Search

Two-dimensional van der Waals (2D vdW) heterostructure photodetectors have garnered significant attention for their potential applications in next-generation optoelectronic systems. However, current 2D vdW photodetectors inevitably encounter compromises between responsivity, detectivity, and response time due to the absence of multilevel regulation for free and photoexcited carriers, thereby restricting their widespread applications. To address this challenge, we propose an efficient 2D WS2/CuInP2S6vdW heterostructure photodetector by combining band engineering and ferroelectric modulation. In this device, the asymmetric conduction and valence band offsets effectively block the majority carriers (free electrons), while photoexcited holes are efficiently tunneled and rapidly collected by the bottom electrode. Additionally, the ferroelectric CuInP2S6layer generates polarization states that reconfigure the built-in electric field, reducing dark current and facilitating the separation of photocarriers. Moreover, photoelectrons are trapped during long-distance lateral transport, resulting in a high photoconductivity gain. Consequently, the device achieves an impressive responsivity of 88 A W–1, an outstanding specific detectivity of 3.4 × 1013Jones, and a fast response time of 37.6/371.3 μs. Moreover, the capability of high-resolution imaging under various wavelengths and fast optical communication has been successfully demonstrated using this device, highlighting its promising application prospects in future optoelectronic systems.

On-device deep learning technology has attracted increasing interest recently. CPUs are the most common commercial hardware on devices and many training libraries have been developed and optimized for them. However, CPUs still suffer from poor training performance (i.e., training time) due to the specific asymmetric multiprocessor. Moreover, the energy constraint imposes restrictions on battery-powered devices. With federated training, we expect the local training to be completed rapidly therefore the global model converges fast. At the same time, energy consumption should be minimized to avoid compromising the user experience. To this end, we consider energy and training time and propose a novel framework with a machine learning-based adaptive configuration allocation strategy, which chooses optimal configuration combinations for efficient on-device training. We carry out experiments on the popular library MNN and the experimental results show that the adaptive allocation algorithm reduces substantial energy consumption, compared to all batches with fixed configurations on off-the-shelf CPUs.

Low-count positron emission tomography (PET) imaging is challenging because of the ill-posedness of this inverse problem. Previous studies have demonstrated that deep learning (DL) holds promise for achieving improved low-count PET image quality. However, almost all data-driven DL methods suffer from fine structure degradation and blurring effects after denoising. Incorporating DL into the traditional iterative optimization model can effectively improve its image quality and recover fine structures, but little research has considered the full relaxation of the model, resulting in the performance of this hybrid model not being sufficiently exploited. In this paper, we propose a learning framework that deeply integrates DL and an alternating direction of multipliers method (ADMM)-based iterative optimization model. The innovative feature of this method is that we break the inherent forms of the fidelity operators and use neural networks to process them. The regularization term is deeply generalized. The proposed method is evaluated on simulated data and real data. Both the qualitative and quantitative results show that our proposed neural network method can outperform partial operator expansion-based neural network methods, neural network denoising methods and traditional methods.

A dedicated brain positron emission tomography (PET) scanner can achieve higher-spatial resolution, sensitivity, and cost-effectiveness than whole-body PET scanners. In this study, we present the software system for a dedicated brain PET scanner, encompassing data acquisition, detector calibration, sinogram generation, imaging reconstruction, and data correction. The dedicated brain PET scanner features 224 depth-encoding detectors, each with a depth of interaction (DOI) resolution of approximately 2 mm. The electronics and data acquisition system of the scanner can be configured in different modes for detector calibration or image acquisition. Procedures for obtaining detector calibration parameters, including crystal look-up tables (LUTs), crystal depth-of-interaction LUTs, crystal energy, and timing calibration parameters, were developed. A novel virtual crystal-based sinogram generation method was developed to reduce sinogram size while preserving positioning accuracy. We also introduced a graphics processing unit-accelerated ordered subset expectation maximization imaging reconstruction method. The spatial resolution of the scanner was assessed using a point source at both the center and 1/4 axial field of view with varying radial offsets. We measured singles and prompt count rates at different activities using a monkey-sized phantom. Furthermore, we conducted scans on a 3-D Hoffman brain phantom and a rabbit, demonstrated the imaging capabilities of the PET scanner.

