Federated learning (FL) with fully homomorphic encryption (FHE) effectively safeguards data privacy during model aggregation by encrypting local model updates before transmission, mitigating threats from untrusted servers or eavesdroppers in transmission. However, the computational burden and ciphertext expansion associated with homomorphic encryption can significantly increase resource and communication overhead. To address these challenges, we propose FedBit, a hardware/software co-designed framework optimized for the Brakerski-Fan-Vercauteren (BFV) scheme. FedBit employs bitinterleaved data packing to embed multiple model parameters into a single ciphertext coefficient, thereby minimizing ciphertext expansion and maximizing computational parallelism. Additionally, we integrate a dedicated FPGA accelerator to handle cryptographic operations and an optimized dataflow to reduce the memory overhead. Experimental results demonstrate that FedBit achieves a speedup of two orders of magnitude in encryption and lowers average communication overhead by $60.7 \%$, while maintaining high accuracy.

Large language models (LLMs) have demonstrated impressive capabilities in generating software code for high-level programming languages such as Python and C++. However, their application to hardware description languages, such as Verilog, is challenging due to the scarcity of high-quality training data. Current approaches to Verilog code generation using LLMs often focus on syntactic correctness, resulting in code with functional errors. To address these challenges, we present AutoVeriFix, a novel Python-assisted two-stage framework designed to enhance the functional correctness of LLM-generated Verilog code. In the first stage, LLMs are employed to generate high-level Python reference models that define the intended circuit behavior. In the second stage, these Python models facilitate the creation of automated tests that guide the generation of Verilog RTL implementations. Simulation discrepancies between the reference model and the Verilog code are iteratively used to identify and correct errors, thereby improving the functional accuracy and reliability of the LLM-generated Verilog code. Experimental results demonstrate that our approach significantly outperforms existing state-of-the-art methods in improving the functional correctness of generated Verilog code.

Chip manufacturing relies on rigorous verification to prevent costly design errors before fabrication and deployment. Branch coverage, a key metric for Register-Transfer Level (RTL) validation, ensures thorough testing of decision points in the design. However, RTL designs often contain numerous hard-to-activate branches, which can lead to hidden bugs and security vulnerabilities. While concolic testing addresses the memory explosion issues associated with formal methods, it relies on heuristics that may get stuck in local optima. In this paper, we propose a novel approach that reformulates Concolic testing as a reinforcement learning problem. Our method utilizes the agent that takes into account RTL structural characteristics and runtime simulation states to select strategies for guiding the simulation path toward target branches. Experimental results demonstrate that our approach effectively directs simulations toward branch targets, reduces search redundancy, and significantly increases branch coverage, thereby improving the efficiency and effectiveness of the test generation process.

General Matrix-Matrix Multiplication (GEMM) stands as the most ubiquitous operation in machine learning applications. However, performing GEMM within Fully Homomorphic Encryption (FHE) is inefficient due to high computational demands and significant data migration constrained by limited bandwidth. Additionally, the inherent limitations of FHE schemes restrict the widespread application of machine learning, as standard activation functions are incompatible. This incompatibility necessitates alternative nonlinear functions, which lead to notable accuracy reductions. To address these challenges, we introduce a polynomial encoding methodology for GEMM under the Brakerski/Fan-Vercauteren (BFV) scheme and extend the method to inference with packing inputs and weights for different sizes. Furthermore, we design specialized hardware to accelerate the inference process through optimized scheduling between the hardware and the host system. In experiments, we implemented our hardware on an FPGA U250 platform. Compared to existing solutions, our method achieves superior performance, achieving the highest $4.22 \times$ and $3.99 \times$ speedups on MNIST and CIFAR-10.

Simulation plays a crucial role in the verification of hardware designs, ensuring that they behave correctly before fabrication. However, traditional simulation methods can be inefficient when dealing with complex designs, especially in corner cases. To mitigate this inefficiency, Concolic testing has emerged as a promising technique, utilizing symbolic execution to guide the simulation process. However, the heuristics used in path exploration for Concolic testing often struggle with local optima, resulting in suboptimal verification outcomes and incomplete coverage of the design space. In this paper, we propose an agent-based framework to dynamically adjust path exploration strategies by leveraging beam search and large language models (LLMs). Experimental results demonstrate that this approach significantly improves branch coverage, especially for hard-to-detect branches, while also optimizing the use of computational resources.

Functional verification is critical in ensuring the correctness of register transfer level (RTL) models. Formal methods, such as model checkers, are powerful tools that help achieve high coverage in functional validation by transforming the coverage problem into property verification tasks. However, these methods typically demand significant memory usage and long verification times. One major issue is that the satisfiability problem for each unsolved property always starts from the reset state of a design, leading to repeated solving of the same subset of clauses across different properties. In this paper, we propose an open-source semi-formal framework based on model checkers that accelerates test stimulus generation through two techniques: assertion ordering and strategic selection of starting states. These techniques enable model checkers to intelligently select starting states that are much closer to the final state, thereby reducing unnecessary computations. Through comprehensive experiments on ITC'99 benchmarks and modern complex processor designs, including OpenCores 1200 and Rocket-Chip, we demonstrate that our proposed techniques can achieve higher coverage with less than half of the test generation time.

