Multilayer Q-BIC-like Optical Filters with High Throughput Direct-Write Multilayer Lithography

TL;DR

The paper addresses the fabrication bottlenecks of multilayer metasurfaces in the visible/NIR and introduces a high-throughput direct-write platform using an Sb-BDCA precursor that in situ converts to $Sb_2S_3$, combined with HSQ planarization to enable multilayer integration. By engineering three-layer q-BIC–derived metasurfaces with independently tunable resonance wavelengths and $Q$ factors, the work demonstrates spectrally decoupled layers and multi-resonant filters suitable for compact on-chip spectroscopy. It further constructs decorrelated filter arrays for compressive sensing, achieving average pairwise Pearson correlations of $|r_{xy}|=0.11$ (9 filters) and $|r_{xy}|=0.21$ (36 filters) across 1100–1700 nm, outperforming prior metasurface and photonic-crystal implementations. Overall, the approach provides a scalable route to multilayer spectral filters with programmable spectra for hyperspectral imaging and computational spectral reconstruction.

Abstract

Multilayer metasurfaces provide substantially greater spectral design freedom than single-layer devices, yet their implementation in the visible and near-infrared remains limited by the complexity, cost, and low throughput of conventional nanofabrication. Here, we establish a recently proposed direct-write electron-beam lithography approach as a high-throughput fabrication platform for multilayer resonant metasurfaces, based on an antimony precursor that decomposes in situ into high-index antimony sulfide. This method eliminates deposition-etch cycles and reduces each layer to only two fabrication steps, enabling efficient realization of multilayer architectures. Using this platform, we demonstrate multilayer q-BIC-derived metasurfaces with independently tunable resonance wavelengths and linewidths, allowing the construction of compact multi-resonant filters with spectrally decoupled layers. We experimentally demonstrate three-layer devices supporting three resonances and show independent control of resonance wavelength and Q factor across layers. Leveraging this capability, we generate decorrelated filter arrays for compressive sensing and hyperspectral reconstruction, achieving sets of 9 and 36 filters with average absolute Pearson correlation coefficients of 0.11 and 0.21, surpassing prior metasurface and photonic-crystal implementations. These results establish a practical route toward scalable multilayer resonant metasurfaces for spectral filtering, on-chip spectroscopy, and computational imaging.

Figures (9)

• Figure 1: (a) An example flow of standard nanofabrication processes to fabricate a multilayer metasurface. (b)Proposed Sb$_2$S$_3$ direct-write process for multilayer metasurfaces.
• Figure 2: (a)The unit cell of the metasurface made of two Sb$_2$S$_3$ tilted posts on top of a 1.3 um thick SiO$_2$ spacer on Si substrate and the definition of parameters $\alpha$ and $\theta$. (b)Simulated transmission for the square lattice ($\alpha$=1, $P_y$=$P_x$=887 nm, $\theta$=35$^circ$), and the norm of the electric field for the inter-post (top) and intra-post (bottom) modes given in the orange boxes. (c)Simulated transmission for the rectangular lattice ($\alpha$=1.34, $P_x$=673nm, $P_y$=$\alpha P_x$=902nm, $\theta$=35$^o$), and the norm of the electric field for the inter-post mode given in the green box. (d)Simulated transmission shown for $\theta$=35$^o$ and $\alpha$=1.202, 1230,.., 1.477.The solid lines refer to the parameter configurations used in Figure \ref{['fig:fig_3layer_3wvl']}.b. (e)The change in the resonance linewidth with $\theta$=5$^o$,12.5$^o$,..,80$^o$. (f)SEM images of the fabricated Sb$_2$S$_3$ elliptical posts for $\theta$=20$^o$ (left) and $\theta$=35$^o$ (right).
• Figure 3: (a)The supercell for the rectangular lattice 3 layer device simulation; the side view shows the different number of iterations per layer and the periodic boundary conditions used on sides in FDTD. (b)Single layer simulations for ($\alpha$=1.23,$\theta$=35$^o$), ($\alpha$=1.34,$\theta$=35$^o$), and ($\alpha$=1.45,$\theta$=35$^o$) on the left; ($\alpha$=1.23,$\theta$=20$^o$), ($\alpha$=1.38,$\theta$=65$^o$), and ($\alpha$=1.45,$\theta$=35$^o$) while $P_x$=673nm is kept constant. (c)Simulated device transmission normalized with respect to the transmission through SiO$_2$/Si substrate for the 3 layer supercell simulation with the parameters given in b respectively. (d)Measured device transmission normalized with respect to the measured transmission through the SiO$_2$/Si substrate, for the devices simulated in c. (e)The norm of the electric field for the resonances given in c, the largest wavelength resonators sit at the bottom layer and the smallest wavelength resonators sit at the top layer.
• Figure 4: (a)The supercell for the square lattice 3 layer device simulation. (b)Single layer simulations for ($P_y$=$P_x$=802nm,$\theta$=35$^o$),($P_y$=$P_x$=893nm,$\theta$=35$^o$), and ($P_y$=$P_x$=1002nm,$\theta$=35$^o$). (c)Simulated device transmission normalized with respect to the transmission through SiO$_2$/Si substrate for the 3 layer supercell simulation with the parameters given in b. (d)Measured device transmission normalized with respect to the measured transmission through the SiO$_2$/Si substrate, for the device simulated in c.
• Figure 5: (a)Pairwise correlation coefficients for the selected 9 devices with average absolute value of the pairwise correlation coefficients 0.11 (left) and the heatmap for the transmission spectra of these filters (right). (b)Pairwise correlation coefficients for the selected 36 device with average absolute value of the pairwise correlation coefficients 0.21 (left) and the heatmap for the transmission spectra of these filters (right).