High-Reliability Ultrascaled IGZO FETs for Multibit 2T0C DRAM toward Monolithic 3D Integration of an AI Accelerator
ACS Nano · 06.11.2026
DOI: 10.1021/acsnano.6c08923 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACS Nano, online erschienen am 06.11.2026. Verfasst von Kexin Wang und 17 weiteren Autor:innen. Die Arbeit nennt 68 Literaturangaben. Gefördert von: National Natural Science Foundation of China, National Science and Technology Major Project.
Abstract (Originalsprache)
The continued scaling of silicon-based dynamic random-access memory (DRAM) slows down due to increased process complexity and reduced storage capacitance. Oxide semiconductors (OS), represented by InGaZnOx (IGZO), offer an attractive alternative owing to their ultralow leakage and low thermal budget, making them promising candidates for next-generation DRAM as well as monolithic three-dimensional (M3D) architectures. However, realizing IGZO devices with well-balanced overall performance, including high performance and high reliability under ultrascaled dimensions, remains challenging. Here, we report ultrascaled IGZO field-effect transistors (FETs) fabricated using an optimized O2 plasma-enhanced atomic-layer deposition (PE-ALD) process. The IGZO FETs deliver state-of-the-art performance, including an on-state current (Ion) of 785 μA/μm at a drain voltage of 0.5 V and a transconductance (gm) of 620 μS/μm at a drain voltage of 1.0 V alongside a contact resistance (Rc) of 164.5 Ω·μm with positive threshold voltage (Vth). Reliability degradation is effectively suppressed, with bias-temperature instability (BTI) down to 8 mV even at extremely scaled contact length (Lc) of 50 nm. Leveraging these optimized devices, we further demonstrated a scaled two-transistor-zero-capacitor (2T0C) DRAM capable of up to 6-bit per cell, long data retention over 4000 s, and write time down to 10 ns. We proposed an M3D-DRAM architecture with vertically stacked multilayer embedded memory for artificial intelligence (AI) accelerators, achieving 77× faster weight loading and 9.6× lower energy consumption compared with mature high-bandwidth memory (HBM) counterpart, while maintaining graphics processing units (GPU)-equivalent inference accuracy of 92.25%. This work demonstrates an effective approach for the co-optimization of performance and reliability in ultrascaled IGZO FETs, providing critical foundation for next-generation DRAM and M3D applications.
Förderung
- National Natural Science Foundation of China
- National Science and Technology Major Project
Zitieren
APA
Wang, K., Tang, J., Zhang, Y., Su, Y., Tang, Z., Li, Y., Fan, Y., Yang, H., Yan, G., Ma, X., Xiang, J., Wang, G., Zhao, C., Xi, Y., Gao, B., Deng, N., Qian, H., & Wu, H. (2026). High-Reliability Ultrascaled IGZO FETs for Multibit 2T0C DRAM toward Monolithic 3D Integration of an AI Accelerator. ACS Nano. https://doi.org/10.1021/acsnano.6c08923
Vancouver
Wang K, Tang J, Zhang Y, Su Y, Tang Z, Li Y, et al. High-Reliability Ultrascaled IGZO FETs for Multibit 2T0C DRAM toward Monolithic 3D Integration of an AI Accelerator. ACS Nano. 2026. doi:10.1021/acsnano.6c08923
BibTeX
@article{Wang2026,
author = {Wang, Kexin and Tang, Jianshi and Zhang, Yibei and Su, Yanbo and Tang, Zhidong and Li, Yuankun and Fan, Yijia and Yang, Huimin and Yan, Gangping and Ma, Xueli and Xiang, Jinjuan and Wang, Guilei and Zhao, Chao and Xi, Yue and Gao, Bin and Deng, Ning and Qian, He and Wu, Huaqiang},
title = {High-Reliability Ultrascaled IGZO FETs for Multibit 2T0C DRAM toward Monolithic 3D Integration of an AI Accelerator},
journal = {ACS Nano},
year = {2026},
doi = {10.1021/acsnano.6c08923}
}
EUVIMED