In some seismic processing steps such as reverse time migration (RTM), full waveform inversion (FWI), and rock physics modeling of seismic response, numerical solution of wave equation is commonly used. The usual method of solving the wave equation is finite difference, which increases the dimensions of the matrices involved and the calculation time due to the stability constraints for the grid size and time step. This issue is a deterrent to their use, especially for iterative processes. On the other hand, due to the efficiency of parallel computing using GPUs, they have been widely used in recent years. Here, we have introduced a physics-informed deep network that, in addition to parallel computations in matrix multiplication, is capable of simultaneously solving of independent shots which are mapped on mini-batches of the network inputs. So, decreases run-time by those remedies. Efficiency of the method was evaluated by applying it on synthetic models.