Time‐Dependent Operations in Molecular Gap Atomic Switches

Abstract

Learning by human beings is achieved by changing the synaptic weights of a neural network in the brain. Low frequency stimulation temporarily increases a synaptic weight, which then decreases to the initial low state in the interval after each stimulation. Conversely, high frequency stimulation keeps a synaptic weight at an elevated level, even after the stimulation ends. These phenomena are termed short‐term plasticity (STP) and long‐term potentiation (LTP), respectively. These functions have been emulated by various nonvolatile devices, with the aim of developing hardware‐based artificial intelligent (AI) systems. In order to use the functions in actual AI systems with other conventional devices, control of the operating characteristics, such as matching a decay constant in STP, is indispensable. This paper reports an electrochemical method for controlling the characteristics of time‐dependent neuromorphic operations of molecular gap atomic switches. Pre‐doping of Ag+ cations into an ionic transfer layer (Ta2O5) changes the amount of shift in an electrochemical potential in the time‐dependent operation, which drastically improves the decaying characteristics in STP mode.

Non-Covalent Free-Standing Monolayer Nanoarchitectonics of Heterocoronene-Based Discotic Liquid Crystal Molecules

“…Next, an I-V measurement was carried out on the Ag-precipitation which revealed the Ohmic nature of the filament (Supporting Information, Figure S7b). Since the filament is localized, a bias sweep (−50 to 50 mV) executed at a point 1 μm away showed very low current (Supporting Information, Figure S7c), indicative of the high resistance of the DLC monolayer . However, the DLC monolayer allows Ag + ion migration and electron transport under the effect of bias application.…”

Influence of unique behaviors in an atomic switch operation on hardware-based deep learning

“…s is a constant that determines the limit of decay. As reported in the previous study, 32) the limit of the weight decay differs in different situation, so s is used to modelized these changes.…”

“…variation in the obtained resistance even with the same bias application 31) and autonomous decay in resistance even without bias application. 32) Similar to the other devices mentioned above, these characteristics should cause unintentional changes in resistance from the ideal value during a learning process. Especially, an atomic switch shows unique decay soon after the bias application, [27][28][29][30] which therefore becomes more remarkable in a learning session.…”

Intelligent Nanoarchitectonics for Self‐Assembling Systems

“…Atomic switches could also work in physical devices but their actual operation mechanisms are mostly working with chemical mechanisms. Hasegawa and coworkers recently reported the effect of predoping ions on operation time dependence with drastic improvements of decay behaviors in short‐term plasticity operations, for example, the predoping of Ag ions within the ionic transfer layer (Ta 2 O 5 ) gap of atomic switches. The concentration of Ag ions in the Ta 2 O 5 layer shifts the levels of electrochemical potential, which modifies necessary bias and speeds of cluster formation and dissolution upon changes in precipitation and diffusion of atoms and ions.…”