rapid inspection.
This study uses 〔1〕 corrected parameter B-Score and the standard deviation formula of this study to determine the randomness and yield homogeneity. The standard deviation formula is calculated from the Rattleback obtained through the new scattering point simulation method. According to this formula, the corresponding standard deviation can be calculated for the results of different numbers of diesize and different yields, and then abnormal wafers can be screened.
Previously, there have been papers for dicing wafers for discrimination. For example, 〔2〕 discriminate after simulating each substrate with a variety of diesize, and 〔3〕 the homogeneity test is based on the difference in yield. This study will use statistics to test the homogeneity of the cut part, and use two methods to divide it into two parts for verification, which are the up and down cut and left and right cut. The verification method is to calculate the characteristic parameters of each of the two parts and bring them into their own verification formulas. The higher the verification score, the more inconsistent it is, and the wafers with special features can be screened.
In this study, a homogeneity test was established for randomness and yield. Finally, WM-811K data provided by TSMC was used to verify the test. Observing the pieces with the highest test scores showed obvious error characteristics. This paper will use two modes for homogeneity testing, namely wafer partition and wafer-to-wafer, and present the results. The simulation software used is Matlab.
| en_US |