Wafer fabrication is a complex manufacturing system. Using complex network models to describe the correlation between parameters in the wafer fabrication process is effective in finding the critical influencing factors related to the causes of defects. However, many couplings, redundancies, and spurious associations exist in the wafer fabrication complex network, and it is challenging to clarify the direct and spurious associations in the network. Therefore, this paper proposes a data-driven framework of direct correlation identification for wafer fault detection. Firstly, the correlation model between the parameters based on the Copula function is designed for the nonlinear relationship and the non-normal distribution characteristics. Then the goodness-of-fit test method based on the Euclidean distance theory is designed to check the fitting effect of the proposed model. After that, the complex network correlation graph of parameters in the wafer fabrication process is constructed with each parameter as a node and the correlation coefficients between parameters as joint edges. The network deconvolution method identifies and removes the spurious correlation, and the critical factors directly related to the causes of low yield are found. Finally, a case study using actual data from semiconductor wafer fabrication systems is conducted to verify the effectiveness of the proposed method.