New approach to global alignment in IC manufacturing based on a neural network model

One of the most crucial emerging challenges in lithography is achieving rapid and accurate alignment under a wide variety of conditions brought about by the different overlying films occluding the marks. The problem is exacerbated by planarizing processes such as chemical mechanical polishing that reduce the topographical contrast used to view the marks and by distortion of the wafer and of the stage. Thus an effective learning process is needed to rapidly acquire the best possible positional information from an array of the marks across the wafer. In this paper, a neural network model for global alignment is described. This method has significant advantages including robustness to measurement noises and other random disturbances. Stage distortion can also be easily included in the model, resulting in more accuracy in the presence of a known repeatable distortion. The process can learn to identify and ignore any alignment marks that yield significantly erratic signals. Different approaches for training of the system will be discussed. A few common distortion functions will also be used to test the model. Preliminary simulation results show very accurate alignment in the presence of noise and distortion. In addition the simulations show that stage distortion has no effect on the results. In other words, increasing stage distortion amplitude will not change the positional accuracy in a given time.