4 ms·

HP and Hynix - Bringing the Memristor to Market in Next-Generation Memory

32 points by hornokplease 16y ago

wcarss 16y ago
I have a burning question with regards to memristors: I've read that memristor technology can be used both for memory and for processing; does that imply both? If there is just one fabrication process that builds a "memristor chip", and fpga-like systems can be built where memory can be converted into processing power, and back into memory freely, then this is by far the most revolutionary technology advance I've witnessed in my life. If not, it's still neat for many reasons - but I haven't ever seen clarification on this. Even if it's not being done in this batch or with this fab process, would it be possible/not stupid hard to do, after some investment?
- izendejas 16y ago
  Yes, the link above links to an hp page with the following anchor text: "[with memristors] both compute and memory functions could be conducted within the same chip".
izendejas 16y ago
I've been following this work for years now and it's exciting to read this announcement. I hope I'm right when I predict this technology will be a catalyst for huge economic growth in years to come just like transistors and the internet. Low-power devices that remember their state even when powered off? Imagine the possibilities.
ewams 16y ago
Prith Banerjee, HP's senior VP of research at HP Labs gave a talk about the memristor at HPtech 2010 back in June. He did mention quite a bit of the information that was in the article linked, except of course the purchase of Hynix. This is very exciting technology, hopefully it comes out in a positive manner that allows the adoption in to many new (and existing) technologies. Our planet could be improved with this technology if handled correctly.
DanielBMarkham 16y ago
This is great to hear. There are 3 hot new technologies in the wings, waiting to enter: OMAP4, PixelQi, and now memristors? Can't wait to see the convergence of these three. Now we just need a miracle in energy storage technology for a perfect storm.
adbge 16y ago
I remember really becoming excited about memristors after reading this post a while back on HN: http://highscalability.com/blog/2010/5/5/how-will-memristors-change-everything.html http://highscalability.com/blog/2010/5/5/how-will-memristors...
izendejas 16y ago
Wikipedia has a nice section on the history of said memristor. Apparently, its discovery goes back to 1960 with work by Bernard Widrow (Professor Emeritus, Stanford). A few patents are coming out of this. http://en.wikipedia.org/wiki/Memristor#Timeline_of_memristor_and_memristive_systems_developments http://en.wikipedia.org/wiki/Memristor#Timeline_of_memristor... One of the patent awardees, Blaise Motett, seems to have a different view of HP's memristor, namely that their sr is not actually one. http://www.slideshare.net/blaisemouttet/mythical-memristor http://www.slideshare.net/blaisemouttet/mythical-memristor
tocomment 16y ago
So what happens between 2007 (when they were discovered) and 2013 (when manufacturing starts)? What work has to be done that takes that many years? I'm genuinely curious why bringing something to market takes so long. What are the tasks involved? What are the labor and time-intensive points?
- nooneelse 16y ago
  I'm just guessing, but a bit of experience in research and fabrication is behind the guesses. The first successful devices are usually the result of a more finicky and expensive set of processes. Also, all the fabrication steps have been ordered, and if necessary had their process parameters pushed out of otherwise optimal zones for the machines in question, to make the desired device come out right. So, after making the first ones they had to find a way to make them with better, commercially viable, yields. Also they had to develop a process flow to make memristors at the same time as, or at least compatible with, the other steps needed for the other devices that will be needed on a marketable chip. It is common to push necessary support circuitry for a device off the initial test wafers (if for nothing else, it helps reduce the variables, if the devices don't appear to work, was it the new device under test or the amp you put right next to it on the chip; better to put another test device in that space, have two to test, and use an off chip amp to test them). In addition, there may have been issues with the materials used or waste products produced. You can play around with small amounts of more expensive materials in research. Similarly, disposing of small amounts of very nasty waste products can be fine on the research scale, but dealing with the amounts produced after scaling production up might make the venture unprofitable. Also, many of the process steps may have been far more "hands on" than mass production demands. So all of those have to be automated and process parameters for them optimized. All of this is based on academic research experience, which, afaik, doesn't constrain itself as much to sticking as close as possible to the established tools as industry research might. But with a whole new class of devices like this, I imagine even an industry research group wouldn't let a little thing like lack of an established, trustworthy tool for a job stand in the way of a promising avenue toward success. I'd say six years is doing very well considering all these issues and many of them are probably still in the process of being ironed out.

