Resist coating optimization on 8-in. deep-UV litho cell: modeling and application to 0.25-μm technology

One of the main problem for 0.25 micrometers features will be the depth of focus; Deep UV source will be necessary, but anyway DOF will be in the one micron range. So will it be possible to have CD in specifications all over the field? Many parameters have to be checked and monitored but the photoresist thickness on wafer is an important part due to the interferences franges created inside the photoresist during exposure. In order to reduce this phenomenon, we are used to choose the thickness on the flat part of the swing curve, for maximum or minimum coupling. As the wavelength is being lower, from 365 nm on i-line steppers to 248 nm on Deep UV steppers, this flat part length is reduced, so to obtain the same CD's repeatability the photoresist thickness uniformity must be improved in Deep-UV. This paper presents an effective method to obtain a Deep UV photoresist uniformity allowing 0.25 micrometers CD's in specification on the whole 8' wafer. A great number of process parameters may affect the resist coating uniformity, so we have decided to use the Designs Of Experiments to solve such a complicated problem with a minimum number of experiments. Due to the centrifuging process used to coat the resist on the wafer, the main uniformity default that can appear is a radial default. So we have chosen a combination of Taguchi and Response Surface Methods, in order to minimize the thickness variations versus the radius. First of all, to evaluate the effects of the 8 process parameters on this response, we have performed a screening design of 16 experiments. A statistical analysis has shown that the thickness variations depend on only 5 parameters. Then a Response Surface design of experiments has been achieved, with 27 experiments. A very accurate modelization of the uniformity default has been carried out from this design, and the optimization has led to a thickness standard deviation of only 3.3 angstrom on the whole wafer. Moreover the quality of the modelization, predictive in the complete domain, will allow to work out further problems.