nonsingular is a horribly nonconvex constraint, and there is no good strategy
X+X'>=0 does not enforce X to be non-singular.
If you want to define X o be positive semidefinite, you have to define X as symmetric (which is default for square matrices) and then add X>=0. To get it positive definite, you add a lower bound, X>=margin*eye(n). Of course, positive definite is a much smaller set than the set of non-singular matrices