A magnetic tunnel junction (MTJ) that avoids electrical shorts and has improved data retention is disclosed. An uppermost capping layer has a first sidewall that is coplanar with an interface between outer oxidized portions and a center ferromagnetic portion of a free layer (FL) that has a FL width (FLW). A dielectric spacer is formed on the first sidewall and oxidized outer FL portions. The pinned layer (PL) has a width (PLW) substantially greater than FLW, and a second sidewall thereon is formed by a self-aligned etch using the dielectric spacer and capping layer as an etch mask. A sidewall layer may be formed on the second sidewall and dielectric spacer but does not degrade MTJ properties since the sidewall layer does not contact the FL and PL center portions responsible for device performance. PL width>FLW ensures greater capability for data retention especially for FLW