EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Tiki-Taka-Inspired SOT-MRAM In-Memory Computing Architecture for Long-Term Edge Learning

Yu Li, Fengjun Dong, Guozhong Xing

Applied Sciences · 29.04.2026 · 16 (9) · 4326

DOI: 10.3390/app16094326 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 29.04.2026. Verfasst von Yu Li und 2 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key R&D Program, National Natural Science Foundation of China, Strategic Priority Research Program of the Chinese Academy of Sciences.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Spin–Orbit Torque Magnetic Random-Access Memory (SOT-MRAM)-based in-memory computing (IMC) offers a transformative solution for energy-efficient edge intelligence, yet the deployment of robust online learning remains challenging due to memristive non-idealities and the “memory wall” inherent in Von Neumann architectures. In this work, we propose a Tiki-Taka-inspired SOT-MRAM IMC architecture with a cross-layer co-design that decouples gradient accumulation and weight storage. A time-multiplexing (Time-MUX) strategy is introduced to alleviate interconnect overhead, achieving over 28 times area reduction while maintaining comparable energy consumption to spatial mapping schemes. By integrating a canted-type x SOT-MTJ compact model into the NeuroSim framework, we perform system-level evaluation under realistic constraints. The results reveal that the proposed approach enables an effective trade-off between area, latency, and accuracy. Furthermore, the 1-bit/4-bit hybrid precision scheme achieves accuracy comparable to 8-bit configurations with reduced hardware cost while supporting a five-year cumulative write load of 1.87 × 107 cycles, significantly surpassing the endurance limits of conventional memristor devices. This work provides key insights into cross-layer optimization and establishes a practical design framework for resource-constrained edge learning systems.

Förderung

  • National Key R&D Program
  • National Natural Science Foundation of China
  • Strategic Priority Research Program of the Chinese Academy of Sciences

Zitieren

APA

Li, Y., Dong, F., & Xing, G. (2026). A Tiki-Taka-Inspired SOT-MRAM In-Memory Computing Architecture for Long-Term Edge Learning. Applied Sciences, 16(9), 4326. https://doi.org/10.3390/app16094326

Vancouver

Li Y, Dong F, Xing G. A Tiki-Taka-Inspired SOT-MRAM In-Memory Computing Architecture for Long-Term Edge Learning. Applied Sciences. 2026;16(9):4326. doi:10.3390/app16094326

BibTeX
@article{Li2026,
  author = {Li, Yu and Dong, Fengjun and Xing, Guozhong},
  title = {A Tiki-Taka-Inspired SOT-MRAM In-Memory Computing Architecture for Long-Term Edge Learning},
  journal = {Applied Sciences},
  year = {2026},
  volume = {16},
  number = {9},
  pages = {4326},
  doi = {10.3390/app16094326}
}
Kostenlos starten →