We have developed new metrology to characterize and understand the behavior of nanoscale magnetic memory elements in disparate environments. For instance, we developed ways to monitor individual switching events in spintronic devices, perform ferromagnetic-resonance spectroscopy of individual devices, and quantitatively determine the roles that stray fields and exchange coupling play in hybrid spin-valve/Josephson-junction cryogenic memory. This work has direct relevance to companies developing memory for both conventional and superconducting supercomputers. We also developed new techniques to characterize the nonlinearities that cause phase-locking and phase noise in nanoscale magnetic oscillator arrays. These measurements are helping companies develop bio-inspired, non-Boolean computing architectures for low-power and approximate computing.
The Spin Electronics Group develops metrology to determine how spin currents can be generated and used to control and manipulate magnetization in new ways at timescales less than 1 ns. We provide and disseminate advanced high frequency measurements, analysis, and fabrication of nanoscale magnetic structures and materials to enable the development of novel spin-based devices. Focus areas include (1) the fundamental understanding of the interactions between spin and magnetic materials, superconductors, and materials with large spin-orbit scattering; (2) the nonlinear dynamics of both individual and interacting nanoscale magnetic systems; and (3) the role of thermal noise in nanomagnetic systems. We study how these effects can be controlled with applied spin currents and electric fields. Applications include nonvolatile magnetic data storage, non-Boolean computation architectures, active magnetic nanoscale devices, and novel radiofrequency communications.
Additional Technical Details
Spin-Transfer-Torque Magnetic Random-Access Memory
The ubiquitous dynamic random-access memory (DRAM) for computers must be refreshed every 50 ms or so and therefore requires power to retain information. A promising alternative to DRAM is a nonvolatile, radiation-hard, low-energy-loss, nanoscale, magnetic memory: spin-transfer-torque magnetic random-access memory (STT-MRAM). An STT-MRAM element consists of two ferromagnetic layers separated by a thin (about 1 nm) MgO layer, which allows the two layers to behave independently and contributes to the electrical transport through the device. The resistance of such a structure depends on the relative orientations of the magnetization of the two layers; it changes by roughly a factor of two when their magnetizations are switched between parallel (P) and antiparallel (AP), thus storing "0" and "1" bit states. The structures are fabricated so that the two ferromagnetic layers in the three-layer sandwich are asymmetric, which allows one of the layers to be switched freely while the other is fixed.
In conventional MRAM, switching between these states is achieved with applied magnetic fields. Spin-transfer torque, on the other hand, allows switching between the P and AP states to be achieved with low-loss, spin-polarized currents, which impart their angular momentum to the magnetization of the free layer in the film stack. We developed new methods to measure switching statistics for over one million events per device, which showed that many experimental devices do not necessarily switch reliably at the levels required by the microelectronics industry. We found that certain devices (or bits) behave differently than their seemingly identical counterparts, and we developed a frequency-domain method to quickly identify these individual outliers.
- R. Heindl, W. H. Rippard, S. E. Russek, and M. R. Pufall, "Time-domain analysis of spin-torque induced switching paths in nanoscale CoFeB/MgO/CoFeB magnetic tunnel junction devices," J. Appl. Phys. 116, 243902 (Dec. 2014); doi: 10.1063/1.4905023.
- E. R. Evarts, R. Heindl, W. H. Rippard, and M. R. Pufall, "Correlation of anomalous write error rates and ferromagnetic resonance spectrum in spin-transfer-torque-magnetic-random-access-memory devices containing in-plane free layers," Appl. Phys. Lett. 104, 212402 (May 2014); doi: 10.1063/1.4879847.
Ferromagnet-Based Josephson-Junction Memory for Superconducting Computing
Large-scale computing will increasingly be used to model complex systems such as weather and climate, biology, astronomy, finance and economics, weapons simulations, and cryptography. However, completely new computer architectures will be required to achieve high clocking speeds and low energy dissipation. The Intelligence Advanced Research Projects Activity (IARPA) is considering superconducting computing as a promising technology for large-scale computing and has started a multiyear program to investigate its viability. One major hurdle is the lack of a nonvolatile memory element that is scalable (down to nanometers), fast (on the order of nanoseconds), and energy efficient (consuming femtojoules).
