New RAM Based On CD-RW Film On Horizon

SiliconShaman writes: "Ovonyx Corporation is reporting initial success at producing a very cheap non-volatile RAM. Supposed to be cheaper, faster, etc. than conventional Flash memory. Applications include satellite and aerospace, cell phones, the list goes on. Details can be found here: Short Technical Presentation in HTML." "OUM memory technology," says Ovonics' Web site, "is similar to DRAM and many orders of magnitude faster than Flash write." The proliferation of devices which will need copious nonvolatile memory should mean a healthy market if this technology reaches the production line. I know my camera could use some more storage ...

Comments

This is a Good Thing

[XChemie]

For a little background (I'm a chemistry grad student working in this area), the maturing of this technology is good to see, after many years of battle vs. magnetic storage for both CD-RW and non-volatile memory. I think the potential here is very great because of the relative prodcution simplicity and cost benefit compared to competing technologies. This stuff is not vaporware! The technology of phase change media (based upon optically/electrically induced reversible amorphous to crystalline transitions) has been researched extensively over the last two decades by big names like Philips, Micron, 3M, IBM, etc. But magnetic storage had more research money for quite awhile for alot of reasons; however, it's phase change media that eventually got used for your CD-RW (not CD-R though...) discs. Making non-volatile RAM was the next obvious target, but latency was a problem until recently (and the inertia problem of changing manufacturing methods). For some more info, goto the website of the parent company Energy Conversion Devices. It's based out of Troy, Michigan and was founded by Stan Ovshinsky, who's somewhat of a rogue in the physics community. But it's a cool company that also uses a similar technology to make surge protectors against the EMP from a nuclear blast! :)

Will your memory wear out?

[jbuhler]

The presentation states that OUM is only good for about 10 trillion writes. If I bang on the same bit for a week or two at a rate of (say) 10^7 writes per second, won't it wear out?

Of course, most bits won't see this kind of punishment, but I can imagine data acquisition apps that might trash a few bytes of storage within a year or two or ten. Hopefully the machine's L2 cache would effectively rate-limit writebacks to main memory, but it should still be possible to set up an aliasing situation that would cause a line to be flushed on every write.

Re:May be serious.

[Jage]

AmigaOS, or should I say Kickstart, partially implemented recovering memory state after reboot (yeah, and this was done way back in 1985.) After rebooting, it wouldn't erase memory, but the OS would look for some system vector tables and do some basic sanity checks to see if the tables are corrupted. Some nice uses were for example a RAM-disk that could survive rebooting.

Sometimes the system hanged in repeating crashes and reboots, though. Then your only option was to really erase memory, for example by toggling The Most Significant Bit (ie. power switch).

Other bad point was that also viruses liked to hook themselves to those vectors, enabling them to survive reboots.

For the rest of the state, maybe you could log some state changes to for example display adapter, network card and etc. Or maybe your devices would have to expose their internal state in some compact structure you could just copy to log areas periodically (when the system would be in special recoverable state.)

I don't know. :)

Re:Some Background on this Company

[orpheus]

Thanks for the update. I'd seen those implications in the ECD website, but...

I am not sure that NiMH batteries were invented by ECD. The ECD website only claims a) all "significant" manufacturers of NiMH have a license with them (implying that some don't - why?) and b) the first commercial NiMH manufacturer 'used' Ovionics technolgy. This company (Gold Peak) was also 'one of ECD's early licensees' -- this seems like an odd way of saying "we invented NiMH". It is almost as if they are trying to imply that without actually saying it

I tried checking the USPTO database, but if he had the groundbreaking patents, they are too old to show up on the database. I concluded this by checking the bibliographic entries for current Ovshinsky NiMH patents (which contain the patent numbers of preceding/related art) I presume he would cite his own groundbreaking patents. It is possible that I'd learn more by reading the complete applications, but IANAL (and no one's paying me for *that* depth of research)

I was aware of their early flexible amorphous solar cells in the 70's, but I didn't know these were a major part of today's market. Are they?

