Session: 06-08: Bioinspired Networks and Neurons

Paper Number: 111189

111189 - Optimization of Biomolecular Neuristor Action Potentials to Mimic Biological Response 

Biological neurons generate action potentials for communication and computational tasks. Along with firing frequency, the shape and nonlinear response curve of action potentials carry essential information and significant computational power. Many spiking neuron models take an integrate-and-fire approach to represent spike timing and frequency with computational simplicity but lack the biological relevance of more complex action potential dynamics. Therefore, implementing biologically relevant action potentials made of soft materials in engineered systems could have a significant impact on their ability to replicate biological behaviors with high efficiency and energy density. This paper introduces a model-based approach for optimizing the parameters of a soft neuristor circuit to match the action potential shape of Hodgkin-Huxley neurons. The neuristor consists of two biomolecular memristors coupled in parallel, which are governed by Hodgkin-Huxley-like dynamics. Features of the device include generating biologically relevant action potential shapes at adjustable frequencies, along with power efficiency and scaling in networks. Typically, designing these devices for experimental implementation requires either exhaustive hand-tuning of key parameters or optimization over a large parameter space, where gradient-based optimization methods perform poorly due to strong nonlinearities and nonconvexities that make global minima difficult to find. This paper instead employs genetic algorithms to optimize design parameters, with the objective of matching neuristor response to biological action-potential-like references while respecting constraints on parameters to ensure the final design is experimentally achievable. The approach is demonstrated through three numerical examples. The first seeks to match a reference action potential with a single neuristor, including both response shape and frequency. The second example seeks to tune several neurons composing a neural circuit, where the structure and synapses are fixed and the neuristor is again optimized to match a biologically-inspired reference. Specifically, a central pattern generator is selected as the reference circuit, as this generates oscillating behavior and can control rhythmic motions such as walking or swimming. The final example optimizes the circuit to control a vehicle navigating toward a target. The proposed optimization tool provides the ability to match both action potential shape and frequency in selecting device parameters for lab experiments. Furthermore, this tool enables the creation and tuning of biologically inspired controllers for engineered systems.  

Presenting Author: Ahmed Salah Mohamed The Pennsylvania State University, Department of Mechanical Engineering

Presenting Author Biography: I am a fourth-year mechanical engineering PhD student and graduate research assistant at the mechanical engineering department at Penn State University. My work focuses on the development and use of bio-molecular materials for neuromorphic computing research and applications.