Intel & Micron Show 34-nm, 32-Gbit Flash Memory Chip

Lucas123 writes "IM Flash Technologies, a joint venture between Intel and Micron, announced it has developed a 32-gigabit NAND flash memory chip that is expected to enable the production of cheaper solid-state drives with twice the storage capacity of today's products. The 34-nanometer, multi-level chip is smaller than Intel's latest CPUs. Samples will be available in June with production by the end of the year."

Smaller is bigger

[tepples]

Why do chipmakers put emphasis on ever-smaller processes with the notion that smaller == better?

Smaller is better because smaller is bigger. If you can make the flash chips smaller, you can put more chips in a module and stripe more data across them, giving a bigger capacity per module.

Re:

[Anonymous Coward]

smaller is better because smaller is cheaper (higher yield from wafer) and lower power.

Re:Phirst Spot

[tirerim]

Because smaller is more energy efficient, which is useful on a number of levels: for one, it saves electricity, and it also means that the chips produce less heat, which lets them run better.

Re:

[SmlFreshwaterBuffalo]

This is already common practice in the industry when you can't get your yield up.

There are two methods that I've seen:

• Redundant cells - extra cells in the array that can be swapped in to replace bad cells
• Error-correction - extra cells are used for each word (for example, 12 bits total for every byte), and an algorithm is implemented during the read cycle to determine the correct values

The key is determining how much to add. Having too few won't allow you to hit your yield targets, and adding too many is a

Re:Phirst Spot

[Sentry21]

More importantly, smaller allows it to fit into smaller devices, meaning larger-capacity USB drives, cellphones, and iPods.

Re:

[mako1138]

Smaller size does not necessarily imply greater energy efficiency. When Intel debuted Prescott at 90nm, it was infamously toasty.

At the 65nm node and below, gate leakage is a big concern, and it can increase power consumption beyond the savings gained from reducing the operating voltage.

The Price of Flash

[Jah-Wren Ryel]

FWIW, another article covering this same press-release noted that most flash costs $2.50/Mbit to manufacture, but this new stuff by Intel costs just under $1/Mbit to manufacture. So the rapid downward spiral of flash storage pricing should continue for at least the short term.

Re:The Price of Flash

[j1m+5n0w]

That doesn't sound right. Perhaps you mean $/Gbyte?

Re:The Price of Flash

[Kjella]

Probably this [eweek.com]:

(...) Jim Handy, an analyst with Objective Analysis (...) "At a die size [they are using], the price of a 32Gb chip will be just shy of $4, which works out to about 99 cents/GB. The companies will be the first to break the $1/GB barrier with this product," Handy said. Today's NAND prices are hovering near $2.50 per gigabyte, Handy said.

So an external analyst said thar, nor Intel/Micron. It sounds rather nice if you can get a good boot disk for 1$/GB+margins though. The bulk multimedia will probably still go on HDDs though, but I'd definately get one at those prices.

Re:

[cnettel]

Well, for a few-GB drive the interface logic will dominate the cost. Compare how a single-platter, single-side disk doesn't really get that cheap.

Re:

[SporkLand]

I rhink rhere is somerhing wrong wirh your 'r' key.

Re:The Price of Flash

[John Whitley]

Those prices are insanely high. I'm seeing current retail prices for high speed compact flash cards at 60 to 100 USD for 8GByte (varies with speed, rebates, etc.). Taking one of the lower-end prices for a top-tier part (since I want an upper bound on part manufacturing costs), that works out to a bit less than 0.001 USD per Mbit. Even if those numbers are per Gbit, that still leaves the cited current manufacturing cost at more than twice actual retail.

Re:

[Finch42]

I think you need to re-visit the local elementary school. A little bit of 3rd grade math might do you some good. Current NAND manufacturing cost = a bit WAY less than 0.001 USD per Mbit. (~$0.0003/Mb,$0.0024/MB,$2.50/GB) 34-nm NAND manufacturing cost = ~$0.0001/Mb,$0.001/MB,$0.99/GB $100 8GB CF = $12.50/GB. Factoring in controller and assembly that's about 60% profit.(Yes I did research assembly costs) $900 64GB SSD = $14.06/GB. Again about 60% Profit (Controller and assembly for an SSD is much higher due

Re:

[John Whitley]

My math is just fine, you need to revisit your basic English, logic, and social skills. I was trying to debunk the absurd prices cited by the parent. My "punch line" was to observe that back-of-the-napkin calculations using conservative bounds yielded (as you observe and agree with) completely unrealistic results.

