Slashdot Mirror
Why Micron/Intel's New Cross Point Memory Could Virtually Last Forever
Lucas123 writes: As they announced their new 3D XPoint memory this week, Micron and Intel talked a lot about its performance being 1,000X that of NAND flash, but what they talked less about was how it also has the potential to have 1,000X the endurance of today's most popular non-volatile memories. NAND flash typically can sustain from 3,000 to 10,000 erase-write cycles — more with wear-leveling and ECC. If Micron and Intel's numbers are to be believed, 3D XPoint could exceed one million write cycles. The reason for that endurance involves the material used to create the XPoint architecture, which neither company will disclose. Unlike NAND flash, cross point resistive memory does not use charge trap technology that wears silicon oxide over time or a typical resistive memory filamentary architecture, which creates a statistical variation in how the filaments form each time you program them; that can slow ReRAM's performance and make it harder to scale. Russ Meyer, Micron's director of process integration, said 3D XPoint's architecture doesn't store electrons or use filaments. "The memory element itself is simply moving between two different resistance states," which means there's virtually no wear.
Wouldn't it be way obsolete by the 1 millionth write cycle ?
The last thing that was supposed to last virtually forever was those overpriced lightbulbs that were supposed to last for decades. I'm lucky if I get a year out of them. Let's revisit the new memory when it has a track record.
Is this memory based on silicon, or something else, like GaAs or Germanium or Graphene or something else?
long live RM films.
I believe they've also indicated the "1000x endurance" figure isn't based upon write cycles -- meaning it has to be based on typical lifetimes. So yes, you're talking many years of service.
Kythe
that there is always a catch.
So, what's the catch?
Overhyped technology that will never ship as a part of real product.
That is a!!
I've had a couple CFLs go out early, but I do have a few that are going on 7 years, but I've only had one LED bulb die and that was within a week...mfg replaced it. And I'm always a fairly early adopter so my oldest LED bulb is about 4 years old....and yes I paid way to much for it.
If you are having bad luck, I'm guessing you have crappy power. I'm lucky that my power is pretty clean (I borrowed a powersight monitor from work for a month).
"Crappy power" is normal. Manufacturers need to design for that. How long something lasts in a lab is irrelevant.
This sounds like this may be some sort of phase change memory where controlled pulses of current can switch a material between a high and low resistance phase. The memory can then be read by passing low current through the bit to determine which phase it is in without flipping the phase.
... that says that Google must erase this memory after 500,000 write cycles.
After all, bits have a 'right to be forgotten'.
Crossbar vs XPoint (Crosspoint) can't wait to see how this plays out. :-)
Once these hit consumer devices life could be alot faster and last longer
But what is its behavior re data remanance? Not the first concern of course, but certainly one of them these days...
Wear-leveling, levels the demand for erase-write a cell, between all available cells. It does not increase their durability. It just makes them equally worn out, so part of them don't fail, while the others still have a lot of juice in them. ECC catches errors caused by the environment. It also does not increase the durability limit of the cells. And after they're worn out - they are worn out. ECC will not make them continue to store info for you. 8GB x 1million write cycles = 8 000 000 GB changed. According to Wikipedia, DDR3-1600 has a Peak transfer rate of 12,5GB/s. At comparable speeds, the 8GB of Intel's new memory will be worn out in about 178h, which is a little over a week of crunch. And this is not nearly forever.
CFL, much like incandescent experience most of their wear when you first turn them on and during warmup when they are overdriven. (They also consume more power during this time.) Putting them in a frequently used closet or bathroom and turning them on and off many times per day will kill them very fast compared to their rated lifespan. LED fixtures are far more tolerant of power cycling.
If information wants to be free, why does my internet connection cost so much?
Your Internet connection is not information. It is a complex system of wires/tubes/fibers run by computers, and uses electricity, occupies land. It is operated by a corporation who pays people, negotiates with other corporations, and deals with/pays for many subtle and not so subtle political aspects of the whole thing.
This issue is a bit more complicated than you think.