Besides I'm not sure that one major product (or maybe two) in 40 years is a "decent track record in products" (ECD was founded in 1960) ECD seems to have some viable products, but I'm surprised that they seem to be incremental improvements over the state of the art, while their past (and present) technology always seems billed as 'revolutionary'

Of course, that can happen. I think Harry Bede is a good example. He designed and produced the BD-5 airplanes and an interesting 'inline' automobile -- both products I'd have bought instantly (well, once I could afford them) but his companies can't seem to stay in production very long.
_____________

Interesting stuff.

[dbirchall]

I especially liked the read/write cycle lifespan, since for anything to replace existing memory technologies (like flash ram and CD-RW), it needs to improve on that performance parameter.

I'm a little curious, though, whether this is best suited for compact applications like flash RAM, or as a means of replacing both RAM and fixed-disk technologies. If the size reduction is substantial enough, and it's as fast as RAM, it's possible that a system might ship with a single fixed rotating rewritable drive (like a hard drive) using OUM technology to provide a significant chunk of storage space (say, multiple gigs) with high access and transfer speeds. Imagine your swap partition running at the same speed as your RAM. That'd be pretty spiffy.
--

Re:RAM

[mr3038]

Just wanted to add that if you have (cheap enough) memory with low latency you already have memory with low latency and high bandwidth. Just interleave those memory chips and you are done.

For example if I have SDRAM (64 bit bus) PC133 10 ns memory with bandwidth of 1.06GB/s and RDRAM (16 bit bus) PC800 20 ns memory with 1.6GB/s which ones I can interleave to get memory with latency of 12 ns and bandwidth of 4GB/s (256 bit)? (I have really no idea of real latencies of those memory types and it hugely depends how you measure it - time it takes for chip to read bits or processor to receive required bits. But what I know is that SDRAM has much lower latency)

I'm not aware of motherboards supporting SDRAM interleaving but Intel already does this with RDRAM in its new mobos so it should be possible. I think that this is because SDRAM already provides enough bandwidth in most cases.

Note that when using DDR SDRAM you can double the bandwidth (for SDRAM) and one can only wonder why is/was Intel trying to push something like RDRAM into market?

For those still wondering the answer to second paragraph I will give a hint: you cannot reduce latency with interleaving.
_________________________

One of these days...

[zpengo]

I can't wait for the day when I can get something like this installed in my skull, so that when I think to myself, "Mental note: remember to...", I'll actually do it!

Some Background on this Company

[orpheus]

As soon as I heard the name "Ovonyx" I was reminded of Stanford Ovshinsky, a physicist who claimed to revolutionize amorphous silicon technology, reportedly 'inventing' a field of Materials Science called "ovonics". He was a minor hero to me as a teenager (the 70's). I'm not such a fan now.

Though I can hardly blame /. for not checking further, I think readers might want to.

Sure enough, on the "About the Corporation" page, it confirms: "The Corporation Ovonyx, Inc. (Ovonyx) was formed in 1999 as a joint venture between Energy Conversion Devices, Inc. (ECD) and Messrs. Tyler Lowrey and Ward Parkinson. ECD is Ovshinsky's company (founded in 1960). Surprisingly, no link or further information is provided on ECD, which according to the Ovonyx Corporate page "has been actively pursuing phase-change material and device structure development,optimization and characterization for many applications - but principally for optical and electrical memories."

From what I recall from the 70's/80's, ECD wasn't 'focused' in these areas, but on solar panels (hence the name "Energy Conversion Devices"). This may explain why "ECD's contribution to the Ovonyx joint venture was all of its intellectual property (IP) in this area" [optical and electrical memory]. Perhaps ECD didn't feel it could make use of this IP, and turned it over to two former top Micron Technology (chip) execs. Read on for details

THE BACKGROUND
Ovshinsky was a darling of the Japanese in the 70's/80's when Americans were in shock because the Japanese were using American findings like the management theories of Edwards Deming (who was revered like a god in Japan), 'process control', and "just in time" to turn out cars that the US auto giants couldn't match. American business was also awash in culture shock from dealing with the Japanese business systems (kureitsu, etc.) as equals and as a potential market.