So your issue is... what?

Re:

[Finch42]

My issue is that I am sad, angry, and confused.
Sad and angry because my recently dis titled girlfriend is a lying cheating whore (but still incredible in bed) and genuinely confused as to how you came to the conclusion that cited current manufacturing cost is more than twice actual retail. While I remain completely and entirely baffled by this conclusion I do wish to apologize for unjustly insulting you without provocation.

P.S.
<pointless_ranting>
The reasoning behind my seemingly incoherent
message was

Re:

[Finch42]

Excerpt from my previously miss-formatted post...
(Slashdot really needs an "Edit Post" button)

Current NAND manufacturing cost = a bit WAY less than 0.001 USD per Mbit. (~$0.0003/Mb,$0.0024/MB,$2.50/GB)
34-nm NAND manufacturing cost = ~$0.0001/Mb,$0.001/MB,$0.99/GB

$100 8GB CF = $12.50/GB. Factoring in controller and assembly that's about 60% profit.(Yes I did research assembly costs)
$900 64GB SSD = $14.06/GB. Again about 60% Profit (Controller and assembly for an SSD is much higher due to low demand)
So a 64GB SSD based on the 34-nm process would cost about $420, dropping our cost to $6.56/GB.

Sorry to burst your bubble but the only thing out there that has a manufacturing cost that is double it's retail price is the PS3.

Re:The Price of Flash

[matt21811]

Anyone can check the spot price for flash anytime by looking at this site:

http://www.dramexchange.com/ [dramexchange.com]

Scroll down to the flash section.
SLC is the good stuff used in the big fast SSD's you get from people like Apple.
MLC is the slower, less long lasting, stuff commonly used in thumb drives.
$2.08 for a Gigabyte in MLC
$6.70 for a Gigabyte of SLC

If you want to know the long term price improvement rate for flash, you can join that site for $1000 a year or if you want the cheap version, I've been tracking retail flash (MLC) prices for 9 years at my site here:
http://www.mattscomputertrends.com/flashmemory.html [mattscomputertrends.com]

Re:

[Anonymous Coward]

Do we also get an increase in R/W speeds as the chip shrinks.

If so how much? Potential speed increase over magnetic disk could accelerate the payback for IO bound database applications.

Currently magnetic disks have such effective caches that they are able to nullify a lot of the speed advantages of SSD. In fact some benchmarks I've read still show magnetic disks leading speedwise in realistic application benchmarks.

But this could also be partially due to new unoptimized drivers for such devices.

Re:

[Courageous]

All the high speed flash devices coming out now are essentially 'sort of' native RAID solutions. They feature multiple lanes, akin to RAID-0 striping. That's how the FusionIO product gets it's 600MB/s write speed. It has 160 lanes.

All the 'latest' SSD's are dominating the hard drive now. FusionIO's special case, utterly devestates the hard drive.

C//

Re:

[afidel]

Uh, the 120K IOPS is great (about what I can get per shelf from my SAN vendor with their current offering) but the estimated cost of $30/GB [blogspot.com] is insane! Not to mention I don't see any management software for running large numbers of them so they are in no way a replacement for a real SAN.

Re:

[Courageous]

I've met with these folks personally. The upcharge for the card is, essentially, the custom ASICs that do the 'RAID'. With a little volume, the pricing will be much more in line with other flash.

Anyway, no, they're not a replacement for a SAN. That's just marketing hyperbole.

You'll find them in your SAN soon enough, though. Their real target market is things like embedded devices inside SAN/NAS controllers. Think "journal device" or MRU unit for SAN blocks.

They would also make for a great swap file device.

Re:

[afidel]

I wonder if IBM's chipkill chips are flexible enough to do the RAID for flash, those are already in large volume production and can do full bandwidth RAID for RAM so they could easily keep up with flash. That would probably lower the cost significantly.

Re:

[Courageous]

The fusion IO system stripes 160 wide, the way I understand it...

C//

Re:

[default luser]

Do we also get an increase in R/W speeds as the chip shrinks.

If so how much? Potential speed increase over magnetic disk could accelerate the payback for IO bound database applications.

Sure they do. The performance of flash is constricted by how long it takes to charge the cell.

Q=CV => Charge = Capacitance * Voltage

Translation: a higher voltage charges a given capacitance more quickly. A lower capacitance means you reduce the overall charge needed to get the cell into the "written" state you want. Thu

Size matters

[russoc4]

"You have to give Intel and Micron a lot of credit; going to 34 nanometers is pretty big move."

