Sign in with
Sign up | Sign in

HP, Hynix Not Commercializing ReRAM Until 4Q13

By - Source: The Kavli Foundation | B 17 comments

HP doesn't plan to go commercial with ReRAM until the end of 2013 due to business reasons.

Stan Williams, Hewlett-Packard senior fellow and director of the company's cognitive systems laboratory, recently said that the company has decided to postpone manufacturing commercial memristor-based ReRAM (resistive random access memory) products until late 2013. The company originally planned to go into production during 2Q/3Q 2013.

"We'll have something technologically viable by the end of next year. It's sad to say, but the science and technology are the easy part," he told The Kavli Foundation. "The economics, investment, and market readiness are harder."

He went on to explain that Hynix, a major producer of flash memory that co-developed ReRAM with HP, will see its existing business "cannibalized" by memristors because they will replace some of its flash memory products. That means timing the introduction of memristors will be extremely important for the entire industry.

"There's a lot more money being spent on understanding and modeling the market than on any of the research," he said. "Development costs at least 10 times as much as research, and commercialization costs 10 times as much as development. So in the end, research -- which we think is the most important part -- is only 1-percent of the effort."

Short for "memory resistor", memristors can retain information even when powered off, requires less energy to operate, and is faster than current SSD technology. As current flows through the memristor in one direction, the electrical resistance increases – it decreases when current flows in the opposite direction. When the current stops, the memristor retains the last resistance that it had. Once the flow resumes, the circuit returns to what it was when it was last active.

HP began developing a switching memristor back in 2008, based on a thin film of titanium dioxide. At the time, the company predicted the tech would be commercially available 18 months later. Even more, HP said it will serve as a replacement for DRAM, SRAM, Flash and SSDs.

"One important aspect of memristors is that they are simple structures made from materials and processes already used in semiconductor foundries," Williams added. "This is critical, because we would face tremendous resistance if we asked the industry to change technologies. But they don't have to change. If you know what you're doing - and there's a lot of intellectual property involved - literally any foundry could make memristors tomorrow."

With commercial manufacturing of ReRAM not starting until the end of 2013, products utilizing this new tech – such as smartphones, tablets and more – likely won't show up on store shelves until 2015 or 2016.


Contact Us for News Tips, Corrections and Feedback

Display 17 Comments.
This thread is closed for comments
  • 0 Hide
    tanjo , October 1, 2012 2:43 PM
    Replace SSD, flash, yes... and maybe DRAM for gadgets. Unfortunately, memristors, like flash have write endurance. So, no, I don't want that in my desktop's RAM slot.
  • 0 Hide
    Anonymous , October 1, 2012 2:51 PM
    Interesting, can't wait! :D 
  • -1 Hide
    ojas , October 1, 2012 2:57 PM
    That's all good, just give me my DDR4-2800 in time for Haswell and i'll be happy. :) 
  • -3 Hide
    Estix , October 1, 2012 3:07 PM
    tanjoReplace SSD, flash, yes... and maybe DRAM for gadgets. Unfortunately, memristors, like flash have write endurance. So, no, I don't want that in my desktop's RAM slot.

    A good point, but, personally, I don't care about write endurance, so long as the expected life of the component is at least as long as the service life of the device.
  • 2 Hide
    alidan , October 1, 2012 3:47 PM
    tanjoReplace SSD, flash, yes... and maybe DRAM for gadgets. Unfortunately, memristors, like flash have write endurance. So, no, I don't want that in my desktop's RAM slot.


    for me, depends entirely on how many cycles it has.

    seeing as you could do a fullshutdown to full windows boot exactly how you left it in less than 10 seconds, im at least intrested, granted, it needs to either have a few million writes, or be a backup for ram, like if i "shut down" the computer, ram gets dumped to that ram, and it boots from that ram while everything is moved to normal ram.
  • 1 Hide
    Mathos , October 1, 2012 4:24 PM
    If it's usable to replace SRam the endurance is likely going to be pretty high. Considering Sram is what they use for Cache memory in most processors.
  • 4 Hide
    ethanolson , October 1, 2012 6:50 PM
    Well... everything has endurance. I hope the ReRAM has very high endurance so I can use it for 5+ years.
  • -1 Hide
    A Bad Day , October 1, 2012 10:28 PM
    tanjoReplace SSD, flash, yes... and maybe DRAM for gadgets. Unfortunately, memristors, like flash have write endurance. So, no, I don't want that in my desktop's RAM slot.


