Xulei Qin

Neuromorphic imaging sensors (event cameras) offer a promising paradigm for computational imaging and human pose estimation (HPE) under extreme illumination conditions. Nevertheless, dark-scene background activity originating from photodiode dark current and circuit thermal noise, together with hot-pixel noise, severely corrupts event streams and impedes reliable HPE in low-light scenarios. To address this issue, we propose an adaptive event denoising framework built upon a spatiotemporal Gaussian-weighted neighborhood model with a dynamic thresholding mechanism. It can effectively suppress background activity and hot-pixel noise while preserving edge and motion details critical for pose estimation. Leveraging this denoising front-end, we construct a complete dark-scene neuromorphic HPE pipeline by transferring the pre-trained MediaPipe model onto event-based time-surfaces. Quantitative and qualitative evaluations on public and self-collected datasets demonstrate that our approach outperforms state-of-the-art denoising methods with an improvement of over 20% on public benchmarks and over 30% on self-collected dark-scene data. We expect our work to pave the way toward reliable dark-scene human–robot interaction through robust neuromorphic pose estimation.

This work demonstrates a high-performance vertical GaN p−i−n diode based on a buried p-layer n-p−i−n epitaxial structure. The post-etch magnesium (Mg) diffusion process is applied to suppress the etch-induced surface damage on the p-GaN layer. The Mg diffusion effectively reduces the valence band barrier from 2 to 1.1 eV, yielding a low specific contact resistivity of 6.521 × 10 ⁻⁴ Ω·cm ² . As a result, the fabricated devices exhibit markedly enhanced forward characteristics, including a reduced turn-on voltage of 3.3 V and a specific on-resistance of 0.92 mΩ·cm ² . Temperature-dependent forward I−V measurements indicate that the dominant carrier transport mechanism evolves from defect-related tunneling in the etched devices toward transport dominated by intrinsic p–n junction conduction after Mg diffusion. In addition, the devices exhibit excellent stability in forward conduction, with a voltage variation of approximately 0.028 V. These results indicate that Mg diffusion effectively improves the contact characteristics degraded by ICP etching and provide a viable approach for achieving high-performance and reliable vertical GaN power devices.

In this work, AlN films were grown using gallium (Ga) as surfactant on 4° off-axis 4H-SiC substrates via microwave plasma chemical vapor deposition (MPCVD). We have found that AlN growth rate can be greatly improved due to the catalytic effect of trimethyl-gallium (TMGa), but AlN crystal structure and composition are not affected. When the proportion of TMGa in gas phase was low, crystal quality of AlN can be improved and three-dimensional growth mode of AlN was enhanced with the increase of Ga source. When the proportion of TMGa in gas phase was high, two-dimensional growth mode of AlN was presented, with the increase of Ga source results in the deterioration of AlN crystal quality. Finally, employing a two-step growth approach, involving the initial growth of Ga-free AlN nucleation layer followed by Ga-assisted AlN growth, high quality of AlN film with flat surface was obtained and the full width at half maximum (FWHM) values of 415 nm AlN (002) and (102) planes were 465 and 597 arcsec.

In order to effectively remove the residual Ge atoms at the surface of channel and improve the interfacial characteristic of gate‐all‐around (GAA) Si nanosheet field effect transistors, a low‐temperature atomic‐level trimming approach using ammonia peroxide water mixture (APM) solution treatment followed by diluted hydrofluoric acid is proposed and carried out. The capacitor samples with SiGe epitaxy layer, similar thermal budget and channel release process were fabricated using advanced high‐k/metal‐gate. An 83% reduction in surface roughness at atomic level is obtained by increasing APM treatment time. Moreover, there are 99.45% reduction in the interface state density ( D _it ) and 96.8% leakage reduction in current density ( J _g ) after APM treatment, indicating a promising method for future GAA NSFET performance optimization.

In order to improve the charge collection efficiency of EBCMOS, we predicted and experimentally confirmed the doping structure of the electron multiplication layer. In this study, we simulate the charge collection efficiency of EBCMOS using various doping strategies by modeling the collisional scattering of electrons with solid atoms in semiconductor materials and combining the travel paths of electrons in the electron multiplication layer. The simulation findings demonstrate that the electric field distribution in EBCMOS could be optimized by using a layered doping structure to decrease the thickness of the index strongly doped layer. The enhanced doping structure directly contributes to the enhanced charge collecting efficiency. Based on the results of the simulations, doped samples were created and evaluated. The charge collection efficiency of EBCMOS devices might be significantly increased by decreasing the thickness of the index highly doped layer to 0.1 μm, reaching 86.27%. The testing was done on a 7 μm thick P-type silicon substrate with a 1 μm thick index highly doped layer.

