Quick Wafer Alignment Using Feedforward Neural Networks

This paper proposes a wafer alignment method using a feed-forward neural network (FNN). A wafer is placed on a vacuum table and its misalignment is inspected by a frame grabber. The alignment of the wafer is repeatedly corrected by adjusting its kinematic positions until the misalignment becomes zero. The training set is composed of the misalignment and the xy theta compensation. Fifteen sets of misalignment data were measured by the conventional alignment method. The FNN has four normalized inputs as mark locations in the field of vision and three normalized outputs for the kinematic compensation. The macro and micro steps share the FNN by means of a multiplexer (MUX) and a scaler. The training rule was back-propagation (BP). The proposed method was applied to a wafer manufacturing machine and its alignment performance was compared with that of conventional methods. The alignment time was reduced by the FNN model which was especially efficient in the case of macroscopic alignment. Note to Practitioners-The idea of this study originated from a generalized alignment algorithm irrespective of kinematics and detectors. Wafer alignment systems are usually composed of mechanical parts and machine vision. Alignment algorithms can be derived from geometric analysis of mechanism and misalignment in common. However, an alignment algorithm should be redesigned whenever there are any structural changes in a machine and their performance must be verified by a lot of tests. So, we thought that designing of the algorithm will be unnecessary if we can train the relation between misalignment and kinematical movement using a FNN. This study suggested to obtain 15 training sets for xy theta alignment, but the positional deviation is less than 5%similar to 10% of the full range in an image. The training sets are applied to a simple FNN which has 4 inputs, 3 outputs, two hidden layers and 5 units in each layer. The numbers of the hidden layers and the units in this study is the least value found by experiments. The proposed method showed the better performance than those of other methods, but is applicable to mass production because of setup time and manpower for gathering the training sets. We will present an intelligent alignment algorithm to reduce this setup cost in future research