    There are plans of pushing memristors into other microchips to replace transistors, possibly including CPU and GPU.
  • -1 Hide
    nurgletheunclean , October 1, 2012 10:37 PM
    tanjoReplace SSD, flash, yes... and maybe DRAM for gadgets. Unfortunately, memristors, like flash have write endurance. So, no, I don't want that in my desktop's RAM slot.

    I have serious doubts their DRAM implementation would include desktop system memory replacement. There's no way it will be fast enough. Consider that DDR3 1600 is 12,800MB/s vs a fast toggle NAND is ~60MB/s If they could get even close to 5% that level of performance from ReRAM, they would release it as soon as possible since you wouldn't need any complicated controller (marvell/sandforce) to fully saturate a SATA3 interface. A USB3 thumb drive would be as fast at the fastest SSDs today.

    And for the people thinking that write endurance wouln't be an issue for system memory you are flat out wrong. DDR's number of write operations are far more frequent than disk writes, you don't "feel" them because they happen so fast. Typically you would write to memory 10x+ as often as a disk, as all disk reads are going to be memory writes, as well as all the different transformations, moves, copies, decompression, etc all are memory write operations.
  • -1 Hide
    tanjo , October 2, 2012 1:52 AM
    @nurgletheunclean:
    I've read that Elpida made a ReRAM prototype (64Mb) with a write speed of 10ns (which they say is 10x faster than the best NAND) with a million times write endurance.

    Then again no, I really don't want that as RAM. But they do obliterate flash hands down. Ultra low power plus x times faster. The problem is compatibility with current devices with flash based card readers. But upcoming tablets/smartphones can use this as internal storage no problem. (And maybe make it viable to turn off tablets if they can make it boot under 10 secs :) )

    I wonder what the price/GB would be?
  • 0 Hide
    izmanq , October 2, 2012 1:52 AM
    10x+ ? more like 1000x+ even more :D 
  • 0 Hide
    Anonymous , October 2, 2012 3:47 AM
    sdram from like 15 years ago was ~10 ns latency. DDR was slower, ddr2 was slower still, ddr3 was slower still.

    So, no its not 10x faster, nor is it 1000x faster....

    It would be great to have faster latency ram. Bandwitch has been going up a lot over the last 1-2 decades, but latency has suffered for it.

    If its 10ns at the same bandwitch as ddr3 then that would be great. But im not holding my breath.
  • -1 Hide
    saturnus , October 2, 2012 5:11 AM
    Current memristor technology based on titanium dioxide will not and is not targetted at replacing DRAM but flash RAM. It's far too slow to be used as DRAM, both in latency and potential bandwidth. Both are faster than flash by a factor of at least 10 though.

    The crucial part is that memristor with current technology easily shrinks die sizes as each memristor can easily hold at least 8 bit, and with adequate refinement of the technology potentially much more.

    Write endurance is also in the factor of 100s or 1000s above flash RAM. Seeing that current flash drives offer MTBF of 2 million+ hours, it's going to outlive any of us us by many years.
  • 0 Hide
    billyboy999 , October 2, 2012 5:39 AM
    akljk2lkl2lk2sdram from like 15 years ago was ~10 ns latency. DDR was slower, ddr2 was slower still, ddr3 was slower still.So, no its not 10x faster, nor is it 1000x faster....It would be great to have faster latency ram. Bandwitch has been going up a lot over the last 1-2 decades, but latency has suffered for it.If its 10ns at the same bandwitch as ddr3 then that would be great. But im not holding my breath.

