Abstract
The bonding energy is an important parameter to evaluate the quality of bonded wafers in the semiconductor industry. The most important currently used method to measure the bonding energy is the so-called crack opening method. Unfortunately, the infrared cameras used for the wafer inspection with this method have limiting resolutions, and the derived direct crack length reading error is relatively large. To solve the reading error and adaptability problems, in this study, we improve upon the conventional image processing method and propose a crack length identification method that uses function fitting. The effectiveness and feasibility of the method are verified through experiments.
Highlights
Wafer-level packaging and 3D integratedcircuit are excellent technologies to improve the performance of electronic components, reduce costs, and reduce size
(1) Import the coordinate data of the crack (X, Y) (2) Set calculation iteration = M and the initial calculation iteration m=0 (3) Execute the RKJL algorithm to obtain the specific formula for the function (4) Evaluate whether it is an opening-upward function, whether it has only one extreme value in the range of the light source, and whether its minimum value lies within five pixels of the lowest point of the coordinate extracted during digital image processing
We used the RJKL method for function fitting and to calculate the coordinates of the lowest point of the crack
Summary
Wafer-level packaging (wafer bonding) and 3D integratedcircuit are excellent technologies to improve the performance of electronic components, reduce costs, and reduce size. Measurements of the bonding energy are typically used to assess the quality of a wafer bond.1–3,15–17 These measurements can be done using methods that involve the following: crack propagation, static liquid oil-pressure measurements, four-point bending delamination, MC testing, tensile testing, the particle method, and others.. The main innovations of this study can be summarized as follows: (1) an adaptive algorithm was designed, which can effectively process photos and wafer pairs of varying quality, and (2) function fitting was used to improve the accuracy and stability of the image analysis. The third section introduces the new improved digital image processing flow, which is tailored for the subsequently performed function fitting step. The sixth section summarizes the conclusions of this article, which are followed by the acknowledgment
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call