Process control for micro-lithography

Claim 1

A method of controlling a process to be applied to a patterned structure in a production run, the method comprising the steps of: (i) providing reference data including data representative of diffraction signatures corresponding to a group of different fields in a structure similar to said patterned structure in a production line and data representative of a control window for the process parameters corresponding to a signature representative of desired process results, said group of different fields being characterized by different process parameters used in the manufacture of these fields; (ii) providing an expert system in the form of an artificial neural network trained to be responsive to input data representative of a diffraction signature to provide output data representative of corresponding effective parameters of the process; (iii) applying optical measurements to at least one site on said patterned structure in the production line to obtain at least one diffraction signature of said patterned structure in the production line and generate data representative thereof; (iv) supplying the generated data to said expert system, which analyses the said data to thereby directly determine effective parameters of the process applied to said patterned structure in the production line; (v) analyzing said effective process parameters to determine deviation thereof from corresponding nominal values to thereby enable process control; and (vi) verifying said effective process parameters resulting from the neural network analysis by searching in a signature library that includes results of a plurality of measurements applied to the structure and indicative of process parameters corresponding to various signatures, to thereby ensure a high reliability score of the neural network, wherein said searching is aimed at finding the diffraction signature, which corresponding process parameters within said control window are the closest ones to those of the neural network results; said verifying comprising comparing the found reference signature with the signature measured in the patterned structure in the production line to determine whether a merit function between said found signature and said measured signature is sufficiently low, being thereby indicative of the high reliability score of the neural network results.

Claim 2

A method of controlling a process to be applied to a patterned structure in a production run, the method comprising the steps of: (i) providing reference data including data representative of diffraction signatures corresponding to a group of different fields in a structure similar to said patterned structure in a production line and data representative of a control window for the process parameters corresponding to a signature representative of desired process results, said group of different fields being characterized by different process parameters used in the manufacture of these fields; (ii) providing an expert system in the form of an artificial neural network trained to be responsive to input data representative of a diffraction signature to provide output data representative of corresponding effective parameters of the process; (iii) applying optical measurements to at least one site on said patterned structure in the production line to obtain at least one diffraction signature of said patterned structure in the production line and generate data representative thereof; (iv) supplying the generated data to said expert system, which analyses the said data to thereby directly determine effective parameters of the process applied to said patterned structure in the production line; (v) analyzing said effective process parameters to determine deviation thereof from corresponding nominal values to thereby enable process control; and (vi) verifying said effective process parameters resulting from the neural network analysis by searching in a signature library that includes results of a plurality of measurements applied to the structure and indicative of process parameters corresponding to various signatures, to thereby ensure a high reliability score of the neural network, wherein said searching is aimed at finding the diffraction signature, which corresponding process parameters within said control window are the closest ones to those of the neural network results; said verifying comprising comparing the found reference signature with the signature measured in the patterned structure in the production line to determine whether a merit function between said found signature and said measured signature is sufficiently low, being thereby indicative of the high reliability score of the neural network results.