Active Neuromorphic Metasurfaces for Embedded Mechanical Computing in Extreme Environments
Abstract
Dr. Daniel P. Cole, Mechanical Behavior of Materials, U.S. Army Research Office Active Neuromorphic Metasurfaces for Embedded Mechanical Computing in Extreme Environments The concept of analog computing has recently gained traction both in the photonics and mechanics communities. Contrary to digital systems where a signal varies virtually via discrete values of both time and amplitude, analog mechanical computing predominantly relies on the continuous variation of physical parameters to correlate an output of a computational task to an input which takes the form of the same physical domain. With recent advances in adaptive material behavior, metasurfaces, and fabrication techniques, there is a unique opportunity for mechanical computing to exploit novel features of elastic wave scattering and active material response to conduct complex computations at efficiency levels that are not feasible via classical digital approaches. This work proposes the foundational modeling and synthesis of a class of dynamically-modulated neuromorphic metasurfaces which operate within multiple frequency and momentum (directional) channels. Since these channels are independently tunable, they can be assigned distinct computational tasks, thereby enabling concurrent mathematical operations and pioneering parallel processing in mechanical computers. The latter is a critical first step towards allowing mechanical neuromorphic systems to conduct computationally-complex tasks by breaking them into smaller bits; analogous to multi-core processing in high-performance supercomputers. In the presence of active phase modulation, the scattering culminating within the different layers of trained metasurfaces becomes equivalent of digital computations in the layers of a neural network, and a readout mechanism can be used to interpret the results at a detection plane at the far end via displacement sensing and scanning laser vibrometry. By utilizing the controlled, frequency-selective directional beaming taking place in the modulated metasurface, we envision an intelligent mechanical computing system that is able to classify multiple decoupled inputs by encoding their features into the elastic wave mechanics using distinct frequency signatures. Not only are such capabilities unprecedented in the mechanical domain, they also pave the way for structurally-embedded computations in remote environments where digital alternatives are either inaccessible or simply infeasible due to energy or hard-wiring constraints. The project will also seek to exploit unique dispersive features of anelastic media, specifically the interplay between metadamping emergence and energy localization in viscoelastic materials to realize a degree of versatility in the neuromorphic system which enables it to adapt to changing loading forms, energy availability, or computational requirements.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 08, 2023
- Source ID
- W911NF2310078
Entities
People
- Mostafa Nouh
Organizations
- Army Contracting Command
- United States Army
- University at Buffalo