We prove that for analytic functions in low dimension, the convergence rate of the deep neural network approximation is exponential.