Ningfeng Bai

Terahertz (THz) metamaterials provide a promising platform for polarization control, which is essential for imaging, sensing, and communication. However, realizing independent manipulation of orthogonal linear polarizations and circular polarization control with high quality ( Q ) factors under normal incidence in a THz platform remains a major challenge. Here, we numerically and experimentally demonstrate a bound state in the continuum (BIC)-based THz metamaterial that achieves multi-polarization control under normal incidence via stepwise symmetry breaking. With in-plane symmetry preserved, the transverse magnetic (TM) resonance is supported by guided mode resonance (GMR). Breaking the in-plane C ₂ rotational symmetry excites a high- Q transverse electric (TE) response via quasi-bound states in the continuum (q-BICs), while leaving the TM mode unaffected, thereby realizing independent control of orthogonal linear polarizations. By further breaking the in-plane mirror symmetry, a high- Q , near-unity, and reversible circular dichroism (CD) effect is realized under normal incidence by leveraging tunable topological polarization, allowing selective control of right-handed and left-handed circularly polarized waves (RCP and LCP). This work provides a practical route to integrating linear and circular polarization control within a single platform, opening new avenues for multifunctional THz polarization devices.

This letter proposes a sandwich metamaterial window (SMW) for a W-band TE01 mode gyrotron travelling-wave tube (gyro-TWT), which is optimized using machine learning (ML) to achieve high transmissivity and low reflectivity in a wideband range. This SMW has three layers, BN-metal-BN, where the BN has a thickness of 1.05 mm. The metal layer is composed of hexagonal units allowing to improve the cold-test characterization of SMW. The ML optimization makes the SMW have a high simulation performance with S11 less than −20 dB and S21 higher than −0.14 dB within 12.15 GHz. Compared to conventional designs, the proposed SMW provides a thicker one to achieve comparable performance through a simplified fabrication process. Experimental validation confirms alignment with simulation results, demonstrating the feasibility of this approach. This work provides a scalable and cost-efficient methodology for high-frequency metasurface window design.

Nonreciprocal transmission, a critical functionality for terahertz (THz) isolators, remains challenging to achieve in the compact and bias-free systems. This work proposes a THz metamaterial leveraging Brillouin zone folding (BZF)-induced bound states in the continuum (BZF-BICs) to realize robust nonreciprocal transmission. At an incident intensity of 60 MW/cm², the transmission spectra for opposite propagation directions exhibit stark asymmetry at 0.854 THz, yielding 44.87% and 0.18% transmission, where the isolation is 23.97 dB. In contrast, quasi-bound states in the continuum (Q-BICs) under identical conditions show negligible nonreciprocal transmission, where the transmission approaches zero in both directions at 0.856 THz. The enhanced performance of BZF-BICs arises from their high-quality (high-Q) resonance, enabled by reduced radiation losses through Brillouin zone folding. Further analysis reveals that the nonreciprocal intensity range (NRIR) depends on out-of-plane structural asymmetry, with optimized geometries achieving NRIR values up to 2.06. This work establishes BZF-BICs as a promising platform for integrated THz isolators.

This article presents an inverse design with transfer learning based on neural network (ID-TL-NN) for the rapid design of electron guns, which expands the range of the structural parameter designs through TL. This ID-TL-NN method can quickly predict electron beam trajectory envelopes and beam waist radius based on given structural parameters. Moreover, it has inverse design function, which can rapidly design corresponding electron gun structures based on given target electron beam envelopes and beam waist radius. The simulation results show that the beam waist radius error is less than 5% compared with the value of the target radius. Furthermore, through TL, the proposed model can extend the range of the structural parameters of the electron gun, achieving high-precision design with only a small number of samples. The model trained with a limited sample set predicts a beam waist radius error of 5%. Compared with traditional methods, this approach significantly increases the efficiency and accuracy of the electron gun design.

The metasurfaces are able to flexibly control electromagnetic waves. However, their design necessitates strict phase matching, which requires high computational load, and its control method has low flexibility. This paper proposes a novel method for the customization of the electric field of metalens with high degrees of freedom based on an improved Pixel Generative Adversarial Network (I‐pixGAN). This I‐pixGAN can design the phase distribution of the metalens according to the customized target electric field, and the constructed metalens based on this phase can modulate the target electric field. The results show that the proposed I‐pixGAN can arbitrarily control the position of the focus in three‐dimensional space, where the focus position is offset within one wavelength. Therefore, it is not need to strictly follow the electric field distribution characteristics, and it can flexibly construct unknown electric field, realizing a highly free electric field customization method in three‐dimensional space, providing a new way for electric field operation. This paper promotes the development of electric focusing systems, not only improving focusing accuracy but also enhancing system flexibility and promoting the automation and intelligence of optical and terahertz systems.

The high speed, wide bandwidth, and parallel processing capabilities of a diffractive optical neural network (DONN) stimulate its applications in computer vision for image recognition and information processing tasks. This paper presents a joint loss function (J-SCE) that combines classification performance and diffractive power efficiency, thereby improving the power efficiency of the DONN classifier from 0.92% to 12.89% while maintaining a classification accuracy of 95.36%. The J-SCE function improves the overall power efficiency of the system by directing energy more effectively toward the target area. Furthermore, the J-SCE function enhances the system's robustness to noise and overall stability. This work significantly contributes to the application of DONN classifiers in practical image recognition and other information processing scenarios.

The integration of tunable characteristics and dual functions into a single terahertz (THz) metamaterial is quickly becoming a major focus of research. This paper presents a nonvolatile switchable dual-functional THz metamaterial with selective transmission and adjustable absorption based on quasi-bound states in the continuum (q-BICs), where the q-BICs are achieved through periodically arranged silicon disks tetramer. By introducing asymmetry perturbation through manipulating the radii of diagonal silicon disks of the proposed THz metamaterial, resulting in the transition from symmetry-protected bound states in the continuum (SP-BICs) to q-BICs. When the middle layer Ge₂Sb₂Te₅ (GST) is in the amorphous state, selective transmission is achieved, which can be applied as a bandstop filter. Additionally, by transforming the phase of GST into the crystalline state in the way of heating, the proposed THz metamaterial behaves as an adjustable absorber due to the generations of q-BICs and Fabry-Pérot resonance (FPR) mode. Multipole decomposition and temporal coupled-mode theory (CMT) are applied to explain the underlying physics. Our work provides valuable insights into the integration of the nonvolatile switchable and dual-functional THz metamaterials empowered by q-BICs.