The relentless pursuit of miniaturization, high frequency, and efficiency in electronic equipment has set new and stringent benchmarks for soft magnetic materials (SMMs). Key among these are the demands for higher saturation magnetic induction intensity (B_s) and lower coercivity (H_c). Yet, a longstanding trade-off has persisted: higher B_s typically increases H_c. For over a century, this dilemma has remained unsolved, with no SMM exceeding a B_s of 1.9 T and maintaining a H_c below 2 A/m, significantly constraining the capabilities of modern electronic devices. Facing the vast materials composition space, traditional trial-and-error approaches to develop SMMs have proven inefficient and labor-intensive. To overcome these limitations, this study employs machine learning techniques to construct artificial neural networks, forecasting with approaching 95% precision. From a compositional space of 486,000 permutations and combinations, the optimal FeCoBSiCu alloy was identified and synthesized, achieving an unprecedented B_s of 1.96 T and an extreme-low H_c of 1.2 A/m. This performance breaks the B_s-H_c trade-off and surpasses all previously reported SMMs. The magnetic domain evolution and unique heterostructure were applied to the soft magnetic mechanism. This work may pave the way for a paradigm shift in the compositional design of SMMs, accelerating the advancement of electronic devices.

中文翻译：

---

机器学习辅助在 FeCoBSiCu 合金中集成超高饱和磁化强度和极低矫顽力


电子设备对小型化、高频化和效率的不懈追求为软磁材料 （SMM） 设定了新的严格基准。其中关键是对更高饱和磁感应强度 （B_s） 和更低矫顽力 （H_c） 的要求。然而，一个长期的权衡仍然存在：更高的 B_s 通常会增加 H_c。一个多世纪以来，这一困境一直未解决，没有 SMM 的 B_s 超过 1.9 T，并且 H_c 保持在 2 A/m 以下，这极大地限制了现代电子设备的能力。面对巨大的材料组成空间，传统的试错法开发 SMM 已被证明效率低下且劳动密集型。为了克服这些限制，本研究采用机器学习技术来构建人工神经网络，以接近 95% 的预测精度进行预测。从 486,000 次排列和组合的组成空间中，确定并合成了最佳的 FeCoBSiCu 合金，实现了前所未有的 1.96 T 的 B_s 和 1.2 A/m 的极低 H_c。此性能打破了 B_{s-H c} 的权衡，并超过了以前报告的所有 SMM。将磁畴演化和独特的异质结构应用于软磁机制。这项工作可能为 SMM 组合设计的范式转变铺平道路，加速电子设备的进步。