The invention relates to a method for designing a nonlinear corrector of a true random number generator Construct a type of corrector function with n input and m output. The technical solution is to select a set of bases in the finite field and establish an isomorphic mapping from to a [n ‑ m, m, t] equidistant linear code. The component functions of the corrector are constructed so that each component function is cascaded by 2m n-m element affine functions. The coefficients of these affine functions are the same as the 2m in the [n-m, m, t] equidistant linear code. There is a one-to-one correspondence between vectors, and the constant term of each n-m element affine function in which the i-th component function is cascaded is equal to the value of the i-th input symbol. Using this design method, the nonlinear corrector obtained has an algebraic order of m+1 and an elastic order of t-1. It has the current optimal correction order of t, which is larger than the functional elastic order. The corrector obtained by the invention has the ability to resist linear attacks, related attacks, and algebraic attacks, and has broad application prospects in cryptography and stochastic simulation.