Robust Noise Characteristics in Carbon Nanotube Transistors Based on Stochastic Resonance and Their Summing Networks

Abstract

Robust noise characteristics in carbon nanotube field-effect transistors (CNT-FETs) based on stochastic resonance (SR) were demonstrated to detect small signals in noisy environments. When weak pulse trains were applied to a CNT-FET in the subthreshold regime, the correlation coefficient between the input and output signals increased upon adding an appropriate intensity of noise. Offset-voltage dependences were investigated, and moreover, a virtual summing network was formed using CNT-FETs having different offset voltages. The measurement indicated that responses correlated with the input signals were enhanced in a wide range of noise intensity. Therefore, the summing network based on SR is a promising candidate for highly sensitive label-free sensors which are to be utilized in unintentionally noisy environments.

Bi2O2Se-Based Bimode Noise Generator for the Application of Generative Adversarial Networks

“…To alleviate those disturbing effects, many efforts in the electronic field endeavor to diminish those “detrimental” noises; for example, introducing a double fullerene layer for trap passivation to suppress the shot noise and flick noise in a perovskite photodetector; the shunting electrode, working as a blanking unit, exhibited a record low dark current in a perovskite X-ray detector; the memristor in the threshold switching mode could be utilized as a selector, which largely suppressed the sneak path current in a memristive crossbar array; and applying small-voltage pulses could denoise the random telegraph noising in a memristive crossbar array to precisely control the conductance levels . Interestingly, those “detrimental” noise effects in electronic and photonic devices turn out to be constructive in many emerging fields: stochastic resonance in carbon nanotubes, GaAs nanowires, and MoS 2 photodetectors for detecting low-amplitude signals; − entropy sources for random number generation, including tungsten diselenide (WSe 2 ) field-effect transistors or hexagonal boron nitride (h-BN) and Bi 2 O 2 Se-based memristors (BMs); − as well as some emerging reported pioneering works that implemented the memristive intrinsic noise for accelerating the neuromorphic computing, such as realizing the stochastic gradient descent, improving the evolution, and inducing a diversity of solutions for a Hopfield neural network, suppressing the overfitting and realization of generative adversarial networks (GANs). − Compared to fabricating artificial synapses, implementing noises in neural networks, such as GANs, could be another feasible approach for neuromorphic computing and require further research effort.…”