Thermodynamics 2.0 Program: Sessions and Abstracts
Mon - Wed, June 22 - June 24 , 2020 , Massachusetts, USA
Session T14: Theoretical advances III
10:45-12:00. Wednesday June 24, 2020
Chair: Gábor Balázsi
Title: Spike Based Computing on Self Assembled Lipid Monolayers
Bio-sketch
I am a Senior Research Associate at the Rosalind Franklin Institute and the department of engineering science at the University of Oxford, where I am developing photonic and acoustic tools for monitoring and controlling the state of biological materials. The vision is motivated by the extraordinary properties of 2D shock waves in lipid interfaces that were established for the first time in my Ph.D. thesis at Boston University. The research strongly supports the possibility of such waves being a physical basis for the phenomenon of nerve pulse propagation. The research has led to the invention of a fundamentally new platform for performing brain-like sensing and computing with brain-like efficiency.
Author(s):
(University of Oxford, UK )
Abstract:T14.W160
Abstract
A new physical platform inspired by the physics of biological neurons is presented, which will allow testing and implementing thermodynamically optimized computing schemes at the hardware level. While the remarkable energy efficiency of neuronal processes is inspiring the next generation hardware, the physical basis of this efficiency within neuroscience is surprisingly a matter of debate. The electrical spikes in neurons, also known as action potentials, are believed to be purely electrical phenomenon. The standard model of the phenomenon assumes that the information, in the form of a spike, travels via directed mass transfer or diffusion as it irreversibly depletes an electrical gradient across the outer membrane of a neuron. However, recent research strongly suggests that information can also travel in such systems, i.e. the outer membrane of a neuron, via momentum transfer, a reversible process similar to the propagation of sound.
Such spikes have now been realized experimentally as two-dimensional nonlinear sound waves that propagate within single molecule thin films of lipids, which provide the physical basis for the new computational platform [1]. These waves can mimic all the computationally relevant properties of spikes in neuron, such as a threshold for excitation, solitary wave propagation, its dependence on environmental variables like pressure and temperature, and annihilation of spikes upon collision. The lipid monolayers that support the phenomenon are extremely robust and quickly self-assemble at the air/water interface and provide a practical platform to develop brain-like computing. The device operates at energy and timescales comparable to real neurons, and is also highly scalable as the platform employs nonlinear wave mechanics in a planar continuous medium, which allow for massively parallel computing. Furthermore, as the phenomenon occurs at a material phase transition in the lipids, the platform provides unique capabilities to implement self-organized criticality (SOC) and momentum based optimization, as well as dissipation based learning at the hardware level.
Keywords: self-organized criticality, SNN hardware, 2D materials, shock waves, avalanches