We introduce several endpoint algorithms that map real-time, in situ proces
s signals to a via etch process endpoint, Some of the mathematical techniqu
es include: Andrews plots (Fourier series), Chebyshev polynomials, Legendre
polynomials, wavelets, singular value decomposition, and neural networks.
We show that many of the techniques work to varying degrees of success for
a via etch process on 0.16-mum technology. Based on our observations from m
any lots of manufacturing wafers and experiments with all the endpoint meth
ods, we believe the Chebyshev polynomial area-time curves perform the best,
but this statement should be taken with a caveat. It is really best to emp
irically test the various methods for a given etch process to deduce the en
dpoint algorithm for that application.