Neural network based on parametrically-pumped oscillators

Abstract

We demonstrate that sub-harmonic injection locked oscillators (SHILOs) can serve as building blocks of neural networks. After numerically studying the locking properties of injection-locked ring-oscillator models, we show that resistively or capacitively interconnected networks of such oscillators fall into well-defined ground states, which ground states, in turn, depend on the strength of interconnections. We argue that these networks may serve as efficient hardware implementations for emerging neural network-based processing devices.