基于光信号在线掩码及电信息注入下VCSELs复用网络的储备池计算

储备池计算(RC)是可在硬件实施的非传统信息处理方法，能高效处理语音识别等复杂任务，其中基于半导体激光器(SL)实施的延时型RC(SL-DRC)因算法简单、速率快、能耗低而倍受关注。已报道的SL-DRC在输入层均采用离线掩码方式而不能摆脱昂贵的数据注入设备，也不能进行实时的掩码操作；储备池通常采用单个SL带延迟反馈环的结构，无法突破单非线性节点RC的速率瓶颈（目前在MSa/s量级）。本项目拟构建光信号在线掩码及电信息注入下基于垂直腔面发射激光器(VCSELs)复用网络的RC系统，研究低成本、在线掩码，高数据处理速率(≥1GSa/s)的SL-DRC。具体为：在输入层研究基于SL在光反馈下产生优质光掩码信号实现在线掩码，基于现场可编程门阵列注入数据；在储备池层构建基于VCSELs的复用网络，突破SL-DRC的速率瓶颈。本项目有望开发出在线掩码的高速RC系统，促进RC在人工智能领域的应用。

Reservoir computing (RC) is a non-traditional information processing method, which can be implemented in hardware. It can efficiently perform complex tasks such as speech recognition. The delay-based RC implemented by semiconductor laser (SL-DRC) has attracted much attention due to its simple algorithm, fast speed, and low power consumption. The previously reported SL-DRCs adopt offline mask method at the input layer, which cannot get rid of expensive data injection equipment, nor can realize masking operation in real time. Meanwhile, the reservoirs usually adopt a structure of single SL with delay feedback loop, which cannot breakthrough the speed bottleneck (about MSa/s) of RC system with single nonlinear node. In this project, we plan to construct a RC system of multiplexed network composed by vertical-cavity surface-emitting lasers (VCSELs) under optical signal online mask and electrical information injection. We plan to investigate the SL-DRC with low-cost online mask and high data processing rate (≥1GSa/s). Firstly, at the input layer, we investigate online mask by high-quality optical mask signal based on a SL subject to optical feedback, and inject the data into reservoir by a field programmable gate array. Secondly, at the reservoir layer, we plan to construct a multiplexed network based on VCSELs to breakthrough the data processing rate bottleneck of SL-DRC. Through the implementation of this project, it is expected to develop a high-speed RC system with online mask, which will promote the application of RC in the field of artificial intelligence.