Modeling magnetically enhanced RIE of aluminum alloy films using neural networks

Aluminum (Al) and its alloy films are widely used for fabricating VLSI interconnections. The discharge behavior of a magnetically enhanced reactive ion etching (MERIE) of Al(Si) has been modeled using neural networks. A 2^6-1 fractional factorial experiment was employed to characterize etch variations with RF power, pressure, magnetic field and gas mixtures of Cl₂, BCl₃, and N₂. Responses of an Al(Si) film etched in a chlorine-based plasma include etch rate, selectivity to oxide, anisotropy and bias of critical dimension (CD). The generalization accuracy of the models, measured by the root-mean squared error (RMS) on a test set, are 285 Å/min for etch rate, 5.58 for oxide selectivity, 0.08 for anisotropy, and 3.82 Å/min for CD bias. Al(Si) etch rate was found to be chlorine-dependent with significantly affected by magnetic field variations. For the other etch responses, RF power was dominant. Gas additives such as BCl₃ and N₂ were seen to have conflicting effects on etch outputs. Predicted Al(Si) etch behaviors from neural process models were in qualitative good agreement with reported experimental results