Neural Network Modeling of Acid-Catalyzed Degradation of Photosensitive Polycarbonates

There is a growing need to reduce processing temperatures used in the manufacturing of microelectromechanical systems (MEMS). Polycarbonates, which can be used as sacrificial layers in the fabrication of MEMS devices (Reed, 2001), typically decompose at 200-300 degC. Adding a photoacid generator (PAG) to a polycarbonate makes it photosensitive and reduces its processing temperature to 100-180 degC. A study has been conducted to examine how various weight percentages (1-5%) of the photoacid generator, diphenyliodonium tetrakis-(pentafluorophenyl) borate (DPI FABA PAG), added to a solution of the polycarbonate, polypropylene carbonate (PPC) and gamma butyroacetone (GBL), effects the temperature and time of development of the photosensitive solution after exposure to UV radiation through a chrome mask and subsequent heating for a specific amount of time. Data from the experiment is used to model this process using neural networks. The inputs to the neural networks are initial solution thickness, PAG weight composition, and development time. The output of one of the networks is average residual remaining, and the output of the second network is development rate. To validate the neural network models, the root-mean-square (RMS) error is used as a performance metric. RMS errors on the order of 12% are achieved when comparing the outputs of the neural network models with test data