Bruno, Ugo (2024) Neuromorphic organic electrochemical transistors for bio-inspired computation, learning and biological interfaces. [Tesi di dottorato]

Anteprima

Testo Bruno_Ugo_36.pdf Download (9MB) | Anteprima

Abstract

The rise of artificial intelligence and artificial neural networks revolutionized information technology and our daily lives. Such progress was inspired by the parallel organization and computation paradigm of the brain, allowing to build networks able to learn from experience and, very recently, actively process the natural language, to generate text and images, summarize content, translate text in different language and offer a human-like assistance. This progress, though, comes with a price. While the software is inspired from the brain parallel and efficient computing, the supporting hardware is still based on classical von Neumann architecture, which, unlike the human brain, was designed to run simple operation in a linear succession. This mismatch between software and hardware causes an excessive energy consumption when training and operating such architectures. In this framework, neuromorphic engineering aims to take inspiration from the brain, to design the hardware of the future. It harvests from all possible biological computational primitives, such as single synapses, ionic fluxed, plasticity and sparse coding to develop novel principles to design hardware. Material science, with the study of innovative materials, has been of utmost importance in continuously providing a wave of innovation, allowing researchers to demonstrate innovative brain-derived architectures. Among the plethora of available materials, organic semiconductors emerged for easy processability, solution processing and easy processability, along with low voltage operations. Nonetheless, such materials can offer a mixed conduction mechanism, relying on both electrons and ions to modulate their conductivity. This last feature was indeed crucial in interfacing such materials with biology, where electronic signaling leaves space to ions moving in an electrolytic environment. Moreover, the soft nature of these materials minimized the mechanical mismatch between electronical devices and biological tissue, offering a seamless interface with the human body, and without eliciting any inflammatory response. The thesis work here presented envisions the merging of these two concepts: the possibility to optimally interface with biology, while achieving brain-inspired computation. Organic electrochemical transistors were designed to demonstrate bio-inspired in loco computation of biologically relevant signals, as pressure, light, or neurotransmitters. By engineering the alternance of conducing and non-conducting thin films, a non-volatile pressure sensor, inspired by the sense of touch was designed. OECTs were subsequently endowed with light-sensitivity, through an ad hoc synthetic strategy. Here, the obtained transistors were able to emulate light-processing of the human retina, while demonstrating memory compartmentalization, as formulated by classical neuroscience theories. Neurotransmitter-mediated synaptic plasticity was then demonstrated and exploited to build a closed-loop neuromorphic system, in which organic transistors could communicate and cooperate with standard CMOS technology to endow a well-established mechatronic system with brain-derived adaptive features, as autonomous reinforcement learning. Lastly, biomimicry was enforced in OECTs, by embedding bio-inspired membranes in the electrolytic environment of the transistors. Notably, while offering a seamless interface with biological tissue, the membranes were able to modulate both short- and long-term plasticity of the artificial synaptic device.

Downloads

Actions (login required)

Modifica documento