32 points by hornokplease 16y ago

I have a burning question with regards to memristors: I've read that memristor technology can be used both for memory and for processing; does that imply both? If there is just one fabrication process that builds a "memristor chip", and fpga-like systems can be built where memory can be converted into processing power, and back into memory freely, then this is by far the most revolutionary technology advance I've witnessed in my life. If not, it's still neat for many reasons - but I haven't ever seen clarification on this. Even if it's not being done in this batch or with this fab process, would it be possible/not stupid hard to do, after some investment?

izendejas 16y ago
Yes, the link above links to an hp page with the following anchor text: "[with memristors] both compute and memory functions could be conducted within the same chip".

izendejas 16y ago

I've been following this work for years now and it's exciting to read this announcement. I hope I'm right when I predict this technology will be a catalyst for huge economic growth in years to come just like transistors and the internet. Low-power devices that remember their state even when powered off? Imagine the possibilities.

ewams 16y ago

Prith Banerjee, HP's senior VP of research at HP Labs gave a talk about the memristor at HPtech 2010 back in June. He did mention quite a bit of the information that was in the article linked, except of course the purchase of Hynix. This is very exciting technology, hopefully it comes out in a positive manner that allows the adoption in to many new (and existing) technologies. Our planet could be improved with this technology if handled correctly.

DanielBMarkham 16y ago

This is great to hear. There are 3 hot new technologies in the wings, waiting to enter: OMAP4, PixelQi, and now memristors? Can't wait to see the convergence of these three. Now we just need a miracle in energy storage technology for a perfect storm.

adbge 16y ago

I remember really becoming excited about memristors after reading this post a while back on HN: http://highscalability.com/blog/2010/5/5/how-will-memristors-change-everything.html http://highscalability.com/blog/2010/5/5/how-will-memristors...

izendejas 16y ago

Wikipedia has a nice section on the history of said memristor. Apparently, its discovery goes back to 1960 with work by Bernard Widrow (Professor Emeritus, Stanford). A few patents are coming out of this. http://en.wikipedia.org/wiki/Memristor#Timeline_of_memristor_and_memristive_systems_developments http://en.wikipedia.org/wiki/Memristor#Timeline_of_memristor... One of the patent awardees, Blaise Motett, seems to have a different view of HP's memristor, namely that their sr is not actually one. http://www.slideshare.net/blaisemouttet/mythical-memristor http://www.slideshare.net/blaisemouttet/mythical-memristor

tocomment 16y ago

So what happens between 2007 (when they were discovered) and 2013 (when manufacturing starts)? What work has to be done that takes that many years? I'm genuinely curious why bringing something to market takes so long. What are the tasks involved? What are the labor and time-intensive points?

nooneelse 16y ago
I'm just guessing, but a bit of experience in research and fabrication is behind the guesses. The first successful devices are usually the result of a more finicky and expensive set of processes. Also, all the fabrication steps have been ordered, and if necessary had their process parameters pushed out of otherwise optimal zones for the machines in question, to make the desired device come out right. So, after making the first ones they had to find a way to make them with better, commercially viable, yields. Also they had to develop a process flow to make memristors at the same time as, or at least compatible with, the other steps needed for the other devices that will be needed on a marketable chip. It is common to push necessary support circuitry for a device off the initial test wafers (if for nothing else, it helps reduce the variables, if the devices don't appear to work, was it the new device under test or the amp you put right next to it on the chip; better to put another test device in that space, have two to test, and use an off chip amp to test them). In addition, there may have been issues with the materials used or waste products produced. You can play around with small amounts of more expensive materials in research. Similarly, disposing of small amounts of very nasty waste products can be fine on the research scale, but dealing with the amounts produced after scaling production up might make the venture unprofitable. Also, many of the process steps may have been far more "hands on" than mass production demands. So all of those have to be automated and process parameters for them optimized. All of this is based on academic research experience, which, afaik, doesn't constrain itself as much to sticking as close as possible to the established tools as industry research might. But with a whole new class of devices like this, I imagine even an industry research group wouldn't let a little thing like lack of an established, trustworthy tool for a job stand in the way of a promising avenue toward success. I'd say six years is doing very well considering all these issues and many of them are probably still in the process of being ironed out.