Finite element method (FEM) is widely used in the boundary value problems (BVP) in engineering practice. However, implicit FEM usually come to the convergence issues when dealing with some specific problems. Encouraged by recent success of physics-informed deep-learning method as PDE solvers [1-2], a physics-informed DL-aided FEM (PIDL-FEM) framework is developed by integrating deep learning into FEM for boundary value problems in this study. PIDL-FEM converts the algebraic equations solving problem into optimization problems, which avoids the convergence issues. In PIDL-FEM, the sampling dataset depends on spatial discretization by finite elements. The DL will represent the relationship between coordinates and displacement. For different coupling scheme proposed in this study, the strain will be estimated through autograd operator in DL or conventional displacement-strain matrix in FEM. The forces of nodes derived by stress also can be evaluated by means of either autograd operator in DL or conventional scheme in FEM. A physics-informed loss function then can be established in terms of governing equations in FEM. By minimizing the loss function, the DL model intend to predict the corresponding displacement field to the given boundary and initial conditions. Both strong form and weak form of governing equations are utilized and discussed on the computational accuracy and efficiency of PIDL-FEM. According to the order of autograd operators, the weak forms of PIDL-FEM can further be divided into 0-order weak form, 1-order weak form, and 2-order weak form. The result show that the 1-order weak form PIDL-FEM takes advantages in both computational accuracy and efficiency.