Nonlinear 2D materials such as 3R-phase molybdenum disulfide (3R-MoS2) offer strong second-order optical nonlinearities in an atomically thin platform, making them attractive for on-chip frequency conversion, quantum light generation, and integrated nonlinear nanophotonics. However, the second harmonic generation (SHG) efficiency of monolayer or few-layer 3R-MoS2 deposited on planar substrates remains fundamentally limited by weak light–matter interaction, poor phase matching, and small interaction volumes. Multi-layer metasurfaces (MLMs) provide expanded geometric and material degrees of freedom, enabling dual-resonant cavities that can independently and simultaneously enhance the fundamental (pump) and second-harmonic fields at the 3R-MoS2 location. Yet, traditional trial-and-error or forward-only optimization becomes impractical for designing such dual-resonant, multi-parameter structures due to the vast design space and the need to co-optimize linear and nonlinear responses. Here, we introduce NanoPhotoNet-PINL, a physics informed AI-driven inverse design framework based on a hybrid one-dimensional convolutional neural network and deep neural network autoencoder, tailored for nonlinear MLMs metasurfaces. The model numerically maps target dual-resonant reflection spectra at the fundamental and second-harmonic wavelengths to the required multi-layer geometries and material compositions that maximize the effective nonlinear overlap with an embedded 3R-MoS2 sheet. By integrating Maxwell-based nonlinear electrodynamics into the inverse design loop, we compute the second-harmonic conversion efficiency and modal overlap factors for each predicted MLMs design, enabling physics-guided training and evaluation. Our approach achieves an inverse-design prediction efficiency exceeding ∼99.2% along the linear spectral manifold, while the optimized dual-resonant MLMs yield more than three orders of magnitude enhancement in SHG intensity compared to a bare 3R-MoS2 flake on a planar substrate. These substantial increases in nonlinear conversion and radiation efficiency position the 3R-MoS2/MLMs platform as a scalable route towards compact, CMOS-compatible nonlinear metasurface devices for on-chip frequency converters, quantum light sources, and all-optical signal processing. Consequently, NanoPhotoNet-PINL establishes a generalizable paradigm for intelligent inverse design of nonlinear multi-layer metasurfaces and phase-matched dual-resonant cavities for high-efficiency second-order processes.
Lookup-table-based digital predistortion (LUT-DPD) provides a low-complexity approach to mitigating pattern-dependent transmitter nonlinearities. However, conventional LUT-DPD relies on discrete symbol levels for table indexing and is therefore difficult to apply directly to continuous-valued multicarrier waveforms such as discrete multitone (DMT) signals. In this paper, a novel LUT-DPD scheme based on delta sigma modulation (DSM) is proposed. The DMT waveform is first quantized using a fourth-order DSM, which enables pattern indexing with fewer amplitude levels and subsequent LUT predistortion, followed by signal reconstruction through low-pass filtering and downsampling. The proposed scheme is experimentally evaluated in a 132-Gbit/s C-band DMT 64-ary quadrature amplitude modulation (QAM) optical back-to-back intensity modulation/direct detection (IM/DD) link with a 25-GHz signal bandwidth. At the 7% HD-FEC BER threshold of 3.8e-3, receiver sensitivity improvements of 0.5 dB and 0.7 dB are achieved for LUT memory depths of 3 and 5, respectively. The results indicate that DSM-assisted indexing can extend LUT-DPD to DMT signals while maintaining manageable table sizes.
Optical asymmetric cryptosystems based on Qin and Peng’s algorithm, phase truncated Fourier transform, were found to be vulnerable to specific attacks due to an inherent deficiency in the transform domains. This paper presents an asymmetric encryption algorithm which uses fractional Hermite transform (FrHMT) that can resist these specific attacks. The proposed method integrates amplitude and phase-truncation techniques with FrHMT, where the fractional orders, Hermite polynomial orders, and Gaussian window parameters act as additional security parameters. Extreme sensitivity is provided to the simple scheme by FrHMT, as even a slight deviation in a single pixel of the ciphertext results in completely unsuccessful decryption. A detailed sensitivity analysis has been performed to evaluate the scheme’s robustness under varying encryption parameters, confirming its adaptability to diverse security requirements. Experimental results demonstrate that the proposed algorithm effectively resists known plaintext attack as well as specific attack, and maintains high computational performance.
This paper presents the numerical investigation of the temporal photonic quantum well using the transfer matrix method. The proposed structure consists of a temporal Bragg mirror-temporal disorder stack-temporal Bragg mirror, where a disordered PTC stack is sandwiched between temporal Bragg mirrors. The structure’s transmission peaks, reflection, and absorption are examined as the disorder strength and refractive index are varied. This combined effect of periodic and disordered structures leads to the emergence of controlled localized modes within the temporal bandgap. It provides control over light propagation in the time domain, including transmission, reflection, and absorption. Statistical analyses were performed using 1000 disorder realizations: the histogram of i) transmission, reflection, and absorption and its resonances of the Hermitian system with mean values of 2.4, 2.5, and 2.6, with four different standard deviations. ii) transmission, reflection, and absorption and its resonances of the non-Hermitian PTC with mean values of 2.4, 2.5, and 2.6, and a standard deviation of 0.5 for four different complex permittivities for each group. This structure features a clear momentum bandgap with controlled localization and tunable amplification, giving sharp transmission peaks, broadening its range of applications in tunable temporal filters, adaptive amplifiers, absorbers, and modulators.
To address the complex structural coupling and computationally intensive iterative parameter optimization of existing multi-Fano metasurfaces, as well as the insufficient training of inverse-design models caused by the limited availability of original samples, a cross-coupled ring all-dielectric metasurface supporting triple Fano resonances is designed. A limited-sample data augmentation and inverse-design method based on physics-constrained conditional flow matching (PCFM) is subsequently proposed. First, the transmission response of the designed structure is calculated using the finite-difference time-domain (FDTD) method, and the formation mechanism of the triple Fano resonances is elucidated through analyses of the magnetic-field distributions and multipole decomposition. To address the difficulty of conventional generative models in maintaining the physical consistency of augmented data under limited-sample conditions, a PCFM-based data augmentation model is developed to generate a highly reliable augmented dataset. The mean absolute relative error (MARE) and mean absolute error (MAE) between the wavelengths of the augmented samples and the corresponding reference wavelengths are 0.0097% and 0.089 nm, respectively. To represent the approximate one-to-many relationship between the target wavelengths and structural parameters, a PCFM-based inverse-design model is further established to generate multiple sets of candidate structural parameters from prescribed resonance wavelengths. The MARE and MAE between the wavelengths corresponding to the candidate structures and the target wavelengths are 0.0134% and 0.124 nm, respectively. The results indicate that the proposed method improves the physical reliability of augmented samples and the wavelength-matching accuracy of inverse design under limited-sample conditions, thereby enabling target-driven design of multiple candidate structures for triple-Fano all-dielectric metasurfaces.
