面向6G移动通信Massive MIMO系统的深度学习光子芯片研究

本课题面向6G移动通信技术展开研究，针对Massive MIMO系统预编码矩阵求逆过程所需算力随规模呈指数增长的问题，力求解决传统算法和硬件面临的计算性能、速度和能耗等限制瓶颈。本课题计划构造深度学习算法，降低矩阵求逆运算复杂度，提高并行处理性能。在此基础上，通过超表面光芯片直接架构深度学习硬件，从而提出高速度、高效率的深度学习矩阵求逆硬件设计方案，从而进一步推动光子计算芯片在空天地海立体化的6G通信技术领域的发展。

This project is to address the challenges in developing 6th generation mobile communication technology, in which massive MIMO pre-coding requires inverting large-scale matrices that leads to exponential growth in demand of computing power. To solve the bottlenecks of performance, speed and energy consumption faced by traditional algorithms and hardware, this project plans to construct Deep Learning algorithms to reduce the complexity of matrix inversion and improve parallel processing performance. Based on this, we want to propose a high speed, high efficiency photonic deep learning matrix inversion hardware architect based on all-dielectric metasurfaces, and paves the roads for developing photonic computing chips for 6G mobile communication technology in land, sea, sky and even space.