<p>Double-sided chemical mechanical polishing (DSCMP) is a mainstream processing method that simultaneously achieves low surface roughness on both sides of ultra-thin crystal substrates. However, due to the complexity of the polishing process, research on substrate surface roughness has mainly relied on experimental studies, and the processing adopts a “measurement-processing” iteration, which reduces processing efficiency. In this study, based on probability theory and the Monte Carlo stochastic method, a surface roughness model for double-sided polished substrates was established. Notably, this model considers the relationship between formation of micromorphology of the substrate and micromorphology of the polishing pad as well as the particle size of the grains for the first time. The surface morphology of the substrate during the double-sided polishing process was simulated under varying processing parameters, enabling subsequent calculation of the surface roughness (Sa). The experimental results of the surface roughness align with the trends observed in the simulation results, thereby validating the correctness and validity of the model. On this basis, the influence of processing parameters on surface roughness was revealed, and the processing parameters were optimized to enhance machining efficiency while achieving lower surface roughness. This study provides theoretical support for substrate surface roughness processing and control in the fields of lasers and semiconductors.</p>