Vowel recognition with four coupled spin-torque nano-oscillators

We present a method for learning nonlinear systems, echo state networks (ESNs). ESNs employ artificial recurrent neural networks in a way that has recently been proposed independently as a learning mechanism in biological brains. The learning method is computationally efficient and easy to use. On a benchmark task of predicting a chaotic time series, accuracy is improved by a factor of 2400 over previous techniques. The potential for engineering applications is illustrated by equalizing a communication channel, where the signal error rate is improved by two orders of magnitude.

There is currently much interest in the development of 'spintronic' devices, in which harnessing the spins of electrons (rather than just their charges) is anticipated to provide new functionalities that go beyond those possible with conventional electronic devices. One widely studied example of an effect that has its roots in the electron's spin degree of freedom is the torque exerted by a spin-polarized electric current on the spin moment of a nanometre-scale magnet. This torque causes the magnetic moment to rotate at potentially useful frequencies. Here we report a very different phenomenon that is also based on the interplay between spin dynamics and spin-dependent transport, and which arises from unusual diode behaviour. We show that the application of a small radio-frequency alternating current to a nanometre-scale magnetic tunnel junction can generate a measurable direct-current (d.c.) voltage across the device when the frequency is resonant with the spin oscillations that arise from the spin-torque effect: at resonance (which can be tuned by an external magnetic field), the structure exhibits different resistance states depending on the direction of the current. This behaviour is markedly different from that of a conventional semiconductor diode, and could form the basis of a nanometre-scale radio-frequency detector in telecommunication circuits.

The spin torque effect that occurs in nanometre-scale magnetic multilayer devices can be used to generate steady-state microwave signals in response to a d.c. electrical current. This establishes a new functionality for magneto-electronic structures that are more commonly used as magnetic field sensors and magnetic memory elements. The microwave power emitted from a single spin torque nano-oscillator (STNO) is at present typically less than 1 nW. To achieve a more useful power level (on the order of microwatts), a device could consist of an array of phase coherent STNOs, in a manner analogous to arrays of Josephson junctions and larger semiconductor oscillators. Here we show that two STNOs in close proximity mutually phase-lock-that is, they synchronize, which is a general tendency of interacting nonlinear oscillator systems. The phase-locked state is distinct, characterized by a sudden narrowing of signal linewidth and an increase in power due to the coherence of the individual oscillators. Arrays of phase-locked STNOs could be used as nanometre-scale reference oscillators. Furthermore, phase control of array elements (phased array) could lead to nanometre-scale directional transmitters and receivers for wireless communications.