信息门户 工大邮箱 教务管理 新闻投稿

通知公告 报告讲座 文明创建 图书馆




您当前所在位置: 首页 > 报告讲座 > 正文

石媛媛: Building the hardware of future artificial intelligence systems--2D m...

日期:2019-01-04  稿件来源:  

时间 地点
职称 学院
报告人 工作单位
举办单位 报告人简介



:石媛媛 博士后研究员




石媛媛,女,博士,比利时微电子研究中心(IMEC)博士后。2015年硕士毕业于西班牙Rovira i Virgili University, 2018年7月以最高荣誉“Excellent Cum Laude”获得西班牙巴塞罗那大学博士学位,师从Mario Lanza 教授。博士期间,于美国斯坦福大学进行过为期一年的博士生交流学习,师从H.-S. Philip Wong教授。在硕博期间以第一作者或共同作者发表论文42篇,包括Nature Electronics, IEDM, Nano Letters, Advanced Functional Materials, Nano Energy, Nano Research, Journal of Materials Chemistry A, ACS Applied Materials & Interfaces , Advanced Electronic Materials 等国际期刊,撰写一篇Wiley-VCH 书籍章节,申请两项国际专利(其中一项已获得一百万美元投资),担任Scientific Reports, Thin Solid Films, ChemElectroChem等多个国际期刊的审稿人。在斯坦福大学期间,所开展的基于二维材料电子突触器件的研究,被2017年国际顶级电子器件会议IEDM接收并在该会议上作口头报告,这一课题的更深入研究成果被发表在Nature Electronics,并被Nature Electronics News & Views作为亮点工作报道。2018年获得IEEE EDS Excellent PhD Student Fellowship (全球仅三位),并入选参加亚洲顶尖工学院院长论坛–高等教育工程学界女性明日之星研讨会 (The Rising Stars, Women in Engineering Workshop)。


Current AI systems rely on advanced computers to process a massive amount of data and carry out complex operations very fast (<1 ns/operation), and by using sophisticated algorithms. therefore, until now, progress in ai has been strictly linked to: i) the computing power of the systems used, and ii) the efficiency of the algorithms used to process the data. in order to create more powerful and efficient ai systems, electronic engineers have started to consider the possibility of designing new hardware, i.e. fabricate new electronic circuits that act as artificial neural networks, and that can compute the data as the human brain does. several different electronic components have been suggested as the hardware implementation of electronic synapses for artificial neural networks, and among them memristors based on metal/insulator/metal (mim) nanocells are the ones that have shown the best performance. in this talk i will present our work on mim-like memristors that can be used as electronic synapses. by introducing h-bn in the structure of the memristors we have been able to emulate both volatile or non-volatile rs depending on the amplitude, duration and interval of the electrical impulses applied. this allows emulating several synaptic behaviors, being the ultra low variability of the relaxation process the most remarkable performance. the power consumption in standby and per transition in volatile regime can be as low as 0.1 fw and 600 pw (respectively), and the switching time is <10 ns. these behaviors are related to a novel resistive switching mechanism in the h-bn stack, which is based on the generation of b-vacancies that can be filled by metallic ions from the adjacent electrodes.



分享到 :
XML 地图 | Sitemap 地图