Abstract
Self-assembly of block copolymers (BCPs) is an alternative patterning technique that promises high resolution and density multiplication with lower costs. The defectivity of the resulting nanopatterns remains too high for many applications in microelectronics and is exacerbated by small variations of processing parameters, such as film thickness, and fluctuations of solvent vapor pressure and temperature, among others. In this work, a solvent vapor annealing (SVA) flow-controlled system is combined with design of experiments (DOE) and machine learning (ML) approaches. The SVA flow-controlled system enables precise optimization of the conditions of self-assembly of the high Flory–Huggins interaction parameter (χ) hexagonal dot-array forming BCP, poly(styrene-b-dimethylsiloxane) (PS-b-PDMS). The defects within the resulting patterns at various length scales are then characterized and quantified. The results show that the defectivity of the resulting nanopatterned surfaces is highly dependent upon very small variations of the initial film thicknesses of the BCP, as well as the degree of swelling under the SVA conditions. These parameters also significantly contribute to the quality of the resulting pattern with respect to grain coarsening, as well as the formation of different macroscale phases (single and double layers and wetting layers). The results of qualitative and quantitative defect analyses are then compiled into a single figure of merit (FOM) and are mapped across the experimental parameter space using ML approaches, which enable the identification of the narrow region of optimum conditions for SVA for a given BCP. The result of these analyses is a faster and less resource intensive route toward the production of low-defectivity BCP dot arrays via rational determination of the ideal combination of processing factors. The DOE and machine learning-enabled approach is generalizable to the scale-up of self-assembly-based nanopatterning for applications in electronic microfabrication.