Feeds

back to article Flash memory made immortal by fiery heat

Taiwanese flash memory non-volatile memory manufacturer Macronix is set to reveal technologies is says will make it possible for flash memory to survive 100 million read/write cycles. Today's flash memory wears out over time, because moving electrons in and out of transistors wears them down. That problem has led manufacturers …

COMMENTS

This topic is closed for new posts.
Bronze badge

Local and Remote...

This could also turn out to become a favorite exploit for dangerous crackers or criminals. Sniff, assess, access, and manipulate. Just add heat.

I wonder just HOW HOT a phone can be made by this process. I suppose that some protection, such as a heat sink or some built-in heat-suppressing gel might be stuffed into the limited space in the phone...

I am not a phone designer, so, my thoughts could easily be faulty...

0
15
Anonymous Coward

Re: Local and Remote...

I wouldn't imagine it would even be a function available to anything other than the memory chip itself. There's no reason for it to be externally accessible. Even if it were, it'd take some pretty serious mis-engineering to have a device designed to run for milliseconds at a time, using microjoules of energy in total, be capable of actually being turned into a heater.

11
0
Boffin

Re: Local and Remote...

"heat-suppressing gel"

1
0
Bronze badge

Re: Local and Remote...

INT WTF?

Why so many down thumbs? Is it NOT possible that malicious outside exploit can happen? Look at power plant exploits that have already happened. Other feats have, too. Granted, they were not along the lines of Chernobyl...

11? I feel like i hit the raw nerve of that company's publicist. Wow, one with an informed explanation would have sufficed. Is there something about the process that will not render the phone or tablet or similar device useless?

I hope this is bookmarked, because i certainly recall in 1995 or so predicting to some programmers that in i few years people would be hacking computers via the modems. They ridiculed me because i was not a developer.

Just bookmark it. In a few years, tho tpit will not be dramatic, we may start seeing non-battery-related killing of phones.

0
6
Silver badge
Thumb Down

Re: Local and Remote...

> I feel like i hit the raw nerve of that company's publicist.

I felt like you hit the raw nerve of baseless fear mongering.

> in 1995 or so predicting to some programmers that in i few years people would be hacking computers via the modems.

You must be very, very young. 1995 or so? The Shockwave Rider was written '75, pretty sure it wasn't the first.

> We may start seeing non-battery-related killing of phones.

Other than throwing the mobe down the loo? Might as well argue that one kill phones by running extreme tasks on the CPU up until thermal excursion happens. Or that DRAM refresh cycles will be switched off until the charge has been drained. Not happening.

5
0
Bronze badge

Re: Local and Remote...

I am not talking about the Gorn lockin on to Spock's tricorder type of destruction. I am talking about simple, malicious, expensive destrynction. But, then, since it appears to already be posdible, then, I guess. I AM overreacting. Still, does not justify the ass ailing onslaught of down thumbs.

If ~ 20 counts as very young, then I guess I was, back then.

0
3
Bronze badge

Re: Local and Remote...

when this exploit happens, remember this day, and recall the shitty bullying or thinly disguised censorship attempts. I will not wish evil... I wil just ask forrestoration of my votes from negative to positive....

0
0
Paris Hilton

Re: Local and Remote...

> Sniff, assess, access, and manipulate. Just add heat.

Why do you start by sniffing asses? Seems like a perverse exploit.

0
0
Gold badge

Remember that we're not talking about baking the entire drive at 800c here. This is talking abotu spot-heating the individual gates to 800c. This is something like 80 silicon atoms. It is not only not that energy intensive, it shouldn't damage the rest of the drive, if done right.

Phase change memory can do this today, albiet only to 500-600c. It should be something easily made available for mass production.

12
0
Silver badge

Way to ruin my dream Trevor!

And there I was heating my Kingston DTSE9* with my Zippo (TM) to make it immortal!

*a nice lil' drive, mind. The sturdiest max-bang-for-bucks I've seen since they invented that variety of bread that comes pre-sliced.

3
0
Paris Hilton

How do you send a correction on something this bad?

