| Title |
Fabrication, Characterization, and Applicability to Hardware Artificial Intelligence of Ultra-Thin-Channel (UTC) Si 2T DRAM with Enhanced Data Retention |
| Authors |
김수민(Soomin Kim) ; 이예지(Yeji Lee) ; 육찬기(Chan-Gi Yook) ; 심원보(Wonbo Shim) ; 조성재(Seongjae Cho) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.3.35 |
| Keywords |
Hardware-oriented artificial intelligence (AI); All-Si CMOS process integration; Two-transistor (2T) dynamic random-access memory (DRAM); Processing-in-memory (PIM); System-level simulation |
| Abstract |
For realizing the highly energy-efficient hardware-oriented artificial intelligence (AI) integrated chip, low-power, high-speed, and high-density memory plays a crucial role in efficiently emulating the synaptic functions. In this work, a novel structure of all-Si 2-transistor (2T) dynamic random-access memory (DRAM) cell was proposed and fabricated for processing-in-memory (PIM) applications. Based on its non-destructive read operation capability, the proposed novel 2T DRAM cell achieved an outstanding data retention over 100 seconds and synaptic conductance modulation with more than 8 levels, demonstrating its plausible suitability to neuromorphic and processing-in-memory (PIM) systems. Furthermore, system-level simulations revealed a high inference accuracy of 89.5% and a superb energy efficiency of 3.72 TOP/W with CIFAR-10 workloads. A forward step towards ultra-wide bandwidth DRAM-based AI chip has been made. |