arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2407.14633physics.comp-phcond-mat.mes-hall
收藏

ELEQTRONeX:面向纳米材料非平衡量子输运的GPU加速百亿亿次计算框架

ELEQTRONeX: A GPU-Accelerated Exascale Framework for Non-Equilibrium Quantum Transport in Nanomaterials

Saurabh Sawant, François Léonard, Zhi Yao, Andrew Nonaka

AI总结:

针对非平衡量子输运器件模拟多尺度计算成本高的问题,提出GPU加速的ELEQTRONeX框架,自洽求解非平衡格林函数与静电问题,实现量级提速与良好扩展性,可处理复杂3D纳米器件结构。

摘要:

非平衡电子量子输运对现有及构想中的电子、光电子和自旋电子器件的运行至关重要。传统上,由于高保真耦合多物理场与多尺度需求带来的计算成本,在同一种非平衡器件模拟方法中覆盖从原子尺度到介观尺度的终极目标一直颇具挑战。在本研究中,我们提出ELEQTRONeX(百亿亿次规模纳米材料静电量子输运建模),这是一个大规模并行GPU加速框架,用于自洽求解复杂器件几何结构中的非平衡格林函数形式与静电学问题。通过为GPU多线程定制算法,我们实现了计算时间数个数量级的提升,并且在至多512个GPU与数十亿空间网格单元上具备出色的可扩展性。我们通过计算碳纳米管场效应晶体管各类3D构型的能带结构、电流-电压特性、电导以及漏致势垒降低效应,验证了代码的有效性。我们还证明ELEQTRONeX适用于周期性方法不可行的复杂器件/材料几何结构,例如需要全3D模拟的未对准碳纳米管阵列建模。

英文摘要:

Non-equilibrium electronic quantum transport is crucial for the operation of existing and envisioned electronic, optoelectronic, and spintronic devices. The ultimate goal of encompassing atomistic to mesoscopic length scales in the same nonequilibrium device simulation approach has traditionally been challenging due to the computational cost of high-fidelity coupled multiphysics and multiscale requirements. In this work, we present ELEQTRONeX (ELEctrostatic Quantum TRansport modeling Of Nanomaterials at eXascale), a massively-parallel GPU-accelerated framework for self-consistently solving the nonequilibrium Green's function formalism and electrostatics in complex device geometries. By customizing algorithms for GPU multithreading, we achieve orders of magnitude improvement in computational time, and excellent scaling on up to 512 GPUs and billions of spatial grid cells. We validate our code by computing band structures, current-voltage characteristics, conductance, and drain-induced barrier lowering for various 3D configurations of carbon nanotube field-effect transistors. We also demonstrate that ELEQTRONeX is suitable for complex device/material geometries where periodic approaches are not feasible, such as modeling of arrays of misaligned carbon nanotubes requiring fully 3D simulations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