"He's also said the technique has the Macronix intends to commercialise the technology,"

Umm....What? Granted I'm tired right now but I cant even begin to unravel this.

4
0
Silver badge

Re: How do you send a correction on something this bad?

For great justice!

2
0

Re: How do you send a correction on something this bad?

I think it was meant to read something like:

"He's also said Macronix intends to commercialize the technology,"

Probably started out thinking something like, "He's also said the technique will be commercialized," but then suffered from a brain fart in the transition.

0
0
Bronze badge

Hurry Up

"But Macronix's innovations, which it will reveal at the 2012 International Electron Devices Meeting in San Francisco on December 11th, look like making wear levelling irrelevant."

It's a neat and impressive idea, but too late I fear. They'd better get a move on before HP's memristor makes FLASH itself irrelevant.

0
0
Silver badge
Trollface

Re: Hurry Up

Dont you mean before HP itself becomes irrelevant?

3
0

A seemingly rare thing

This seems like something genuinely innovative that can be licensed for others to use because of patent law.

4
0
Boffin

Radiation hardness

Could this technique be used for "healing" chips exposed to ionising radiation?

0
0
Gold badge

Re: Radiation hardness

Depends on whether or not the oxide substrate is buggered too.

0
0

I'm pretty sure my laptop already gets to ~400C when doing anything in 3D, so my SSD must be invincible. Shame my legs aren't.

2
0
Silver badge

Is that the famous Iranian exploding laptop?

0
0
Anonymous Coward

"moving electrons in and out of transistors wears them down"

Facepalm.

1
0
Silver badge

Well, those tiny tweezers don't last for ever you know.

2
0
Silver badge
Flame

sounds great

A laptop that self heats to 800C with lots of IO writes. What can go wrong? Plus since the heater will use lots of power (I suppose) then how about incorporating a hydrogen fuel cell for added fun.

0
9
Silver badge
Boffin

This tends to imply that this memory could survive running hot

Which is something my previous testing tends not to support. I've had commercial Micro SD flash memory working at high temperatures - over 150C - but not for many [classified, sorry] hours: certainly not as long as I need. That said, many microcontrollers with flash memory will work up there but lack the capacity for storage.

If they get this working reliably at high temperature, there's a small but high-paying market for the stuff. High temperature memory is thin on the ground.

0
1
Ru
Silver badge

Re: This tends to imply that this memory could survive running hot

It'll be a very small heating area, heated for a very short period of time. The bulk of the control systems and packaging will be the same as everything else on the market.

4
0

Re: This tends to imply that this memory could survive running hot

If heating to higher temperatures than that briefly *resets* the chip life, it would make sense that it wouldn't work very well at temperatures approaching the threshold. Do the devices work OK afterwards once they cool down, or is it permanent damage?

In effect, regular flash erase builds up an effect which stops the gate working reliably after ~10k cycles; it seems this 'thermal erase' clears that cumulative effect, but will itself stop being effective after ~10k cycles.

"100M erase cycles should be enough for anybody"?

2
0
Anonymous Coward

Re: This tends to imply that this memory could survive running hot

No, if you write or read the data at higher temperatures you increase the rate of loss of electrons which greatly reduces the endurance of the flash. Even at room temperature, the smaller geometry flash suffers from read disturbance - you have to move data around fairly often as its read to avoid losing the ability to distinguish the bits accurately. Clever signal processing can tell the controller when this is needed.

I suspect in this case, it's a short reset cycle when the block has been erased and therefore isn't under any read/write stress and is in its most energetic state. Presumably, the electron pool is somehow replenished at these high temperatures.

All in all, a very interesting development.

1
0
Bronze badge

Re: James 100

"100M erase cycles should be enough for anybody"

You're really Bill Gates aren't you?

0
0
Bronze badge

No different to the temperatures inside some printers, on tiny elements. Some thermal printers get up to the same temps on the tiny heads that look like little ribbon-cables that do the printing. It doesn't take much power because it's SO small an area heated (and so rapidly heated/cooled), doesn't heat to the point you can't touch things afterwards, and doesn't present a danger because there's not a lot of wattage there to do anything more than heat the very edge of the surface by the temperature (and even the other side of the paper/tape/label probably won't change in temperature AT ALL, or all your receipts would feel hot).

