Slashdot Mirror
Samsung Develops 16Gb Flash Memory
nofrance writes "As promised earlier this year, Samsung has unveiled the world's first 16-gigabit flash memory chip. These chips, when combined in a 16x16 configurations, will allow 32 GigaByte flash cards. Using 50-nanometer manufacturing technology, these chips will be in production by the second half of 2006, with Samsung promising that their 32Gb team will impress next year." From the article: "According to the company, the cell size of the fingernail-sized flash chip has been reduced about 25 percent from that of the 60 nm 8 Gbit NAND: The new 50 nm flash memory contains cells that measure 0.00625 square microns per bit. The 16 Gbit device holds 16.4 billion functional transistors, Samsung said. "
Woah, that's a relief. I was afraid that I might be buying a device with billions of non-functional or even disfunctional transistors.
Just a note...
Flash is not perfect. It is typical for a small percentage of bits to be bad right off of the line. All of the devices contain error correction circuitry in order to compensate for bad bits. There are actually many more than 16.4 billion transistors on board. Many of them will be marked as bad, however.
More
Yes, but individual memory chips often do not come rated in bytes but in bits and are configured in parallel to complete the byte. Hence "16x16 config" making 32GB.
Camera for video/etc. An audio recorder..
/var/tmp was mounted in a ram drive] but you could get away with a binary only Gentoo/Debian install just fine.
At 2Mbit/sec [250KB/sec] ~34 hours of recording with no moving parts other than the shutter. Current video recorders don't last super long on batteries and reporters in the field have to lug them around [or have their camera crew do that].
If they could make it last a while [e.g. handle wear] you could use it as a laptop hard drive. I probably wouldn't run Gentoo on it [unless
And in most laptops the harddrive is the second most waste of energy anyways. So while making the CPU take less power you can make the storage take less too.
Granted I wouldn't use this where I was doing many re-writes or needed quick writes. But for a normal user [e.g. email, web, im, word processing] it's more than adequate.
Tom
Someday, I'll have a real sig.
Err, it is Gb. The individual chips are rated in Gigabits, and only the final 16 chip products in gigabytes.
Thus the 16 Gb chip is 2GB and when you have 16 of those you get, you guessed it, 32GB.
East Coast Brewers
Next, the 2GB has Toshiba Flash Memory Soldered to the board, whereas the 4GB has a daughterboard with 2x2GB Samsung chips. Therefore, it is possible that someone will reelase an upgrade to the 4GB Nano at some point in the future, but Apple may well have disabled support in the (closed) Nano sofware for flash support above 4GB in the current generation.
Get a free iPod Nano 4GB!
Problem is, flash cards don't work quite as well as regular hard drives because you cannot flash them with consistent information all the time. I am not sure about what these chips can handle nowadays, but a couple of years back, such memory could be flashed about 10,000 times and that's it.
Full Tilt
(a) Do a cost analysis. Even if they shrink the gatelength to 25 nm (which will not happen because FLASH memories WILL not work at 25 nm gate lenght, regular transistors will), you will be still be limited to say 100 GBit. Yield is another issue which will drive cost. Debugging such large memory arrays is NOT trivial.
(b) Reading mechanism for FLASH memories is different from Harddisks. Larger the memory arrays, slower it becomes. Make arrays smaller ? You will have lot of peripheral overhead which will drive your cost up. Why is peripheral hard to make ? Because peripherals are made in regular CMOS technology as compared to FLASH technology - integrating them together is a pin in the ass. This is one place which requires more improvement, the memory controller on the FLASH chips is still slow (even if access time from the individual cell is fast).
(c) Will 25 nm FLASH be any faster ? Not necessarily. The gate length scales, but interconnect capacitance doesn't. Smaller transistors will have smaller parasitic capacitance but they may not be necessrily able to drive the long bit/word lines. Solution : Make individual cells bigger. What do you lose ? Your memory becomes bigger.
In short there is a reason why magnetic HDD will stay. Yes there are applications where 10-20GB is enough, but not everywhere. That is why digital MP3s are swept by FLASH based drives. And don't forget that FLASH drives have rated endurance of 100,000 write/erase. Do you want such a thing for your laptop ? probably not.
Can you imagine that? Hard-drives without spinning parts!
:D
They will have to quadruple the throughput and we will have competitive hard-drives with seek rates to the order of nanoseconds.
