Spansion sees profits in flash memory's future

A lot of companies have proposed future technologies to lead the flash memory market. But which company will ship the first products?

Spansion, created from a flash-memory joint venture started by AMD and Fujitsu, says that the "charge-trapping" technology that underlies its MirrorBit and Ornand flash chips provides a path for the industry to continue to shrink the size — and hence increase the performance — of flash chips.

Spansion already makes flash chips that rely on charge trapping, which is similar but slightly different to conventional "floating gate" flash. Now it has begun a campaign to license some of its ideas to other manufacturers, Spansion chief executive Bertrand Cambou has said.

"Samsung, Toshiba, and Hynix have all announced publicly that they have a strong interest in charge trapping," Cambou said. "We have some fundamental patents for it. It is not an easy technology... We have been working on it for years."

Charge trapping, moreover, can be used in NOR flash chips, a smaller market historically dominated by Intel and Spansion, and NAND flash chips, found inside phones and MP3 players. Although AMD started making charge-trapping chips back in 2002, a lot of the basic technology comes from Israel's Saifun Semiconductors, which Spansion bought in October for $328m (£159m).

If Spansion can pull this off, it could turn into a financial boost for the company, which chronically suffers financial losses. In the third quarter, Spansion reported a $72m net loss on revenue of $611m. Chipmakers typically are loath to license technology, particularly from their direct competitors, but it occurs. Toshiba has generated significant revenue from its technology formulas for NAND.

NAND patents are valuable because of the still-growing demand for consumer electronics and phones. Although NAND prices swing up and down, NAND volumes continue to increase annually. In fact, many chipmakers now upgrade their factories to accommodate NAND volumes first. In the past, many launched new factories with DRAM or other chips. Some NAND makers are now trying to push their products into notebooks and blade servers, displacing hard drives.

"NAND is the killer app," Mike Splinter, chief executive of Applied Materials, which makes semiconductor equipment, said in a recent interview.

Most executives and scientists in the booming flash memory market say that the technology is headed toward a Moore's Law-style crisis. Simply put, manufacturers won't be able to continue to shrink their chips without major architectural changes. Without shrinking, chipmakers can't cut costs, drive up performance or increase volumes with the same underlying economics as the past.

So what is charge trapping? The memory cells in conventional floating-gate flash consist of a sandwich of materials: polysilicon is encased between two layers of silicon dioxide, or glass. Once an electron is trapped in a cell, it will stay inside for 10 years, which is why your pictures stay on memory cards until you deliberately erase them.

By contrast, in charge trapping, the middle section consists of an oxide-nitride-oxide layer. Charge trapping is also known in the industry as "Nitride", or "ONO".

"Once you get down to a certain cell size, you have real problems with conventional flash," said Jim Handy, president of analysis firm Objective Analysis. "It seems like at 45 or 35nm [nanometres] conventional flash starts to run into a limit. They won't be able to use the same materials."

In the next one to two years, 45nm flash chips are due. Some types of non-volatile memory already having shrinkage problems. MRAM, a type of memory supported by Freescale Semiconductor and IBM, may not get beyond the 65nm processes in widespread use today.

Many approaches
The big problem for Spansion will lie in getting the industry to go along with it. Everyone has their own ideas for the future of memory and, at this point, it's far from clear which will survive. Some of the competing ideas include Toshiba's 3D memory chips; the phase-change chips from Numonyx, a joint venture between Intel and STMicroelectronics; and STT-RAM, a sort of miniature magnetic memory from Grandis.

Charge trapping has two significant advantages over those listed above, Cambou asserted. Firstly, people already know how to make it, which removes some of the risk. Secondly, lab tests show that charge-trapping memory can be shrunk.

"We have [laboratory] cells that are functional at 20nm," Cambou said. "If we can shrink it, we can be the cheapest." The figures refer to the average feature size on chips. A nanometre is a billionth of a meter. The smaller the features, the better the performance and cheaper the cost, generally.

Still, manufacturing of 20nm memory isn't expected to happen until 2012 or later.

Charge-trapping memory cells, Cambou added, are also small compared with other types of memory cells on the same manufacturing dimensions, which further cuts costs.

"If your cell size is big, that is a niche," Cambou said. "Charge trapping is going to be the big technology in the next 10 years."

As for phase-change memory, Cambou stated: "In 2016 maybe it is a good idea, but it will take 10 years before it gets cheap." Some phase-change advocates say that the memory could start popping out in the next few years. Then again, the concept of phase-change memory has been around since the early 1970s.

Handy agrees, to a point. Charge-trapping will probably be easier to adopt than phase-change or rewritable-3D memory. It's more of an evolutionary step than a rewrite of the cookbook. Big companies have also expressed quite a bit of interest, but whether they will have to pay Spansion is an open question.

"Everybody says ONO is the way we will have to go. Samsung has said they are going to use ONO. The question is whether they are going to be tripping over the [Spansion] patent portfolio," Cambou said.

Ironically, Spansion's own experience with Saifun may be an indicator that licensing could be in the cards. When it was an independent company, Saifun tried to license the technology to AMD. AMD came out with its MirrorBit memory. A lawsuit ensued that later settled.

Spansion will, of course, also continue to make chips and increase the amount of manufacturing it outsources to foundries. The company has signed deals with Taiwan Semiconductor Manufacturing Company, as well as China's SMIC. Currently, foundries make about half of Spansion's chips, Cambou said. Spansion's portion will ultimately drop to 40 or 30 percent.

In these arrangements, Spansion will make the chips requiring the most advanced processes, while the foundries will make chips on manufacturing equipment that is one generation older.