New Manufacturing Technology Enables Vertical 3D Transistors 75
MrSeb writes "Applied Materials has taken the wraps off a new etching system meant to turn vertically stacked, three-dimensional transistors from lab experiments into commercial reality. The new Centura Avatar solves multiple problems facing manufacturers who are interested in 3D NAND but find their current equipment not up to the task of actually building it. According to the folks at Applied Materials, trying to build 3D NAND structures in real life would be like trying to dig a one-kilometer-deep, three-kilometer-long trench with walls exactly three meters apart, through interleaved rock strata — and that's before we discuss gate trenches or the staircases. While this machine specifically targets 3D NAND today, a number of the challenges to scaling flash memory apply to scaling CPU logic as well. As for when 3D chips will be available for commercial purchase, Applied Materials was vague on that point, but personally I would expect to see companies adopting the new etch equipment in the next few years."
Re: (Score:2)
Only Two Questions: (Score:2)
1. When will this be in consumer product use?
2. How will this change the price per GB?
Re: (Score:3)
1. Random ballpark guess, I'd say 5 years, and it will probably be extremely expensive then (likely for enterprise and professionals who need extremely high density NAND).
2. It probably won't, not for a while anyways. It should allow much (much) higher density, but only after they start mass producing it. So, don't expect to be buying 1TB flash cards for $100 anytime soon. It will also almost certainly have major reliability issues for a long time to come, due to the difficulties with the process. Seriousl
Re: (Score:3, Insightful)
I would think this tech will come sooner, and while expensive, it should also increase performance while increasing density. Shorter traces = faster signals and less problems trying to coordinate synchronization between multiple paths since the difference between longest and shortest traces is reduced.
Re: (Score:2)
Re:Only Two Questions: (Score:4, Interesting)
Five years are barely enough to get a small modification of a process from a research fab to a real one, if it works flawlessly. A couple more years are typical for technologies that don't work flawlessly at the first try. This process needs an antire new fab, with much more layers than normaly available, and their special etching tech. I wouldn't expect it to get mainstream soon.
Re: (Score:2)
Friedman Units (Score:2)
The next six months will be crucial to the development of this technology.
Re: (Score:1)
Re:wait wut? (Score:5, Informative)
This technology bears about the same resemblance to what Intel is doing as anaglyph 3D does to a hologram. Intel basically just stuck a 3rd gate on top of two others, stacking normal planar transistors. It's "3D" in the technical sense, but only barely. This new(ish) technology takes essentially a single block and molds it into arbitrarily many levels of transistors, so you can have a stack of dozens or hundreds deep. Much more difficult, and potentially far more rewarding.
Re:wait wut? (Score:5, Insightful)
Re: (Score:3)
Re:wait wut? (Score:4, Informative)
Yeah, I'm not sure what they plan on doing about that. Seems like it could be a major issue. They might not even have a solution to that yet, although TFA seems to be thinking this is going to be used for Flash memory, rather than CPU transistors, which makes heat considerably less of an issue.
NAND memory, and memory in general, is generally first in adopting new process technologies. It's far easier to make (relatively simple) memory circuits (generally consisting of a tiny number of transistors) than it is to make (relatively complex) logic circuits (consisting of orders of magnitude more transistors).
Re: (Score:2)
Yeah, I'm not sure what they plan on doing about that. Seems like it could be a major issue. They might not even have a solution to that yet, although TFA seems to be thinking this is going to be used for Flash memory, rather than CPU transistors, which makes heat considerably less of an issue.
http://www.gizmag.com/ibm-supermuc-supercomputer/23086/ [gizmag.com]
Any other questions?
-AI
Re:wait wut? (Score:5, Informative)
They aren't talking about non planar FETs ... their 3D NAND requires a far higher anisotropy (the whole 1 km deep 3 m wide bit).
oh really? (Score:5, Funny)
"would be like trying to dig a one-kilometer-deep, three-kilometer-long trench with walls exactly three meters apart, through interleaved rock strata..."
No problem dudes, let me fire up Minecraft and I'll show you how its done!
Re: (Score:2)
Re: (Score:2)
Step 1: Manufacture a diamond-tipped rotary saw blade three meters thick and approximately one kilometer in radius.
Step 2: Purchase a used generator from a hydroelectric plant and wire it up as a motor.
Step 3: Purchase all the power from an active hydroelectric plant.
Step 4: Throw the switch and run like hell.
Re: (Score:2)
probably not fast to market (Score:3)
I foresee this going at about the pace that perpendicular recording did with hard drives. Remember how we heard about this whiz-bang great new idea years ago, and look how long it took to actually come to the practical market.
