中心博士生易思成同学的工作——32-bit photonic processor beyond noise limitation based on parallelized bit-slicing的相关成果近期被Optics Express期刊接收发表，该工作得到了国家自然科学基金（T2225023，62205203）的部分资助。
光学计算已成为大规模数据信号处理的一种高效方法，与传统电子架构相比具有显著优势，包括卓越的并行性和能效。然而，光学计算芯片的实际实现受到了几个固有挑战的阻碍，例如器件加工的不均一性，光路不均一性和系统噪声，所有这些都会降低计算精度。本工作展示了一种超越光计算极限的光学方案，该方案通过位切片技术，将高精度计算任务分解为多个低精度计算的组合，从而在低精度光学模拟设备上实现高精度计算。基于这一方法，本工作在实验中成功实现了32位高精度光子计算，并在图像分割、识别任务中较传统光计算显著提升。这一突破不仅证明了光学计算在高精度计算任务中的潜力，也为光学计算芯片在人工智能、大数据处理等领域的实际应用提供了新的技术路径。
摘要：
Photonic processors have emerged as powerful platforms for large-scale signal processing, offering distinct benefits over traditional electronic computing, such as large bandwidth, superior parallelism, and energy efficiency. However, their practical deployment is limited by inadequate computational precision owing to device insertion loss, link noise, and uneven chip fabrication. This study introduces a 32-bit photonic processor that overcomes the limitations of conventional photonic processors. The proposed approach employs the parallelized bit-slicing principle to decompose high-precision operations into multiple low-precision operations, thereby enabling high-precision computing using low-precision analog photonic devices. This study demonstrates 32-bit photonic computing and yields substantial performance improvements over conventional approaches in computing tasks such as image segmentation and recognition. This breakthrough mitigates the impact of nonideal conditions on computational precision and facilitates the practical deployment of photonic processors in artificial intelligence-driven and large-scale data processing scenarios.