But a small boot routine in ROM that erases a range of RAM (persistent or not) isn't that hard to conceptualize. Besides, depending on the type of volatile RAM, it doesn't always come up as all zeros at power-on either... I mean, what do you think happens when you press a reset button? Everything is still in RAM at that point.
What would make things different is a software architecture change that gets rid of the separate permanent storage layer and makes everything RAM-persistent. That would be kinda strange to imagine.
"Crappy power" is normal. Manufacturers need to design for that.
Yeah, see physics doesn't care what you think is "normal". While it's possible to design for a reasonable range of power conditions it is not economically possible to design for all of them. Frankly if your power quality is so poor that they are constantly blowing out light bulbs the answer is to fix the power, not the bulbs. You take the bullets out of the gun rather than insist everyone wear a bullet proof vest.
If XPoint memory can only do 1,000,000 read/write cycles...then in no way could this stuff ever substitute for DRAM. In BASH you could easily program a loop that counted from 1 to 1,000,000 - wouldn't this write some place over a million times...and now that piece of memory would not work? DRAM can easily take millions of reads and writes!
Sounds like someone who is completely ignorant of what constitutes laboratory testing. They better test response to crappy power than you possibly could in your uncontrolled tin shed, or whatever rat hole in which you dwell.
Please don't use 'virtually' like that in an article about memory. It's just going to confuse the issue.
When someone says, "Any fool can see
""Crappy power" is normal. Manufacturers need to design for that. "
Get a UPS.
Is this just Intels version of a memristor? The change in resistance is do to the movement of oxygen atoms if it's a memristor. HP has a large amount of information available on memristors and seems to lead in the development of memristor cross bar technology.
If it was ever true that power cycling is hard on incandescents, it hasn't been for decades. The primary wearout mechanism is evaporation of the filament.
Contribute to civilization: ari.aynrand.org/donate
Can someone explain how exactly they're "moving between two different resistance states"? Because I think that that in itself does not guarantee lower wear.
Has anyone heard anything technical about how this works?
I think people notice incandescents fail right when being turned on rather than just crapping out while they were already running. I don't know if this is selection bias and these events are just more noticable than the other types of failure.
Whether the power cycling accelerated the degradation over the long term, they also don't know without controlled experiments. But there *seems* to be a at least a final triggering effect that might be as simple as thermal expansion/contraction causing it to flex and break at the weakest point in the filament.
Just look for "industrial Flash". The problem is that it does not get the density and is way more pricey than consumer Flash. Hence it is quite possible that this is just another "wonder" storage technology that will fail. BTW, I am still waiting for those holographic tapes that will solve all bulk-storage needs forever.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
"Crappy power" is normal. Manufacturers need to design for that.
No, you need to design for that. Install a power conditioner, or a UPS (not SPS). You can fix this problem.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
The last thing that was supposed to last virtually forever was those overpriced lightbulbs that were supposed to last for decades. I'm lucky if I get a year out of them.
Don't buy overpriced and overhyped. Just buy good quality. There's no reason you should have failing lightbulbs anymore unless you're driving them with really nasty power or installing them in some wonderfully heat-trapping light fittings.
I think people notice incandescents fail right when being turned on rather than just crapping out while they were already running. I don't know if this is selection bias and these events are just more noticable than the other types of failure.
I'm quite sure that they fail right when turned on more in this house, but I've got almost all of them replaced with Cree lights now that they are dirt cheap at the home despot. Except I've got a $1 CFL over the stove, where the most lights died, and it's been a peach.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
"Crappy power" is not normal. Many western power grids have strict standards for voltage spikes, dips, sags, frequency errors, and THD that they will provide you. If you have crappy power it's because either your wiring is stuffed or you're running devices which are absolutely ancient (30 year old fridges make for some nasty power spikes), or have electronics which are failing / fake / never had the right certification to begin with and are spewing noise back onto your powerline.
It is? I'm genuinely curious why you say that. My impression is that in the developed world the power supply is quite reliable and carefully regulated, and that even in the less-developed parts of the world, you can find relatively inexpensive solutions for normalizing your home power, if it's that important.
