Full Waveform Inversion (FWI), as an advanced method for velocity reconstruction, stands out in the current stage for its high-precision results. Typically, FWI employs the L2 norm of the residual between observed and synthetic seismic records as the objective function, iteratively updates the velocity model to make the objective function get small. However, due to the limited amount of observation data compared to unknowns, the inversion problem is ill-posed. Applying regularization constraints to the objective function is crucial for mitigating the ill-posedness. This paper introduces an exponential regularization term to the L2 objective function, thereby partially alleviating the ill-posedness. This method introduces layer boundaries during the updating process, resulting in a more favorable blocky structure of the model. Comparison to conventional L1 regularization, the method proposed in this paper does not need to address the problem of non-differentiability of the L1 norm regularization term with respect to the model parameter when computing the gradients, while directly calculating the gradient of the regularization term with almost no additional computational cost. The application testing on synthetic data of the salt dome model and Marmousi model demonstrate that the proposed method can more accurately depict velocity models, which validates its effectiveness.