Modeling of dry development in bilayer-resist process for 140-nm contact hole patterning

The dry development process using SO/sub 2/-containing plasma for 140-nm hole patterning in a bilayer-resist system was described by a neural network technique in conjunction with a statistical modeling method. The SO/sub 2/ addition into O/sub 2/ plasma gave the enhanced anisotropy of 140-nm holes. However, defect (polymeric residue) generation at wide-open areas has been observed as a side effect, driving the need for an accurate model to optimize the process. Empirical nonlinear modeling using neural networks was conducted on the basis of data obtained from statistically designed experiments to investigate the effects of relevant parameters (chamber pressure, plasma-source (TCP) power, bias (bottom power), and O/sub 2/-SO/sub 2/ flow rate ratio) on defect (polymeric residue) generation and the anisotropy. The authors have found that the reduction of chamber pressure to 10 mT allows independent control of anisotropy and the defect generation. In addition, neural network modeling in a Bayesian framework gives superior accuracy (R/sup 2/: /spl ges/0.9) over statistical modeling using limited experimental data and requires fewer training data to accurately describe the process.