I've got a whole whackload of Cree LED bulbs. I had a 100W completely die, my other 100W flashes on occasion (apparently their 100W bulbs have a notoriously high failure rate), I had to get rid of a 60W from a lamp fixture because it periodically switched back and forth between full brightness and lower brightness. So far, they have not been substantially more reliable than CFL or incandescent. They may not burn out as often, but they "soft-fail" more often than CFL or incandescent did, those tended to either work fine or not at all.
Their warranty is basically worthless. They are only sold by one store (Home Depot), and warranty replacements can't be done where you bought it. You need to ship them the broken lightbulb (at your own cost) to get a replacement, and shipping something that size/weight costs more than buying a new LED bulb in the first place, making the warranty completely useless.
I was not aware that LED light bulbs have been having significant problems (those are the only types of bulbs I can think of that are supposed to "last for decades," correct me if I'm misunderstanding). My impression is that LED bulbs are improving in longevity and dropping in price very rapidly. If we're just trading anecdotes, I have not had to replace a single LED bulb at my house since I started phasing them in 6 years ago.
Comment removed based on user account deletion
He lives in the US.
Hmm, Cree are not cheap enough that you should have to put up with that. Maybe try Osram next time you replace one? I have 16 of them, without a single problem for two years. Usage pattern: I obsessively turn the lights off and on when I move between rooms, so compared to a normal person I probably keep them turned on a bit less in average, but with a lot more on-off cycles.
I blame children. A couple of years ago mine discovered that if you flip the switch up and down fast you get a strobe effect. It sort of works with CFLs but it does end their life pretty quick. Previous to that I think I had replaced 1 CFL bulb in the bathroom in the previous 7 years. Also the quickest way for them to quit doing that is to make them pay for a new bulb and tell them they can do it some more but they get to keep buying bulbs.
Time to offend someone
Any sufficiently advanced technology is indistinguishable from magic.
The mechanism is quite well known.
Hint: imagine a more-or-less thermally coupled series string of PTC thermistors with equal power ratings. Now consider what happens when one of them has a significantly higher cold resistance (= a thin section in the filament).
the 40W and 60W equivalent I've had no issues with, has anyone tried the 75W equivalent.
Maybe 100W equivalent (13.5 watts) is just too much heat in too small an area for LEDs
I even had an incandescent explode (or more accurately implode) when turning it on.
I had three of the 40w TW Cree bulbs crap out with flickering and eventually dying. I contacted Cree support, took a photo of the packaging, and they, no questions asked, fedexed me three new bulbs. I didn't have to send the old ones back. Worth a shot.
If latency is 1,000 times lower and endurance is 1,000 times higher then, under continuous load, endurance measured in real time is unchanged. Not by any means a hypothetical scenario.
When all you have is a hammer, every problem starts to look like a thumb.
Phase-chase resistive memory (crosspoint resistive memory is described in decent detail towards the end of this presentation):
https://www.micron.com/~/media/documents/products/presentation/gatwood_current_emerging_memory_landscape_fms2011_final.pdf
It's just like the grandparent said: the filament gets thinner over time as the metal evaporates off during use. The moment of death for the bulb is typically when someone turns it on. Why? There is a surge of power sent thru the bulb's filament when is turned on from an already-off state, and if the filament is very thin (from having been "on" for say 1000 or 2000 cumulative hours, or whatever the rated life of the bulb is), then it will break the excessively thin filament at that moment and the bulb is no more. That same excessively thin filament could've handled *staying on* for a while longer, but a clean restart is what it *can't* survive.
Yeah, I've got about 30 Phillips LEDs around my house and none has failed since I bought them five years ago.
>The last thing that was supposed to last virtually forever was those overpriced lightbulbs that were supposed to last for decades. I'm lucky if I get a year out of them.
