Samsung: PRAM to push mobile battery life

Samsung has begun producing a new chip that one day may replace flash memory and that is expected to increase cell phone battery life by more than 20 percent.

Samsung PRAM chip

(Credit: Samsung)

The world's largest maker of memory chips said that it is now manufacturing phase-change random access memory (PRAM) in 512-megabit (Mb) capacities.

Phase change memory has been discussed for decades. Intel co-founder Gordon Moore, for instance, wrote an article about the technology that was published in the September 1970 issue of Electronics magazine. And the basic way the technology works hasn't changed. In phase change memory chips, a medium called chalcogenide--the same stuff as used in CD-RW rewritable disks--gets heated up to very high temperatures, the heat changes the physical state, and the two resulting states become the ones and zeros used by computers for data storage.

PRAM has promise because it can read and write data at lower power than conventional flash memory and single bits can be changed to either 1 or 0 without the need to first erase an entire block of cells--a shortcoming of flash.

Phase change memory is also "executable," which is particularly useful in cell phones for handling application code.

"By using PRAM, the battery life of a handset can be extended over 20 percent," Sei-Jin Kim, vice president of the mobile memory planning and enabling group in the Memory Division at Samsung Electronics, said in a statement. "We expect it to become one of our core memory products in the future."

The 512Mb PRAM chip can erase a small memory segment more than 10 times faster than NOR flash memory. In data segments of 5MB, PRAM can erase and rewrite data approximately seven times faster than NOR flash, Samsung said.

The chip is produced using 60-nanometer manufacturing technology, the same process technology used in NOR flash production today. Finer technology nodes will be applied in future generations of PRAM to accelerate wider commercial adoption, Samsung said.

Market researcher Gartner said in a research note published Monday that it is taking a wait-and-see stance. "Samsung said that the PRAM samples it provided to chipset and phone makers have shown much-better performance than NOR flash," Gartner said. "However, before a final judgment can be made, Gartner is waiting for the reactions of...chipset makers and the first commercial product to confirm the practical advantages that PRAM offers."

Gartner continued. "Samsung has also demonstrated that the power consumption of its mobile DRAM + PRAM is 22 percent lower than that of mobile DRAM only. If Samsung can show such power savings and other benefits in final products...the company will find itself in a commanding position in the memory segment for the entire mobile handset industry."

[ecdfan]

I am sorry, but have you ever switched on a heater? Does it use less or more energy than a stereo?<br /><br />You say that "PRAM has promise because it can read and write data at lower power than conventional flash memory and single bits can be changed to either 1 or 0 without the need to first erase an entire block of cells--a shortcoming of flash." Hello? Melting stuff sounds like low power consumption to you? The truth is that the ACTUAL datasheet for the phase change memory (that of Numonyx, as Samsung's is still "secret") reveals NO SUCH POWER SAVINGS for this 40-yr outdated memory technology. The write process in phase-change is horribly slow (0.3 megabytes per second, based on Numonyx datasheet) and power consuming. Mr. Sei-Jin Kim better provide a public datasheet, so we can verify his claims of 20% savings (which, by the way, he probably meant vs DRAM, as NOR Flash already has the same, if not superior, power savings as far as execute-in-place is concerned).!<br /><br />And please do not quote things that you don't understand, just because Samsung told them to you! Why should we care how quickly the data get erased? We, the users, care about how quickly we can take a picture, transfer an MP3 files, etc. Care to tell us how quickly the Samsung chip does these things (that is, the sustained write and read throughput in megabytes per second)? I thought so. At least Numonyx had the guts to finally reveal the dismal performance characteristics of the chips on which they (and, previously, Intel and STM) had been working for almost 10 years. 0.3 megabytes a second write throughput. Hello again? To save a picture, it would take 30+ seconds. (and just a tad shorter to save an MP3 file). That's much worse than the first digital cameras! The truth is Flash (NAND, NOR, and various combinations of those with DRAM) allow up to 200 megabytes per second write throughput, or 500x better than phase change memory.<br /><br />Phase-change memory is a techno-Ponzi, and you fell for it. Phase change memory will never be commercialized in volume, as it sucks in terms of write speed, density, and cost.<br /><br />You should demand from Gartner and Samsung to see the actual datasheet for this mystery chip (by the way, it looks nothing like the picture of the chip you have in your article, which is an old, 2005, prototype), and then ask them to explain to you the sustained read and write throughput (in MB/s) and the respective power consumption (in mW), and then you should verify the claims by carefully reading the datasheet. Then, maybe, your article will become informative. Because, right now, it is simply a planted piece.

