An experimental demonstration of the viability of energy minimizing computing using nano-magnets

Abstract

There has been recent proposals for the use of nano-magnets to directly solve quadratic minimization problems, especially those arising in computer vision applications. This is unlike proposals for using nano-magnets to represent binary states. A collection of nano-magnets, when driven to their ground states, can be seen to optimize a quadratic energy function that is determined by their relative placement. By controlling the relative placement of nano-magnets, we can change the energy function being minimized. In this work, we experimentally demonstrate this capability by fabricating and testing an example of a quadratic optimization problem that accomplishes line grouping.

Simulated annealing with surface acoustic wave in a dipole-coupled array of magnetostrictive nanomagnets for collective ground state computing

“…Ideas of ground state computing with nanomagnets available in the literature [11][12][13][14][15][16] always presuppose an assured final relaxation of the nanomagnetic system to the ground state in order to solve a problem. This, of course, may not happen if there are deep metastable states that trap the magnetization, If and when that happens, simulated annealing will be required to reach the ground state.…”

“…Recently, there has been a spurt of interest in solving challenging software tasks (image processing, computer vision, combinatorial optimization, etc) with application-specific hardware (instead of software), which does not need to execute instruction sets and hence will be much faster than software based approaches [7][8][9][10]. Some of these hardware emulators involve interacting nanomagnetic systems [11][12][13][14][15][16] because they are particularly suitable for this approach. The problem variables are mapped into the magnetic states of the interacting nanomagnets and the whole system is allowed to relax to the many-body ground state.…”

“…Fortunately, this problem is not insurmountable. The work presented here shows how to perform simulated annealing in nanomagnetic systems with timevarying strain and therefore makes an important contribution to nanomagnet-based ground state computing approaches, espoused in [11][12][13][14][15][16].…”

“…To our knowledge, experimental demonstrations of solving optimization problems with interacting nanomagnets [13][14][15][16] are (until now) somewhat rare and we are not aware of any demonstration of simulated annealing in such systems triggered by an external agent. Here, we experimentally demonstrate emulation of simulated annealing in a small interacting nanomagnetic system comprising a two-dimensional array of 3 × 3 dipole-coupled magnetostrictive nanomagnets.…”

Applications of nanomagnets as dynamical systems: II

“…Fortunately, there are a host of non-Boolean computing applications for which magnetic devices may be eminently suitable. They include application-specific-integrated-circuits (ASICs) for image processing [16,17], simulated annealing in energy minimization computing for solving combinatorial optimization problems [18], probabilistic computing with p-bits [19][20][21][22], computer vision [23,24], Bayesian inference engines for belief networks and machine learning [25][26][27][28], restricted Boltzmann machines [29], ternary content addressable memory [30] and non-volatile and reconfigurable equality bit comparators [31] for electronic locks and hardware countermeasures against cyberattacks.…”