If it can be automated away from the user (because you just know they'll refresh cells more than necessary if they have any control over it, and it'll become one of those "data recovery" techniques like putting it in the oven/freezer), and provide that length of read-write cycle expansion, then it's got to be worth it.

Hurry up and get it to the point where it's all built into a single chip, with Flash and heaters and control circuitry and you'll be a millionaire.

1
0
Bronze badge
Holmes

So many catches, so many orders of magnitude...

You have to be careful estimating the life of flash memory, because some processes use multiple writes, but even when you allow for that, running the input/output at full bandwidth can take a huge amount of time to wear out the chip. This tech will, practically, give us 3 orders of magnitude, with some margin left between lab results and normal use.

There are other limits, things such as the slow erosion of the metallic tracks as they carry current, and if you want really long life you might have to combine the component size of a couple of generations back with this heating system. So you might not see it in the highest capacity chips. That sort of compromise is what engineering is often about. And the useful life might also be limited by the service life of the interface. Look at how motherboard connector standards have changed over the last decade.

On the other hand, if you have a 100-year SSD, that's an incentive to keep supporting the interface, but what sort of expensive converter will you need to read it in 2112?

1
0
Gold badge
Thumb Up

*Licensable* design techniques *multiply* memory life by 10 000x

Is there any business where that would be greeted by not much more than a "So what?"

In (US) car terms that would be a car that can do an average 174 000 mpg. Call it a top up every 50 years.

Thumbs up for a clever hack that could be retro-fitted to other mfg's design flows.

2
0
Bronze badge

Re: *Licensable* design techniques *multiply* memory life by 10 000x

The fact is, that in computing my current computer is something like 500-1000 times faster than the first computer I ever owned, and has four cores (eight hyperthreaded) capable of that speed without even struggling. It has something like 300,000 times as much RAM as my first computer and a nearly a thousand times more storage, in units that didn't even exist when I was younger. My internet connection is something like 5000 times faster than my first Internet connection.

When an industry is this young, it's not hard to have huge multiples in performance. The first car was infinitely faster than any car before it. The second car probably went ten times or more faster. It's after that initial "burst" that things settle down (and thus things like CPU speed and RAM have stagnated in recent years). The first plane barely got off the ground, modern planes break the speed of sound if required.

And this isn't a factor of speed, or limited by aerodynamics, or some huge feat of science. It's making a relatively new and unresearched product (that, in one form or another, preceded even my first computer) to be more tolerant to long-term use. That's an entirely different statistic to increase, and far less interesting. The first car probably had a life-span of 10 minutes before it went boom. The second probably lasted a couple of days. Now we routinely get 100,000 miles out of them.

Multiples are a bad statistic to compare too, because they bias towards smaller improvements at lower figures. You won't get a 10-fold increase on the "new" MTBF if it turns out to be true, because that's so much more difficult and would be almost untestable (and thus, unlikely to ever happen). But you might well get an increase of 5-10 times in capacity of SSD's in the next couple of years.

0
0

Re: *Licensable* design techniques *multiply* memory life by 10 000x

Hey Lee,

Are you the new AmanFromMars?

;-)

I do like your postings

0
0

I Think Wear Leveling Will Still be Needed

> look like making wear levelling irrelevant.

I don't think so.

Image a block being written once a second (quite possibly pessimistic for something like control information in a database file on a busy server), that's ~31.6 writes a year.

So 100 million is reached in a little over three years.

So unless the ">" in "to > 100M Cycles" is really order of magnitude greater than, wear levelling is still going to be needed for server (ie. enterprise grade) SSDs.

1
0
Gold badge

Re: I Think Wear Leveling Will Still be Needed

100 million writes *is* plausible, but would probably cause trouble for a piece of spinning rust as well. Existing strategies for bad block management might therefore suffice.

1
1
Bronze badge

Re: I Think Wear Leveling Will Still be Needed

wear levelling seems like a lot of overhead for general use in this case.

