Abstract
Neuro-inspired computing architectures are one of the leading candidates to solve complex, large-scale associative learning problems. In addition to the widely studied spiking neural network architecture, other neuro-inspired architectures, such as the oscillatory neural network, can be used to solve such special purpose computing problems. In both cases, voltage- or current-controlled relaxation oscillators are considered a minimal representation of neurons for hardware implementations. For an oscillatory neural-network architecture, such oscillators should demonstrate synchronization and coupling dynamics to achieve collective learning behavior, in addition to desirable individual characteristics such as scaling, power, and performance. To this end, we propose the use of nanoscale, epitaxial heterostructures of phase-change oxides and oxides with metallic conductivity as the fundamental unit of a low-power, tunable electrical oscillator capable of operating in the microwave regime. We perform simulations to show that an optimized heterostructure design with low thermal boundary resistance can result in an operation frequency of up to 3 GHz and a power consumption as low as 15 fJ/cycle with rich coupling dynamics between the oscillators on a 100-nm channel. This study provides an alternative device design of tunable relaxation oscillators for neuromorphic computing applications.
1 More- Received 21 May 2018
- Revised 23 October 2018
DOI:https://doi.org/10.1103/PhysRevApplied.11.014020
© 2019 American Physical Society