All-optical convolution utilizing processing in memory based on a cold atomic ensemble
Ying-Hao Ye, Jia-Qi Jiang, En-Ze Li, Wei Zhang, Da-Chuang Li, Zhi-Han Zhu, Dong-Sheng Ding, Bao-Sen Shi · 2025
Read All-optical convolution utilizing processing in memory based on a cold atomic ensemble by Ying-Hao Ye, Jia-Qi Jiang, En-Ze Li, Wei Zhang, Da-Chuang Li, Zhi-Han Zhu, Dong-Sheng Ding, Bao-Sen Shi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-14716.txt.
Introduction
Ying-Hao Ye and seven co-authors run a classic image task inside a cloud of cold atoms, in a 2025 physics paper. Computing inside memory, rather than shuttling data to a separate chip, promises speed and low delay. For light-based computing, a coherent memory is the missing piece.
The team stored photon images in the atomic cloud using a transparency effect that maps light patterns onto atoms and back. While stored, the image was processed to sharpen its edges, a convolution step basic to vision. The memory held its coherence for over 320 microseconds, long enough for the operation. Retrieved images, caught by a photon-timing camera, showed the enhanced edges. The feat points toward optical processors that think where they store.
Catalog record
| CREATOR | Ying-Hao Ye, Jia-Qi Jiang, En-Ze Li, Wei Zhang, Da-Chuang Li, Zhi-Han Zhu, Dong-Sheng Ding, Bao-Sen Shi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-14716
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
All-optical convolution utilizing processing in memory based on a cold atomic ensemble Preprint: APS/123-QED Ying-Hao Ye, Jia-Qi Jiang, En-Ze Li, Wei Zhang, Da-Chuang Li, Zhi-Han Zhu, Dong-Sheng Ding, Bao-Sen Shi Abstract Processing in memory (PIM) has received significant attention due to its high efficiency, low latency, and parallelism. In optical computation, coherent memory is a crucial infrastructure for PIM frameworks. This study presents an all-optical convolution experiment conducted within computational storage based on a cold atomic ensemble. By exploiting the light-atom phase transfer facilitated by the electromagnetically induced transparency, we demonstrated spiral phase contrast processing of photon images in memory, resulting in the edge enhancement of retrieved images recorded using time-correlated photon imaging. In particular, adopting state-of-the-art atomic techniques provides a coherent memory lifetime exceeding 320 µs for PIM operations. Our results highlight the significant potential of cold atomic ensembles as computational storage for developing all-optical PIM systems. I Introduction Cold atomic ensembles present promising light-matter interfaces for all-optical processing, primarily due to their collective enhancement effects Lukin 2003 , consistent spectral properties, and reduced Doppler broadening. The enhanced strength of light-matter interactions simplifies the otherwise stringent experimental requirements, such as high-Q resonators, for increasing the absorption cross-section of a single-atom system Kimble 1998 and facilitates operations in specific optical modes Zhao and Pan 2013 . Moreover, the consistent spectral characteristics, combined with the suppression of Doppler broadening, offer significant advantages for constructing and scaling atomic-based networks Sangouard et al. 2011 . One of the most compelling applications of cold atomic ensembles is their use as storage media for photonic information Chanelière et al. 2018 ; Lvovsky et al. 2009 . The long coherence times of atomic metastable states and the reduced particle drift in low-temperature systems support robust optical storage Zhao et al. 2009 , provided that ambient magnetic fields are precisely controlled Matsukevich et al. 2006 ; Jenkins et al. 2006 ; Matsukevich et al. 2006 . Researches have demonstrated long-lived and faithful memory across various photonic degrees of freedom, including polarizations Vernaz-Gris et al. 2018 , wavevector directions Wang et al. 2021 , and temporal modes Heller et al. 2020 . Among these, the transverse modes of the stored field are particularly promising for image-correlation applications, offering tremendous flexibility for encoding information Wang et al. 2012 . To date, studies have reported successful storage of complex intensity Shuker et al. 2008 or phase patterns of photons and have even extended into the realm of single-photon storage Nicolas et al. 2014 ; Ding et al. 2013 . A widely used protocol for memorizing transverse multimode data employs the optical modulation technique known as electromagnetically induced transparency (EIT) Fleischhauer et al. 2005 . In this approach, photonic information is actively transferred to the profile of metastable collective atomic excitations, or spin waves, in a manner that is lossless and reversible Lukin 2003 . The processes of storage (write-in) and retrieval (read-out) can be mathematically described using Heisenberg equations Fleischhauer and