CVX’s Successive Approximation method was invoked to deal with functions which invoke the exponential cone. It is an unreliable method, and has been superseded by Mosek, which has a native exponential cone capability, or if that is not available, CVXQUAD.
If you have access to Mosek, use it as solver. Otherwise, follow the instructions at CVXQUAD: How to use CVXQUAD's Pade Approximant instead of CVX's unreliable Successive Approximation for GP mode, log, exp, entr, rel_entr, kl_div, log_det, det_rootn, exponential cone. CVXQUAD's Quantum (Matrix) Entropy & Matrix Log related functions
It may or may not also the case that the input data your problem is badly scaled, i.e., non-zero numbers which are not within a small number of orders of magnitude of one. If so, that is bad no matter what solver or solution method is used.