cieonline.co.uk
21 January 2008 | 10.43:09 am The leading information resource for electronics industry professionals
Article January 2008
Life after flash
Freescale's 56F8000
Could flash memory be coming to an end as shrinking geometries make the technology less reliable? Steve Rogerson looks at the alternatives.
As geometry sizes shrink, flash memory is reaching a barrier and the race is on among developers to find the next technology that can take standalone and embedded storage forward. And the problem is with us now as chip companies take geometries below 65nm; flash has serious problems at 45nm and the feeling is that it has no long-term future.
Putting this into perspective, if flash will not go to 45nm, it still has at least five to ten years of life in higher geometries, and probably more, but given the time it would take to move a memory technology from the design board to being a commodity, companies are looking at alternatives, and there are some front runners.
One is ferroelectric RAM, or FRAM. This fits well with current semiconductor processes as it just involves adding a couple of mask steps. And its proponents claim that it would have no problem scaling down to 45nm and below, though that has yet to be put to the test. Ramtron, the main company behind the technology, has produced products at 130nm, but hasn’t dipped into the 65nm market yet, let alone 45nm.
“We don’t tend to start a new memory technology on the latest processors but on tried and tested and reliable technologies,” explained Duncan Bennett, Ramtron’s strategic marketing manager. “But we don’t see a problem taking it down to new processes.”
The company though is not rushing to try to reduce the geometries until it is sure there is a market for it and that it would be economical to produce, but Bennett said that there was no reason why it would have to follow the normal steps.
“We don’t have to come down to 90nm,” he said. “We can miss that and go straight to 65nm, but it would be at least five to ten years before we would be looking at 45nm. We could go straight to 45nm if it was a stable and cost effective process, but the processes are not cost effective yet and I couldn’t say when they will be.”
New kid on the block
FRAM is not new – Ramatron has been round for about 20 years pushing the technology – but the new kid on the block is Freescale’s magnetoresistive RAM (MRAM). The first commercial MRAMs went into production in 2006 but though it shows promise, industry pundits believe it has at least five to ten years development before it could be classed as a commodity product.
Its main advantage is that it is fast, faster than standard flash. And Freescale says that it has unlimited endurance, a claim that is not totally accepted in the market.
“They claim it has infinite endurance, but we will see if it turns out to be true,” said Doug Mitchell, senior marketing manager at Cypress Semiconductor. “But there is a lot of reluctance to make a commitment to it because there is no real infrastructure.”
“MRAM is fairly interesting in that they can achieve high densities and be non volatile,” added Michael Ching, senior marketing manager at Rambus. “It’s a technology that requires more development work before it can be in the mass market.”
Freescale admits freely that this is the case. Andreas Wild, Freescale’s director of research, said: “It is not a commodity product at this point in time. For one thing, there is only one supplier. We need to find the application that will take advantage of MRAM in an innovative way so that a volume market can be guaranteed. It will be several years before we have a commodity product, maybe three to five years.”
One problem MRAM has is that the power consumption is quite high compared with other memory technologies because of the current needed to generate the magnetic fields. While development work continues to reduce this power consumption, it does mean that the high volume mobile and battery-powered sectors are outside its target market.
“MRAM is not the preferred solution for portable applications,” said Wild, “but that will change.”
Another criticism of MRAM is that it is susceptible to magnetic fields but Wild pointed out that that the technology is the same as used on traditional discs, portable music players and credit-card magnetic strips.
“There is no practical difficulty with this technology,” he said. “There are screen layers to shield the magnetic field and if it is needed in an application with a high magnetic field, there are other ways to shield it.”
As to the unlimited endurance, Wild pointed out that everything deteriorated eventually but he said that MRAM had no degradation mechanism. Signs of degradation, he said, are visible years before a device stops functioning.
“We have run this through 10*15* to 10*16* cycles and have seen no signs of a degradation mechanism,” he said. “Therefore we can say unlimited endurance.”
But Bennett said: “They say there is no evidence of it wearing out, though some disagree with that. How can you prove it? Flash can be written to 5000 to 25,000 times, with EEPROM it is a million times. But when you are talking 10*18* writing to a memory at 10MHz would take many years. So it is hard to prove that it really has unlimited endurance.”
He said that with FRAM on older technologies, the endurance cycle was of the order of 10*16* or 10*18*, but on 130nm the firm only claims 10*14*.
“We think it is 10*15* or 10*16* but it hasn’t been around long enough to test,” he said.
Alternative to flash…..
Another alternative to flash is flash, well flash done a different way. Traditionally, flash uses a floating gate process with a piece of polysilicon that stores the charge sandwiched between the oxides, and it is this process that is having problems scaling down. But flash can also be made using a process called SONOS (silicon oxide nitride oxide silicon), which describes its composition. Instead of polysilicon, it uses nitride as the storage element, and the physics of this process scale down better than with floating gate.
As an added advantage, SONOS is more radiation tolerant and as such has been used in high reliability applications for many years but production problems have stopped it being considered for commercial applications. However, those production problems have mostly been solved and given the problems with floating point flash, the major players are again taking an interest in the technology.
“We believe SONOS will go into 45nm and below,” said Mitchell. “It is also more compatible with standard CMOS. Floating gate flash takes ten to 12 additional steps, with SONOS it only takes four additional steps.”
To find why SONOS is thus not more accepted means going back 25 years when it required a processing step that needed very tight controls, meaning it couldn’t be made very reliably in high volume. The yield was too low, and thus the product too expensive. For high reliability applications such as in space, cost was not as important and thus its radiation tolerant properties earned it a niche market.
However, advances in DRAM that uses similar processing technologies have solved what were the yield problems with SONOS and that is why some wafer fabs today have some SONOS production, including a few of the traditional large south-east Asian subcontractors.
“Over the next four to five years, we will see a transition from floating gate to SONOS,” predicted Mitchell. “It will also become standard for new fabs.”
Whatever technology is taken up, the next few years looks interesting as the major players battle to have their technologies accepted. And some pundits still believe that a bit more can be squeezed out of traditional flash.
Ramtron
www.ramtron.com
Freescale
www.freescale.com
|
