An emergent attractor network in a passive resistive switching circuit

Area of Science:

• Materials Science
• Computer Science
• Neuroscience

Background:

• Resistive memory devices exhibit significant conductance changes and rapid switching.
• Nonvolatile bipolar switching events in these devices act as nonlinear activation functions with hysteresis.
• Device interactions in crosspoint arrays suggest emergent network behavior.

Purpose of the Study:

• To demonstrate that passive resistive switching circuits function as attractor networks.
• To establish an energy function for these networks and analyze their dynamics.
• To explore the potential of resistive switching circuits for associative memory.

Main Methods:

• Modeling resistive switching circuits as attractor networks with binary artificial neurons.
• Defining pairwise voltage differences as an anti-symmetric weight matrix.
• Constructing an energy function to analyze network state transitions.

Main Results:

• The resistive switching circuit operates as an attractor network where device switching reduces network energy.
• The hysteretic function introduces a thresholded energy change for bit flips, differing from classic Hopfield networks.
• Experimental demonstration with 3- and 4-neuron circuits validated network dynamics and external voltage modulation.

Conclusions:

• Passive resistive switching circuits can be viewed as attractor networks, analogous to neural networks.
• The unique energy dynamics allow for stable state storage, offering a compact associative memory solution.
• Experimental results confirm the network dynamics and controllability for memory applications.