Motivation and Objective: Concussion due to traumatic head impact is a major health problem worldwide. The brain is a nonlinear viscoelastic soft solid; where large deformation produces shear shock waves [1] that require real-time predictions for deployability in detecting concussion. Physics informed neural networks make instant predictions once trained, but if the initial condition, source term, equation coefficients, etc., change then it must be trained again. Physics informed neural operators give real-time predictions to a family of solutions so that the trained model can be used with any initial condition, i.e., type of head impact, or equation coefficients, i.e., viscoelastic material parameters. Statement of Contribution: We implement and compare two algorithms, i.e., physics informed neural operators, to learn three different initial boundary value problems with a mapping between initial condition and solution. The standard neural operator architecture is known as DeepONet [2], as shown in Figure 1. We implement an architecture with a modified multilayer perceptron in the branch and trunk [2], along with other changes in the algorithm, which is our contribution to the methodology. To the best of our knowledge, there is currently no physics informed neural operator models of cubic nonlinearity in viscoelastic media. We solve a novel operator learning problem, but improving the methodology is a pre-requisite because neural networks struggle to model shear shock fronts. Results and Conclusion: We have three problems with increasing order of nonlinearity to compare the different physics informed neural operators: advection equation (linear), viscous Burger’s equation (quadratic), viscoelastic shear shock waves (cubic). Our model outperforms the standard physics informed DeepONet in each case, however, this is more noticeable as the problem nonlinearity increases.