Current-driven domain wall motion and magnetomemristance in FeRh-based nanostructures

Lead Research Organisation: University of Leeds
Department Name: Physics and Astronomy


This project will study current-driven motion of antiferromagnetic/ferromagnetic (AF/FM) domain walls (DWs) in FeRh-based nanostructures. This will both elucidate the fundamental physics of the phase transition and also explore the potential for (magneto-)memristor devices, based on our prior demonstration of temperature/field controlled DW motion in an FeRh epilayer with a doping gradient. Memristors are devices suitable for ultradense non-volatile memory and also show many of the characteristics of an artificial synapse, opening the door to novel neuromorphic memory and logic architectures that promise enhanced functionality and low energy operation in future generations of ITC hardware.

To achieve this goal we first need to know what doping materials and densities are required to control the AF-FM phase transition temperature, provide appropriate hysteresis in the transition (to store information), and largest possible resistivity change (for readout) between the AF and FM phases, as well as what parameters (materials and densities) would describe an ideal doping gradient. Next, we will need to establish how small a magnetic nanostructure can be formed from FeRh and retain a suitable AF/FM transition, and the precise conditions and requirements that permit the smallest nanostructures to be stable. Then it will be necessary to establish the current densities needed to move the domain walls that separate ferromagnetic and antiferromagnetic regions in the phase-separated regime of FeRh. Last, we will need to demonstrate a memristive action under current-driven domain wall motion in a nanostructure with a suitable doping gradient.

Our project will combine state-of-the-art magnetic materials growth, characterisation, direct imaging of these novel DWs, and device fabrication and test, taking us from basic materials development to a fully operational magneto-memristor prototype nanostructure. We will begin by sputter-depositing uniformly- and gradient-doped FeRh epilayer materials and measuring their magnetic and magnetotransport properties, which will tell us the dopant materials and doping levels needed to achieve optimal memristive action. We will then fabricate nanostructures down to the few 10s of nm scale from the optimally doped FeRh layers, either as individual nanostructures (for microscopy) or as large-scale arrays of nanostructures (for magnetometry), which will reveal the minimum size at which a phase transition that is useful for memristive action is retained. Next, we will carry out world-first experiments on current-driven AF/FM domain wall motion in lateral FeRh nanowires, using magnetic microscopy to track the motion of DWs in response to current pulses, revealing the efficiency of spin injection for DW motion and the degree and nature of DW pinning arising from different sources. We will then pattern nanopillars from gradient-doped layers and study DW motion driven by a vertical current in a prototype memristor device, using both magnetotransport measurements and direct imaging of the internal structure of the device. The key result will be the magneto-memristance as a function of device size, temperature, and magnetic field.

The results we shall obtain will not only lead to high impact publications and conference presentations by shedding light on the still poorly understood fundamental problem of the nature of the phase transition in FeRh, but also reveal the performance characteristics of the world's first magneto-memristor, developing potentially valuable knowhow in the field of novel neuromorphic computer architectures.

Planned Impact

The work we shall carry out will have impact in two areas identified by BIS as among the Eight Great Technologies that 'support UK science strengths and business capabilities': these are "Advanced Materials" and "Big Data and Energy Efficient Computing".

FeRh is a remarkable functional material that responds to doping, stoichiometry, strain, temperature, and magnetic fields. It has potential applications in heat assisted magnetic recording, energy harvesting, and smart sensors as well as the memristive devices that we propose here. It also forms a model system for the study of magnetocaloric materials. The cost of rhodium precludes its widespread use in bulk form, but nanoscale applications can be realistically targeted: other precious metals such as platinum and iridium are used in the magnetic recording industry, for instance. Nevertheless, processing FeRh at the nanoscale is tricky, partly due to the need to maintain the ordered alloy structure, and partly due to its extreme sensitivity to impurities and strains. Although we are targeting an application in nanoelectronics, the progress that we shall make here in enhancing our present methods to prepare, pattern, and characterise this remarkable material will have impact by opening the door to the use of FeRh in all these different sectors. It will also have impact on the nanoscale use of chemically ordered alloys more generally, where crystal structure must be controlled truly at the atomic scale to realise proper functionality. As well as fitting in with one of the Eight Great Technologies, this approach matches well to the EPSRC Physics Grand Challenge "Nanoscale Design of Functional Materials".

Our results will also have impact in the ITC hardware sector, where memristors can find applications as non-volatile memory cells and artificial synapses. Spintronics has already enabled the huge amounts of extremely cheap data storage needed to provide social media, such as Facebook, Twitter, and Youtube, free to users. Nevertheless, the reality is that the world's server farms are consuming 30 billion watts of power. Furthermore, only about 10% of this energy is actually used for computation. The remainder is used to keep servers available should an urgent demand be requested, and to run cooling systems to dissipate this enormous amount of waste heat. As long ago as 2008, it was pointed out that the carbon footprint of the internet exceeds that of commercial air travel. As the rates of data production and consumption increase, this is clearly not sustainable. As the data volumes are unlikely to reduce, we need to search for new materials that will permit new devices and architectures to make much more efficient use of energy. Fast, dense non-volatile memories are a crucial technology to permit the development of normally-off/instantly-on computers to avoid the wasted energy of keeping idling servers ready for use. Further in the future, tremendous impact can be expected if neuromorphic computing realises its potential. At the time of writing the world's most powerful supercomputer is China's Tianhe-2 in Guangzhou, which consumes 17.8 megawatts of power. In spite of this, its computational ability in most respects lies far below that of the human brain, which requires only 20 watts, about one million times less. Indeed, a recent estimate by Dharmendra Mohda, leader of IBM's Cognitive Computing group, is that replicating the performance of the human brain in a current-day supercomputer would require 12 gigawatts of power, which is approximately one-third of the entire electricity consumption of the United Kingdom. This emphasises the point that moving away from conventional von Neumann architectures to those that mimic neural architectures found in the brains of animals can not only enhance the performance of computers in tasks at which they currently perform poorly, such as pattern recognition, but offers the prospect of truly radical reductions in power consumption.


10 25 50
Description That Ir and Pd are the optimal dopants for creating tranisition temperature gradients, that we can create FM domains in an AF background by local Joule heating, and that we can grow FeRh on NiAl buffers. That a transition can be maintained in a sub-micron nanostructure, albeit suppressed within about 250 nm of the edge. That the mixed-phase state in the midst of the transition can be reversibly affected by current pulses in an electrically connected nanopillar.
Exploitation Route Neuromorphic computing
Sectors Electronics