Exploiting the stochasticity of memristive devices for computing

Many memristive devices are prone to partly unpredictable or stochastic behavior. For most applications this point constitutes a challenge that needs to be overcome. However, it does not necessarily need to be. In this chapter we start by reviewing scientific phenomena and methods that are able to harness randomness. Most of these techniques have been known for long, but have never delivered their promises of leveraging noise for computing. We articulate how finding an appropriate stochastic device can unlock the potential of these methods. In consequence we then review various memristive and spintronic technologies, whose intrinsic stochastic behaviors have been proposed for computing. We then present several modern projects that harness memristive randomness for computing, and how their energy efficiency can be benchmarked to more conventional computing approaches. We conclude on a discussion of the tasks and applications for which exploiting stochasticity is a good strategy.