Abstract
Fast, direct correction kernel generation by neural network application was shown. Estimation of the proximity effects by experimental method is used. Neural processor consists of two dedicated neural networks. Training set for Network 1 is based on convolutional Fourier pre compensation method with optimal filtering over the test structure MAT1. Second step of correction - ORG mode - made by Network 2 is a multivariable function approximation in which exact weight vector for arbitrary chosen technological parameters - proximity parameters - in prepared for actual corrector - CORR - mode of Network 2. Learned in such way neural processor is ready to correct for any patterns for any proximity function in almost real time. Modelled comparison of analytical correction method for real sub-quarter micrometer design rules chip with proposed here assure shortening of the process from 6 years to 17 hours. Some further improvements of the hardware solutions of the neural processor are also suggested.
Published Version
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