Structural plasticity for neuromorphic networks with electropolymerized dendritic PEDOT connections

Area of Science:

• Neuroscience
• Materials Science
• Computer Science

Background:

• Neural networks excel at complex problems, but optimal topology design is challenging.
• Biological systems utilize neurogenesis and structural plasticity for network development.
• Current artificial neural networks primarily rely on synaptic plasticity, lacking hardware for bottom-up development.

Purpose of the Study:

• To introduce a hardware-based approach for structural plasticity in neural network development.
• To enable bottom-up development of artificial neural networks mimicking biological processes.
• To investigate the impact of dendritic growth on network topology and performance.

Main Methods:

• Utilizing PEDOT:PSS-based fibers and AC electropolymerization for dendritic growth.
• Implementing structural plasticity during network development via fiber growth.
• Validating the approach through software simulations.

Main Results:

• The dendritic growth strategy follows Hebbian principles.
• Achieved network topologies with improved computing performance and sparse synaptic connectivity.
• Demonstrated up to 61% increased network sparsity in classification tasks.
• Showcased up to 50% increased network sparsity in signal reconstruction tasks.

Conclusions:

• Dendritic growth of PEDOT:PSS fibers offers a viable hardware solution for structural plasticity in artificial neural networks.
• This bio-inspired approach enhances network efficiency and performance for non-trivial tasks.
• The method provides a pathway towards more biologically plausible and efficient artificial intelligence.