You know, they could even replace CDs and DVDs:
- Data rate high enough for HD-DVD or BR quality
- Put them into a good plastic case (ala zip disks, but smaller)
- No scratches!
Sounds like the 21st century to me.
If Samsung plays its cards right, they can make some serious dough with that technology. We're almost there.
Giggidy-Giggidy
"All you have to do is be fragile and grateful. So stay the underdog." Chuck Palahniuk, Choke
The third comment on this page (if they know anything) says that the 2GB modules in the Nano are made of 4 500MB chips, so these new 16Gb (2GB) chips would allow for 8 and 16GB Nanos.
You guys really should try at least to read the TFA. It says the chip is 16Gb, running in a configuration of 16x16 yielding 32GB (yeah, thats Bytes).
(No I'm not new around here, but comeon, lets start a trend and at least read some of the posted stuff before bashing)
On hearing a heckler in the front row question his sanity, George Carlin replied... "Nice...I see you've been given the gift of a functional brain - please let us know when you unwrap it and take it out of the box."
...for something to be a carburetor is for it to mix fuel and air in an internal combustion engine--carburetor is a functional concept. In the case of the kidney, the scientific concept is functional--defined in terms of a role in filtering the blood and maintaining certain chemical balances.
Ok, I'll spot you this one, but next time, do yourself a favor and pay attention during class...
Functionalism has three distinct sources. First, Putnam and Fodor saw mental states in terms of an empirical computational theory of the mind. Second, Smart's "topic neutral" analyses led Armstrong and Lewis to a functionalist analysis of mental concepts. Third, Wittgenstein's idea of meaning as use led to a version of functionalism as a theory of meaning, further developed by Sellars and later Harman.
In the world of transistors, during manufacturing, we have functional and non-functional dies, where the non-functional are discarded, and the rest are further tested and assigned a 'functional level'.
This is where we end up with ram (or processors, etc.) being 'speed' rated, such as 80ns, 70ns, 60ns... These different speed components can all surface as part of a yield from the same 'batch', when some refuse to lock at one speed, and then pass inspection at another. Samsung doesn't run production of 60's on one day, and then 70's on the next. It's all about their 'functional' status as they come out of the oven...
A 'functional' transistor can go thru as many as 300 steps before it earns that title.
All current flash devices have a built in wear leveling algorithm that ensures (within limits) that the whole device, on a block by block basis, sees even erase cycles (the damaging part). In addition to that, a flash drive will fail more gracefully than a hard disk would under most conditions. All in all a flash drive will wear out after the PC went through a refresh cycle (4 year cycles) anyway so it doesn't matter all that much.
The limits to the wear leveling are that the flash device will not move data in order to wear level, thus if you have a flash drive with all but one block full of data and you then constantly update a single file on that disk, it will alternate between the block it was on and the unused block while all the other blocks are untouched. In the real world this would be less of an issue because windows bombs when it's disk is that full anyway.
Some of the benefits are that the OS can be stored on blocks given hardware level protection against erasure, making it more difficult to get a virus that damages the host OS. Defrag is completely unnecessary, and access times should be awesome. I already run a tablet PC off only Flash memory, and while it is somewhat limited with current capacity drives, a 32gig drive would be awesome.
-nB
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
Replacing disk with flash RAM is not feasible: flash isn't fast enough, and doesn't survive enough re-writes to the same blocks.
It's not only feasible, it's been done. It's horrifically expensive, but it works. A "wear leveling" algorithm is used to ensure the same flash cells aren't erased and re-written continuously. Heck, even the flash keychain drives and digital camera cards do that. No, it probably won't hold up to as many write cycles as a magnetic disk will, but writes are much less common than reads, especially in some database and web applications. The drive doesn't need to last forever anyway, since the computer it's part of won't either. I've heard that these guys have had one of their flash drives on a continuous rewrite cycle for a few years now - no errors yet.
Where do you get the notion that flash is slow? It's slow compared to RAM, but it's way faster than a hard disk. That's one of the selling points of these things.
Where do you get the notion that flash is slow?
Right here:
Timing cached reads: 1584 MB in 2.00 seconds = 791.72 MB/sec
Timing buffered disk reads: 20 MB in 3.20 seconds = 6.26 MB/sec
Timing cached reads: 1568 MB in 2.00 seconds = 784.12 MB/sec
Timing buffered disk reads: 118 MB in 3.03 seconds = 38.92 MB/sec