Swapping some blocks out when they look likely to fail after three years constant use would just require a bad block/relocation table like regular hdds have.

Whereas an "old" SSD block would fail in a couple of hours in the same situation, so wear levelling makes alot more sense.

0
0

Re: I Think Wear Leveling Will Still be Needed

Since MLC flash must be erased before a write and erasure is slower and uses a larger block size, erasure is generally handled as a background process (so some fresh blocks are kept ready at all times). This means that some redistribution of writes will occur anyway. Whether this redistribution is sufficient for enterprise use with a 100M endurance is questionable, but the endurance behavior is not as simple as 100M writes to the same (virtual) location.

Since decent wear-leveling techniques are relatively mature, providing adequate wear-leveling should not require special algorithms when write endurance is relatively high. Such high endurance flash *might* reduce the benefit of clever flash management (or trade-off cost of the flash chips vs. cost of the flash controller).

Since write endurance is one of the factors limiting flash cell scaling, this technique *might* help flash scale down further than previously expected.

1
0
Anonymous Coward

Big waste of time

Unless you desperately need more speed now (ie. it is your business to run things fast) then you're better off waiting for memristor technology.

0
0

magic smoke

I've heated semiconductors to >400C before but the result was less good. There was less self healing and more molten jets of plastic and smoke. I must be doing it wrong.

0
0
Go

SSD live rocks!

Been running an Intel-X25 24/7 as a high traffic, high I/O web server for over 2 years now. Works perfectly no slowdown as far as I can tell. Not even using TRIM on it.

SSDlife software reports 97% health levels and a 33 year projected lifespan.

Who knows how long it will last but its doing fine so far!

0
0
Silver badge

Re: SSD live rocks!

> a 33 year projected lifespan.

And tomorrow is another day.

0
0
Silver badge
Boffin

This is important as flash gets denser

For each Moore's Law generation (34nm, 25nm, 18nm and so on) flash cells have a shorter life by something approximating an order of magnitude. In addition, multi level cells have a reduced life. One bit per cell SLC lasts longer, then 2 bit per cell MLC, with three bit per cell TLC bringing up the rear. Again by a factor approximating an order of magnitude.

The industry used to think it would go to 4 bits per cell (QLC) but abandoned such plans when they saw the trends. There have been worries that 13nm might be the last flash generation as TLC would be useless when shrunk this far and MLC will have reached its limits.

So if this technology can add FIVE orders of magnitude it'll buy approximately a decade of further process shrinks - assuming we can go that far - which would be nice, in case memristors don't pan out. You know, exactly like PCM that was supposed to replace flash a decade ago didn't pan out.

People who are saying this doesn't matter because memristors will save us are being foolish. I don't know of anything in the field of computers that with more failed promises than memory technology. Bubble memory was supposed to replace RAM in the 80s. In the 90s holographic storage would replace both optical drives and hard drives (you still hear about this even today once in a while, could have sworn I saw a Reg article recently that I didn't read when I saw the subject) Then we saw PCM in the 2000s and memristors look to be the great white hope of the 2010s (I'm keeping my mind open here, but the track record in the memory field isn't good) These are just the major ones that everyone talked about as almost sure fire bets. I'm not even getting into things everyone knew was a long shot like ZRAM and its ilk.

0
0
Fai

So memristors will have more time to mature before mass adoption (if they are indeed going to fulfill their potential to be more capable than current memory tech), just as SSDs were slowly reaching affordability vs spinning platters of glass. It'll perhaps prolong the transition period of existing NAND-based memory.

I think what will happen is that if thermal annealing really does become a prominent feature of flash storage, it'll start to look like a hybrid of PCM and current flash where thermal annealing is a regular activity used with TRIM and wear levelling to manage the drive state - especially if the cells lives are reduced to something like 10 writes through process shrinks and high MLC levels. Perhaps this is what keeps Moore's Law alive for flash memory.

Incidently I read about this same topic on Ars, and it was surprising to see the number of commenters there who'd not read the article properly about the localised heating - so to Reg's credit it's refreshing not having to read comment after comment of 'ovens in your pocket fail'.

0
0
This topic is closed for new posts.