Keynote on memristors by R. Stanley Williams of HP Labs | Hacker News Reader

Keynote on memristors by R. Stanley Williams of HP Labs

youtube.com

35 pointsmodeless10 comments

This 45-minute keynote presentation has pretty much everything you could possibliy want to know about memristors. If you want the 6-minute condensed version instead, try http://www.youtube.com/watch?v=rvA5r4LtVnc

It seems as if memristors have the potential to be very important to the future of computing, but the one question I haven't seen answered yet is endurance. The operation of memristors involves atoms physically migrating back and forth. If memristors wear out like flash memory after a few thousand switchings then the talk of them replacing DRAM is just hot air, and building artificial synapses seems unlikely to work either.

If I'm not misunderstanding, in the full 45-minute keynote, he mentions endurance a bunch of times. Apparently, on the newer memristor devices they came up with, the lifetimes of the devices are on a "geological" timescale, and could "theoretically" last forever. Of course, theory != practice. But there was quite a bit of time spent talking about this issue.

Edit: Also, I would highly recommend those with the time to watch the 45-minute keynote, and not just the condensed explanation. It really is worth it.

When he was talking about geological timescales, what he meant was how long a memristor retains it's state after being switched, while I'm more interested in how many times can he safely switch the device.

He indicates at 0:33.30 that the endurance is slightly better than flash.

Yes, and what I'm wondering is if that's a fundamental limitation or if it's something that can be easily improved. The endurance of DRAM is essentially infinite and I have a feeling that's going to be tough to reach with memristors.

Williams discusses briefly here (at 50 min) his hope that endurance would be significantly improved by moving from the lab equipment they use for fabrication to a more commercial quality system. http://www.youtube.com/watch?v=BHJvp5MybkM&feature=chann...

Sounds a bit wishful in thinking, but I have no experience in the matter.

The center panelist then seems to indicate that Intel is more interested in the phase change version because they want 'a rock that can switch 10^6 times..'.

So I think you might be right that this is a significant issue (particularly into DRAM or computation).

This is astounding! To think that circuit component as fundamental as resistors/capacitors/inductors has been hidden from us till now..

And the applications are equally mind-boggling. If he's right about what is potentially possible using memristors then we're in for an amazing ride this decade. I for one hope we can soon say our farewells to HDD's, DRAM and SSD's.

Er ... Josephson junctions, bubble memory, etc. ? Sometimes things do not pan out. Either because there's no significant advantage to the new tech, or because of manufacturing costs/problems, or ...

Having hope is good, but keep in mind that the guys coming up with the new stuff point the upsides, of course. Then the hard-nosed spoilsports figure out the problems.

Personally, I've been keeping an eye on this one for over a year - we'll probably figure out if it's a 'yea' or a 'nay' in another year. But I'm not seeing as much third-party excitement as expectable.

I'm wary as well. Think of how many revolutionary ram technologies have been just around the corner for ages. Mram, Feram, the "Upcoming" section of Wikipedia's memory pages lists more. Some of them were discovered in the 70s and are now 3-5 years from mass market! Just like they were three years ago!

This also seems a perfect fit for FPGA.

Reply on news.ycombinator.com