Ovshinsky raised money by chiding US corps that they would miss the Next Big Thing if they didn't invest in his work, but his primary business relationships were with the Japanese.

The PBS show NOVA even did a one hour documentary on him ("Japan's American Genius" 10/27/87) that seemed to promise that cheap high efficiency amorphous solar cells would be around any day now. I've kept my eyes peeled ever since (I instantly made the Ovonyx connection in 5/2000!), and never heard anything about ECD. The advances in amorphous technology always seemed to come from elsewhere.

I haven't been impressed by ECD's ability to bring *any* product to market in 40 years, and I don't know of any breakthroughs they have made or licensed -- and all their 'background' and 'product' links seem to point to nonexistent documents in a 'drafts' folder. However, I may be wrong, or perhaps the documents are being updated, so I refer you to ECD's home page, which is (perhaps unsurprisingly, considering Ovshinsky supposedly pioneered 'ovonics') http://www.ovonic.com not to be confused with the OVONYX site in the /. article

Perhaps I am a little jaded, because I was once a big fan of Ovshinsky (as he was presented), but what I've read in the OVONIX and ECD (OVONIC)web sites seems to have an unusual, almost Microsoftian degree of 'spin'. Caveat emptor.

However, if anyone out there knows of some concrete product or technical advance that proceeded directly from ECD or Ovshinsky, I would be *most* happy to hear about it. You have to reclaim those childhood heroes when you can! It's good for the soul.

_____________

Re:Some Background on this Company

[Animats]

However, if anyone out there knows of some concrete product or technical advance that proceeded directly from ECD or Ovshinsky, I would be *most* happy to hear about it. You have to reclaim those childhood heroes when you can! It's good for the soul.

Ovshinsky has a decent track record in products:

• His companies manufacture amorphous solar cells in volume, but they cost about the same as silicon solar cells. However, they're flexible, and so they're widely used on boats. Some sailboats even have them on the sails.
• Nickel-metal-hydride (NiMH) battery technology was developed by Ovshinsky's company, and all 16 makers of NiMH batteries license his patents. Camcorders, laptops, electric scooters, and the GM EV-1 use this technology.

Ovshinsky's big problem has always been with his manufacturing processes. Since he's just about the only player in amorphous semiconductors, he has to develop his own fab techniques. That's a big job; many billions of dollars have gone into making silicon semiconductor technology work as well as it does. ECD has to match that, working alone. They never really have, which is why amorphous transistors and amorphous ICs never took off, although both did work. (Burroughs even shipped some machines with his early non-volatile memory in the 1970s.) His battery technology also has fabrication problems. But slowly the stuff is starting to work.

Questions -

[Taco Cowboy]

Questions -

1. What effect will this type of memory
device has under radiation bombardments?

If this memory is to be used on spacecrafts, it has to be able to withstand radiation bombardments.

2. What effect will magnetic forces (and/or
electro-magnetic forces) has on this
type of memory?

If this memory is to be applied in miniature appliances, the chances that these appliances be in contact with (or being put next to) magnets or under magnetic fields is very large. So, will the amorphous (disordered atomic structure) and crystalline (regular, highly repetitive, and ordered atomic structure) stage be reversed/change/effected due to the magnetic effects?

We do have to be concerned with the effects, because we just do not know what this type of memory device will ultimately be used on - it could be used on embedded devices to control anti-lock brakes on cars, for example - and if the memory device is "affectabled" by either magnetic or radiations, then, this type of memory device should not be used to power computers which controls critical functions that may affect the safety of human lives.

