1/f Noise and Machine Intelligence in a Nonlinear Dopant Atom Network

Abstract

Noise exists in nearly all physical systems ranging from simple electronic devices such as transistors to complex systems such as neural networks. To understand a system's behavior, it is vital to know the origin of the noise and its characteristics. Recently, it was shown that the nonlinear electronic properties of a disordered dopant atom network in silicon can be exploited for efficiently executing classification tasks through "material learning." Here, we study the dopant network's intrinsic 1/f noise arising from Coulomb interactions, and its impact on the features that determine its computational abilities, viz., the nonlinearity and the signal-to-noise ratio (SNR), is investigated. The findings on optimal SNR and nonlinear transformation of data by this nonlinear network provide a guideline for the scaling of physical learning machines and shed light on neuroscience from a new perspective.

Waveguide-integrated PdSe2 photodetector with high responsivity and speed in the near-infrared range across the O+E+S+C bands

“…The measured 1/f noise spectral density in Fig. 4d indicated that PdSe 2 photodetectors possess typical characteristics of 1/f noise, which can be determined by the Hooge's empirical relationship (𝑆𝑆 𝑙𝑙 = 𝐴𝐴 • 𝑖𝑖 𝛼𝛼 /𝑓𝑓 𝛽𝛽 ), 61,62 where i, f, A are the channel current, the frequency, and noise amplitude, respectively. Current noise induced by 1/f noise decreases gradually with the increasing applied frequency.…”

Emergence of In‐Materio Intelligence from an Incidental Structure of a Single‐Walled Carbon Nanotube–Porphyrin Polyoxometalate Random Network

“…Intrinsic electrical properties of this nature that are nonlinear with network wide high-dimensionality and obey the 1/f γ power law of maximized information processing, and are manifested in a recurrently connected random network structure, have recently proven to play a key role in in-materio RC tasks of binary classification and prediction. [42,45,46] Thus, by utilizing the innate dynamics emergent from our SWNT/Por-POM reservoir, we proceeded to establish the RC benchmark tasks of waveform generation and object classification, which we present in the next sections.…”

In‐Materio Reservoir Computing in a Sulfonated Polyaniline Network

“…[ 14–16 ] One of the biggest challenges for material‐based computing is the control of the output by external stimuli, because materials exhibit dynamic fluctuations at all times. [ 17 ] Among the potential applications of different architectures of material‐based neural networks, reservoir computing (RC) is a promising avenue. [ 18 ] In RC, a randomly connected network, the “reservoir,” creates nonlinear projections of inputs into high‐dimensional space.…”