CCF-FET: Small: Enhanced Quantum Reservoir Computing via Non-Markovian Environment

Artificial intelligence systems today demand enormous computational resources, consuming vast amounts of energy and requiring large-scale hardware. This project addresses a fundamental question: can a single atom, interacting with light inside a tiny waveguide, perform the complex computations normally reserved for large networks of processors? The research team will investigate a radically minimalist approach to machine learning hardware by exploiting quantum-mechanical memory effects — the ability of the atom’s environment to store and feedback information over time. If successful, this work could lay the foundation for ultra-compact, energy-efficient computing devices suitable for real-time processing and edge applications where power and size constraints are critical. The project also supports workforce development through a quantum boot camp for high-school and undergraduate students, integration of research results into university coursework in quantum optics and machine learning, K–12 outreach in collaboration with local organizations, and open-access dissemination of software tools and publications. This project develops a quantum reservoir computing framework in which a single two-level atom, strongly coupled to a mirror-terminated waveguide, serves as a universal computing reservoir. The waveguide supports a continuum of optical modes that interact through the atom, and the mirror creates coherent time-delayed feedback that renders the dynamics non-Markovian. By adjusting the waveguide length, the degree of memory in the environment can be continuously tuned, transitioning the system between memoryless and strongly non-Markovian regimes. The research pursues two objectives. The first objective tests whether non-Markovian quantum environments yield greater reservoir complexity than their Markovian counterparts, as quantified by new complexity metrics developed within the project. The second objective tests whether the atom-waveguide system can approximate arbitrary input–output mappings — the hallmark of universal computation — by measuring quadratures of an increasing number of optical modes. Numerical simulations employ a matrix product state approach using time-bin bosonic fields to efficiently handle the exponentially large Hilbert space. Performance is benchmarked against classical reservoir computers on tasks including chaotic time-series prediction, nonlinear waveform classification, and handwritten digit recognition. The results will clarify the computational role of environmental memory in quantum systems and may establish a path toward single-device universal quantum reservoir computers. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2529700 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Daniel Soh | Institution: University of Arizona, TUCSON, AZ | Award Amount: $600,000 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2529700 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2529700.html