No pun intended, I'm sure.

Re:

[niceone]

It's one small stepping for mankind.

Costs

[TubeSteak]

http://sst.pennnet.com/display_article/329434/5/ARTCL/none/UPFRN/1/IDM-economics-at-32nm-and-beyond/ [pennnet.com]

As the industry moves to 32nm and beyond, the sharply escalating costs of both IC product development and fab equipment may combine to slow down the historic chip cost reduction trendline.

The total cost to develop a chip product -- including all EDA functions as well as maskmaking -- has been nearly doubling each node from 90nm to 65nm to 45nm. Moving on to 32nm is projected to raise costs only ~50% over 45nm, but the absolute numbers are now making design-teams pause to consider their choice of manufacturing node. Kinugawa predicted that neither Japanese fabless nor customers nor IDM-internal designers are prepared to jump to the next node -- such that a "several year gap" will appear between the availability of 32nm node fab capacity and substantial demand!

They're kinda jumping the gun with 32nm.
New tech is expected every 2-3 years.
32nm was expected for 2009-2010 and 22nm is expected in 2011-2012

Memory is almost always ahead.

[Anonymous Coward]

Memory is almost always ahead of the curve when it comes to silicon manufacturing. These were the guys who were at 55nm when the processor industry was temporarily stuck at 90nm.

All kinds of memory can use smaller processes because the logic is much simpler; you're basically laying the same thing out over and over and over again on a die. For the same exact reason, most companies use SRAMs to test their processes before moving up to higher level logic like processors.

Re:

[Enderandrew]

Color me stupid, but if AMD is struggling with manufacturing processes to get to smaller sizes, shouldn't they purchase a memory company that can fab at that size, and use said technology? I thought a decent chunk of their budget went into researching and developing fabrication processes at smaller sizes. By purchasing a memory company, couldn't they just pay for the research once and use it in memory, CPUs, and GPUs?

Re:

[Firehed]

That works great the first time, and might be a viable catch-up strategy. What happens for the next die shrink?

Re:

[Anonymous Coward]

Unfortunately most memory companies are worth a lot more than AMD is, and have their fabs booked solid for years in advance, so it's not likely they would cancel orders to be acquired by some company like AMD (which barely had the money to buy the fabless ATi, and is still considered to be a bad move by many investors).

Secondly, the technology is considerably different. Because RAM is so simple to build, there are a lot more corners you are able to cut. With some RAMs you can even get away with selling non-

Re:

[i.of.the.storm]

I guess you didn't know, but AMD used to have a large flash memory division that was spun off to form the company called Spansion since their main cash cow is processors. http://en.wikipedia.org/wiki/AMD#Corporate_history [wikipedia.org]

Re:

[rahvin112]

AMD and Intel have both used flash memory as their process "proving" ground. Intel produces tons of flash, but if you look at their profit statements they make very little money on it. Why? Because they use it as the proving ground to bring new process technology to CPU's. AMD's spinning off of their memory division wasn't a good decision IMO. It's going to show up (if you argue it hasn't already) in their process transition. Where before they were 1 year behind Intel on Process technology IMO they will slo

Re:

[i.of.the.storm]

Hmm, that's an interesting point. It does seem like AMD is falling further and further behind in process technology. But I think their partnerships with IBM and stuff might help them improve, although I don't really know anything about their relationship with IBM.

Re:

[Kjella]

Memory cells are much simpler to lay out with the same over and over again and much more resistant to flaws in the manufacfuring process by switching off bad cells. Intel says it's on track for doing 32nm processors in 2009 which I guess means RAM/flash early in 2009. Still, yes this is technology 6mo+ away I think. Maybe they surprise us, new generations of flash seems to have been coming all over the place. Or maybe smarter, cheaper controllers or whatever, I see a lot of room for improvement elsewhere.

Re:Costs

[Iron Condor]

As the industry moves to 32nm and beyond, the sharply escalating costs of both IC product development and fab equipment may combine to slow down the historic chip cost reduction trendline.

Celebration! 2008 marks the 25th anniversary of this claim being made at a die-shrink -- or at least 1983 is the first time I heard it. People were talking about the 1u (1000nm) "holy grail" and how it would likely never happen because it wouldn't be worth the cost...

Re:

[zippthorne]

True, but at some point, your lithography process has to use gamma radiation, and you end up with a significant risk of turning your mild mannered employees into angry, green-skinned clones of Lou Ferrigno.

that is a small chip

[FudRucker]

they should have included a photo of it on someone's finger tip...

Reliability

[Anonymous Coward]

I've had lots of flash memory go bad lately, including a brand name 1Gb SD card.

What about the reliability? It's great that they can manufacture on a 32nm line, but given that this is a new process, what reliability testing has it undergone?

I'll go with the proven technology, thank you very much, especially for something where the 'smaller is not necessarilly better' physics side of things kick in.

Re:Reliability

[Darkness404]

HDs though, go bad quickly. For me not a single flash chip has ever catastrophically failed like many HDs I have had. Also, do you really want to carry a hard drive in your cell phone? And flash chips are much, much, faster then most hard drives. And really, HD speeds are one of the biggest bottlenecks in high-speed computers, RAM is cheap enough to get 1 GB for less then $50, CPUs are multi-core, Linux has a fast and usable OS, USB is fast enough for most devices, so all we need is faster HD read/writes and we have a much faster computer, problem is, the way to speed up a HD is only via either A) RAID 0 which costs reliability or B) increase RPMs which add price and chances are, decrease reliability over time. So as of now, the only way to get large amounts of space, without spending a fortune and having it be reliable (no moving parts) and fast is with Flash chips and SSDs.

Re:

[Gavagai80]

And for me, I've never had a hard drive fail in my life (16 years of hard drives, sometimes several at once, and with a CPU dying in that time). Personal experience doesn't mean anything.

Re:

[PitaBred]

True. But failure modes for hard drives are more likely to be catastrophic than they are for flash. A head crash on a hard drive has no analogue with flash unless you count, say, hitting a flash drive with a hammer.

Re:

[toddestan]

When I've had flash drives fail, it's almost always been catastrophic - I can't read anything off the flash at all. Furthermore, it's always been sudden - it simply stops working. My harddisk failures generally are not - the drive gives warnings (becomes loud, starts clicking, slow spin up) before it fails completely. For the flash, I think it's almost always been the controller chip that fails and not the actual flash memory, but unless you are a soldering wizard, it might as well be the same thing.

Re:

[cheater512]

Correct me if I'm wrong but arent hard drives faster than flash when it comes to raw speed?
Flash obviously has better seek of course.

Oh and as a guy with a 4 drive RAID 5 array which can hit 200mb/s, hard drive speeds are not a big bottleneck.

And hard drives will always beat flash when it comes to raw data storage.
My raid array gives me a terrabyte of usable space with redundancy for a few hundred bucks (back 2 years).
Flash today would cost thousands for the same amount of storage.

Re:

[tirerim]

Not sure what you mean by raw speed. Currently, iIrc, hard drives still beat flash for sequential writes, but that may not last. Given the way flash prices have been plummeting and sizes have been increasing, hard drives' advantage in total storage space, even at the top end, will only last a few more years, too.

Re:Reliability

[quanticle]

Sure, you're getting 200Mb/s, but you need four drives to do it. A SSD can give you the same performance and reliability as a RAID array in a single drive. Sure, right now, that single drive will cost as much as your entire array, but that situation will improve as manufacturing volumes increase and prices come down.

Re:

[Glonoinha]

Yea - when I can get 750MB/s sustained throughput to / from hard drives, count me in. Until then, I'm saving my pennies for a rig like Nostromo5 over on Overclockers. Six SSDs (Gigabyte iRAM) running RAID 0, using a second computer (actually just a powered motherboard) to host and power them. That's three Gigabytes of data every four seconds, sustained. Best part - he did it almost a year ago - not sure what's come of it since then.

Details here [ocforums.com]

Re:

[afidel]

No, a 4 drive array that will do 200MB/s can be had for a couple grand and will store at least a TB. The equivalent in an SSD will cost you tens of thousands and will probably take up nearly as much space. SSD's are great for niche applications like log files for a high transaction DB or for a laptop where the idle power savings are significant, but they have a long way to go before they eliminate spinning disks.

Idea

[Anonymous Coward]

It just occurred to me while reading this that we can come up with a brand new metric for storage capacity.

When speaking with most non-tech people I know their eyes glaze over whenver I mention megabytes, gigabytes etc.

But now I think this can be solved with MP3, Kilo-MP3, Giga-MP3 etc metrics.

This:

me: "I have 64GB of RAM in my PC"
listener :

Becomes this:

me: "I have 2K-MP3 of RAM in my PC"
listener : wow!

