Melting Memory Chips In Mass Production

chill writes "Nature is reporting that 'South Korean manufacturer Samsung Electronics announced this week that it has begun mass production of a new kind of memory chip that stores information by melting and freezing tiny crystals. Known as phase-change memory (PCM), the idea was first proposed by physicists in the 1960s.' With transistor-equivalent cells only 20 nm wide, switching time is around 16 ns. The first target market is cell phones, but the companies behind the technology see applications in PCs, servers, and other devices as well."

Comments

awesome

[Anonymous Coward]

i've been waiting for pcram to show it's head in consumer electronics for a while now. it has the advantages of being hundreds of times faster than flash along with having at least ten times the write-cycle life. it could turn out to be the OLED to DRAM's LCD.

the main disadvantage is that it's rather heat-sensitive, since writing is accomplished by melting crystals with a low melting temperature.

Re:awesome

[icebike]

Rather heat sensitive, in comparison to other technologies, but the critical temperature of GeTe grystals is around 446 Celcius.

At room temp this stuff is rhombohedral structure, at at 446C it changes to a cubic structure. The size of these tiny crystals is so small that this temperature is easily reached.

Note that no liquid phase is involved here, its simply changing from a glass structure to a crystal structure.

This 446C temperature is not likely to be reached in the absence of other heat related destructive events, regardless of how tight your jeans are.

Re:awesome

[davester666]

Yes, and they've genetically engineered tiny little sharks to swim around zapping the crystals on and off.

Re:awesome

[Tweenk]

Note that no liquid phase is involved here, its simply changing from a glass structure to a crystal structure.

No. Glass is a form of liquid - it has no translational symmetry at the molecular level. Rhomboedral or cubic structures are not glasses because they do have translational symmetry and are therefore crystalline. The change is between two types of crystal lattice.

Re:awesome

[qeveren]

Glass is an amorphous solid, not a liquid.

Re:awesome

[fractoid]

This 446C temperature is not likely to be reached in the absence of other heat related destructive events, regardless of how tight your jeans are.

Well... unless you count lithium battery fires... :P Although I suppose they'd count as 'destructive events'.

Kidding aside, thanks for the rundown. It now makes sense. :) One thing I don't understand, though, is the write cycle life. Does the phase change substance gradually settle into a third state? Or does the heating mechanism 'wear out'?

Re:awesome

[networkBoy]

The crystals are melted by passing a current through a BJT transistor. The heat given off melts the calcoginide(sp?) material. Reducing the current quickly causes it to freeze in an amorphous state, cooling it slowly produces a crystal. The resistance of the two phases is different, thus having memory.

Pros:
*Naturally rad hardened since it is a physical state, not a charge like flash and DRAM.
*Easy to erase in manufacture (the reflow temp is high enough to erase the whole memory)
*3D memory arrays are possible. The same material can be used (with metal) to make a transistor, thus you can make layers of arrays. Traditional flash is one layer deep as it requires doped Si for the transistor.

Cons:
*In prototypes we've seen cell phones erase themselves when left in a closed up black car on a black dash with a black interior on a hot Phoenix AZ summer day.
*you can't factory program the memory, it must be programmed after reflow onto the device. (flash can be ordered from the factory pre-programmed in large unit orders).
*Manufacturing defects have been an issue (bubbles in the calcoginide material.

I used to work for a company developing this stuff. We had prototype units in modified production cell phones as long as two years ago (when I left). Not sure if some of the cons have been fixed since then.
-nB

Re:awesome

[TheRaven64]

Because it's a very old question. People used to look at old glass, notice that it was thicker at the bottom, and assume that it had flowed. It turns out that, back when this glass was made, you made sheets of glass by spinning a blob until it became a disc. You then let this cool and cut it into the right shapes. The parts along the edge were thicker. Any glass that was installed with the thick bit at the top had more weight on weaker glass than sheets installed the other way up, and so cracked, broke, and was replaced. The only bits that survive are the ones where the glass was installed with the thick part at the bottom. More modern glass sheets are made by floating the molten glass on top of another liquid and so do not have this deformity and do not appear to flow.

