A new entrant to the universal-memory contest--Grandis

Santa Clara, Calif.--A universal form of computer memory that can replace all of the different breeds of chips in computers and electronics today--MRAM, Spintronics, , Zettacore, silicon nanocrystals--has been a holy grail for component monkeys for a long, long time. Donovan sang about it, I think.

Several solutions have been proposed, but each one has failed to become a solution to everyone's memory needs for every application.

The latest entrant is Grandis, which has developed a magnetic type of memory chip called Spin-Transfer Torque RAM (STT-RAM to his friends). Grandis has made samples and chips based on its technology will hit the market late next year.

Grandis essentially places a small magnet on top of a transistor and caps it with a layer of sensitive material. An electric current is applied. If the current goes through the magnetic junction in a bottom to top direction, that creates high resistance and that registers as a "1" in the computer. If it runs from left to right, resistance drops and the computer recognizes this as a "0",

"It is the cheapest solution out there for universal memory," said Farhad Tabrizi, CEO during a presentation at the Flash Memory Summit this week. "It is basically a transistor plus a magnet."

As an added bonus, the chips can be made on regular silicon manufacturing lines. It takes two additional steps. The additional equipment required to make STT-RAM runs about $10 million, he said.

Tabrizi, who used to work at Hynix and has been around the memory business for years, added that the company's chips will start competing against flash in a few years and could start replacing flash at 25-nanometer manufacturing, which will come around 2013 or so.

The first application will be for airbags. Grandis chips will replace the SRAM-NOR memory structure in those. It will then try to get into mobile and computing.

Renesas has licensed the technology and other companies are examining it, he said. Sevin Rosen, among other VCs, have invested in the company. Grandis has 30 patents and 50 pending applications.

"Our rates are much more reasonable than Ovonyx," he said. Ovonyx licenses technology for phase-change memory.

If they succeed, consumers and manufacturers would benefit, he said. Consumers would get smaller, faster memory, leading to cheaper devices. Manufacturers, meanwhile, wouldn't have to juggle factory capacity between different types of memory. Just put on the STT-RAM lithography masks and let 'er rip. The ultimate prize--conquest of the memory world--is a longshot. Large companies hate licensing technology, something Rambus can talk about. Magnetic memory has also had some hiccups. MRAM, promoted by Freescale, will likely run out of steam at 65 nanometers, according to Freescale. Competing ideas such as Spintronics from IBM are out there, although the word is that is having some issues.

Nonetheless, the company has a shot.

[David Bondurant]

MRAM Scales Beyond 65 nm

In your recent blog (?A new entrant to the universal memory contest?? ? August 10), you state that ?MRAM?will likely run out of steam at 65 nanometers, according to Freescale.? This statement warrants clarification. During my MRAM roadmap presentation at last week's Flash Summit in Santa Clara , I noted that Freescale has demonstrated MRAM at the 90 nm node and that our data suggests that it will scale to 65 nm. Furthermore, I stated Freescale has evaluated Spin Momentum Transfer (SMT) technology, and we believe that it has potential at 45 nm and beyond. SMT and Spin Torque Transfer (STT) are both forms of MRAM so it is not accurate to conclude that MRAM cannot scale beyond 65 nm. We certainly believe that MRAM has potential when scaled to smaller feature sizes. One of the benefits of MRAM is that it uses magnetic storage (like hard drives) rather than charge storage (like flash memory). MRAM is engineered for unlimited endurance and long-term data retention, two of the issues with flash technology today and potential roadblocks to scaling flash in the future.