Main > Advanced Photonics >  Volume 2 >  Issue 2 >  Page 026003 > Article
• Figures
• Abstract
• Figures (7)
• Tables (0)
• Equations (5)
• References (89)
• Suppl. Mat.
• Get PDF
• View Full Text
• Paper Information
• Received: Mar. 15, 2020

Accepted: Apr. 13, 2020

Posted: Apr. 30, 2020

Published Online: Apr. 30, 2020

The Author Email: Xu Lei (lei.xu@ntu.ac.uk), Rahmani Mohsen (mohsen.rahmani@anu.edu.au), Ma Yixuan (yixuanma@mail.nankai.edu.cn), Smirnova Daria A. (daria.smirnova@anu.edu.au), Kamali Khosro Zangeneh (khosro.zangeneh@anu.edu.au), Deng Fu (u_deng@foxmail.com), Chiang Yan Kei (y.chiang@adfa.edu.au), Huang Lujun (lujun.huang@adfa.edu.au), Zhang Haoyang (zhangh49@qut.edu.au), Gould Stephen (stephen.gould@anu.edu.au), Neshev Dragomir N. (Dragomir.neshev@anu.edu.au), Miroshnichenko Andrey E. (andrey.miroshnichenko@unsw.edu.au)

• Get Citation
• ##### Copy Citation Text

Lei Xu, Mohsen Rahmani, Yixuan Ma, Daria A. Smirnova, Khosro Zangeneh Kamali, Fu Deng, Yan Kei Chiang, Lujun Huang, Haoyang Zhang, Stephen Gould, Dragomir N. Neshev, Andrey E. Miroshnichenko. Enhanced light–matter interactions in dielectric nanostructures via machine-learning approach[J]. Advanced Photonics, 2020, 2(2): 026003

• Category
• ##### Research Articles
• Share

Fig. 1. (a) Top: (top left) Schematics of the silicon nanobars metasurface and (top right) its unit cell. Bottom: Calculated transmission spectrum of the metasurface with structural parameters $w=316 nm$, $L=580 nm$, $x0=189 nm$. (b) Spherical multipolar structure of the metasurface. (c) Top: Cartesian ED and TD modes excitations. Bottom: The electric energy enhancement $ηE/ηE0$. It is defined as the electric energy inside the two nanobars normalized by the electric energy within the same volume of the nanobars for the pump field. (d) Electric near-field distributions at the resonance. Left: 3-D view. Right: top view.

Fig. 2. The architecture of the TN model, which consists of an inverse-design network connected to a pretrained forward model network. $X$ represents the input and output, which is the transmission spectra data in our case, and $Y$ represents the output in the middle layer which is the structural parameters here.

Fig. 3. (a) Evolution of the training loss for the forward model network. (b) Comparison of the NN approximation to the real transmission spectrum. (c) Evolution of the training loss for the inverse-design model network. (d) Comparison of the spectra between the NN approximation and the input based on Eq. (2).

Fig. 4. Inverse design of Si nanobar metasurfaces with Fano-shape transmission spectra. (a)–(c) $λ0=1450$, 1500, and 1550 nm, respectively. $Δλ=15 nm$, $q=0.8$. (d)–(f) $λ0=1500 nm$, $Δλ=10 nm$, $q=0.3$, 0.5, and 0.7, respectively. (g)–(i) $λ0=1500 nm$, $Δλ=5$, 15, and 25 nm, respectively, $q=0.7$.

Fig. 5. (a) SEM image of the fabricated sample with designed resonance at 1500 nm. (b) Experimentally measured linear spectra. (c) Experimentally measured THG spectra of the samples.

Fig. 6. (a)–(c) Optomechanic vibration under the $y$-polarized pump. (a) Displacement of the nanobars after 1 ns. (b) The transient displacement $Dx$ and $Dy$. (c) Spectral densities of displacement $Dx$ and $Dy$. (d)–(f) Optomechanical vibration under the $x$-polarized pump. (d) Displacement of the nanobars after 1 ns. (e) The transient displacement $Dx$ and $Dy$. (f) Spectral densities of displacement $Dx$ and $Dy$.
Fig. 7. The spectral density of $D$ in the (a) $x$ and (b) $y$ directions for different laser pump wavelengths.