Memristors in Motion

Speaker Name: 
Jason Eshraghian
Speaker Title: 
PhD Candidate
Speaker Organization: 
The University of Western Australia
Start Time: 
Thursday, May 9, 2019 - 7:10pm
End Time: 
Thursday, May 9, 2019 - 8:45pm
J Baskin 156
Prof. Yu Zhang, Prof. Sung-Mo Steve Kang, & Prof. Austin Chen


Artificial neural networks have become ubiquitous in modern life, which has  triggered the emergence of a new class of application specific integrated circuits for their acceleration. ReRAM-based accelerators have gained significant traction due to their ability to leverage in-memory computation. In a crossbar structure, it turns out they are able to parallelize a huge number of simple operations. This is precisely what is needed in hardware implementation of deep neural networks with billions of parameters, making ReRAM processing of neural networks superior over classical digital computers. This talk will explore a variety of methods memristors can be used with  CMOS in accelerating neural networks, the powerful capabilities memristive  neural networks are capable of (such as generating artificial slow-motion footage), and the implications of vertical stacking of crossbars.


Jason received the B.Eng degree in electrical and electronic engineering and the LL.B degree from The University of Western Australia, Perth, WA, Australia in 2016, where he is currently completing his Ph.D. degree. From 2015 to 2016, he was a Research Associate at Chungbuk National University, Cheongju, South Korea working on the Memristive Retina Project. His current research interests include memristive systems, inference acceleration and  
generative adversarial networks. He was presented with the Best Paper Award at the 2019 IEEE Artificial Intelligence Circuits and Systems Conference held in Hsinchu, Taiwan for his work titled “Analog Weights in ReRAM DNN