Mesh-free finite difference(FD)methods can improve the geometric flexibility of modeling without the need for lattice mapping or complex meshing process.Radial-basisfunction-generated FD is among the most commonly use...Mesh-free finite difference(FD)methods can improve the geometric flexibility of modeling without the need for lattice mapping or complex meshing process.Radial-basisfunction-generated FD is among the most commonly used mesh-free FD methods and can accurately simulate seismic wave propagation in the non-rectangular computational domain.In this paper,we propose a perfectly matched layer(PML)boundary condition for a meshfree FD solution of the elastic wave equation,which can be applied to the boundaries of the non-rectangular velocity model.The performance of the PML is,however,severely reduced for near-grazing incident waves and low-frequency waves.We thus also propose the complexfrequency-shifted PML(CFS-PML)boundary condition for a mesh-free FD solution of the elastic wave equation.For two PML boundary conditions,we derive unsplit time-domain expressions by constructing auxiliary differential equations,both of which require less memory and are easy for programming.Numerical experiments demonstrate that these two PML boundary conditions effectively eliminate artificial boundary reflections in mesh-free FD simulations.When compared with the PML boundary condition,the CFS-PML boundary condition results in better absorption for near-grazing incident waves and evanescent waves.展开更多
Large eddy simulations generally are used to predict 3D wind field characteristics in complex mountainous areas.Certain simulation boundary conditions,such as the height and length of the computational domain or the c...Large eddy simulations generally are used to predict 3D wind field characteristics in complex mountainous areas.Certain simulation boundary conditions,such as the height and length of the computational domain or the characteristics of inflow turbulence,can significantly impact the quality of predictions.In this study,we examined these boundary conditions within the context of the mountainous terrain around a long-span cable-stayed bridge using a wind tunnel experiment.Various sizes of computational domains and turbulent incoming wind velocities were used in large eddy simulations.The results show that when the height of the computational domain is five times greater than the height of the terrain model,there is minimal influence from the top wall on the wind field characteristics in this complex mountainous area.Expanding the length of the wake region of the computational domain has negligible effects on the wind fields.Turbulence in the inlet boundary reduces the length of the wake region on a leeward hill with a low slope,but has less impact on the mean wind velocity of steep hills.展开更多
Deep learning has recently gained significant prominence in various real-world applications such as image recognition,natural language processing,and autonomous vehicles.While deep neural networks appear to have diffe...Deep learning has recently gained significant prominence in various real-world applications such as image recognition,natural language processing,and autonomous vehicles.While deep neural networks appear to have different architectures,the main operations within these models are matrix-vector multiplications(MVM).Compute-in-memory(CIM)architectures are promising solutions for accelerating the massive MVM operations by alleviating the frequent data movement issue in traditional processors.Analog CIM macros leverage current-accumulating or charge-sharing mechanisms to perform multiply-and-add(MAC)computations.Even though they can achieve high throughput and efficiency,the computing accuracy is sacrificed due to the analog nonidealities.To ensure precise MAC calculations,it is crucial to analyze the sources of nonidealities and identify their impacts,along with corresponding solutions.In this paper,comprehensive linearity analysis and dedicated calibration methods for charge domain static-random access memory(SRAM)based in-memory computing circuits are proposed.We analyze nonidealities from three areas based on the mechanism of charge domain computing:charge injection effect,temperature variations,and ADC reference voltage mismatch.By designing a 256×256 CIM macro and conducting investigations via post-layout simulation,we conclude that these nonidealities don’t deteriorate the computing linearity,but only cause the scaling and bias drift.To mitigate the scaling and bias drift identified,we propose three calibration methods ranging from the circuit level to the algorithm level,all of which exhibit promising results.The comprehensive analysis and calibration methods can assist in designing CIM macros with more accurate MAC computations,thereby supporting more robust deep learning inference.展开更多
基金supported by the National Science and Technology Major Project(2016ZX05006-002)the National Natural Science Foundation of China(Nos.41874153,41504097)
文摘Mesh-free finite difference(FD)methods can improve the geometric flexibility of modeling without the need for lattice mapping or complex meshing process.Radial-basisfunction-generated FD is among the most commonly used mesh-free FD methods and can accurately simulate seismic wave propagation in the non-rectangular computational domain.In this paper,we propose a perfectly matched layer(PML)boundary condition for a meshfree FD solution of the elastic wave equation,which can be applied to the boundaries of the non-rectangular velocity model.The performance of the PML is,however,severely reduced for near-grazing incident waves and low-frequency waves.We thus also propose the complexfrequency-shifted PML(CFS-PML)boundary condition for a mesh-free FD solution of the elastic wave equation.For two PML boundary conditions,we derive unsplit time-domain expressions by constructing auxiliary differential equations,both of which require less memory and are easy for programming.Numerical experiments demonstrate that these two PML boundary conditions effectively eliminate artificial boundary reflections in mesh-free FD simulations.When compared with the PML boundary condition,the CFS-PML boundary condition results in better absorption for near-grazing incident waves and evanescent waves.
基金supported by the National Natural Science Foundation of China(Nos.51925808 and 52178516)the Natural Science Foundation of Hunan Province(Nos.2020JJ5745 and 2023JJ20073),China.
文摘Large eddy simulations generally are used to predict 3D wind field characteristics in complex mountainous areas.Certain simulation boundary conditions,such as the height and length of the computational domain or the characteristics of inflow turbulence,can significantly impact the quality of predictions.In this study,we examined these boundary conditions within the context of the mountainous terrain around a long-span cable-stayed bridge using a wind tunnel experiment.Various sizes of computational domains and turbulent incoming wind velocities were used in large eddy simulations.The results show that when the height of the computational domain is five times greater than the height of the terrain model,there is minimal influence from the top wall on the wind field characteristics in this complex mountainous area.Expanding the length of the wake region of the computational domain has negligible effects on the wind fields.Turbulence in the inlet boundary reduces the length of the wake region on a leeward hill with a low slope,but has less impact on the mean wind velocity of steep hills.
基金supported in part by the National Key Research and Development Program of China under Grant 2022YFB4400900in part by the Natural Science Foundation of China under Grant 62371223.
文摘Deep learning has recently gained significant prominence in various real-world applications such as image recognition,natural language processing,and autonomous vehicles.While deep neural networks appear to have different architectures,the main operations within these models are matrix-vector multiplications(MVM).Compute-in-memory(CIM)architectures are promising solutions for accelerating the massive MVM operations by alleviating the frequent data movement issue in traditional processors.Analog CIM macros leverage current-accumulating or charge-sharing mechanisms to perform multiply-and-add(MAC)computations.Even though they can achieve high throughput and efficiency,the computing accuracy is sacrificed due to the analog nonidealities.To ensure precise MAC calculations,it is crucial to analyze the sources of nonidealities and identify their impacts,along with corresponding solutions.In this paper,comprehensive linearity analysis and dedicated calibration methods for charge domain static-random access memory(SRAM)based in-memory computing circuits are proposed.We analyze nonidealities from three areas based on the mechanism of charge domain computing:charge injection effect,temperature variations,and ADC reference voltage mismatch.By designing a 256×256 CIM macro and conducting investigations via post-layout simulation,we conclude that these nonidealities don’t deteriorate the computing linearity,but only cause the scaling and bias drift.To mitigate the scaling and bias drift identified,we propose three calibration methods ranging from the circuit level to the algorithm level,all of which exhibit promising results.The comprehensive analysis and calibration methods can assist in designing CIM macros with more accurate MAC computations,thereby supporting more robust deep learning inference.
