Processing information is essential to both living and engineered systems. In living systems, this emerges from physical processes such as chemical reactions and electrical signals. Engineered systems, on the other hand, typically rely on central processors to analyze signals and extract meaningful information, often through the implementation of AI and machine learning algorithms. Harnessing physical processes could allow computation in hardware, which might be more efficient, cheaper, and integrated. While there have been large advances in the ‘quasistatic’ regime, the dynamic regime, in which responses cannot be described by instantaneous equilibrium relations, has been largely unexplored.
The aim of the current research is to explore how physical systems such as mechanical structures or simple electronic networks can perform a given desired dynamical function. Moreover, this research aims to develop a framework for training such systems in hardware, without the use of an external CPU. This line of research could open the door to intelligent materials that process dynamical signals, enabling temporal computations, frequency-dependent filtering, and passive sensing for classification.
