Julie Grollier
spin torque

The Spin Torque Lego

The spin transfer effect allows to manipulate the magnetization of a nanomagnet without the help of an applied magnetic field. This phenomena, theoretically predicted in 1996 by John Slonczewski and Luc Berger, immediately attracted a lot of attention both for its fundamental interest as a new spintronic effect and its huge potential for applications. Spin transfer takes its origin in the transfusion of magnetic momentum from a spin polarized current to the local magnetization. For large current densities, typically of the order of 107 A.cm-2, the spins carried by the conduction electrons can exert a torque large enough to reverse the magnetization of a small magnetic object, typically with lateral dimensions smaller than a few hundred nanometers.

The general principle of spin torque nano-devices is depicted above. A current is injected through a magnetic trilayer structure: a non-magnetic layer sandwiched between two thin nano-magnets. One of the magnetization is usually kept fixed, whereas the second one is free to move. Under the action of spin torque, magnetization dynamics can be generated in the free layer. This magnetization motion is then converted into resistance and voltage variations thanks to the trilayer magnetoresistance, Giant Magnetoresistance (GMR) or Tunnel Magnetoresistance (TMR), depending on the stack.

After more than a decade of intense research, the understanding of spin torqueís microscopic origins and the resulting magnetization dynamics has reached such a level of maturity that it is possible to predict accurately through coupled transport/micromagnetic simulations the device behaviour. By adjusting the input current waveform, by playing with the materials and geometry, we are now able, through spin torque engineering, to implement complex functions at the nano-scale. Many different spin torque functionalities, such as binary memory, telegraphic switching, microwave oscillations, microwave detection, spin wave emission, and memristive effects have been experimentally demonstrated at room temperature.

Spin torque Lego. These different devices can be seen as Lego bricks, each brick with its own functionality that can be assembled to build novel types of computing architectures. Recently, a new class of applications has appeared, that takes full advantage of the spin torque building blocks. The goal here is to assemble different bricks and to combine their various functionalities to build novel types of Hybrid spintronic/CMOS information processing hardware architectures working at room temperature with low power consumption and high performances. We are particularly interested in spin torque bio-inspired architectures, concepts relying on non-Boolean processing of information, avoiding competition with sectors where pure CMOS excels, and opening the way for novel types of spintronic accelerators.

Spin torque neuromorphic architectures. The domain-wall based spin torque memristor can emulate a nano-synapse. But to maximize neural-networks size, and therefore efficiency, shrinking also the neurons size to nanometer dimensions becomes an important issue. We are currently working on the design and fabrication of spin torque neurons.

Synchronization. Neurons can be modeled as non-linear oscillators that adjust their rhythms depending on incoming signals. In the brain, they form a network of coupled oscillators, where the coupling is mediated by synapses. Assemblies of neural oscillators can self-synchronize, in frequency or phase, defining and linking vast areas of the brain where neurons oscillate in unison. We are exploring the synchronization of spin torque nano-oscillators as a means to perform cognitive operations.

Spin torque stochastic computing. The classical way to perform data processing is to reduce all sources of noise to the maximum. An interesting alternative strategy is, on the contrary, to exploit noise for computing. In this trend, stochastic computing has a great potential for the implementation of low power information processing systems. Indeed noise is often seen as a key element of neural computation, beneficial for a number of operations as near-threshold signaling and decision making. And spin torque devices, just like neurons, can exhibit noise induced sensitivity improvement, for example via stochastic resonance. We are working on the development of probabilistic bio-inspired hardware exploiting the controlled stochasticity provided by spin torque.