Reliability of normal-state current–voltage characteristics as an indicator of tunnel-junction barrier quality

We demonstrate that one of the most commonly used criteria to ascertain that tunneling is the dominant conduction mechanism in magnetic tunnel junctions—fits of current–voltage $\mathrm{(I–V)}$ data—is far from reliable. Using a superconducting electrode and measuring the differential conductance below $T_c,$ we divide samples into junctions with an integral barrier and junctions having metallic shorts through the barrier. Despite the clear difference in barrier quality, equally reasonable fits to the $\mathrm{I–V}$ data are obtained above $T_c.$ Our results further suggest that the temperature dependence of the zero-bias resistance is a more solid criterion, which could therefore be used to rule out possible pinholes in the barrier.