Abstract
•Functional network predictive model for film characterization is constructed.•154 experimental data sets were used for training and testing.•It accurately predicts poly-SiGe film thickness, resistivity and deposition rate.•The models can be used to proactively control film properties.
Optimal performances of thin film devices such as those of micro/nano-electromechanical systems like sensors and actuators are possible with accurate and reliable characterization techniques. Such techniques can be enhanced if predictive models are constructed and deployed for production and monitoring. This paper presents functional networks as a novel modeling approach for rapid characterization of thin films such as thickness, deposition rate, resistivity and uniformity based on 8 deposition parameters. The functional network (FN) models were developed and tested using 154 experimental data sets obtained from ultrathin polycrystalline silicon germanium films deposited by Applied Materials Centura low pressure chemical vapour deposition system. The results showed that the proposed FN models perform excellently for all the outputs with minimum and maximum regression coefficients of 0.95 and 0.99, respectively. To further demonstrate the robustness of these models, several trend analyses were conducted. The performance statistics indicates that the mean percentage error for the model, based on the deposition rate, lies between 0.3% and 0.8% for silane, germane, diborane flow rates and pressure. For these deposition variables, the probability or p-value at a significance level of 0.01 implies that no significant difference exists between the means of the predicted and the measured values. The results are further discussed in light of physics of the CVD process.