Subscribe / Unsubscribe Enewsletters | Login | Register

Pencil Banner

Seeking the Fourth Circuit Element

F.Y. Teng | May 25, 2010
A conversation with the czar of R&D at HP Labs about taking memristors from paper to production to mobile computers to enterprise infrastructures.

In 2008, we actually had the light bulb moment and we said: Aha! This thing that many people have been working on is actually memristor! So ReRAM is memristors and that also essentially taught us the basic physics of the operation of the device. Once we understood that we could start engineering the devices a lot better instead of doing Edisonian tweaking, which is what we had been doing before that.

In the past two years we published 17 papers in this area. Many of these papers essentially describe issues of the basic physics of the operations of the devices, while others describe applications. The different types of applications for instant memory, which is obvious I think, as configuration bits and switches in something called field programmable data rays, and as learning networksbeing able to build a hard wired neural map essentially where memristor switching in hardware is actually done to teach a machine to do something, as you would with a neural map but instead of having a computer program in the digital computer, you actually do it in hardware.

The most recent paper that we published was [round the week of April 19] which came out in Nature [international journal of science]. This paper described another new thing that we figured out that memristors can do. The latest new thing that we learnt was that memristors are not only storage devices. They can not only act as memory but they can also compute. They are logic gates, so the same device can either be a bit of storage or it can be a gate to enable us to do computation. Thats rather unique. It opens up some interesting implications that are even further down the path. I mentioned the possibility of flattening out the memory and storage hierarchy earlier; well, eventually somewhere down the road it is possible that one can even do the computation in the memristors themselves. So instead of having to send the data to a processor, youd send the program to the data where it resides in some sea of storage, and do that computation locally, and then just shift out a few bits that represent the answer that people want to know.

When do you expect memristor chips as processors to go into full production for commercial use?
We are just now entering the development stage, so weve gone from what we call lab to fab getting wafers out of production scale fab. That process will take almost 18 months of tweaking and getting several cycles in order to look at manufacture ability and reliability.

After that, we would go through a commercialisation process. That would be at least three years. About three years from now, we hope to have a first product on the market place and that would likely be something simple of course. You dont start at the very top, so were likely to come up with something that would compete with flash memory or phase-change memory for applications in handheld devices, iPod-type devices, various things like digital cameras that currently use flash memory. We think this technology would be superior there.


Previous Page  1  2  3  4  5  6  Next Page 

Sign up for Computerworld eNewsletters.