This letter reports the first cloudy-sky land surface temperature (LST) retrieval from the MEdium Resolution Spectral Imager-II (MERSI-II) onboard Fengyun-3D (FY-3D). In this study, a new cloudy-sky LST retrieval algorithm was adapted for FY-3D/MERSI-II. The method consists of two key components: (1) estimating the hypothetical clear-sky LST using geographically weighted regression (GWR) downscaling and linear model correction, and (2) quantifying the temperature difference (ΔT) caused by cloud radiative effect (CRE) based on surface energy balance (SEB) theory. Validation against in situ LST measurements from the SURFRAD and HiWATER networks indicates that the retrieved cloud-sky LST exhibits a bias of 0.02 K and a root mean square error (RMSE) of 3.77 K. The results confirm that the algorithm achieves reliable retrieval of FY-3D/MERSI-II cloudy-sky LST, with the ΔT correction term making a notable contribution. This method is expected to be used to generate the operational FY-3D cloudy-sky LST product.

The validation of Register-Transfer Level (RTL) models requires achieving sufficient branch coverage. However, automatically activating all branches in RTL models is challenging, considering the complexity of modern designs. While traditional methods, such as model checkers, can achieve high coverage, they typically demand substantial computational resources to solve formal equations. In contrast, constraint-random approaches have better scalability but suffer from inefficiency due to poor heuristics. This paper introduces APE-FV, a Concolic testing framework designed for RTL functional verification to effectively cover rare branches. APE-FV dynamically modifies the path exploration strategy by considering structural information, simulated paths, and states. Additionally, our framework incorporates an incremental exploration technique, which reduces the burden on solvers and enhances efficiency. Experimental results demonstrate that our approach accelerates the verification process and maintains high coverage, outperforming state-of-the-art techniques.

Land surface temperature (LST) is a direct result of earth-atmosphere interactions and has been widely used in earth system science and climate change. Thus, it is recognized as one of the essential climate variables (ECVs). Thermal infrared (TIR) remote sensing is the most effective way to obtain high-quality LST at a large scale. However, TIR cannot penetrate the Cloud and obtain the cloudy sky LST, which seriously hinders its applications. To address this challenge, we proposed a paradigm for estimating the seamless LST at regional and global scales. Validation results showed the root mean square error (RMSE) of the produced seamless LST achieved approximately 3K. The produced seamless LST at China landmass, East Asia, and global have been freely released to the public (https://elite.bnu.edu.cn).

Sea surface temperature (SST) is an important variable in climate and weather research. We utilize two SST datasets to analyze the performance of the Fengyun-4A (FY-4A)/Advanced Geostationary Radiation Imager (AGRI) SST product retrieved from the nonlinear split-window algorithm. Validation results with the in situ SST Quality monitor (iQuam) (version 2.1) show that the overall bias and root mean square error (RMSE) are −0.38 and 1.04 K, respectively, meeting the demand for numerical weather prediction. There is a good agreement between the AGRI SST and the Advanced Himawari Imager (AHI) SST (version 2.0), with an overall bias of 0.09 and an RMSE of 0.95 K. The significant accuracy decreases in AGRI SST since August 2020 could be attributed to an updated operation calibration. This letter will benefit the scientific disciplines that require an SST as input by highlighting the accuracy and uncertainty of the AGRI SST product.

The ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) land surface temperature (LST) product provides LST data with a high-spatial resolution of 70 m$\times70$m. In this letter, the quality of ECOSTRESS LST product was assessed using ground measurements collected from 17 sites, including seven surface radiation budget network (SURFRAD) sites, seven baseline surface radiation network (BSRN) sites, and three National Tibetan Plateau/Third Pole Environment Data Center (TPDC) sites. After outlier removal using the “$3\sigma $-Hampel identifier,” the overall bias and root mean square error (RMSE) of ECOSTRESS LST at SURFRAD, BSRN, and TPDC sites are −1.61 and 3.08 K, −0.75, and 3.50 K, and −0.82 and 4.18 K, respectively. This letter shows the accuracy and uncertainty of ECOSTRESS LST product, and will benefit research fields that require LST with high-spatial resolution.

Atmospheric correction is a key step toward estimating land surface temperature from the sensor with only one thermal infrared (TIR) channel. We use ground radiosounding profiles collected from 163 radiosonde observations to provide insights on how well the ERA5 reanalysis product performs in the atmospheric correction of Landsat series TIR data. Despite the poor performance of the ERA5 product for estimating atmospheric upward radiance, downward radiance, and transmittance of Landsat series TIR data in the Americas and Africa, the performance of the ERA5 product was superior to that of the M2I6NPANA (inst6_3d_ana_Np) dataset (MERRA2) and (Final) Operational Global Analysis data (FNL) products in Asia and Europe. The vertical distribution of air temperature and relative humidity profiles may explain the poor performance of ERA5 in the Americas and Africa. This letter shows the advantages and weaknesses of the ERA5 reanalysis product in the atmospheric correction of Landsat series TIR data and will benefit research fields that require an atmospheric profile as input.