    LOL... troll or ignorant at best. RAM has never been as fast as 10ns, if you're talking about total time to read a word, RAM from 15 yrs ago, or from now. Latency has not gone up much but it certainly hasn't degraded.
    The Pentium Pro was a CPU you could expect to find 15 years ago. Clocked at 150-200mhz, that gives it's L1 cache 7-5ns access time at best. You must be kidding if you think RAM was anywhere as fast as L1.
    SDRAM ran at 100MHz back then. Cas latency of 1 or 2 doesn't mean it takes one cycle to access a value.
  • -1 Hide
    luciferano , October 2, 2012 3:38 PM
    akljk2lkl2lk2sdram from like 15 years ago was ~10 ns latency. DDR was slower, ddr2 was slower still, ddr3 was slower still.So, no its not 10x faster, nor is it 1000x faster....It would be great to have faster latency ram. Bandwitch has been going up a lot over the last 1-2 decades, but latency has suffered for it.If its 10ns at the same bandwitch as ddr3 then that would be great. But im not holding my breath.


    billyboy999LOL... troll or ignorant at best. RAM has never been as fast as 10ns, if you're talking about total time to read a word, RAM from 15 yrs ago, or from now. Latency has not gone up much but it certainly hasn't degraded. The Pentium Pro was a CPU you could expect to find 15 years ago. Clocked at 150-200mhz, that gives it's L1 cache 7-5ns access time at best. You must be kidding if you think RAM was anywhere as fast as L1. SDRAM ran at 100MHz back then. Cas latency of 1 or 2 doesn't mean it takes one cycle to access a value.


    Latencies for system memory have been improving over the years last I checked, not degrading. A timing of say 9 on DDR3-1600 is still lower latency for that timing's operation than a timing of 3 on SDRAM that runs at 133MHz. Timings increase, but the frequency increases are enough to counteract that (although early modules of a new RAM interface often have higher latency than the previous modules of the same transfer rate, probably caused by the chips being more complex and also immature process shrinks, as they mature, they improve).

    I think that billyboy999 is correct in that DRAM access latency for system RAM was never anywhere near 10ns. Iit's something like 150-250 in most systems right now IIRC, but that's probably mostly caused by the system memory being several inches away instead of several micrometers away from components in the CPU, not the difference in performance between CPU cache SRAM and DRAM. A specific timing such as CAS can get around 10ns, but I don't think that it was near 10ns with any of the older SDRAM. SDRAM had higher CAS duration than that IIRC, let alone full access time with the other timings included.

    ReRAM can probably be fast enough to replace SRAM. Remember, a huge part of SRAM's performance comes from the process that it is built on. eDRAM can be pretty darned close and the difference between it and DRAM is just that eDRAM is built on the same process technologies as CPUs and such instead of the processes that are used for DRAM chips that are focused on capacity and density instead of performance. ReRAM, when manufactured in a CPU, can probably be faster than the SRAM that is currently used in the caches. Also, if ReRAM is much denser than SRAM, then interleaving and such can be used to great extents to possibly beat SRAM in at least some metrics. ReRAM might not be fast enough for L1 cache, but even if its not, then it might be great as L2, L3, or L4.
  • 0 Hide
    yourma2000 , May 12, 2013 3:37 PM
    "Hynix Not Commercializing ReRAM Until 4Q13"
    Until? It was pushed back to 2014 so if anything it's back on track
    Also memristor write cycles are meny magnitudes higher than flash
    youtu.be/n3XzuBt54ig?t=51m56s
  • 0 Hide
    yourma2000 , May 12, 2013 3:45 PM
    "Hynix Not Commercializing ReRAM Until 4Q13"
    Until? It was pushed back to 2014 so if anything it's back on track
    Also memristor write cycles are meny magnitudes higher than flash
    youtu.be/n3XzuBt54ig?t=51m56s