Computing Devices Based on Nanoscale Oscillators
Final Report Abstract
We studied oscillator-based computing devices using circuit models for spin torque oscillators and SPICE-based circuit models. We implemented an STO circuit model, used it to study actively and passively interconnected associative memories and studied in detail the most promising actively interconnected schemes. There are very few case studies for oscillatory computing devices, which use realistic, physics based models and our case study yielded to important conclusions about the practicality of various coupling schemes. We developed and verified a simplified phase model of the STO network behavior and studied the possibility of using spin waves to mediate the oscillator-oscillator interactions. Our proposal was born from the recognition that capabilities of modern neural networks (aritifical intelligence) is mainly limited by the lack of efficient computing hardware and oscillator-based computing devices and the construction of special-purpose hardware based on those could be one possible way to increase the efficiency, possibly by orders of magnitude. This was a timely idea: in the past few years, a large number of experimental and theoretical efforts were devoted to oscillator-based computing, with many prestigious publications and large scale projects (especially in the US). While the scope of our project was relatively small, we could achieve important result for particular network studied. Perhaps even more importantly, this project helped to maintain the status of our group at the forefront of the research of magnetic computing devices.
Publications
- (2018) Simulation of coupled spin torque oscillators for pattern recognition. Journal of Applied Physics 124 (15) 152128
Popescu, Bogdan; Csaba, Gyorgy; Popescu, Dan; Fallahpour, Amir Hossein; Lugli, Paolo; Porod, Wolfgang; Becherer, Markus
(See online at https://doi.org/10.1063/1.5042423)
