Stochastic Magnetic Tunnel Junctions for Bio-Inspired Computing
With the rise of nanoelectronics, many novel technologies have emerged, holding the promise to replace or complement the traditional computing building block — the CMOS transistor. However, at the nanoscale, noise significantly affects the behavior of systems, inducing random fluctuations. It is thus natural to look for computing techniques that are intrinsically tolerant to noise, variability and errors, or even better, which take advantage of these.
Among the possible solutions, one paradigm has emerged as particularly promising and disruptive: taking inspiration from biology. Indeed, our brain is able to perform computations — while consuming only 20 W — even though its components themselves exhibit stochastic behavior. Bio-inspired computing with stochastic nanodevices should prove to be particularly successful for cognitive tasks such as pattern recognition and classification. Mixing conventional electronic components with emerging technologies could allow performing such tasks at low energy cost.
The focus of this seminar is a specific nanodevice, the magnetic tunnel junction. Because of its endurance, reliability and CMOS compatibility, this bistable system has emerged as the flagship device of spintronics. However, maintaining the stability of this device while reducing its size is a challenge. Unstable magnetic tunnel junctions — called superparamagnetic tunnel junctions — behave as stochastic oscillators. Here, we investigate for the first time how to harness the random behavior of stochastic magnetic tunnel junctions, taking inspiration from biology.
First, an analogy between superparamagnetic tunnel junctions and sensory neurons — which fire voltage pulses with random time intervals — is drawn. Pushing this analogy, it is numerically demonstrated that interconnected populations can perform computing tasks such as learning, coordinate transformations, and sensory fusion. Such a system is realistically implementable and could allow for intelligent sensory processing at low energy cost.
Then, it is experimentally demonstrated that electrical noise can induce the synchronization of a junction to a weak voltage source. A theoretical model is developed and predicts that using noise could allow a hundredfold energy gain over the synchronization of traditional dc-driven spin torque oscillators. This result opens the way to the low-power hardware implementation of synchronization-based computing schemes that can perform tasks such as pattern recognition.
All these results suggest that the superparamagnetic tunnel junction is a promising building block for hardware implementations of bio-inspired computing.