Biological systems are far more complex and robust than systems we can engineer today, and Because of its advantages of stability and specificity, DNA molecules have been used for the construction of nano-scale structures. With the development of DNA molecule self-assembly strategy, lots of programmable DNA tiles are constructed and used for solving NP problems. The tile assembly model, a formal model of crystal growth, is a highly distributed parallel model of natures self-assembly with the traits of highly distributed parallel, massive storage density and low power consumption. In the system, a function can be computed by deterministic assembly and identified two important quantities: the number of tile types and the assembly speed of the computation. Here a DNA self-assembly model for perfect matching problem of bipartite graph is demonstrated, and a 10-vertices bipartite graph is used as an example to illustrate this model. The computation is nondeterministic and each parallel assembly is executed in time linear in the input. The result shows that the successful solutions can be found among the many parallel assemblies, and it requires only a constant number of different tile types: 14.