Recipe synthesis for PECVD SiO2 films using neural networks and genetic algorithms

Silicon dioxide films deposited by plasma-enhanced chemical vapor deposition PECVD) are useful as interlayer dielectrics for metal-insulator structures such as multichip modules. Due to the complex nature of particle dynamics within a plasma, it is difficult to determine the exact nature of the relationship between PECVD process conditions and their effects on critical output parameters. In this study, neural network process models are used in conjunction with genetic algorithms to determine the necessary process recipes to achieve novel film qualities. To characterize the PECVD process, SiO₂ films deposited in a plasma-Therm 700 series PECVD system under varying conditions are analyzed using a central composite experimental design. Parameters varied include substrate temperature, pressure, RF power, silane flow and nitrous oxide flow. Data from this experiment is used to train back-propagation neural networks to model deposition rate, refractive index, permittivity, film stress, wet etch rate, uniformity, silanol concentration, and water concentration. A recipe synthesis procedure is then performed using genetic algorithms, Powell´s algorithm, the simplex method, and hybrid combinations thereof to generate the necessary deposition conditions to obtain novel film qualities, including zero residual stress, 0% non-uniformity, 0% impurities, and low permittivity. Recipes predicted by these techniques are verified by experiment, and the performance of each synthesis method is compared