Lukin 2000 ; Fleischhauer and Lukin 2002 . By incorporating transverse spatial degrees of freedom into these equations, we can identify methods to manipulate the collective excitations’ profile by tailoring the wavefronts of the control fields. Cold atom ensembles provide essential infrastructure for all-optical processing within Processing-in-Memory (PIM) frameworks due to their storage capacity and potential for flexible manipulation. As a burgeoning area of research, PIM systems offer significant advantages, such as breaking the von Neumann bottleneck Zou et al. 2021 , improved efficiency, and parallelism by alleviating the latency and energy cost due to data transmission between processing and memory units. Figure 1: Experimental scheme. (a) Conceptual graph of all-optical convolution in computational storage. (b) Schematic representation of the experimental setup, highlighting key components such as the magneto-optical trap (MOT), image target (Mask), spatial light modulator (SLM), and an intensified charge-coupled device (ICCD) operating in the Time Correlated Single Photon (TCSP) imaging mode. The energy level structure of the EIT protocol is depicted in the upper right corner. (b) An example of time sequence adopted for a storage time of 400 \mu s , together shown the signal pulse registered by a PMT. The transfer of spiral phases between collective excitation and photons has been theoretically proposed in an off-resonant Raman memory protocol Vashukevich et al. 2020 . Here, we generalize this concept to the on-resonance EIT protocol and report an experimental demonstration for optical PIM based on an atomic medium by performing all-optical convolution of images in memory. The feasibility of our approach is illustrated through an edge-detection technique called spiral phase contrast (SPC) Fürhapter et al. 2005 ,in which the vortex phase of control fields is ’imprinted’ onto spin waves during both the write-in and read-out stages, acting as modulation transfer functions on the Fourier spectrum. A 4 f imaging setup performs the convolution between the original patterns and the vortex phases, and the read-out pattern are faithfully captured by an intensified charge-coupled device (ICCD) without sacrificing brightness, thanks to the interferometric and centrally symmetric nature of the SPC process Davis et al. 2000 . For a longer attainable on-demand duration of storage, we minimize destructive interference among different Zeeman sub-levels of atomic coherence by mapping photonic information to a pair of magnetically insensitive states. Additionally, we apply a strong guiding magnetic field to suppress inhomogeneous evolution across the cross-section Ye et al. 2021 . We finally achieved on-demand read-out of processed images for over 320 \mu s , which is, to the best of our knowledge, the most extended duration reported for an atomic-based memory to date. II Principle and Methods FIG. 1 (a) illustrates the concept of all-optical convolution within the PIM framework, wherein a cold atomic ensemble is positioned at the spectral (or Fourier) plane of the 4f optical convolution system. This ensemble serves as the memory medium for all-optical PIM, capable of performing convolution computations, storing results, and subsequently releasing them upon request for further processing. FIG. 1 (b) depicts a more detailed scheme of the experimental setup, where coherence memory is implemented using the EIT protocol based on a Lambda-type level structure, as illustrated in the upper right corner of FIG. 1 (b). The states \left|g\right\rangle , \left|s\right\rangle and \left|e\right\rangle correspond to the hyper-fine energy levels \left|5S_{1/2},F=2\right\rangle , \left|5S_{1/2},F=3\right\rangle and \left|5P_{1/2},F=3\right\rangle of Rb85, respectively. The signal field ( \mathbf{E}(\vec{\mathbf{r}}) ) and the control field ( \mathbf{U}(\vec{\mathbf{r}}) ) resonantly address the transitions \left|g\right\rangle\leftrightarrow\left|e\right\rangle and \left|s\right\rangle\leftrightarrow\left|e\right\rangle . The entire process of storage and retrieval is commonly represented as the propagation of quasi-bosons, known as Dark State Polaritons (DSPs), within the storage medium Fleischhauer and Lukin 2000 ; Fleischhauer and Lukin 2002 . The composition of the photonic and atomic excitation components in a DSP, along with its group velocity, is governed by the Rabi frequency of the control light Fleischhauer and Lukin 2000 ; Fleischhauer and Lukin 2002 . During the write-in stage, DSPs become purely atomic, and their group velocity drops to zero as the control light is adiabatically turned off. This transition facilitates the transfer of the spatial profile from the light fields into atomic excitation. In the subsequent read-out stage, the control light is adiabatically reintroduced, causing the DSPs to become photonic with its group velocity increases until they escape