dramatic change in conductance associated with the activated state and a concurrent
shift from weak to strong memristive behavior. Subsequent hard switching was
observed following the destruction of this highly conductive pathway, as strongly
memristive elements were redistributed throughout the network, increasing the
probability that connecting a given link would create an equivalent highly conductive path (Fig. 6c
0 ). This is an example of a dynamical self-organization process:
different ASNs can have very different initial conditions, yet the basic features of
their functional units and network topology cause similar patterns of behavior to
emerge during activation and subsequent operation.
Based on experimental and simulation results, a description of physical processes
in the ASN was formulated to describe the activation process based on the two
mechanisms described in the preceding paragraph: a bias-catalyzed phase transition
of Ag 2 S, and the subsequent Ag filament formation. A weakly memristive effect is
caused primarily by a distribution of phase-transition driven atomic switches, with a
small fraction of filamentary driven switches. As overall conduction and the fraction
of hard switching elements increases, the electric field intensifies across the
remaining soft switching junctions, encouraging further filament formation. Network
response changes from weak to strong memristive behavior when a percolative
pathway of hard switching junctions forms across the network. Having undergone
this transition, the continuously swept network operates as a hard switching
memristor, as only a few local switching events are required to reconnect an
equivalent path.
5.3 Coupling and Harmonic Generation
Individual atomic switches were shown to be directly coupled in configurations
using a shared ionic conducting layer, even when separated by large distances.
Spatially distributed atomic switch junctions interact through local variations in
ionic concentration and electrochemical potential that depend on the combined
electrical resistance of the entire network as well as the configuration, or state, of
all other electro-ionically interconnected switches.
While ‘weak’ and ‘strong’ behavior can be exhibited by single elements, the most
interesting features of this complex atomic switch device are its network-specific
properties. Infrared imaging was used track Joule heating from current flow during
DC bias sweeps in order to confirm distributed network conductance. The IR images
(Fig. 7) show power dissipation occurring across the network, indicating that the
phase change in network I-V behavior was not attributable to percolation but rather
due to the sum of parallel current flows, meaning that network structure and
connectivity actively influence device function. Additional evidence for the distribution of switch function stemmed from the analysis of the device’s frequency
response. Theoretical simulations indicate that second harmonic generation will
occur under an applied sinusoidal voltage in networks whose percentage of hard
switching junctions exceeds the percolation threshold. Furthermore, the relative
224
R. Aguilera et al.
shift from weak to strong memristive behavior. Subsequent hard switching was
observed following the destruction of this highly conductive pathway, as strongly
memristive elements were redistributed throughout the network, increasing the
probability that connecting a given link would create an equivalent highly conductive path (Fig. 6c
0 ). This is an example of a dynamical self-organization process:
different ASNs can have very different initial conditions, yet the basic features of
their functional units and network topology cause similar patterns of behavior to
emerge during activation and subsequent operation.
Based on experimental and simulation results, a description of physical processes
in the ASN was formulated to describe the activation process based on the two
mechanisms described in the preceding paragraph: a bias-catalyzed phase transition
of Ag 2 S, and the subsequent Ag filament formation. A weakly memristive effect is
caused primarily by a distribution of phase-transition driven atomic switches, with a
small fraction of filamentary driven switches. As overall conduction and the fraction
of hard switching elements increases, the electric field intensifies across the
remaining soft switching junctions, encouraging further filament formation. Network
response changes from weak to strong memristive behavior when a percolative
pathway of hard switching junctions forms across the network. Having undergone
this transition, the continuously swept network operates as a hard switching
memristor, as only a few local switching events are required to reconnect an
equivalent path.
5.3 Coupling and Harmonic Generation
Individual atomic switches were shown to be directly coupled in configurations
using a shared ionic conducting layer, even when separated by large distances.
Spatially distributed atomic switch junctions interact through local variations in
ionic concentration and electrochemical potential that depend on the combined
electrical resistance of the entire network as well as the configuration, or state, of
all other electro-ionically interconnected switches.
While ‘weak’ and ‘strong’ behavior can be exhibited by single elements, the most
interesting features of this complex atomic switch device are its network-specific
properties. Infrared imaging was used track Joule heating from current flow during
DC bias sweeps in order to confirm distributed network conductance. The IR images
(Fig. 7) show power dissipation occurring across the network, indicating that the
phase change in network I-V behavior was not attributable to percolation but rather
due to the sum of parallel current flows, meaning that network structure and
connectivity actively influence device function. Additional evidence for the distribution of switch function stemmed from the analysis of the device’s frequency
response. Theoretical simulations indicate that second harmonic generation will
occur under an applied sinusoidal voltage in networks whose percentage of hard
switching junctions exceeds the percolation threshold. Furthermore, the relative
224
R. Aguilera et al.