Magnetic tunnel junctions are a key component in non-volatile memory technologies such as magnetic random access memory (MRAM). Typical two-terminal MRAM write speeds have been limited to the nanosecond regime. Currently, sub-nanosecond switching has been achieved in three-terminal structures relying on spin-orbit torque switching mechanisms. However, using this mechanism increases the footprint and number of transistors needed, ultimately limiting scalability. Here, we demonstrate that it is possible to achieve reliable sub-nanosecond switching in a two-terminal design by utilizing a Double Spin Torque magnetic tunnel junction. In this design, a second reference layer is added to the standard two terminal MTJ. Using a non-magnetic spacer, we are able to do so without sacrificing magnetoresistance like in double MTJ designs. Figure 1 shows the error rates for a single device, plotted using a normal quantile scale (using the absolute value of the standard deviation from the fifty-percent switching current density), measured at -40 C, 25 C, and 85 C, for pulse widths from 225 ps to 10 ns. We demonstrate 100% successful switching for 1e6 write attempts using 225 ps wide write pulses, over the temperature range -40 C to 85 C. Further, we show reliable sub-nanosecond switching across a few hundred devices. 1e10 write-endurance testing on a handful of devices also demonstrates that reliability can be achieved as well. Lastly, we compare our Double Spin Torque magnetic tunnel junction to three-terminal spin-orbit torque MRAM. We find that with our design, a 3-10X reduction in write power consumption can be achieved.

References