Abstract: Neuromorphic computing and engineering is capitalizing heavily on the new physical properties offered by nantechnologies to engineer biological processes. At the frontiers in between bio-mimetism and bio-inspiration, various solutions have been proposed for synaptic plasticity or neuronal features based on discrete memory elements, bistable switches or transistors circuits. One missing element that has been missing in the neuromorphic toolbox is the ability to reproduce the complex 3D interconnections o bserved in biological neural networks. Here, we propose to take advantage of bipolar electropolymerization of PEDOT dendritic fibers to reproduce the ability of neural networks to generate complex topologies. The electropolymerization mechanism is used t o realize structural plasticity based on Hebbian-like plasticity rules. We explore how such bottom-up process can find optimal topologies for specific computing tasks. We demonstrate that such optimal topologies results in a drastic reduction of intercon nects for classification and reconstruction tasks, thus offering an interesting option for neural network design.
Alibart F., Janzakova K., Scholaert C., Balafrej I., Kumar A., Drouin D., Rouat J., Pecqueur S., Coffinier Y.
Abstract: Neural networks are powerful tools for solving complex problems, but finding the right network topology for a given task remains an open question. Biology uses neurogenesis and structural plasticity to solve this problem. Advanced neural network algorith ms are mostly relying on synaptic plasticity and learning. The main limitation in reconciling these two approaches is the lack of a viable hardware solution that could reproduce the bottom-up development of biological neural networks. Here, we show how t he dendritic growth of PEDOT:PSS-based fibers through AC electropolymerization can implement structural plasticity during network development. We find that this strategy follows Hebbian principles and is able to define topologies that leverage better com puting performances with sparse synaptic connectivity for solving non-trivial tasks. This approach is validated in software simulation, and offers up to 61% better network sparsity on classification and 50% in signal reconstruction tasks.
Janzakova K., Balafrej I., Kumar A., Garg N., Scholaert C., Rouat J., Drouin D., Coffinier Y., Pecqueur S., Alibart F.*
© 2019-2025 Sébastien Pecqueur
