TDI cameras, image processors team up for wafer auditing
To meet wafer manufacturers` demands for a faster, cheaper, and more accurate means of locating and identifying production problems, vision-inspection-system suppliers are developing integrated configurations that can image a wafer at high speeds and also locate pattern and contaminant defects. However, wafer-manufacturing goals are not merely to locate contaminants, pattern defects, and tag defective die. Rather, their objective is to optimize the process of building wafers.
In today`s fab, manufacturers require wafer-auditing systems that can measure the results of the layering, patterning, and doping processes for each wafer. To meet these needs, Techne (Albany, OR) has developed the Quicksilver system series, which uses off-the-shelf time-delay integration (TDI) cameras and image-processing boards. To image the wafers at high production speeds, the system incorporates a TDI camera from Dalsa Inc. (Waterloo, Ontario, Canada) to capture images.
The TDI sensor contains a linear array of 2048 elements installed in 96 adjacent columns that are internally coupled together and sequenced by external timing logic. The array captures a line of image data over a period of time and then transmits these data to several Oculus F/64 ISA-based PC capture and image processing boards from Coreco Inc. (St. Laurent, Quebec, Canada).
During operation, each element across the array is exposed simultaneously, and each column is exposed sequentially over a brief time interval. As the wafers are continuously moved passed the TDI camera, the charge developed in each element of the TDI column is transferred onto the next column in synchronization with wafer motion. The charge collected from the light sensed at the wafer is summed with the light collected at the wafer`s new position. In this way, the wafer image features are collected over a period of time.
On the F/64 board, TMS34020 graphical signal processors clock in 8-bit wide RS-422 digital data from each tap of the TDI camera, and place the data stream in a hardware first-in, first-out buffer. These data are then clocked into the frame buffer where it can be accessed by both the graphics signal processor (GSP) and the TMS32040 digital signal processors (DSPs). In addition to transferring image data into the frame buffer, a GSP-based line-acquire software program detects particular features in the image data by finding features of interest.
Once an approximate location of a feature is detected, adjacent pixels are placed into a special memory array, and the DSPs perform both edge detection and die-to-database comparisons. While the GSP is detecting features and storing arrays of image data, the DSPs simultaneously read the already stored lines to perform such functions as template matching.