Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Computational memory: a stepping stone to non-von Neumann computing?
Abu Sebastian.YouTube,01:20:34,published onMar 8, 2018,stanfordonline,https://www.youtube.com/watch?v=_2Wiql4QSLQ.Type:Video
Date Reviewed: Sep 27 2018

The topic of this video presentation is computational memory (CM), a seemingly fascinating new paradigm in computing. The presenter’s approach to the topic is captivating; when you start the video, you just want to proceed forward and see what will come next. The presenter is a scientific researcher from the IBM Research lab in Zurich, Germany. His words are very clear, his statements coincide with our knowledge on the theory of computation, and his research can do nothing other than stimulate our inquiry on the physical limits of computation and how can we bypass them.

In the first part of the presentation, which is one third of the video, we learn: What is CM? What are the constituents of CM? Why is CM important? What are the computational tasks that can be performed using CM, like logical and arithmetic elementary operations? In the remaining part, the speaker discusses potential applications of CM. First of all, CM finds its challenges in easing the way of accessing and updating the memory of a system that can be used as a training repository in machine learning scenarios. Traditionally, if we have a dataset to be learned and generalized, then a copy of this data is held in the memory and frequent access (read or update) on its constituent blocks are executed by the central processing unit (CPU). However, in the CM paradigm, the access from the CPU to the memory is not just minimized but abolished, because memory updates and changes are performed on the memory units themselves. Of course, not all computational operations are directly performed on the memory, and certain logical and arithmetic operations will still reduce the speed and simplify the bottleneck of the wide complexity of executing operations. Exploiting the physical attributes of the memory device and the memory state dynamics does this. Specifically, the atomic arrangement on a nanoscale volume of material is used to designate a state of memory, which electric pulses can alter. For instance, an atomic arrangement can be changed from a disordered configuration to an ordered configuration by applying specific voltage pulses, which manifests a writing process to the memory. Also, a reading process can be manifested by checking the resistance of the material to electrical pulses, because an ordered configuration has a low resistance while a disordered configuration has a high resistance. (This is the reason that such devices are called resistive memory devices--memristors.)

Two thirds of the presentation tackles the implementation of an exclusive or (XOR) function, a matrix vector multiplication, an image compression scenario, and a deep learning network using memristors. I was shocked because the speaker is just examining the memristor capabilities that are well known in the literature. Thus, labeling the topic of this presentation as “a stepping stone to non-von Neumann computing?” is misleading because one doesn’t notice the question mark. My conclusion is that an eccentric title should not fool you as it did me. However, you can spend your free time watching this presentation to learn about the benefits of memristors.

Reviewer:  Mario Antoine Aoun Review #: CR146254 (1812-0633)
Bookmark and Share
  Featured Reviewer  
 
Von Neumann Architectures (C.1.1 ... )
 
Would you recommend this review?
yes
no

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy