IBM develops 'instantaneous' memory, 100x faster than flash | Hacker News Reader

IBM develops 'instantaneous' memory, 100x faster than flash

engadget.com

265 pointsalvivar34 comments

Whenever I see announcements like this, I try and cap my enthusiasm -- I think reading about solid-state-drives in the 90s and not getting them until... well the last 2 years, I have learned to be cautious with falling in love.

THAT being said, this article suggests mass-production on a timeline of 5-years (give or take) along side other innovations like Intel's 50Gbps Thunderbolt.next() [1]

I'm really gunning for the world not ending Dec 2012 now; I want to see this stuff in a desktop PC ;)

Aside: It is fun to think about what changes to current industries would occur when computing power becomes insignificant -- for example, video editing/production, video games, voice recognition, security, etc.

Just in our industry, with the cost of a GB dropping to damn near $0.00, the first thing we saw was an explosion of apps dealing with huge data sets -- something previously only done by a few select mega-corps.

Anyone, just speculation and fun at this point.

[1] http://www.engadget.com/2011/04/29/intel-touts-50gbps-interc...

Actually, commodity hardware that you can buy today for less than $1000 would have made it to the top tiers of HPC Top500 just 11 or 12 years ago. But most of this was done in a step by step evolution of old technologies, not in a revolutionary way, because most of the revolutionary technologies that make the news are not really feasible for mass production (at least at the time of the discovery - remember the costs of the solid-state drives back in the '90s).

What sort of apps dealing with huge datasets?

Here's a nice intro to "Big Data": http://www.forbes.com/2010/11/02/startup-facebook-twitter-te...

There isn't a business case for getting new technologies in mass production so quickly.

Businesses price to maximise profits, which isn't necessarily anything to do with production cost. It's more profitable to sell at a high price to early adopters first, and then gradually bring the price down.

If a technology is truly revolutionary, then there is most certainly a business case for putting it into mass production quickly. If you can disrupt an old industry or technology and replace it en masse with your new product, you will achieve market dominance in the new market. On the other hand, if you phase your product in slowly from the high end, your competitors will have time to copy you and grow their market share apace with yours.

> I think reading about solid-state-drives in the 90s and not getting them until... well the last 2 years

Lower your expectations, or look in different places. I've had SSDs for at least six years. My first ones were USB flash drives (often kickstarted from floppy, many mobos did not boot usb), industrial flash modules (little stubby thing that plugs right into the IDE socket) and CF-IDE adapters.

Even six years ago the industrial drives were relatively easy to get - they just were just not something you could find at consumer tech shops.

> reading about solid-state-drives in the 90s

Or before: http://en.wikipedia.org/wiki/Bubble_memory

The infant mortality on apparently neat technology is horrific.

From the wikipedia article on this class of memory (http://en.wikipedia.org/wiki/Phase-change_memory), it seems like this technology has been around since the 1950s-1960s. That's the perspective that you should view the IBM announcement though. It seems like the IBM technology might be the final step towards mass adoption.

Faster than flash, much higher write-cycle lifetime than even SLC flash, if these hold up in mass production it'll be like christmas for everyone.

Even if they are expensive, everyone will want these. I know I will be first in line. If these are reliable, and their drift friendly modulation works, we probably won't see spinning disks in 10 years. Thats a great future.

I disagree. If they cost 4x as much as flash, then nobody will use them unless the performance improvement is huge (Remember, it's 100x write latency not throughput, and caching fixes latency problems quite nicely).

To differentiate it from RAM, the word is PRAM.

More: https://secure.wikimedia.org/wikipedia/en/wiki/Phase_change_...

Looks like this is the only pram you can buy right now: http://www.google.com/search?q=pram&tbm=shop

It is easy to buy. I think I've some sitting around unused just because it is so neat.

http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&...

damn, pram.com is already registered

Are there ANY 4-letter domains available anymore?

If there are, there certainly aren't any which are widely-used English words like "pram".

Hope it doesn't have to be regularly zapped. #oldmacuserjoke

It will certainly be interesting to see what happens with technology like this when it gets combined with others such as graphene, memristors, etc.

I agree with rkalla that tempering one's hopes towards it is probably the way to go, but it's always nice to see someone pushing the envelope.

I would have preferred a straight-bat presentation, rather than the strained hangover jokes.

There's a big button with the words Press and Release on it.

Maybe, but the journalist's job is to take the press release and make it better, not worse.

If you ask me, the most interesting question is "So how does this thing actually work?" And the answer according to engadget is "PCM is based on a special alloy that can be nudged into different physical states, or phases, by controlled bursts of electricity", or in other words "Fuck, I don't know, I didn't really understand the press release"

The press release is more informative. You put some undisclosed material between two electrodes. The material can be crystalline or amorphous, and you can tell which it is at any given time by the resistance. To change it from one to the other you simply heat it up with a voltage pulse to one threshold temperature or another. Neato!

Wasn't IBM also responsible for bringing the Giant Magneto-resistive (GMR) technology to the spinning disk drives, thereby increasing their capacity many-fold? It looks like they'll do the same to flash with this PCM memory.

Keeping my fingers crossed.

PS: Did anyone else notice the gratuitous mention of "cloud computing" in the press release? :-D

incidentally, does anyone know how close memristor cells are to mass production?

I saw a talk last week that said 3 years. Supposedly HP is in talks with Hynix currently.

Is there video of the talk available?

They were supposed to be here in 2009 but right now HP is saying 2013. They keep pushing it back so who knows.

Aren't the wear-out numbers for flash understated? I seem to remember that the flash I've worked with is in the tens to hundred thousand cycles already but maybe that's not what's in consumer products?

Also would be interesting hear about power consumption for read and write.

Flash used to last much longer. But flash these days is used in MP3 players etc. I would imagine the average number of writes is something like .... 1. Because when my boys fill up a gadget, they buy a bigger gadget. Like most folks.

Even an SSD drive used as a system volume gets written maybe 1 time. SUre some blocks get thrashed, but wear-leveling means they write different blocks. So an mtbf of a dozen would probably serve.

If you only use it for system volumes it will still get written to multiple times with os upgrades, security fixes, and atime modifications. If you don't pull out the page/swap file/hibernate file you are going to be writing those often as well.

Now the bottleneck will be the CPU in a huge way. Intel we need a few orders of magnitude faster processors now!

Nope, northbridge/southbridge will be the bottleneck. They currently run at half the speed of RAM, 400-700mhz. Cycle speed at least, throughput is good.

It's time to buy some IBM shares

Reply on news.ycombinator.com