Seminars

Model mechanical metamaterials for memory storage and computation

While traditional computing has undeniably demonstrated its effectiveness, notably through the widespread adoption of modern computers, it still fails to solve a range of hard problems efficiently. To address this, innovative unconventional computing platforms are emerging, disrupting traditional computing paradigms. Interestingly, these novel approaches leverage alternative media, such as dynamical systems, for computing. On another side, various disordered media, such as amorphous matter or mechanical metamaterials, can sometimes be described as collections of bi-stable elements, which may be used to represent binary data. Notably, various writing operations and the implementation of simple forms of algorithms have been realized in multiple disordered media, both natural and engineered. These advances have sparked renewed interest in mechanical computing, which uses material media to process information.
Building upon these previous works, this thesis explores the use of model materials for both processing and storing binary data. In particular, we consider systems composed of bi-stable spring-mass units organized into chains or lattices. These systems can be viewed as assemblies of non-volatile memory units, making them suitable platforms for information writing, storage, and processing.
We begin by illustrating the feasibility of encoding binary messages within multi-stable chains, showing that a reinforcement learning agent can effectively learn to navigate their complex energy landscape to inscribe information. However, these landscapes need to be known for such writing processes, which may not be the case in real-life situations. To address this, we show that machine learning techniques can successfully uncover the landscape of multi-stable chains, thereby enabling both writing and reading operations within them. Finally, we demonstrate the implementation of basic sequential algorithms within both lattices of bi-stable springs and Ising models, by harnessing their inherent relaxation mechanisms to execute these algorithms. These findings suggest that simple model materials can be used to write, store, and read binary data, as well as perform algorithms, highlighting their potential as effective computational platforms