But now it's ubiquitous. So I suppose the same will happen with the chip. And I can see this dealing a crushing blow to the already hurting spinning disc hard drives with them being able to vastly increase flash storage density.
Re: (Score:2)
Re: (Score:3)
Hard drives eventually fail as well. The problem is that current flash memory tends to fail an order of magnitude faster than disks (that's just a rough figure, depends heavily on the specifics of the particular technology being used).
But this is purely a reliability issue with relatively new technology. They're still figuring out all the oddities, and also (I believe) making a slight trade-off between reliability and storage density.
Eventually, the failure rates will decrease to the point that flash memory
Re: (Score:2)
I have yet to see a satisfactory analysis of this. Everything I read in the product literature indicates that they have more than enough lifetime. The engineering articles indicate that even if they were to fail prematurely, it would result in a read-only disk. Yet every real-world report on these drives is that the entire contents of the drive suddenly vanishes, or the drive cannot be recognized at all.
Either the users are exaggerating the error reports, or lying; or the engineers and product literature
Re: (Score:2)
Perhaps the actual storage chips have that much of a lifetime, but the supporting architecture may not?
Re: (Score:3)
The SSD's that are failing rapidly are not doing so because the flash has reached its erase limits. They are doing so because something else is breaking, and this is immediately obvious when several different companies produce SSD's using flash chips from the same source (in particular, Intel or Sandisk) but have drastically different first-year
Re: (Score:2)
Re: (Score:2)
But try booting windows off a read-only drive that randomly stopped allowing writes the last time the OS was running. You aren't going to get very far, and I think the majority of people would consider that a dead drive.
I'd consider that a poorly designed OS. What is it that absolutely positively has to be written to the hard drive as part of the boot process?
Re: (Score:2)
I seem to recall booting a computer with no hard drive using a Knoppix disk.
Re: (Score:2)
Did you honestly think the actual claim was that you could go on using the unwritable drive as if nothing happened? Thats essentially the straw man you seem to be fighting.
Re:probably not fast to market (Score:4, Informative)
You have to understand that inside one of these "drives" is really a computer with a CPU running embedded software, I/O controllers talking to the SATA host bus, large RAM for buffering data, and a lower level flash interface to manipulate the actual flash storage. The device is then implementing a kind of filesystem on top of the raw flash, keeping track of free/erased flash blocks, wear levels, and the logical mapping to SATA block addresses.
The entire contents disappearing is not due to flash memory losing all its state at once. It is due to buggy firmware on these "drives" crapping itself and corrupting its own filesystem metadata. This is likely due to it not doing safe journaling, e.g. it performs unsafe flash write sequences that leave flash in an unexpected state if they are interrupted due to power loss or firmware bugs/resets. This is a bit like our old filesystems before they did journaled metadata for crash recovery.
I wish manufacturers would expose the flash storage and allow the OS to manage this layer, but they're too busy profiting from the illusion of firmware as just another part of the hardware. Everything seems to be trending towards more elaborate embedded software that we cannot review or maintain separately from the devices containing it, forcing artificial lifetime limits on the products.
Re: (Score:2)
I wish manufacturers would expose the flash storage and allow the OS to manage this layer, but they're too busy profiting from the illusion of firmware as just another part of the hardware.
They do...such devices are called MTD or memory technology devices. Basically bare NAND/NOR flash exposed to the OS. There are special filesystems designed for working with it, like jffs2.
They're a pain in the butt to deal with, which is why hiding the details behind an abstraction layer and treating it like a disk is appealing.
Re: (Score:2)
It makes a handy table for other things which is why I still have the huge thing (and it will probably work for single tapes even though the autoload
Re: (Score:1)
Puppy Linux refreshes the persistent save file on a USB stick about every half hour, so you're not doing continuous write cycles, but other distros can't do this. Puppy is small enough to run entirely in RAM, even 128 MB.
Re: (Score:2)
That's a software workaround to a hardware problem. I'm talking about SSDs actually lasting as long as a hard drive under the same usage.
Re: (Score:2)
Imagine a typical 500GB drive in a typical PC. Assuming 1000 write cycles and 10 years of life, that is somewhere around 130GB written per day. How many people do you know who write 130GB to their drive every day, and at the same time keep their drives for 10 years?
The write limit is only a problem for small drives, and the only reason that drives are small is that they are expensive right now.
Re: (Score:2)
Imagine a typical 500GB drive in a typical PC. Assuming 1000 write cycles and 10 years of life, that is somewhere around 130GB written per day. How many people do you know who write 130GB to their drive every day, and at the same time keep their drives for 10 years?