This is just my humble 2 cents.

Re:May be serious.

[Mr Z]

IRC, in most cases the cache only mirrors what's in the main memory, so it wouldn't be a problem.

That's only half true. The cache holds a copy of what should be in RAM. There are two possible common scenarios:

• If you have a write through cache, then whenever the CPU writes something, it updates both the cache and memory immediately. Older 486s sported that sort of cache. Some are switchable.
• If you have a write back cache (just about every modern machine falls into this category), writes will stay in the cache until some other cache/bus traffic causes the results to be written to memory.

The latter form of cache performs alot better, for a couple reasons. First, writes to memory are slow and a write-through cache doesn't hide them. Second, alot of writes are redundant or can be combined with other nearby writes and a write-back cache serves as a filtering and combining mechanism. The drawback is that the cache has a different picture of memory than the external memory has.

Anyway, the way you do "recovery" is to checkpoint the OSes state every so often (like, say, every timer tick when you do a task switch). You can save all the registers, flush the cache, and checkpoint off to the OUM. The device issue you mentioned is solved in a similar manner to APM, I'd imagine (not that I know the solution, but I'm saying it's a familiar problem that's already being attacked).

--Joe
--

RAM

[Signal 11]

It's a common misconception that we need more RAM. That is false. We need faster RAM. Repeat after me: Latency is bad. Latency is evil. I must keep latency to a minimum.

My tech friends are usually stunned when I pick up a stick of PC100 that has 4.5ns instead of PC133 at 5.5ns - they say "PC133 is better!".. and nevermind why. Alittle arm-wringing and explanations of latency and what cache hits v. misses do to their processor performance usually helps. For the stubborn, I ask them to bring their "PC133" ram over and we run a Quake demo. Guess who wins?

I don't want to hear about how much more RAM the industry is capable of making - I'm interested in hearing how fast they can make it. Bandwidth * Latency = Performance.. you cannot have high bandwidth and HIGH latency and get good performance - we've used up all our tricks in that area.

That being said, there are alternatives underway. I'll give you a few hints and leave you to it: magnets and optics. Enjoy!

Re:RAM

[multipartmixed]

> Bandwidth * Latency = Performance..
> you cannot have high bandwidth and HIGH latency
> and get good performance - we've used up all
> our tricks in that area.

According to your formula, if you keep bandwidth constant and increase latency, then your performance increases.

Clearly, this is wrong.

How about:

Performance = Bandwith / Latency

Have a nice day.

--

Re:RAM

[MinaInerz]

I think you fail to understand that the PC### rating is based on latency.

If a memory has a latency of, say, 5.5ns, and is able to sustain it, then you get the speed from the following equation. 1000/5.5ns = 181MHz (5.5ns is very fast, very expensive RAM - you'll see it on those new hotshot video cards). More commonly, on PC133, you will see 7ns. 1000/7ns = 143MHz (ergo, PC133).

Nobody, to my knowledge, can even manufacture memory that fast. Even the new NVidia cards don't have memory that fast. (1000/4.5ns = 220MHz!)

More than likely, your better Quake performance is due to a more optimized system, better video card, etc, etc, and certainly not due to your main memory (PC133 is faster than PC100). Additionally, the BX chipset on PC100 is a little bit faster than the VIA Apollo Pro133A chipset on PC133 (in some cases). An overclocked BX chipset at 133MHz(!) is incredibly fast.

May be serious.

[Soko]

Thier news page shows they have deals and investment from Intel.

Very cool stuff, especially for laptops. You could selectively power RAM to save a few watts here or there. The only problem I see with a system with nothing but non-volitile RAM is filling the cache inside the processor. Easy to fix I'm sure, but if the cache was non-volitile too, you'd have the entire machine state saved.

Intel got in early - I'm sure Transmeta would be terribly interested in RAM that you can take power away from, and have it maintain state...