The metathesis of ethylene with 2-butene to propene is currently applied for large-scale on-purpose production of this building block of the chemical industry. MoOx-based supported catalysts are promising alternatives to the industrially applied WOx-containing systems. The activity of such catalysts depends on the amount of active Mo–carbene (MoCHR) sites formed in situupon reaction of supported MoOxspecies with olefins. Herein, we demonstrate a way to control the ability of such species on the surface of AlSiOxto be transformed into Mo–carbenes by promoting the support with tiny amounts of alkali or alkaline earth metal (Li, Sr, Ca, Mg, Na, K, Ba, Rb, or Cs) oxides. The electrophilic properties of these additives were established to be relevant for in situtransformation of supported MoOxspecies into Mo–carbenes. The additive electronegativity (Pauling scale) affects olefin adsorption and therefore the formation of carbenes.

As an urgent problem to be solved, the placement of edge server is the first step in the implementation of mobile edge computing. In practice, service providers always hope to provide services to as many users as they can under the limited funds. In this paper, we aim to study the cost-aware server placement issue for a real-world scenario with access delay guarantee. Considering the network workload, we set an affordable upper limit on the maximum number of base stations. The server placement is then modelled as a mix integer nonlinear programming problem, which is proved to be NP-hard. To solve this problem with reduced computation complexity, we propose a biogeography-based optimisation algorithm. Extensive simulations on the real-world dataset from Shanghai Telecom show that the proposed algorithm outperforms other benchmarks in terms of the total placement cost and average user access delay.

The metathesis of ethylene with 2-butene to propene is a 100%-carbon-efficient process for on-purpose production of this important building block in the chemical industry. Herein, we demonstrate that the activity of Mo/SiO2can be increased up to 270-fold when using ZnO as a cocatalyst, which is inert for the target reaction. An intimate physical contact between ZnO and Mo/SiO2is required to efficiently use the enhancing effect of the cocatalyst. H/D isotopic exchange tests with 2-C4H8and 2-C4D8suggest that the enhancing effect depends on the ability of ZnO to activate 2-butene. A gas-phase promoter (not 1-butene but probably butene radical) formed in this reaction is suggested to contribute to the formation of catalytically active Mo–carbenes under reaction conditions and/or hinder their transformation into inactive ones. This fundamental knowledge opens the possibility for further improvements through catalyst or cocatalyst design and reactor operation. The obtained results are also attractive from a practical point of view. Both ZnO and Mo/SiO2can be regenerated without changing their initial performance as proven in several metathesis/regeneration cycles.

Lower olefins are widely used in the chemical industry as basic carbon-based feedstocks. Here, we report the catalytic system featuring isolated single-atom sites of iridium (Ir1) that can function within the entire temperature range of 300–600 °C and transform alkanes with conversions close to thermodynamics-dictated levels. The high turnover frequency values of the Ir1system are comparable to those of homogeneous catalytic reactions. Experimental data and theoretical calculations both indicate that Ir1is the primary catalytic site, while the coordinating C and N atoms help to enhance the activity and stability, respectively; all three kinds of elements cooperatively contribute to the high performance of this novel active site. We have further immobilized this catalyst on particulate Al2O3, and we found that the resulting composite system under mimicked industrial conditions could still give high catalytic performances; in addition, we have also developed and established a new scheme of periodical in situ regeneration specifically for this composite particulate catalyst.

Positron emission tomography (PET) is an advanced medical imaging technique widely used in various clinical applications, such as tumor detection and neurologic disorders. Reducing the radiotracer dose is desirable in PET imaging because it decreases the patient's radiation exposure. However, reducing the dose can also increase noise, affecting the image quality. Therefore, an advanced image reconstruction algorithm based on low-dose PET data is needed to improve the quality of the reconstructed image. In this article, we propose the use of a direct PET image reconstruction network (DPIR-Net) using an improved Wasserstein generative adversarial network (WGAN) framework to enhance image quality. This article provides two main findings. First, our network uses sinogram data as input and outputs high-quality PET images direct, resulting in shorter reconstruction times compared with traditional model-based reconstruction networks. Second, we combine perceptual loss, mean square error, and the Wasserstein distance as the loss function, which effectively solves the problem of excessive smoothness and loss of detailed information in traditional network image reconstruction. We performed a comparative study using maximum-likelihood expectation maximization (MLEM) with a post-Gaussian filter, a total variation (TV)-norm regularization, a nonlocal means (NLMs) denoising method, a neural network denoising method, a traditional deep learning PET reconstruction network, and our proposed DPIR-Net method and evaluated the proposed method using both human and mouse data. The mouse data were obtained from a small animal PET prototype system developed by our laboratory. The quantitative and qualitative results show that our proposed method outperformed the conventional methods.