In order to improve the charge collection efficiency, we simulated and experimentally tested the doping structure of the electron multiplication layer in EBCMOS. In this paper, we simulate the charge collection efficiency of EBCMOS under different doping methods by modeling the collisional scattering of electrons with solid atoms in semiconductor materials and combining the transport trajectories of electrons in the electron multiplication layer, the simulation results indicate that using a layered doping structure to reduce the thickness of the index heavily doped layer can effectively optimize the electric field distribution in the electron multiplier layer and reduce the recombination rate of electrons. The optimized doping structure has a significant effect on improving charge collection efficiency. Based on the simulation results, doped samples were prepared and tested. The test results showed that the charge collection efficiency obtained under the condition of a P-type silicon substrate thickness of 7μm and an index heavily doped layer thickness of 1μm was 72.65%, reducing the thickness of the index heavily doped layer to 0.1μm, the charge collection efficiency obtained can reach 86.27%, which proves that reducing the thickness of the index heavily doped layer can effectively improve the charge collection efficiency of EBCMOS devices.

We investigated the effect of different gradient doping methods on the charge collection efficiency of the electron multiplication layer of EBCMOS devices. Exponential doping of the electron multiplication layer can form a built-in electric field in the electron multiplication layer that is favorable for photoelectron transport, so exponential doping instead of uniform doping in the electron multiplication layer of EBCMOS can effectively improve the charge collection efficiency. It is shown that exponential heavy doping on the side of the electron multiplication layer near the dead layer and exponential light doping on the side near the depletion region can improve the built-in electric field structure and increase the lifetime of the multiplied electrons, thereby improving the charge collection efficiency of EBCMOS. The optimized device achieves a charge collection efficiency of 94.48 at an incident electron energy of 4keV, an electron beam diameter of 20nm, a dead layer thickness of 60nm, and a P-type silicon epitaxial layer thickness of 10μm.

Regarding the correction of X-ray beam hardening in the current CT imaging system, the traditional method will cause the overlapping of images during use, which will gradually harden the beam, and the image reconstructed by the imaging system will gradually become “cup-shaped” or “striped.” “False images” seriously degrade the quality of the images, while causing more interfering diagnostic problems. In this paper, the cloud computing big data analysis algorithm is applied to the X-ray beam hard correction process. According to the transition energy and its energy absorption, the X-ray beam is used as the cut-in point, and the relationship between the attenuation coefficient of similar materials and the X-ray energy is used to remove the artifact image in the initial image reconstruction process to obtain a clear corrected image. Meanwhile, according to the thickness and gray value of the X-ray penetrable object, the result of fitting using a polynomial function is calculated, and an accurate line fitting can be completed for data with smaller coordinates. Finally, the experimental study shows that the cloud computing big data analysis proposed in this paper can detect X-ray beams in real time. This method uses optical receivers to achieve high noise sensitivity to X-rays, and in long-distance transmission scenarios, the bandwidth of communication transmission can be maximized, and different types of formats can be used to complete modulation for different X-rays. Therefore, X-ray beam hardening correction technology has better advantages and the market application scenarios compared with other technologies.

Accurate acquisition of image contrast is an important factor for measuring modulation transfer function and minimum resolvable contrast of camera. Under low illumination, image from intensified charge-coupled device exhibits ion feedback noise with high brightness, resulting in lower signal-to-noise ratio of the image. Thus, one cannot obtain easily the accurate image contrast results by using the conventional contrast calculation method. In this paper, the ion feedback noise of the rectangular target image under low illumination is first reduced by means of constrained noise screening method and asymmetrical orientation window weighted mean filling algorithm. According to the spatial height correlation of the rectangular target image, the maximum sampling region is then founded by using spatial one-way statistical average algorithm combined with Otsu binarization. The image contrast is finally calculated after removing the fuzzy interval in the boundary between black and white stripes. It is shown that the method proposed can achieve the maximum sampling region acquisition and the contrast calculation for the image captured under low illumination, which reduces the error of calculation contrast caused by manually selecting the sampling region.

According to the interaction model between low-energy electrons and solids, combined with Monte Carlo calculation method, MATLAB software was used to simulate the electron scattering trajectories of a large number of photoelectrons incident on EBCMOS substrates. The energy loss rate of different incident photoelectrons on the substrate was analyzed. The influence of electron incidence depth and the range of electron motion. When the incident photoelectron energy increases, the energy loss rate gradually decreases, and the electron incident depth and the electron motion range increase. The simulation results can provide reference and basis for the design and preparation of the back-illuminated CMOS device substrate.