Re:

[i.of.the.storm]

That's one chip though, and stuff like USB drives and SSDs have multiple chips, not just one. Similarly, if you look at a stick of RAM there are several RAM chips on the stick.

Typo ?

[Noroimusha]

hmm i think they meant 32 gigabyte cause 32 gigabit is only 4 gigabyte if i remember right or are they talking about transfer speed ? or i am missing something ?

Re:

[KillerBob]

I'm going to assume that you actually missed it, and not that it was a facetious remark... if it was, then I apologize.

What you're missing is that this is a single chip. One chip with 32gbit is a 4gbyte single chip. Couple 4 of these on a single thumb drive (and they're small enough to do it), and you've got a 16gbyte USB thumb drive. And it only cost them $16 to build. Well... $20, considering packaging and control chips etc.

Now contrast that against the current cost of a 16GB flash drive. 16GB thumb drive

Re:Typo ?

[Millenniumman]

. 16GB thumb drives don't exist...

Yes they do. [newegg.com] They start at like $50.

You can also get 32 GB ones [newegg.com].

and 16GB SSDs cost over $200

That is true, for the most part. I would imagine they are made to a higher quality than thumb drives, or they are just overpriced. Probably both. Certainly, they are ridiculously priced compared to hard drives.

Re:

[LarsG]

16GB thumb drives don't exist... the biggest I've seen is 4GB...

Where you live, middle of Amazon rainforest? 16GB thumbdrives been round pretty long time. 32GB is the largest that is currently available in the retail channel. Even eensy teensy size of your fingernail microSDHC cards reached 8GB a while ago.

As for SSD having higher cost; more complicated interface logic (pci, sata/pata, etc), they tend to use SLC instead of MLC flash, they tend to be in a parallel or interleaved organization, some have proper on-board logic for wear leveling (compared to the "smartmedia

Re:

[v1]

16GB thumb drives don't exist

Actually ADATA sells the MyFlash 16 [amazon.com]. It's a "double long" flash drive compared to most, and quality is very poor. (I've owned two, one was defective, and the other was unreliable, but WAS 16gb)

My current drive is a SanDisk FireFly 8 [shopbot.com.au], for it's small size (sub-single length) and reliable operation. I would LOVE to replace it with a 16, and I've been waiting what I consider a very long time for this jump to occur. Here's hoping by christmas I can have a FireFly 16. It's my serv

A Bit Of Confusion With Flash Sizes

[rsmith-mac]

Since we're on the subject of flash memory, I find this article confuses me a tad bit. I was reading an article yesterday that mentioned that the current iPhone has room for a single flash chip, which means the current 16GB variety has a 128Gbit chip in it. But then TFA implies that 32Gb chips are as big as flash memory comes right now, which leaves me at an impasse. How does a device like the iPhone fit 128Gbit as a single chip if chips only come up to 32Gb in size?

Re:A Bit Of Confusion With Flash Sizes

[Frenchman113]

You can combine them. Or make larger chips. The achievement in TFA is significant because of the storage density achieved.

Re:A Bit Of Confusion With Flash Sizes

[SQL Error]

Multiple dies stacked in a single package. Very common in the flash business.

Looks like Intel is has big SSD plans

[rrohbeck]

Apparently an SSD will be part of Centrino 2: http://www.theinquirer.net/gb/inquirer/news/2008/05/23/intel-bundle-ssds-centrino [theinquirer.net]

vintage flash memory.

[nawcom]

Maybe I'm lame, but I still have an old 128MB thumb drive that has lasted this long. I wonder when it will reach vintage state? Yeah, I know. probably never.

Re:

[toddestan]

I have a 4MB Sony Memory stick. Still works, though pretty much useless.

always just over the next hill

[MilesNaismith]

Yeah I've been hearing SSD would replace spinning disks for a long time now. Still hasn't happened. Even early this year SanDisk was touting it's next products which I still do not see on NewEgg. When they've actually shipped a FAST 32-gig for under $100 get back to me.

Good news for portable operating environments

[1 a bee]

I've been booting from USB for the better part of a year now. (I'm on such a system as I type this.) It's a lot easier to manage an operating environment on a USB than one on an internal HD. Especially, if you're in the habit of switching machines frequently. The OS I use is FaunOS [faunos.com]. As the price of this kind of hardware drops, it's easier to buy into the vision of portable environments [faunos.com].

Portable environments have to be "live" systems. They present interesting, unique challenges, but as FaunOS and a