In collaboration with the Quantum Voltage Group, we are incorporating nanoscale, nonvolatile, ferromagnetic memory elements similar to STT-MRAM devices into Josephson junctions (JJs) for superconducting supercomputing. This combines two disparate disciplines that are generally mutually exclusive since spin polarized currents are generally not compatible with Cooper-pair electron transport in superconductors. Because of this, the field has a significant need for basic, cross-disciplinary metrological advances. For instance, the JJ critical current in these structures depends on the state of the memory element (P or AP), and hence the critical current can be used to interrogate the memory state. However, the microscopic magnetic reasons for this were, until recently, unclear. Through a series of careful measurements we determined that the change of the JJ critical current depended on the ferromagnetic exchange interactions in the memory element and not stray magnetic fields. This is a fundamental advance in this emerging field because the exchange interaction is scalable to the nanoscale whereas the effects of stray fields are not. We recently showed that the same energy-efficient STT effect exploited in room temperature memory can also be utilized in their superconducting counterparts.
- B. Baek, W. H. Rippard, M. R. Pufall, S. P. Benz, S. E. Russek, H. Rogalla, and P. D. Dresselhaus, "Spin-transfer torque switching in nanopillar superconducting-magnetic hybrid Josephson junctions," Phys. Rev. Appl. 3, 011001 (Jan. 2015); doi: 10.1103/PhysRevApplied.3.011001.
- B. Baek, W. H. Rippard, S. P. Benz, S. E. Russek, and P. D. Dresselhaus, "Hybrid superconducting-magnetic memory device using competing order parameters," Nature Commun. 5, 3888 (May 2014); doi: 10.1038/ncomms4888.
Nanoscale Oscillators for Non-Boolean Logic
We work with the Defense Advanced Research Projects Agency (DARPA) to support their efforts to develop spin-based devices for low power, non-Boolean computing. Both industry and the government are investigating alternative computer architectures to efficiently solve the "best-match" pattern-recognition problem, which has applications in the search of very large databases and the real-time analysis of high-definition video streams. Pattern matching requires the calculation of the "distance" between a test item and each item in a database, and accepting the closest matches. The high precision of conventional complementary-metal-oxide-semiconductor (CMOS) logic wastes energy and time when performing such imprecise calculations. A bio-inspired alternative to this calculation is to use arrays of coupled, phase-locking, nonlinear oscillators to perform the distance calculation in a quasi-parallel fashion, with the relative phases of the oscillators representing the degree of match. In effect, the physics of the phase locking performs the distance "calculation."
To this end, we are developing new methods understand the physics of novel, spin-based, nonlinear, nanoscale oscillators, which are promising candidates for the large arrays needed in such computation schemes, due to the oscillators' nanoscale size, greater than 10 gigahertz frequencies, and inherent nonlinearities. We are exploring ways to efficiently couple these oscillators, quantifying the nonlinearities that mediate the coupling, and developing metrology to characterize their phase behavior and understand the time scales over which locking occurs. We have also started to apply these techniques to measure the properties of stochastic, spin-based, pulsed oscillators for use as potentially more efficient relaxation oscillators for neural networks based on "memristors."
- W. H. Rippard, M. R. Pufall, and A. B. Kos, "Time required to injection-lock spin torque nanoscale oscillators," Appl. Phys. Lett. 103, 182403 (Oct. 2013); doi: 10.1063/1.4821179.
- M. R. Pufall, W. H. Rippard, S. E. Russek, and E. R. Evarts, "Anisotropic frequency response of spin-torque oscillators with applied field polarity and direction," Phys. Rev. B 86, 094404 (Sep. 2012); doi: 10.1103/PhysRevB.86.094404.
- G. Csaba, M. R. Pufall, W. H. Rippard, and W. Porod, "Modeling of coupled spin torque oscillators for applications in associative memories," Int. Conf. Nanotech., Birmingham, U.K., Aug. 2012, IEEE, New York; doi: 10.1109/NANO.2012.6322201.