Scholarlink

<p>High domain wall velocity in magnetic nano stripe is a topic of interest in developing more sophisticated devices for neuromorphic computing and storage. A micromagnetic simulation is an excellent tool for investigating spin dynamics and calculating domain wall velocity, but this process is very time-consuming. So, a computational model has been developed to predict the spin dynamics. The domain wall velocity at different current densities has been generated from the time domain data obtained from the micromagnetic simulation. This data is used to train various machine-learning models. We have explored the Echo State Network (ESN), Long Short-Term Memory (LSTM) model, Seasonal Autoregressive Integrated Moving Average with Exogenous Factor (SARIMAX), and Neural Basis Expansion Analysis Time Series (N-BEATS) forecasting model to predict the domain wall velocity from the post-processed sequence data. We found that Echo State Network outperforms all other models in a small dataset. Echo State Network models achieve a lower Normalized Root Mean Squared Error (NRMSE) of 0.785 and Mean Absolute Percentage Error (MAPE) of 0.083 than the other three models. From the present work, we concluded that ESN is a suitable computational model for predicting the domain wall velocity that follows non-linear spin dynamics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Domain wall velocity prediction in magnetic nano stripe under spin-polarized current using machine learning techniques

  • 期刊论文

  • 发表日期：

    2025年2月18日

查看全文
  • Madhurima Sen,
  • Saswati Barman

摘要

High domain wall velocity in magnetic nano stripe is a topic of interest in developing more sophisticated devices for neuromorphic computing and storage. A micromagnetic simulation is an excellent tool for investigating spin dynamics and calculating domain wall velocity, but this process is very time-consuming. So, a computational model has been developed to predict the spin dynamics. The domain wall velocity at different current densities has been generated from the time domain data obtained from the micromagnetic simulation. This data is used to train various machine-learning models. We have explored the Echo State Network (ESN), Long Short-Term Memory (LSTM) model, Seasonal Autoregressive Integrated Moving Average with Exogenous Factor (SARIMAX), and Neural Basis Expansion Analysis Time Series (N-BEATS) forecasting model to predict the domain wall velocity from the post-processed sequence data. We found that Echo State Network outperforms all other models in a small dataset. Echo State Network models achieve a lower Normalized Root Mean Squared Error (NRMSE) of 0.785 and Mean Absolute Percentage Error (MAPE) of 0.083 than the other three models. From the present work, we concluded that ESN is a suitable computational model for predicting the domain wall velocity that follows non-linear spin dynamics.