In diversity embedded coding, information streams are divided into two sub-streams with different priorities. If the optimal DMT performance of each coded stream can be achieved, then such code is said to be successive refinable. For the cases of SISO, SIMO, and MISO Rayleigh slow fading channels, Diggavi and Tse  had shown that superposition coding with successive cancellation receiver achieves successive refinability in these channels. However, such optimality might not be extended to MIMO channel due to the strictly sub-optimality of successive cancellation receiver. In this paper, we first provide an explicit construction of MIMO diversity embedded codes that is sphere decodable.We then show that the proposed code is approximately universal, if joint ML decoding is used, and hence extend the notion of successive refinability to general MIMO channels.