Thermal stability and switching distributions of MgO magnetic tunnel junction (MTJ) devices, that are being developed for spin-torque-transfer random access memory (STT-RAM), have been modeled using single domain simulations. The simulations incorporate realistic current-voltage characteristics, a field-like torque term and thermal heating. These elements are required to allow the single domain model to fit data from standard STT-RAM devices. Simulations were performed with a stochastic thermal field with 10,000 repetitions for each value of applied current and pulse duration. For a typical STT-RAM device with dimensions of 50 nm x 150 nm, at room temperature, with a 10 ns pulse width, the write error rate (WER = 1- switching probability) falls off at a rate of 40 mV/decade. To achieve a write error rate below 10-9, which is a typical value for current memory, the switching voltage must be 360 mV above the intrinsic switching voltage. The direction of the spin polarizer was systematically varied and it was found that switching probability decreases as the polarizer is moved off the device easy axis. The fall off in WER with pulse amplitude remained roughly constant.

 

The simulations are compared to data from devices consisting of a Co40Fe40B20 free layers, 1.1 nm MgO tunnel barrier, and a synthetic antiferromagnetic fixed layer. The average resistance area product (RA) and tunneling magnetoresistance (TMR) are RA=5 Ohm micrometer2 and TMR=150%, respectively. The devices were patterned using e-beam lithography and ion beam etching into ellipsoids of sizes ranging between 50 nm x 150 nm to 100 nm x 300 nm (all with aspect ratios 1:3). The devices were imbedded in a high frequency coplanar wave guide structure to allow high-speed switching and high-frequency thermal ferromagnetic resonance (FMR) measurements. The high speed switching measurements confirmed that simple thermally activated switching modelscannot accurately fit the measured data. The single domain simulations, with a field like torque term and thermal heating, can replicate several key features of the measured switching distribution. The thermal FMR data, however, often show a complex mode structure indicating that the real devices deviate considerably from simple single domain behavior.

 

These results indicate that slow fall off of the WER may be a key problem for STT-RAM technology. Within the single domain model, the slow fall off of the WER is intrinsic. To obtain a more rapid fall off in the WER, micromagnetic design concepts must be employ to prevent the device from accessing low torque configurations during the switching process.