Slashdot Mirror
Flash Memory to Rival Hard Drives
Skal Tura writes "Samsung will start producing 16 gigabit Nand Flash chips this year, nudging the memory technology towards use in notebook PCs and maybe even edging out hard drives in some products in the next few years."
Some more information about the NAND flash memory can be found here.
One nice thing about this article is that it clearly explains the difference between a gigabit (Gb) and a gigabyte (GB)...something the article referenced in the story seems confused about.
From the article referenced in the story:
And from the article referenced above:
Sorry to be picky, but I'm a stickler for detail.
____
~ |rip/\/\aster /\/\onkey
Seems like they're playing fast and loose with capacity: "will start producing 16 gigabit Nand Flash chips this year" vs. " currently in products such as USB drives and digital cameras in capacities of up to 8GB." Correct me if I'm wrong, but doesn't 16 gigabits = 2GB?
The correlation between ignorance of statistics and using "correlation is not causation" as an argument is close to 1.
i dont know about long lifed, since a flash memory card has a limited number of writes anyway. This is because each one of the memory holding transistors has an extra insulating layer between the gate and the emitter, so electrons are "forced" throug, trapping the bit and therefore the data. There is only so many times (in the 10s of thousands) you can do this, and then is toast. If you use it as a hard drive with crappy memory paging, it will die soon.
"The quality of life is inversely proportional to the number of keys on your keyring."
Ummm.... prolly not.
:-)
I don't know any specifics (but neither did your post, so we are even) but I build computer controlled devices that need to work in a fairly high vibration environment. Our current product runs off Win2K, and boots relatively quickly off a 2GB solid state Laptop size HDD.
We are building a new product that will be running Linux off a 256MB CF card. We are not quite done development, but it seems to run OK. It isn't working that hard though, just polling a USB control panel and outputting control commands based on what the user wants to do with a small amount of very simple graphics.
(disclaimer: I know very little about OS's and software. I am mostly solder jockey, circuit design, system installer, and a little bit of sales. I have been struggling to get my MythTV box working for over a year now
Flashdisks are much more reliable then any conventional harddrive. They claim >5,000,000 write/erase cycles and unlimited reads. Unites States defence department is using them for reliability issue alone.
C SIFFD/IDESCSIFFD
M-Systems (top flash disk producer) states this:
(copied from the website)
Top Reliability & Endurance
** 99.999% reliability
** >1,400,000 hours of actual (in the field) MTBF
** Embedded EDC/ECC, based on BCH Algorithm
** Data integrity under power-cycling
** TrueFFS® technology: bad blocks mapping-out and dynamic wear-leveling algorithms
** >5,000,000 Write/Erase cycles; Read unlimited
5-year warranty
Source link:
http://www.m-systems.com/site/en-US/Products/IDES
MTBF isn't absolute. It's a statistical estimate. A hard drive may have a 500,000 hour MTBF. That particular model of drive wasn't tested for 57 years to see if it failed.
Any type of failure rate is also representive of the collection of all products being tested, not a single one.
Read the Failure Rate Wiki entry for more information.
I think it's worth mentioning that the bottleneck in reading/writing large files is an interface problem (usb et.al.) and not actually an issue with the ram. Currently the thing spinning drives have going for them is cost per GB.
2.4 Terabit Hard Drive, or in layman non-marketing speak, 300GB... I mean 279.4 Gigabytes.
I would like to patent a new measurement unit, called the bi (pronounced "bee"). Not to be confused with the slang used to describe a person with bisexual tendencies, this measurement unit quantifies memory size.
1 bi = 0.001 bit
Fear my 2400,000,000,000,000 bi hard drive!!!111
If I can do it, its probably not worth doing... probably
That's 100k per block, not for the entire drive. The wear-leveling algorithms will make sure that even if you constantly re-write the same file, that part of the memory won't get worn out.
With a 512-byte erase block size, that is 419 billion writes. With a 4K erase block size, that's 52 billion writes. Use a 20GB drive instead of 2GB, and you'll get 10x the writes. And, the computer can warn you before the memory stops re-writing.
5 trillion writes is 10,000 writes/second for 13 years.
HIV Crosses Species Barrier... into Muppets
The traditional flash devices are NOR flash devices. They are by nature more reliable than NAND flash but are expensive to amnufacture. NOR flash devices can withstand more write write cycles than a NAND flash. BIOS flash are NOR, as are most of the solid state flash drives used in telecom and aerospace. This is why NOR flash is marketed to embedded/industrial customers and NAND flash is marketed to the consumer market.
5 million writes? that might be true with the 180nm technology (actually it's not, with 180 it's more like 100k), but with samsung and toshiba pushing to 55nm this year and 45 soon after, they can't even guarantee more than 50k.
worse yet, flash retention is going down too because of the smaller gates and use of multi-level cells. one electron jumping off the gate when you have seventy million of them in your huge gate is no problem, but when you are only holding a few dozen of them things in your tiny little gate, it's a much worse scenario.
you either work for the company you advertise, or is just a fanboy. no way 16Gbit chips can have this much rewrite capability.
NOR flash is extremely slow for writes. This Samsung appnoteo ry/appnote/onenand_features_performance_051104.pdf
http://www.samsung.com/Products/Semiconductor/Mem
compares I/O performance of the various technologies (the chart is on page 28, so scroll down...)
For their test rig, NAND flash yields 8.8MB/sec writes vs NOR at 0.14MB/sec. That's why NOR flash is only used for BIOS memory and other things you don't have to rewrite very often. On the flip side, NAND flash gets reads at 16.5MB/sec vs NOR at 23.9MB/sec (or 108MB/sec, presumably in some kind of burst mode - that part isn't explained).
If their OneNAND performs as well as they claim, I could see using it for a boot drive; 68MB/sec read would be fine there, 9.3MB/sec write would be ok as long as you weren't paging to it or doing much of anything else. Linux would run pretty well with those parameters, its buffer cache is good at absorbing and deferring writes; Windows 2K/XP's memory manager/cache manager purges pages too aggressively though, which would make the write throughput a serious system bottleneck.
-- *My* journal is more interesting than *yours*...
the number of transistors required to make a nand gate is less than the number required to make a nor gate. Therefore, nor would offer no benefit, while requiring more tx's (higher power consumption and less chip density)
For me it was very helpful to see it posted -- I'm very much aware of the differences but missed the word gigabit in the article when I quickly read it. Reading that post was in fact "inciteful" to me and if I'd had mod points at the moment I'd have marked it so.
The problem with quotes on the internet, is that nobody bothers to check their veracity. -- Abraham Lincoln
'' MTBF isn't absolute. It's a statistical estimate. A hard drive may have a 500,000 hour MTBF. That particular model of drive wasn't tested for 57 years to see if it failed. ''
Also important: Products like harddisk have a limited life. That harddisk with 500,000 hour MTBF will wear out after five years or 50,000 hours; no way will it last 500,000 hours. The MTBF only means: If you buy 500 harddisks and run them for 1000 hours, you can expect one to fail.