Next-Generation Memory Faster, but Slow to Take Off
The chip manufacturing community has long dreaded the day when flash memory could no longer shrink in size. But that day continues to be pushed back as chip manufacturers are seemingly slowing their commitment to invest in what many believe will be the primary replacement for flash: phase change memory (PCM), also known as phase change random access memory (PRAM).
Jim Handy, analyst with Objective Analysis in Los Gatos, Calif., doesn't believe the conversion from flash to PCM will gain momentum until 2012. Then he would expect PCM to overtake flash within two years.
"As recently as 2002, Intel was very openly stating that after 65 nanometers nodes, you wouldn't be able to manufacture flash using smaller and smaller processes," said Handy. "In 2003, they found they can burst through that wall and get as far as 25 nanometers. Now it looks like 25 nanometers is the brick wall. But I am of the opinion that they might have another breakthrough between now and 2012 and that will push the brick wall."
But once the brick wall is hit--and the flash technology can no longer be shrunk--he said, "That means you can no longer continue to reduce the cost of it. And all of a sudden it becomes a dinosaur... Until that point, manufacturers will dabble at [PRAM]."
"You can devise ways to turn the crank one more time [with NOR flash memory], but it's generally acknowldeged that it is getting harder to do that. Phase-change memory has better scalability than NOR," said Bill Gallagher, senior manager of exploratory nonvolatile memories at IBM.
IBM, Qimonda AG and Macronix International Co. announced earlier this year that they have entered into an R&D partnership to develop PCM technology, but Ghallager said the three companies are not developing a specific product together. "We showed that the material could be made very small and that the technology could scale for many generations, and that this particular material is very fast--flash is a particularly Slow material to write and to erase," he said.
Gallagher said that if IBM manufactures a PCM technology, it would likely be in the form of an embedded chip for its own products rather than a stand-alone product to be sold through OEMs, but he would not say if there was any time frame for such a product.
Recently, the pace of PCM development appeared to be picking up steam with announcements from two major chipmakers that products will start surfacing in 2008.
PCM is a form of nonvolatile memory that can use electrical charges to change an alloy (germanium-antimony in most cases) from a crystalline state to an amorphous state. PRAM promises, to be more than 500 times faster than flash memory and cheaper, as well as consuming less power, than other forms of non-volatile memory. (see "QuickStudy: Phase-Change Memory").
"[PCM] inherently blends the best attributes of the two types of flash, NOR and NAND, high performance read and high performance write" said Greg Komoto, Intel's manager of strategic planning for the flash memory group.
That type of performance could bring about some important changes in the way systems are designed and developed, he said. "The software has the potential for becoming simpler, more efficient, and therefore, to the user, more responsive."
Building the Chip
But first the chips have to come onto the market. Intel's first demonstration of PCM took place at the 2006 Developer Forum last October, where the company showed a 128Mbit chip built at one of STMicroelectronics Inc.'s plants in Italy.
Since then, the two firms have announced spin-off ventures, which are still in formation. Numonyx, a play on the word, mnemonics (assisting the memory), received approval earlier this week by the European Union. This company is being formed with the money-losing NOR flash division assets of Intel and flash assets of STMicroelectronics (see "Intel and STMicroelectronics combine flash memory").
A third investor, Francisco Partners in Menlo Park, Calif., will receive a share in the company in return for a cash investment. Numonyx is expected to employ 8,000 people transferred from both Intel and STMicroelectronics
Handy believes the timing for that initial demonstration was driven by another products demonstration--512Mbit PCM prototype that had taken place just weeks earlier by Samsung Corp. A press release from Samsung at the time declared PCM "the main memory device to replace high density NOR flash within the next decade." There was no mention of NAND, for which Samsung represents 35% of global wafer output.
Shortly after those demonstrations, in December, a team led by IBM researchers showed a 3 by 20 nanometer prototype that was said to run 500 times faster than current flash memory while using less than half the power to write data to memory. A paper was presented on the findings in February 2007. Since then no additional announcements have come out of IBM on the topic of PCM.
When Could PCM Dominate?
Handy believes the applications for PCM are the same ones that flash has transformed, cell phones, cameras, audio devices. "A lot of applications will couple flash memory and an SRAM or a DRAM [chip] together," he said. "Cell phones are a good example of that. Flash RAM is usually used to store the software that makes the phone work in the first place. Then the DRAM or SRAM is used for data manipulation... One of the promises of PRAM is that it can be written to very quickly. And it can be written to a limitless number of times. So it can be used as a substitute for the DRAM or SRAM in the system. That allows you to go from two chips to one chip."
The move from flash to PCM will require investment in new manufacturing equipment, but Handy believes this restructuring could be slipstreamed. "Whenever semiconductor manufacturers move from 70 nanometers to 55 to 45, they buy lots of new equipment anyway," he said. "The new equipment used for putting on the PRAM layer is likely to be something that's absorbed into the new equipment buys that they do."
Currently, the same chip companies that have announced PCM offerings for 2008 are also investing heavily in 300mm facilities. Samsung dedicated its $220 million 300mm fabrication plant in Austin in June.
However, Jim Elliot, director of NAND flash marketing for Samsung, said the Austin facility "is an investment in flash, not with PRAM around the corner. We're very bullish on the prospects for NAND." He declined to predict if or when the company would commit to a PCM product strategy.
Numonyx, which is expected to open its Switzerland-based headquarters in this year, has a new 300mm fabrication plant in Catania, Italy, which, according to STMicroelectronic's web site, "is ready to be equipped."
Spansion Inc., a NOR flash company based in Sunnyvale, Calif., bandied about in PCM discussions but has made no public announcements to date, is scheduled to begin operations in its new 300-mm fab plant in Japan later this year. [How do we know they were considering PCM manufacturing?]
Whether those 300mm chip plants will eventually be fabricating wafers for flash or for PCM or both, no one would say officially.
But as Komoto--currently with Intel but soon to be with Numonyx--pointed out, "If we're wildly successful, we'll be increasing our capacity to meet demand. But really the beauty of the technology is that it doesn't drive any unique capital; it's highly synergistic with existing manufacturing equipment. That's why you see a lot of interest from major memory manufacturers in this technology. It's not using anything exotic."
Dian Schaffhauser reports on technology and business. Reach her at firstname.lastname@example.org.