Perovskite Microwires as a Platform for Optical Neural Networks

The growing demand for fast and energy-efficient information processing has motivated the search for computing platforms beyond conventional electronics. Among the most promising alternatives are optical neural networks, which can exploit the speed of light and avoid some of the fundamental bottlenecks associated with electronic data transfer and power consumption. In this seminar, I will discuss our recent efforts toward implementing optical neural-network architectures based on exciton-polaritons.

Exciton-polaritons are hybrid light-matter quasiparticles formed in the strong-coupling regime between photons and excitons. Their unique combination of low-loss propagation and strong optical nonlinearities makes them an attractive platform for all-optical information processing. I will present our work on room-temperature polariton neural networks based on large-scale CsPbBr₃ perovskite microwires of arbitrary shape. Using a template-assisted fabrication method, we create high-quality perovskite microstructures that can be bent into predefined geometries while preserving their optical properties.

These microwires support waveguiding and enable the formation of spatially extended coherent polariton condensates at room temperature. In contrast to more conventional implementations, our platform does not require external cavity mirrors, which greatly simplifies fabrication and improves compatibility with integrated photonic technologies. I will show evidence of polariton lasing from the edges and corners of the structures, accompanied by pronounced blueshifts at high excitation powers. Far-field photoluminescence and angle-resolved measurements reveal strong mutual coherence between spatially separated lasing signals, demonstrating long-range condensate propagation and coupling between neighboring wires across air gaps.

These features open the way to using individual or coupled perovskite microwires as nonlinear nodes in exciton-polariton neural networks. I will discuss how this platform could be employed in machine-learning tasks such as classification and object recognition, and why it offers a simple, scalable, and cost-effective route toward room-temperature optical computing.