Encountering an early failure feels especially strange the first time. I was thinking that a little shake-it-like-you-mean-it flashlight (magnet flies back and forth inside a coil to charge a capacitor) would never need a battery. There is no "battery", but the little 5.5 Volt 1 Farad capacitor is dead (open circuit). So much for the one flashlight I thought I could trust to work. I'm tempted to replace the soldered-in cap with a bigger one that'll use the full space the design originally made room for...but something tells me the shipped cost of anything may be as much as another flashlight. Maybe I just just tape a 3.6 Volt lithium-ion cell (from a sick laptop pack) to the side and wire that in.
Those dead CFLs usually have a couple of good high voltage power transistors. Maybe using a lens from a dead big-box projection tv, some homeless person can make a solar soldering iron and build a stun-gun inverter or something? Slashdot poll question - what could you build with a pile of dead CFLs and other consumer scrap? It could be just the trick to make the homeless of Michigan and New Jersey ready for Foxconn. Young people might learn something too, but I'm not sure how many could put down their smart-things long enough to try soldering.
Power cycling an incandescent doesn't wear it out faster, but once it's "worn out" it will more likely fail the moment it's turned on or off because the change in temperature induces mechanical stress. A light bulb can also be broken by water condensing on it while it's off, causing asymmetrical stress that cracks the glass when it begins to heat up after being turned on (I had a whole row of vanity bulbs explode because of this once).
I've been in my house 5 years and replaced one. It broke when I knocked a lamp over.
So... back to either finding a decent bulb fittings which don't overheat the bulb or getting the power quality in your house checked.
I think you just made the GP's argument for them.
- standards are great. You know what is greater though? How often are the standards met. Get back to us with metrics or you are talking about "in a perfect world...";
- lightning. 'Nuff said;
- Lots of equipment plugged into the grid can cause problems;
- You can't see electricity and can only infer power problems. Unless you measure and profile, which the average person will never do. I've literally only seen this once in a quarter century in the tech industry;
- the grid is designed to provide power. The realistic result is that base power is substantial, and therefore variations are often not noticeable. Your fridge, your stove, your lamps, in fact most devices are pretty forgiving if the voltage varies by 10% from nominal;
- I've seen the result of what happens when old-school electricians, with no training or information, tried to wire low voltage digital communications cables. They did a terrible job because they did their work to power grid standards, not signalling standards. Yes they got better over time. But it's telling that new Ethernet lines are routinely tested with a fairly elaborate test meter. New power outlets? Many are never tested. The client "tests" them on (and by) first use. Even the ones that are tested, only get a simple test that tells you if you swapped lines somewhere in the circuit, or ground isn't connected. That standard would/will get you fired in digital communications.
do tell the $500 solution that would "fix the power" for an entire house.
This would be a start ($499):
http://www.fullcompass.com/prod/040433-SurgeX-SX20-NE
I run both of my systems and all their gear through this one below (15A 120v) and it costs $399 (less elsewhere):
http://www.fullcompass.com/prod/067278-SurgeX-SA15
They use Inductance for surge suppression and power smoothing so they don't burn out like all the cheap ones people are used to. They'll last forever.
American already as a nation makes the mistake of confusing "really big" with infinite in terms of military and economic power. Big is not infinite. Nearly every Epic Fail of the United States today can be traced to making this simple mistake.
In electronics, big is not even big if you can count of times you can do anything. Big is SRAM or DRAM where there are no cycles to count and where it's only time of use and that time of use is measured in decades to centuries.
And at least get your numbers correct. NOR flash is close to 10^4. NAND flash is 100x larger. Which puts RRAM at closer to only 10x better than flash.
Of course since 90% of the semiconductor, computer and electronics industries is no longer physically located and economic entities in the US, I suppose I can't be too surprised that no one get these basic concepts right, basic math or technology-wise.
Which sums up were I live about 30 minutes west of St. Louis, Missouri. We get regular thunderstorms and lose power 2 or 3 times a year for hours and probably twice that often for seconds to minutes every year. Light bulbs including incandescent, compact florescent, and LED have half lives here of 3 to 6 months so the later two are not even close to economical no matter how much more efficient they are.
All of my computers, network gear, and test equipment is protected by online UPSes which has worked out well.