Towards Gigabit-scale digital signal storage enabled by a distributed Raman-amplified optical memory

Optical computing promises ultrahigh-speed and energy-efficient information processing, offering unique advantages for data-intensive tasks such as machine-learning inference, image processing, and scientific simulations. Yet, the scalability of such systems remains fundamentally constrained by the absence of high-capacity optical memories capable of storing intermediate computational states without resorting to electronic conversions. Here we propose and experimentally demonstrate an all-optical digital memory based on a distributed Raman-amplified frequency-shifting loop, designed to function as an optical cache within large-scale photonic computing pipelines. The system enables long-duration, low-noise circulation of optical waveforms while maintaining high signal fidelity. We experimentally validate the storage of both grayscale and color image data and achieve an effective number of bits (ENOB) of 3 bits and a storage time of 20 ms. The experimental results show that for a single channel, the ideal maximum storage capacity is 3.2 Mbit and the effective storage capacity is 1.2 Mbit; for RGB channels, the ideal maximum storage capacity is 9.6 Mbit and the effective storage capacity is 3.2 Mbit. According to the theoretical calculations, the ideal storage capacity of the system can reach 3.84 Gbit, while the effective storage capacity is approximately 1.44 Gbit. These results establish a practical route toward scalable intermediate data storage in optical computing systems. By addressing one of the central bottlenecks in photonic computation, the lack of high-fidelity, high-capacity optical storage, this work advances the feasibility of fully optical computing architectures with substantial potential across domains requiring high-throughput, low-latency processing.