• Model-based precursor supply for stable vapor delivery in thin-film processes. • A simple physical model of a tank in a baking vaporization module. • The liquid level in a tank is estimated using the model with a Kalman filter. • Noise in mass flow rate and pressure is reduced by 45.7% and 44.6%, respectively. • Sufficiently simple computation for real-time implementation. Stable delivery of a vaporized liquid precursor is critically important for thin-film forming processes in semiconductor manufacturing. This study proposes a new method for model-based liquid precursor supply in a baking vaporization module to achieve stable vapor delivery. A physical model of the tank in the module is constructed to capture its dynamics. A state-space representation describing the internal state of the tank is formulated. To estimate the internal tank state while maintaining consistency between measurements and the model, a Kalman filter is employed. The liquid precursor is supplied based on the liquid level calculated from the model. We experimentally applied this method to the vaporization of hexachlorodisilane, which is used for silicon nitride thin-film forming processes. The results showed that peak-to-peak noise levels in the delivered vapor mass flow rate and downstream pressure were reduced by 45.7% and 44.6%, respectively, compared with a conventional level-sensor-based supply. With this method, the spike noise that can cause film failure in such processes is suppressed. The computation is sufficiently simple for real-time implementation. This approach introduces model-based control with a Kalman filter to vapor precursor supply and improves delivery stability without additional equipment.
Hayashi et al. (Sun,) studied this question.