Pushing the PRAM: when chips just can't get any smaller

Crammed with cells, memory chips have become about as small as they can physically get, and manufacturers are now experimenting with new technologies likely to shake up a $50 billion industry. Existing chips are expected to hit a physical limit within the next five years, triggering a race among chipmakers to find the technology to eventually supplant NAND flash and dynamic random access memory (DRAM), the two staples that have driven the computer world.

“It’s opening up a whole new window of possibilities that for scientists is incredibly exciting and for business is incredibly frightening,” says Gary Bronner, Vice President of Rambus Labs, a technology licensing company that specializes in memory. “People don’t like uncertainty and change.”

More from News & Analysis

What is the US HIRE Bill and why is India’s $250-billion IT sector worried? Is the internet dead? What's this theory that OpenAI's Sam Altman says might be true?

Shrinking sizes of memory chips

That change won’t happen overnight. Though some companies are already shipping products with new technologies, the quantities are small, and the early promise of some announcements often gives way to silence or scaled-back plans. And leading manufacturers are hedging their bets by investing in the new technologies alongside the old.

“There’s a lot of excitement about new technologies, but you need a dose of reality, too,” said Al Fazio, Director of memory technology development at U.S. firm Intel Corp , which no longer manufactures memory, but is a key player in the industry.

For now, every device relies on RAM and either flash or traditional hard drives as it juggles data and applications between processor and storage. The price and performance of these technologies dictate which are used and in what quantity.

For example, SanDisk Corp founder and then CEO Eli Harari was unable to meet Steve Jobs’ request in 2001 to provide flash memory for Apple’s first iPod, because he couldn’t compete on price with mini hard drives. Within four years, however, flash was cheap and small enough for Apple to design the tiny iPod Nano - and prices have fallen so much since that all phones, tablets and even lightweight laptops use flash.

Its low price and increasing capacity have revolutionized the music, camera and film industries and helped make the smartphone the world’s most popular device.

This demand has helped flash grow into a $21 billion industry last year, according to market researcher IHS iSuppli, double what it was worth in 2005. By 2016, revenues are expected to rise to $31 billion. And all these devices also use DRAM, an industry which was worth $30 billion last year.

But within this success lies the problem.

As manufacturers squeeze more memory into a smaller space they are close to the limits of what is physically possible with so few electrons to play with. Hence the experiments, by main manufacturers and smaller companies, with new technologies - ones that rely not on electrical charge to store data but on changing the structure of materials for data storage. At least in theory, this should make them more scalable.

Few of the underlying technologies are particularly new.

MRAM, PRAM and ReRAM

Magnetoresistive RAM (MRAM) was originally developed by Honeywell for satellites and the military. A later version of the technology was adopted by Chandler, Arizona-based Everspin Technologies, the first, and so far only, commercial MRAM maker.

More promising is a version of MRAM called Spin Torque Transfer MRAM (STT-MRAM), which packs the memory together more densely and is as fast as DRAM.

Samsung Electronics has shown an interest in this area, last year buying U.S.-based Grandis Inc, but there’s not yet been any evidence of products, says Yiran Chen, assistant professor at the University of Pittsburgh and a former memory designer for Seagate LLC. Last year Toshiba said it was working with South Korean rival SK Hynix, and analysts say Toshiba hopes to use STT-MRAM in some personal computers next year.

A competing technology is phase-change memory (PCM, or sometimes PCRAM or PRAM) that uses the unique property of so-called chalcogenide glass, which is already used in recordable CDs, to switch it between two states. PCM can store a lot of data but is slower than DRAM, so it could be used in combination with other technologies or as a possible replacement for flash.

In April 2010, Samsung trumpeted that PCM would be “widely embraced by next year as the successor to NOR flash in consumer electronics designs to become a major memory technology.” A few months later, it quietly inserted PCM memory in at least one brand of cellphone, and just as quietly removed it.

Micron Technology , the oldest surviving high-volume memory company, is also a key player in PCM, having bought Intel spin-off Numonyx in 2010.

The third main memory technology is resistive RAM (ReRAM), which works by using an electrical charge to form and dissolve a filament between two electrodes. ReRAM has its drawbacks - its cells wear out quicker than MRAM - but it should, in theory, read and write 1,000 times faster than flash.

Micron’s interest in ReRAM stretches back to 2002 when it acquired technology from Axon Technologies, one of the first players in resistive memory cells. Toshiba and SanDisk are working together on the technology, with SanDisk founder Harari predicting a version of ReRAM will eventually replace flash.

Troubled Japanese manufacturer Elpida in January announced what it called the world’s first ReRAM prototype - and it’s this technology that is interesting Micron, currently in talks to buy Elpida.

Hewlett-Packard has been collaborating with SK Hynix on a version of ReRAM and has been quoted as promising a product next year. An HP spokesman acknowledged there was some skepticism about the company’s ability to fulfill its pledges to bring what it calls memristor-based products to market, but said it retained its faith in resistive RAM. “We continue to expect that this technology will dramatically improve the performance of computer storage in terms of price, energy efficiency and capacity,” the spokesman said.

FINDING A MARKET

The first step for any of these technologies is to get a toehold in the market, however obscure: flash, for example, started out by replacing cassette tape in telephone answer machines in the 1990s. “I_t wasn’t particularly cheap, but people had figured out how to use it_,” said Rambus’ Bronner. “Flash memory took off from there.”

Likely uses for a ReRAM microcontroller announced last month by Panasonic Corp, for example, include fire alarms and televisions, which would benefit from its low power consumption. Buffalo Inc., a Japanese unit of Melco Holdings Inc., last month launched a line of solid state drives using MRAM as a cache. “You’d like to build the whole drive out of MRAM but that’s prohibitive, so they’ve created a cache that’s impervious to power loss,” said Steve Leibson, who writes a blog called the Denali Memory Report.

Niches needn’t be small devices. Pantelis Alexopoulos, Executive Director of Singapore’s Data Storage Institute, said the government-owned research centre plans to use these emerging memory technologies to reduce the amount of hard drive activity - and power consumption - in data centers. Such moves may upend the economics of an industry which consumed 31 gigawatts of power last year, according to DatacenterDynamics.

But the more likely long-term impact of the new memory technologies, analysts say, is when those who design applications and devices find a new way to use them.

Today, the short-term memory of DRAM is separate from the permanent storage of flash or hard drives. The rise of non-volatile memory technologies will make this redundant. Instead, something experts call “universal memory” will replace both, making for much faster devices as data and applications won’t have to shift between them on their way to and from the processor.

That will come, said Leibson, when these technologies start to find a niche which makes them cheap enough for systems designers to experiment with. “It will enable new types of systems that we cannot currently build and there will be people smart enough to figure out how to build it,” he said.

Reuters