This chapter proposes a silicon-based integrated DONN model, which mainly consists of three parts: the two-dimensional silicon-based SDEPM, the training model of the silicon-based integrated DONNs, and the phase weight mapping model. Among these, the two-dimensional silicon-based SDEPM serves as the foundation for the training model of the silicon-based integrated DONN. This is because the two-dimensional silicon-based SDEPM can analytically describe the propagation process of light in slab waveguides based on an SOI platform. With the analytical expression, it is possible to pre-train the key structure parameters of the silicon-based integrated DONN using algorithms such as gradient descent on a computer. Once the silicon-based integrated DONN key structure parameters are obtained, the phase weight mapping model can accurately map the parameters onto the silicon-based integrated DONN chip. Specifically, Sect. 2.1 will introduce the two-dimensional silicon-based SDEPM. Section 2.2 will introduce the training model of the silicon-based integrated DONN. Section 2.3 will discuss the phase weight mapping model. Section 2.4 serves as the summary of this chapter, providing an analysis and summary of each part of the proposed models in this chapter.