可靠性工程
计算机科学
德拉姆
考试(生物学)
组分(热力学)
可靠性(半导体)
故障排除
失效模式及影响分析
工程类
测试系统
嵌入式系统
试验方法
实时计算
作者
Xiaojing Lu,Daoda An,Shuyang Hu,Yutong Wu,Jingyi Chang,Kangyi Chen,Yundong Liang,Liujia Du,Chong Xu,Hao Wang,Yanfeng Li,Zhigang Ji
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-05-01
卷期号:: 472-477
标识
DOI:10.1109/iseda70083.2026.11751427
摘要
As DRAM scales to higher density with tighter design margins, rapid yield ramp-up increasingly depends on efficient failure analysis (FA). Core-array FA remains a major bottleneck because it involves analog/mixed-signal behaviors and spatially correlated fail bit maps (FBMs), and it often relies on expert experience. This paper proposes an agent-assisted FA framework to speed up hypothesis-driven diagnosis. We first develop DATIS-FBM, a fault-injectable behavioral DRAM simulator that connects circuit-level fault characterization to bank-level FBM generation under test-pattern-driven execution. We then introduce an LLM-driven workflow that coordinates tool execution, supports fault injection, and guides the analysis in a human-in-the-loop manner. A knowledge-graph (KG) module is further integrated to reuse historical cases and enable continuous knowledge accumulation. Experimental results show that the proposed framework can reproduce real failure cases and reduce FA turnaround time.
科研通智能强力驱动
Strongly Powered by AbleSci AI