Asking why old windows are thinner at the top is like asking why the sun moves across the sky to contradict the heliocentric view of the solar system, and so deserves a funny moderation.

And What About...

[Nemyst]

The new chips' lifetime? The impacts on overall computer heat? The energy required to use such memory? What is the expected RAM size to be available at first?

The article looks very scarce on details other than the technology itself which, honestly, doesn't say much about the final product at all.

Re:And What About...

[georgewilliamherbert]

Lifetime - significantly better than Flash, 3 plus orders of magnitude more write/erase cycles before there's degradation.

Impact on overall computer heat & energy required to use - lower read power than Flash, no maintenance power (DRAM requires rewrite cycles as the bits decay)

Expected size - Initial model is 0.5 GB (512 MB) per chip. That's on a much larger fab process than current CPUs or DRAM though - expect that to increase rapidly once demand is established.

Re:And What About...

[mrbene]

Yep. PCM natively supports per-bit writing without the need to re-write the rest of the block.

did you try reseting your memory?

[Anonymous Coward]

help, my computer's frozen! nothings responding!

did you try reseting your memory?

how do I do that?

a few minutes with a hair dryer should do the trick.

CDRW

[afidel]

This sound very similar to the phase change crystals in CDRW disks though obviously they are reading these electrically rather than optically since at 20nm you're well into the x-ray part of the spectrum.

Durability and Other Limitations

[mrbene]

PCM is interesting stuff. Here's some info:

• It's expected re-write lifetime is magnitudes larger than that of Flash.
• It'll be heat sensitive - weak crystal bonds will 'fail' if the module gets too hot. This also means interesting challenges around soldering.
• While Flash memory units become less usable the smaller they are (due to bleeding of info from weird electromagnetic interactions at very small scale), PCMs become more usable, as they require less energy to go through the "melt and refreeze" steps.
• It'll be in the manufacturers best interest to direct the heat directly at the PCMs themselves, since this is the most energy efficient solution. They're probably going to run cold.

Better be careful...

[tool462]

If these things run too hot, you'll literally have vaporware.

Obligatory Strong Bad

[TheSpoom]

And the Compy... just peed my carpet.

Re:hmmmm

[icebike]

Yes you are being paranoid. You already run a serious risk of losing all your data when you drop your Cell, so nothing changes here.

Other things will break first.

Does this mean...

[mdenham]

...that processors that support this type of memory will have to provide a Halt-and-Catch-Fire opcode?

Primary advantage, so far unmentioned...

[KonoWatakushi]

Perhaps the single most important advantage of PRAM has not even been mentioned yet. PRAM does not require the stupid block erase semantics of Flash--you can read or write as much or as little as you want, at whatever alignment, with no impact on performance. This also means that an SSD will be very simple, require no caches at all, and still have blazing fast write performance, even for synchronous writes.

PRAM will still require ECC algorithms, wear leveling, and bad block remapping, but on the spectrum of controller complexity, it is a lot closer to DRAM than Flash. (Incidentally, the same can be said of performance.) Reads and writes would still be buffered for queuing purposes, but this is very different from a cache; it is simply to allow requests to be pipelined from the storage controller.

Compared with the very simple constant time operations with PRAM, Flash is a dog. The controller must cache writes while it reads, erases, and otherwise shuffles blocks around. Moreover, as the controller operates with volatile memory, it must do this very slowly and carefully, or a power failure could severely corrupt the disk. (There are Flash SSDs with an onboard super capacitor to work around this, but they are obscenely expensive.)

Due to their inherent nature, even the best Flash SSDs have severely asymmetric read/write performance. The fact that only one company (Intel) has managed to produce a decent controller also betrays the immense complexity required to eek out even moderately acceptable random write performance. In my opinion, so called "SSDs" made with Flash don't even deserve that moniker, as they are more like a fast hard disk. (They still have a sort of geometry which constrains performance, and aren't anywhere near as fast as DRAM.)

PRAM will fix that, offering performance similar to a DRAM SSD. There are many companies banking on Phase-change RAM to displace Flash memory, Intel included. The wikipedia page has a lot more info, but basically, PRAM is superior to Flash in every way, except that the data on a prewritten chip won't survive a trip through the wave soldering machine.