Abstract:
Spin Transfer Torque Magnetic RAMs (STT-MRAMs) are highly prospective for next-generation data storage due to their non-volatility, fast access time, high data retention capacity, 3-D scalability, low power consumption, and excellent compatibility with conventional CMOS technologies. The primary building block of an STT-MRAM is the magnetic tunnel junction (MTJ), which consists of two ferromagnetic layers separated by an insulating tunnel barrier layer. The static and dynamic characteristics of STT-MRAMs strongly depend on the intrinsic properties of each layer and the interaction between different layers of the MTJ. The impact of intra-cell magnetic coupling on STT-MRAM device-to-device vari- ability has been investigated based on experiments and simulations. Measured switching voltages of CoFeB/MgO STT-MRAMs are found to be directly corre- lated with offset fields extracted from magnetic measurements of the same set of devices. By spintronic and Monte Carlo simulations and statistical analysis incorporating stochasticity and variability of the STT-MRAMs, the origin of the experimentally observed correlation has been traced back to the stray field aris- ing from intra-cell magnetic coupling. It is shown that stray fields resulting from intracell and intercell coupling significantly impact the switching variability of STT-MRAMs devices. Such variations can be controlled or mitigated by exploit- ing the exchange coupling between interlayers of the device stack, employing a Fokker-Planck equation-based numerical model calibrated against the switching probabilities of measured devices. Tapering of the device sidewalls resulting from etching can also be utilized to control the variability of switching characteristics of STT-MRAMs. Write and Read error rates of STT-MRAMs have also been evaluated in the presence of stray fields. The results of the statistical simulation show that depending on the direction and magnitude of the stray field, READ and WRITE error rates preferentially increase in one switching direction while decreasing in the opposite direction. The pulse widths required to bring the WRITE and READ error rates to a nominal value of 10−6 have been estimated considering ±20% variation of the stray field. Depending on the switching di- rection, the percentage change in pulsewidth required during WRITE operations remains within ± 5%, but up to -50% to 150% change of pulsewidth is required during READ operations. Moreover, the percentage change in pulsewidth re- quired to maintain a nominal RDR in the larger devices can range from 50% to 290% or higher but to maintain nominal WER, it can range from 14% to 18%. So, READ operations of larger devices are more susceptible to intra- cell magnetic coupling-induced variability than WRITE operations. Magnetic coupling-induced variability studies on voltage-controlled magnetic anisotropy (VCMA)-assisted STT-MRAMs are also presented and compared with existing studies on STT-MRAMs as they offer faster switching and less write energy.