In this letter, the National Oceanic and Atmospheric Administration Joint Polar Satellite System enterprise algorithm and the quadratic split-window (SW) algorithm were adapted to high spatial resolution thermal infrared (TIR) data of Chinese Gaofen-5 (GF5) to estimate the land surface temperature (LST) and sea surface temperature (SST), respectively. Lacking official calibration coefficients, GF5 TIR data were cross-calibrated by the well-characterized Visible Infrared Imaging Radiometer Suite (VIIRS) data. The coefficients of two SW algorithms were obtained by linear regression from the simulated data set generated via comprehensive radiative transfer modeling. The performance of the two algorithms was first evaluated by independent simulation data and then cross-validated by Moderate Resolution Imaging Spectroradiometer (MODIS) LST/SST, VIIRS LST/SST, and Advanced Himawari Imager (AHI) SST products. The preliminary results show good agreement between estimated GF5 LSTs/SSTs and referenced LST/SST products, with an average bias (root mean square error) of -0.26 (1.74), -2.48 (3.49), 0.18 (2.43), and -1.47 K (2.86 K) for VLSTO, VNP21, MYD11, and MYD21 LST products, -0.79 (1.55), -0.28(1.58), and -1.71 (2.21) K for VIIRS, MODIS, and AHI SST products. This is the first time that both LST and SST are retrieved from the real GF5 data. This letter provides a practical method to estimate LST and SST from Chinese Gaofen-5.

Land surface temperature (LST) is a key parameter in weather forecast, global ocean circulation and climate change research. In this paper, the NOAA JPSS enterprise algorithm was adapted to retrieve LST from high spatial resolution thermal infrared data of Chinese Gaofen-5. The GF5 land surface emissivity (LSE) was determined by a new scheme. Visible Infrared Imaging Radiometer Suite (VIIRS) thermal radiance was used for cross-calibration of Gaofen-5 thermal data. The enterprise algorithm is tested by simulation data produced using CLAR and TIGR profiles. Moreover, VIIRS and Moderate Resolution Imaging Spectroradiometer (MODIS) LST products were used for cross-validation. The evaluation results show that the average bias and RMSE are -0.40 (-0.36) and 1.61K (0.90K), respectively. The validation results show that the bias (RMSE) are between -0.47K (2.18K) and 1.55K (2.59K). This study provides an alternative method to estimate LST from Chinese Gaofen-5 data.

Land surface temperature (LST) is a key parameter for a wide number of applications. The Advanced Himawari Imager (AHI) onboard Himawari-8 has four thermal infrared (TIR) bands in the atmospheric window, which provides the possibility of separating LST and LSE from AHI TIR data. In this paper, the TES algorithm and water vapor scaling (WVS) method were combined to retrieve LST and LSE from the AHI TIR data. The retrieval results are evaluated by the ground measurements collected from one site in the Baseline Surface Radiation Network (BSRN) network. The average Bias and RMSE of the estimated LST are -0.096K and 0.705K, respectively. This study demonstrates the capability of the combinations of TES algorithm and WVS method in retrieving LST and LSE from AHI TIR data.

In this study, a single-channel parametric model (SC-PM) algorithm were used to produce 300m LST product from HJ-1B IRS data. The NCEP atmospheric profiles and a parametric model were used for atmospheric correction. In order to improve the accuracy of the land surface emissivity (LSE), the 1km ASTER Global Emissivity Dataset (GED) and self-developed 5-day 1km vegetation cover product were used for estimating the LSE based on the Vegetation Cover Method. Two years of HJ-1B IRS LST product in Heihe River basin (Gansu province, China) from June 2012 to June 2014 were generated. The LST products were evaluated against ground observations collected during the Heihe Watershed Allied Telemetry Experimental Research (HiWATER) experiment. Four barren surface sites and ten vegetated sites were chosen for the evaluation. The results show that the produced HJ-1B IRS LST products demonstrate a good accuracy, with an average bias of 0.10 K and an average root mean square error (RMSE) of 2.43 K for all the sites during daytime. In addition, the biases are within 1K for the four barren surface sites. This indicate that using ASTER GED can produce reliable LST products from HJ-1B IRS data, especially for the barren surfaces.

Land surface temperature (LST) is a key parameter for a wide number of applications, which include hydrology, meteorology and model validation. In this paper a physical single channel algorithm was developed for retrieving LST from the Landsat 8 TIRS data. ASTER Global Emissivity Dataset (GED) and Vegetation Cover Method (VCM) were chosen to improve the accuracy of land surface emissivity and the fast radiative transfer model RTTOV was utilized for atmospheric correction which uses MERRA reanalysis data as inputs. The algorithm is evaluated by the ground measurements collected from in situ sites during the HiWATER experiment. The LST result shows a dynamical variation with the phenological changes and the average Bias and RMSE of the estimated LST for all sites after remove outliers are 0.09K and 2.20K, respectively. This indicates that the algorithm is suitable for producing LST product from Landsat 8 TIRS data and ASTER GED can be used to improve the accuracy of land surface emissivity in arid and semi-arid area.