Research on a new method for optimizing surface roughness of cavitation abrasive flow polishing monocrystalline silicon

Jun Zhao,Rui Wang,Enyong Jiang,Shiming Ji
DOI: https://doi.org/10.1007/s00170-021-06667-6
IF: 3.563
2021-02-10
The International Journal of Advanced Manufacturing Technology
Abstract:Compared with the traditional abrasive flow, cavitation abrasive flow can effectively improve the kinetic energy and motion randomness of abrasive particles during the polishing process without changing the polishing pressure, thus reducing the scratch and damage caused by hard and large particles on the surface and subsurface of monocrystalline silicon, and efficaciously improve the surface quality of polished silicon wafer. Polishing parameters have a significant influence on the surface roughness of abrasive flow polished workpiece. However, it is more complex to optimize the polishing process parameters for cavitation abrasive flow. The traditional process optimization methods, such as the full factorial experiment and Taguchi experiment, are discrete and time-consuming, making it difficult to meet the requirements of cavitation abrasive flow polishing parameters optimization. In order to solve the above problems, based on cavitation rotary abrasive flow polishing (CRAFP) system, a new process parameter optimization method is proposed in this paper: samples are generated by Taguchi experiment, and then a model between the surface roughness of silicon wafer and the factors affecting the surface roughness is established using backpropagation (BP) neural network. Finally, the optimal process parameters are found by genetic algorithm based on this model. The experimental results show that the optimum process parameters are 0.7 um SiC abrasive particles, 12% concentration, 6 nozzles, 0.8 mm from the tool surface to the workpiece, and 0.45-MPa inlet pressure. Under this condition, the average surface roughness (<i>R</i><sub><i>a</i></sub>) of the workpiece is reduced from 35.44 to 3.43 nm, improved by 90.32%. The final optimization effect is proved superior to the single Taguchi method optimization and genetic algorithm coupling with response surface model.
What problem does this paper attempt to address?