Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture

Mo, Pinghui; Li, Chang; Zhao, Dan; Yu-jia, Zhang; Shi, Mengchao; Li, Junhua; Liu, Jie

doi:10.1038/s41524-022-00773-z

Cited by 16 publications

(14 citation statements)

References 99 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To meet this requirement, we developed a MLFF (DeepCNT-22) using the Deep Potential Smooth Edition method [54,55], as implemented in DeePMD-kit [56]. This approach has been proven to be both accurate and computationally efficient [57], with the potential for further acceleration through the utilization of non-von Neumann architectures [58].…”

Section: Resultsmentioning

confidence: 99%

“…22 and LAMMPS enables us to simulate growth on Fe 55 catalysts at a rate of ∼70 ns/day using a single Nvidia V100 GPU and ∼1200 ns/day using a non-von Neumann architecture [58]. While the non-von Neumann architecture achieves a speedup of ∼17 times, all simulations were ultimately performed on Nvidia K80 and V100 GPUs due to their greater availability.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Dynamics of growing carbon nanotube interfaces probed by machine learning-enabled molecular simulations

Hedman¹,

McLean²,

Bichara³

et al. 2023

Preprint

View full text Add to dashboard Cite

Carbon nanotubes (CNTs) are currently considered a successor to silicon in future nanoelectronic devices. To realize this, controlled growth of defect-free semiconducting nanotubes is required. Until now, the understanding of atomic-scale CNT growth mechanisms provided by molecular dynamics simulations has been hampered by their short timescales. Here, we develop an efficient and accurate machine learning force field, DeepCNT-22, to simulate the complete growth of defect-free singlewalled CNTs (SWCNTs) on iron catalysts at near-microsecond time scales. We provide atomic-level insight into the nucleation and growth processes of SWCNTs, including the evolution of the tubecatalyst interface and the mechanisms underlying defect formation and healing. Our simulations highlight the maximization of SWCNT-edge configurational entropy during growth and how defectfree CNTs can grow ultralong if carbon supply and temperature are carefully controlled.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Dynamics of growing carbon nanotube interfaces probed by machine learning-enabled molecular simulations

Hedman¹,

McLean²,

Bichara³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…In addition, as shown in Figure 3a, Zhang Yan is the researcher with the most collaborations (60) in Scopus, followed by Li J. (51) and Wang J. (48) (see Table A1 from Appendix A).…”

Section: Most Cited Authors and Their Collaborationsmentioning

confidence: 99%

“…In the case of molecular dynamics, artificial intelligence is employed to contribute to understanding materials' properties by simulating the interaction of atoms and molecules. Even though some studies lead to considerable differences between accuracy and efficiency, machine learning and deep learning are still considered helpful tools to match efficiency with ac-curacy in molecular simulation [51].…”

Section: Most Contributing Countries and Their Collaborationsmentioning

confidence: 99%

Bibliometric Analysis of Fourth Industrial Revolution Applied to Material Sciences Based on Web of Science and Scopus Databases from 2017 to 2021

et al. 2023

View full text Add to dashboard Cite

Material science is a broad discipline focused on subjects such as metals, ceramics, polymers, electronics, and composite materials. Each of these fields covers areas associated with designing, synthesizing, and manufacturing, materials. These are tasks in which the use of technology may constitute paramount importance, reducing cost and time to develop new materials and substituting try-and-error standard procedures. This study aimed to analyze, quantify and map the scientific production of research on the fourth industrial revolution linked to material science studies in Scopus and Web of Science databases from 2017 to 2021. For this bibliometric analysis, the Biblioshiny software from RStudio was employed to categorize and evaluate the contribution of authors, countries, institutions, and journals. VOSviewer was used to visualize their collaboration networks. As a result, we found that artificial intelligence represents a hotspot technology used in material science, which has become usual in molecular simulations and manufacturing industries. Recent studies aim to provide possible avenues in the discovery and design of new high-entropy alloys as well as to detect and classify corrosion in the industrial sector. This bibliometric analysis releases an updated perspective on the implementations of technologies in material science as a possible guideline for future worldwide research.

show abstract

“…Although some special-purpose computers have been developed to accelerate MD calculations [24][25][26], they are all based on CMD, which makes their accuracy questionable in many important applications [27,28]. Recently, by leveraging MLMD algorithms and NvN architecture, an MD computing system named NVNMD has been developed by Mo et al [29][30][31]. The NVNMD proves that the specially designed NvNbased MLMD computing system has higher computational speed and higher energy efficiency than the vN-based system by deploying on field programmable gate array (FPGA), providing a good hardware solution for accurate and efficient MLMD calculations.…”

Section: Introductionmentioning

confidence: 99%

A Heterogeneous Parallel Non-von Neumann Architecture System for Accurate and Efficient Machine Learning Molecular Dynamics

Zhao

Tan

et al. 2023

IEEE Trans. Circuits Syst. I

View full text Add to dashboard Cite

This paper proposes a special-purpose system to achieve high-accuracy and high-efficiency machine learning (ML) molecular dynamics (MD) calculations. The system consists of field programmable gate array (FPGA) and application specific integrated circuit (ASIC) working in heterogeneous parallelization. To be specific, a multiplication-less neural network (NN) is deployed on the non-von Neumann (NvN)-based ASIC (SilTerra 180 nm process) to evaluate atomic forces, which is the most computationally expensive part of MD. All other calculations of MD are done using FPGA (Xilinx XC7Z100). It is shown that, to achieve similar-level accuracy, the proposed NvN-based system based on low-end fabrication technologies (180 nm) is 1.6× faster and 10 2 -10 3 × more energy efficiency than state-of-the-art vNbased MLMD using graphics processing units (GPUs) based on much more advanced technologies (12 nm), indicating superiority of the proposed NvN-based heterogeneous parallel architecture.

show abstract

Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture

Cited by 16 publications

References 99 publications

Dynamics of growing carbon nanotube interfaces probed by machine learning-enabled molecular simulations

Dynamics of growing carbon nanotube interfaces probed by machine learning-enabled molecular simulations

Bibliometric Analysis of Fourth Industrial Revolution Applied to Material Sciences Based on Web of Science and Scopus Databases from 2017 to 2021

A Heterogeneous Parallel Non-von Neumann Architecture System for Accurate and Efficient Machine Learning Molecular Dynamics

Contact Info

Product

Resources

About