[Hunnter2k3]

Hah, someone woke on the wrong side of the bed today.<br /><br />Boy will your face be red in a year.

[tech_crazy]

@ecdfan<br /><br />I do not know the details of this Samsung phase change RAM but believe your comments about phase change RAM in general are way off. <br /><br />For one, these are miniscule in size and the on time is in fraction of seconds, so the power issue is negligible. e.g. Have you ever touched a CD-RW or DVD-RW after "burning" to feel the "heat"? As for the speed, you get 16xDVD-RWs (21 MB/s) that are made from the same material. As if that is not enough, the bits are written serially. If there is a parallel interface, it would be even faster.

[Seaspray0]

@ecdfan. "We, the users, care about how quickly we can take a picture, transfer an MP3 files, etc. Care to tell us how quickly the Samsung chip does these things (that is, the sustained write and read throughput in megabytes per second)?" <br /> <br /> <br />From the article... "In data segments of 5MB, PRAM can erase and rewrite data approximately seven times faster than NOR flash, Samsung said." <br /> <br />It seems that Gartner has provided an answer to your question (7 times faster) and it doesn't match your claim of "it sucks in terms of write speed." The average digtal camera uses aproximately 1 MB to save a picture in JPG format, so 5 MB isn't out of the balpark by a long shot. The data you provided as 0.3 megabytes/sec comes from Numonyx and previously Intel and STM. When was this speed reported? Your knowledge on the subject suggests you have followed this in the past, so my intention isn't to question whether you were right at one time. It's just that things may have changed recently and you are not aware of those changes (which would explain the difference of opinion between you and the story). Any clarification by you or the Gartner would be appreciated. How about the datasheet?

[richard993]

It does not take much to see that there is validity to these claims. A quick glance on some of Samsung's private research, some of which is already available from the US Trademark and Patent Office (search for Samsung and PRAM) shows some rather promising techniques which are very different to previous approaches to PRAM. Application no 10/827,687 indicates that normal PRAM is slow to change state and that Samsung has developed methods to increase the speed significantly by reducing the surface area between the electrodes through the use of different compounds to create finer pores. The problem is that the author of the article did not include additional research into the topic hence the skepticism that follows. A poorly written article for such a significant breakthrough.

[ecdfan]

To tech_crazy: Sorry, but melting with a laser (and having plenty of moving air around to cool) is one thing. Melting with a heater within a chip is something else. You better learn how PRAM works before making such funny statements about performance.<br /><br />To Seaspray0: 7x times? So, what is that in megabytes per second? Can you tell? There are all kinds of NOR - some are slow like a molasses, some can write at 2 megabytes per second. Which one was Samsung talking about? The fact is, Samsung is too shy to publish a datasheet. The most likely reason - the performance sucks. Samsung needs to come clean!<br /><br />Numonyx was too shy as well, but they were finally forced to publish a datasheet this Summer - they could not keep it a secret anymore - their chip P8P, formerly called Alverstone, was revealed to suck big time, once the datasheet came out.<br /><br />To richard993: Research is one thing. A public datasheet and a chip that anyone could buy and test is another. Samsung has had their 512Mbit (90nm) prototype since 2005. But they have failed to publicly disclose performance data. Samsung does not always make all things right - just note their current performance issues with the 32nm NAND.

[luke_marsh]

ecdfan Yes there may be a lot of problems but you have to consider down at the nano-scale when you want to alter the state of something one option is to heat it. We are of course talking long term storage not Ram. Think about it this way the smaller you make an electric device the greater the chances of leakage if something is small and holds its state in solid form and not charged gate form then the leakage issue doesn't exist. Now this idea of PRAM might have been 40 years to early but that only time will tell If Samsung are making the right moves and to be honest it shows guts for Samsung to be so bold as not to work with the standards at consumer level maybe their guts will drive more types of storage beyond that standard storage types of today from others.<br />40 years late but at least their doing it and are ready for it now.

[ecdfan]

luke_marsh: Bubble memory (http://en.wikipedia.org/wiki/Bubble_memory) was another bold choice. Let us know how that turned out! The truth is, phase change memory suffers from similar problems - slow (write) throughput, low density, so-so reliability, high cost. The outcome is inevitable - it will never be mass produced.