Slashdot Mirror

The S29GL032N on the main system board is a 4 megabyte Spansion flash chip. Not luxurious; but well within the realm of a router-sized embedded linux(though it neither implies nor excludes a bunch of embedded OS options).
 
As for speed, Linux can be made to be quite snappy; but it wouldn't surprise me if enough of the lag is in starting up network-related stuff, along with whatever server program the device uses to allow the client to connect to it, that you wouldn't be able to readily distinguish between Linux, Vxworks, BSD or WinCE on speed alone: sure, an embedded OS booting from solid-state storage on known hardware should move like lighting; but then it has to bring up an external USB device, do the WPA dance, send a DHCP request and receive a reply, and then start up whatever server program the firmware guys threw together for the client to connect to. And then we don't actually know how often the client side of things actually polls the IP where it thinks the device is supposed to be, or whether the device sends out some sort of broadcast when it comes up, or what. Too many variables to even say how fast the OS comes up.

What baffles me is that the author of TFA is apparently geek enough to take a screwdriver to a $150 toy; but is making dumb guesses about OS type based on boot time even though he found a populated serial header, with RX and TX labelled, no less... C'mon, man, you can be pretty sure that the thing is 3.3v(based on the flash IC and lack of visible level converters, might be 5v or 5v-tolerant, highly unlikely to be RS-232), the pins are labelled for you, and it'll probably boot-spew something at you, why are you guessing based on boot time?

AMD has a long history of poor managment decisions.

Back in the late 1990s AMD's CPU busines was a money pit, and the company was kept afloat by their profitable NOR flash joint-venture, but they spent way too much money on their CPU business (for example, those two huge fabs in Dresden that they just divested), and there was nothing left to foster new flash technologies. In the 2000s, NAND became the flash architecture of choice, and since AMD was caught without spare cash, they couldn't afford a crash-course investment in the new chip tech, they spun the unit off as Spansion and left it to slowly rot. Last time I checked, Spansion had only offered competitive NAND THIS YEAR, because they didn't have the cash, and they're still making mostly NOR products. Spansion's sales have dropped to about a quarter of the peak since the spinoff with no end in sight, and they can thank AMD for putting them in that spiral of doom.

Intel (also a MASSIVE NOR flash maker) did not see NAND flash coming, but once they realized their mistake they had the massive piles of cash to make a quick transition. Now just seven years later they are one of the world's premiere NAND providers, all because of their diverse portfolio (and common sense on where they should spend their money).

AMD *had* the money to fix Spansion before they went on their fab spree chasing Intel. And given the peak revenue of over 1.5 billion yearly (nearly as high as their CPU revenues today!) it was worth saving. But they spent the earnings from Spansion on Fabs 36 and 38. They wasted 2.5 billion on Fab 36 alone! And then they spent way too much money (that they didn't have) on ATI, which really has not improved their fortunes.

Hmm, I didn't know MLC used a stronger ECC. It seems you're right

http://www.spansion.com/application_notes/Types_of_ECC_Used_on_Flash_AN_01_e.pdf

But actually look what's happening here. The generations of flash I've used have gone like this

NOR. No ECC. Low density
SLC NAND. Hamming code ECC. Higher density (2xNOR).
MLC NAND. BCH or Reed Solomon. BCH seems to be the most common. Highest density (2xSLC)

It seems like the trend is to have less reliability in built reliability in return for smaller cells and add smarter error correction in the controller. Which was my point - you can live with any bit error rate that you can correct.

I bet the cost of silicon for the controller is negligable compared to the NAND array anyway.

I'd thought AMD had sold off their flash-based production... Yeah, it was previously a joint venture with Fujitsu and is now called Spansion.
http://www.spansion.com/

Actually, AMD spun off it's NOR flash business some time ago:
http://www.spansion.com
Intel is in the midst of doing the same with its NOR flash, but keeping it's NAND flash going.

I think AMD will continue to be around -- but I think they will seriously consider going fabless. It's just too expensive to keep up with Intel and IBM. They already use TSMC for the ATI portion of their business. Certainly going there for the microprocessor wouldn't be a stretch.

It can be up to 20 years under ideal conditions. Flash that's been erased many times will have a shorter data retention time, as will Flash stored at higher temperatures. A Google search on Flash Data Retention will lead to many pages from various semiconductor companies discussing the performance of their particular parts. There's a nice easy-to-read table in this document:
http://www.spansion.com/application_notes/EndureRe tentn_AN_A0.pdf

Check out Spansion if you want to support an American company. They are a spin off of AMD, and have some impressive technology when it comes to flash memory.