A system that average that much write volume per day isn't typically using commodity hard drives. Its using enterprise drives so that also changes the comparison quite a bit.
On that note, you often see anti-SSD slashdotters harping on "$0.05 per gigabyte" or some such, a price only seen in commodity drives. The shittiest enterprise SAS drives start at $0.15 per gigabyte, while prices higher than $1.00 per gigabyte (more than the current generation commodity SSD's) are actually considered normal.
Re: (Score:2)
I've only got SAS drives for the arrays where it makes a difference. In many cases application speed is more of a bottleneck or more than adequate speed is supplied by a cheaper array. If only one person is hitting an array at a time getting a few hundred megs of data and then going you don't need gold plated gear to s
Re: (Score:2)
Swap generally does not actually get written all that much. If it did, you would notice that performance sucked (swap to SSD is still dead slow compared to RAM). It is most likely mostly full of stuff which hardly ever gets used.
sar -d on my own laptop shows that I write 90kB/s on average when the laptop is on. Assuming it is active 24/7, that is a measly 8GB/day. A look through older logs shows that some days I manage to reach 170kB/s, or 15GB/day worst case. Try sar and see for yourself.
Obviously it would
Re: (Score:1)
Wrong units. Please correct the summary. (Score:3)
dig a one-kilometer-deep, three-kilometer-long trench with walls exactly three meters apart
Please avoid these strange and esoteric units and use units that are familiar to us. The approved units are football fields for lengths, Olympic size swimming pools for volume and libraries of congress for data volume, Rhode Island or Delaware for area.
Re: (Score:2)
Texas and Alaska are also acceptable units.
Re: (Score:2)
Re: (Score:2, Funny)
In the UK, we use Wales.
Especially when describing the size of natural disasters for some reason
Re: (Score:2)
Wishful thinking?
Re: (Score:2)
Oddly enough I don't even need a calculator to help you with that.
Two-thirds of a mile deep, two miles long, walls 10 feet apart.
Metric isn't strange and esoteric, it's all closely related. 1 Ampere is the current needed to generate 1 Newton of force between two 1 meter long sections of wire, 1 meter apart in a perfect vacuum.
Now try that with pounds-force, feet and Amperes. (0.2248 lbf, 3' 3.37" , 1A) (Yes, I needed a calculator for that..)
Re: (Score:3)
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
It looks like there are almost exactly 11 football field lengths in a kilometer. So that's an 11-football-field-length-deep, 33-football-field-length-long trench, with walls separated by roughly the average length of a rushing play. (Or 1/3 of a first down.)
Pleonasm at twelve o'clock.... (Score:4, Insightful)
There's two semantic goofs in this submission, one in the title and the other in the first sentence, and neither was noticed or corrected by Soulskill. The phrase "vertical 3D transistors" is misleading, since a literal interpretation doesn't describe z-axis stacking and instead describes objects whose most significant dimension is oriented vertically; it would be more accurate to write "stacked 3D transistors". In the first sentence, the adjective phrase "vertically stacked" is certainly a pleonasm if there ever was one; the definition of "stacked" already describes a z-axis or "vertical" state. The use of the word "vertical" in both of those instances is ineffective semantics.
Re: (Score:2)
>pleonasm
"Gee, Porgie, I didn't know you masturbated!" - Mudhead
--
BMO - Defoliating a victory garden sure works up an appetite.
Re: (Score:3)
Wow, you really had to work hard to come up with a grammar Nazi response.
That is actually rather impressive. You get an A+
Re: (Score:2)
The observation was easy, since I wasn't looking for a pleonasm... it leapt off the screen and attacked my eyes before I could look away. The hard work, as always for me, was organizing what needed to be said and how best to say it. Even though there's a novel's worth of material upstairs, I could never write a novel because it would take me decades to complete a first draft and I'd get distracted by some other shiny and wander off long before I finished. I don't think there's enough methylphenidate on t
Re: (Score:2)
It's all about the quota and quantity, baby.
Re: (Score:2)
[Martin Short voice:] Your funny perversion is more lyrical, I must say!
"...the next few years" (Score:1)
Anywhere between 3 and 100... nice vague PR job there
Not the first... (Score:4, Informative)
I don't believe this is actually the *first* fab process using vertical structures (having actually RTFA). I worked at Texas Instruments in the mid-80's and most of the ALS (Advanced Low power Schottky) devices were of vertical well construction (as opposed to planar process or lateral junction bipolar construction). Looks like the sizes are a lot smaller, and the ratio of depth to width is a lot higher (a lot more junctions stacked in one well).
they probably won't even need investors (Score:2)
These are for stacked memory cells (Score:2)
Re: (Score:2)