New Die Leveling Method Based on Passive Auto-focus in Automatic High Precision Flip-chip Bonders

摘要

A new die leveling method based on passive auto-focus in automatic high precision flip-chip bonder is presented in this paper and is used in our project. Automatic high precision flip-chip bonders ,in which the alignment system are the one of the key components, are used for die placement and die bonding in IC, MEMS and MOEMS assembly process. The alignment is divided into traverse alignment and longitudinal alignment—focusing alignment. Simultaneously, the die must be parallel to the substrate before being bonded for well-proportioned force on bumps. Die leveling and, sometimes, auto-focus are completed by aided complex optical devices other than imaging system in the bonders, where several different components must be alternated orderly in time axis. Firstly, just a set of imaging optical component, which used to merely complete traverse alignment and now could be used for the all three procedures, is designed and adopted in our method. Auto-focus is completed by estimating the sharpness of images and leveling by focusing for several locations covering the feature. As a result, the complexity of system is reduced; the resolution of system is also satisfying; time-consuming can be compensated to some extent by high-speed processors and algorithms. The optical system consists of a dual direction microscope system (with two 10X objectives), which can be inserted between die and substrate to provide visual images of both die and substrate. A 1/2" CCD (cell size, 8.6um x 8.3um) receives alternately images of the die and substrate by controlling the illumination (blue coaxial light). The resolving power of the optical system is up to 1.7um and the depth of view is not beyond 8um. Secondly, passive auto-focus method is discussed as an emphasis. The method determines the distance to the object by computer analysis of the image itself. Changing focal distance or the object distance is then responded to obtain the sharpest image. These two steps above are the sharpness calculation and the searching focus. After a review and comparison of the existing sharpness calculation methods, a grads-based sharpness calculation method for a row or a column of the feature images is suggested for our application object and small depth of view of optical system. Because the height of feature is beyond the depth of view, a large-numbered result of sharpness calculation doesn't mean an exact focusing on searching plane but perhaps on a mid-layer of the feature. The feature recognition is necessary and also could be solved through the change of grads. Lastly, a group of focusing results is obtained by experiment. Passive auto-focus precision can reach in near 1 micron, which could meet the need of die leveling. Except that, the limitations of the system are analyzed, and some potential improved thoughts are also discussed.