PCM: If Not Now, When?

If you are talking MLC (e.g., 2 bits), PCM has NO retention. As said earlier, the drift mixes up the states, within an hour. It's a problem for PCM, as its competitors can do MLC.

What's the PCM retention at 100 C. Not 200K years. Not even 10 years. At 150 C, try 10 minutes.

1. current density exceeding 10 million amps per sq. cm. even spin-based MRAM uses 10x less current density. 2. resistance is not stable, keeps increasing at varying rates after it's been programmed. 3. GeSbTe forms troubled interfaces with most materials in use today. When it oxidizes it becomes a different phase change material.

Let me guess: Aha, so Enron was not a fraud. And neither was Madoff. And Lehman did not make stupid investments with money they did not have. Is that it?

Todor Mitev: Well, why stop here! Tell us more.

Are you crazy: Madoff financial Ponzi scheme operated for 30 years. The PCM techno-Ponzi has been in "usage" for just a bit longer. PCM has always existed ONLY in nice looking presentations and/or press releases authored by the perpetrators and articles written by gullible people with no tech common sense. But, hey, since you are so sane, you are welcome to point me to at least one product that uses (or have used) PCM chips. Can you?

He's short on ECD stock ? Give me a break. It guys like you that ran the market into the ground. Get a life and try to contribute to society you leach.

For more on ECD.Fan, contact Todor Mitev. Todor Mitev Temujin Fund 140 Broadway, 38th Floor New York, NY 10005 212-299-5446 tdm@thales.com, or tdm@temujin.com He may know ECD fan personally. He might even see him daily.

You say that PCM has a "techno ponzi" and then you say "it's been in usage since the 1960's. Clearly a sign of insanity. Are you insane? Isn't Numonyx a private company anyway? If what you are saying is true, who are they trying to scam, themselves? This is hilarious!! You make no sense at all

nvm.expert@gmail.com: If Alverstone is a NOR replacement, why is the author of this article telling us fairy tales about using PCM for storage of digital pictures (one picture per chip)? Well, did you sign an NDA for the 1Gb datasheet? If you didn't, why don't you tell us where we can get it. If you did, what are you talking about? Does a 1Gb sample exist? Maybe you have seen that too? Let us know! In the mean time, surely some products must be using this PCM thing. After all, PCM has been commercial since 2006 (BAE) and 2008 (Numonyx). Not even one product uses these things? How come? PCM is a techno-Ponzi. Its only purpose is to create new ways of siphoning money out of clueless top management and investors. And it is not a new usage - that has been the usage since the 1960s. Now, when Flash can't scale anymore, maybe you will sound more convincing. In the mean time, Flash is readily available in 32Gb densities at 34nm, and DRAM is plentiful, and NOR looks more and more redundant. As I said, so much for an NVM expert.

"This ECD Fan is an idiot:" Does PCM lose data or not? NASA did not find errors in other chips. They found errors in PCM. You have done testing? What testing have you done? Care to share the details?

PCM fan: Yes, the glimpse of the future. The 1970 future. The future that never happened and never will. Activation energy? When these things change state spontaneously due to impurities, then it will be fun to see how long the picture (remember, one chip per picture is the limit!) lasts. But, hey, it is your money, feel free to do whatever you want with it.

Steve Leibson: What early PCM parts? Didn't the September 1970 issue of Electronics show the early part? So you haven't read the datasheet yet? Well, why don't you try. It is fun! It throws Mr. Doller's Nirvana out of the window on the first few pages. And then it gets worse. Promoting products you don't understand and you know nothing about must be also fun these days...

Nothing I've seen seriously suggests that PCM is going to replace DRAM. Something probably will some day because DRAM is reaching the end of the scaling road. Maybe it'll be memristors? Who knows. Replacement won't happen tomorrow or next year, that's pretty clear. Meanwhile, I'll be discussing the memory hierarchy here on the blog in a day or two. It's my last post (I think) from last week's terrific Flash Memory Summit.

Did you read what I said? Obviously not. Their 128Mb product is a NOR replacement and has NOR features which include slow write performance. It's a good start. Their 1Gb data sheet, WHICH I HAVE SEEN, shows 5-6MB/ sec. Let me spell that out so you understand - five to six MEGA BYTES per second. Have you seen the data sheet? Obviously not. You are right, it's certainly not DRAM but it's fast enough for many applications that value latency and non-volatility. And who says PCM will "replace NOR, NAND or DRAM"?? My guess is PCM will create new usages and complement the above memory technologies until the others fail to scale. Do you work for their competition? oh... you do.

Are you for real? What does JPL's radiation testing of BAE's PCM have to do with intrinsic retention of PCM? Don't' you know that most radiation failures are due to periphery failures then the cell itself? The same holds true for DRAM, NOR and NAND... I've done the testing myself!! Have you? Get a life you moron.

ECD Fan: Did you not read the post above??? 200K years at 35C is intrinsically reasonable to anyone who understands the material set - it's thermally activated!! As Steve stated, this by no means implies at 100C anything unless you understand the activation energy. I'm actually surprised it's only 200K years at 35C as I would have expected it to be higher, yes higher. Why? Because the literature states the activation energy is roughly 3 and if the retention at 35C is "200K years", this means the retention at 70C is ~2 years. Once again Mr ECD Fan, you have proven your ignorance... I think Steve hit it right on the head you are a shill and a bad one at that. Thanks for the great article Steve... I love your work and this was a great glimpse into the future. I hope they are successful.

