Analysis of the Vapor Phase Dynamics during iCVD: Free Radical Initiation by <i>tert</i>-Butyl Peroxide

While initiated chemical vapor deposition (iCVD) has been used in a wide range of fundamental and applied studies in the past two decades, little is known about the reactive vapor phase, leading to inconsistent treatment of the initiation process in the existing literature. Di-<i>tert</i>-butyl peroxide (TBPO) is commonly used as an initiator in iCVD by leveraging radical intermediates produced during thermal decomposition. Despite its widespread use, no models exist for TBPO decomposition in iCVD. To bridge this knowledge gap, we integrated the Pirani pressure gauge into the iCVD setup to measure the real-time vapor-phase composition in the reactor. After establishing the relationship between the Pirani gauge pressure and composition from batch data, we use it to quantify the fractional conversion of pure TBPO in the reactor in the continuous mode. We find that under iCVD conditions, TBPO fully decomposes to acetone and ethane following a first-order Arrhenius kinetics. This means that (i) the partial pressures of radicals are small (below the detection limit of the Pirani gauge) and (ii) there is a stoichiometric increase in the number of gas molecules in the reactor, which has not been accounted for in existing models. This causes the overestimation of monomer pressure, which is a key parameter in the current understanding of iCVD. We analyze the decomposition of TBPO in the presence of cyclohexyl methacrylate (CHMA) monomer suggesting that the monomer does not react appreciably in the vapor phase. To show how this model of the vapor phase is related to the deposition kinetics, we compare our data to the classical linear polymerization theory, which describes free radical polymerization of CHMA in solution and finds an acceptable fit. Notably, in Arrhenius plots of the deposition rate, there is a well-documented but unexplained decrease in the apparent activation energy at high filament temperature (>300 °C), which is predicted by our models.