From population to coherence: modeling, optimization, and application of atomic vapor for information storage

Abstract

Multimode optical memories are a fundamental requirement for several emerging technologies. In optical neural networks, classical optical memories provide low-latency data input and buffer optical information awaiting processing. In quantum information processing, memories are key for storing and synchronizing quantum states to enable high-fidelity, secure communication and authentication protocols. This thesis investigates the use of an Optically writable Random Access Memory (ORAM) based on warm cesium (Cs) vapor across both classical and quantum applications. A Cs vapor presents an ideal candidate for both classical and quantum memory due to its potential for long-lived ground-state storage, readily accessible multimode capacity, and reduced technical overhead. Our investigation begins in the classical regime, exploring the operation of an incoherent ORAM based on atomic population pumping. We experimentally implement this memory and integrate it into a machine learning architecture as a novel optical reservoir computer. By evaluating the memory against standard benchmarking tasks, we identify the physical constraints responsible for its limited memory capacity and critically assess its practical viability for reservoir computing. Following this classical demonstration, the focus shifts to the coherent storage of quantum information. We introduce a quantum channel formalism to model warm vapor quantum memories, and use it to evaluate the critical figures of merit necessary for secure applications such as quantum tokens. To address the performance requirements set by different quantum information processing applications, we present a series of experiments optimizing memory efficiency, noise, and multimode capacity. This includes deploying an in-situ machine learning algorithm to optimize temporal control pulses, experimentally confirming theoretical models of pulse shaping. By contrasting both the incoherent and coherent platforms, we outline the specific experimental advancements required to reach application-ready optical memories in a warm Cs vapor. The contributions presented in this thesis can be summarized as follows: - The formulation and investigation of incoherent atomic pumping as a means of classical optical storage. - The first use of a warm Cs vapor as a randomly addressable classical memory in a reservoir computing system. - A channel formalism implementation of atomic ensemble quantum memories, with performance analyses comparing the behavior of different state-of-the-art memory experiments. - The simultaneous optimization of several experimental quantum memory parameters, including the demonstration of a random-access spatially multiplexed coherent memory in a warm Cs vapor. - The first demonstration of an in-experiment machine learning algorithm simultaneously optimizing the efficiency and pulse energy in a warm atomic vapor. This thesis provides new insights into the practical capabilities of multimode warm vapor memories, establishing a foundation for their continued development across both classical and quantum information processing regimes.