ECD Fan: Not reading the posts too carefully, are you? I said in my post I didn't expect the chip's metallization to last 200,000 years. That ought to tell you something about how strongly I think the chips will be around that long. Doller's claim is based on an activation-energy calculation for the glass bit cell. The last line in your last comment again indicates that you are a shill (look it up on Wikipedia)and your demeanor strongly suggests you're a troll. Looking at data sheets for today's early PCM parts will tell me exactly what about the future of the technology? Bye.

To Steve Leibson: Really? Your BS filter did not flag the "200,000 years" thing? Shill or no shill, the datasheet tells the whole story. But I guess you are above reading datasheets these days. Reading nicely prepared "reports" and "presentations" is much easier. That is exactly why techno-Ponzi schemes like this work so successfully for such extended periods. And why OUM aka PCM aka PRAM is still considered a serious NVM contender. By some.

To nvm.expert@gmail.com: Really? 200K years at 35C? But maybe just 3 years at 36C? Have you actually seen a PCM chip? Have you kept it for 3 years so you can verify? Or do you rely on the statements of a perpetrator of a techno-Ponzi, who has claimed PCM gets "pretty close to Nirvana," and that "mass production could begin before the end of 2007." Well, NASA did some tests of PCM (the rad-hard version by BAE). You know what they found: "JPL has experienced one error in six BAE CRAM devices tested. The cause of the error has not been determined; recall that the devices checked were engineering evaluation samples provided as a courtesy to JPL and were unscreened at the factory." How can you be happy when valuable resources are diverted to a tecno-Ponzi that has basically no chance of ever getting commercialized in volume? Low latency is good, but writing at 0.3 megabytes per second and at 66 MHz is no good. DRAM? Do you know what DRAM's write throughput is? Well, here are data for you: a low-end netbook can do it at 4,000 megabytes a second. Put a tiny li-ion battery and you have a NVM "solution." The truth is, PCM cannot be a replacement for NOR, NAND, or DRAM. The combination of its performance, density, and cost makes it impossible. Another bubble memory. Nothing more. So much for nvm.expert!

Just wait until tomorrow's blog post on NAND Flash read disturb problems. Then you'll really think I've sold out. However, the truth is, I provide you with the information that gets past my BS filters. You've got to make all the puzzle pieces fit. Mr. ECD Fan hides behind his nom-de-Net and throws FUD on Numonyx while his handle suggests he's a shill for a competitor. As for the statement that NAND never lost one of my pictures, that's just not true. If you haven't had the pleasure of a USB drive failing, it's only a matter of time. MOS-incompatible, your statement is strange considering that PCM vendors that plan to succeed aren't going to make you drive the PCM array directly. You'll never shape the write pulses properly from outside of the chip. You can bet they'll have CMOS-compatible inputs or outputs or they won't be in business long. Ditto if their accelerated life testing shows that the memory array potato chips on the silicon. Don't the PCM vendors have enough technical problems making their stuff manufacturable and competitive without you guys creating more problems from pure vapor?

I don't think a standard MOS transistor can drive a PCM without N+M generation lithography. I also wonder if the thermal stresses will cause the structure to peel apart?

Did you attend the FMS? I did and His presentation at FMS showed data retention at 200K years at 35C (I repeat 35C) - not bad if you're looking to archive data. I believe their 128Mb device is a NOR focused replacement which we all know has inherently slow write rates. I'm amazed that you seem to point out all the negatives when in act we should all be happy someone is investing in technology other then NAND and DRAM. I have seen their data sheet for their 1Gb device and while I don't recall the exact write rate, I believe it's in excess of 6MB/sec. Lastly, I'm not sure why you are comparing it to NAND? It's not a "data device" per se. Sure it can be used to store data but it's inherent value is latency, like DRAM. You claim to be an ECD fan... I'd highly suggest you get your head out of your ass and speak with data.

Somebody better teach Mr. Doller how to calculate, before he ruins that fine company, Numonyx. Data retention for 200,000 years? Not when the chip is anywhere near a Pentium (which gets hot, 100 degrees centigrade hot). According to the public datasheet for the so-called Alverstone, "at the current time, PCM technology appears to have a write cycling endurance better than that of NAND or NOR flash, but less than that of RAM" and "[although] the device reliability across the operating temperature range is typically superior to that of floating gate flash, the P8P device may be subject to thermally-activated disturbs at higher temperatures." Max operating temperature is 70 degrees C, according to the datasheet. Beyond that, the data may start evaporating. Numonyx will gladly "sell" chips to people like the author. He will have to buy a chip for each picture he stores (as Alverstone's capacity is just 16 megabytes or 128Mbits). So for 100 pictures, he will need 100 chips. And he will have to wait almost a minute to write each picture to the chip and about 1 and 1/2 hours for the 100 pictures (as this thing writes at 0.3Mbytes per second, according to the datasheet). Of course, Numonyx has to learn first how to make these chips at reasonable yield. PCM is a techno-Ponzi. It will likely never be commercialized in volume. It is too slow in write, has poor density, and is too costly.

Are you always this anxious to pay double for a product you have never seen working - anywhere? If so I will gladly solve all your health care needs just as soon as you send me a check for twice what you are now paying. cb