Metallic plasmonic nanosensors that are ultra-sensitive, label-free, and operate in real time hold great promise in the field of chemical and biological research. Conventionally, the design of these nanostructures has strongly relied on time-consuming electromagnetic simulations that iteratively solve Maxwell's equations to scan multi-dimensional parameter space until the desired sensing performance is attained. Here, we propose an algorithm based on particle swarm optimization (PSO), which in combination with a machine learning (ML) model, is used to design plasmonic sensors. The ML model is trained with the geometric structure and sensing performance of the plasmonic sensor to accurately capture the geometry-sensing performance relationships, and the well-trained ML model is then applied to the PSO algorithm to obtain the plasmonic structure with the desired sensing performance. Using the trained ML model to predict the sensing performance instead of using complex electromagnetic calculation methods allows the PSO algorithm to optimize the solutions fours orders of magnitude faster. Implementation of this composite algorithm enabled us to quickly and accurately realize a nanoridge plasmonic sensor with sensitivity as high as 142,500 nm/RIU. We expect this efficient and accurate approach to pave the way for the design of nanophotonic devices in future.
Yan RuoQin, Wang Tao, Jiang Xiaoyun, Zhong Qingfang, Huang Xing, Wang Lu, Yue XinZhao
machine learning, nanosensor, particle swarm optimization, plasmonic