Noise Immunity of Oscillatory Computing Devices

Abstract

We compare oscillator- and level-based analog computing devices in a noisy environment. The computational model of a Hopfield network is implemented using both level- and phase-based signal representations, and these models are simulated assuming both 1/f and thermal noise. The oscillator-based computing architecture turns out to be more immune against 1/f noise and operates reliably at a significantly lower voltage and power levels than the level-based devices. We argue that these results point toward a fundamental benefit of oscillatory architectures and provide an important rationale for their research and utility.

Keywords

• Artificial neural networks
• circuit noise
• Hopfield neural networks
• ring oscillators (ROs)