Information dynamics in neuromorphic nanowire networks

Area of Science:

• Neuromorphic engineering
• Complex systems science
• Information theory

Background:

• Neuromorphic systems utilize self-assembled nanowires with synapse-like junctions and complex network topologies.
• These systems exhibit neural-like dynamics and have demonstrated capabilities in various information processing tasks.

Purpose of the Study:

• To investigate information processing dynamics in neuromorphic nanowire networks using information-theoretic metrics.
• To understand the relationship between network topology, internal dynamics, and information processing performance.

Main Methods:

• Employing Transfer Entropy (TE) and Active Information Storage (AIS) to quantify information flow and short-term memory.
• Analyzing how topological centrality influences information flow within the networks.
• Correlating network dynamics (quiescent to active transition) and connectivity density with performance metrics.

Main Results:

• Topologically central network components significantly contribute to information flow.
• Transfer Entropy and Active Information Storage peak during the transition from quiescent to active states.
• Neuromorphic network performance in memory and learning tasks is contingent upon internal dynamical states and topological structure.

Conclusions:

• Optimal performance in neuromorphic nanowire networks is achieved when pre-initialized to a transition state maximizing TE and AIS.
• An optimal range of connectivity density exists for efficient information processing.
• Information dynamics provide a valuable framework for studying and benchmarking neuromorphic systems.