Moore's Law and memory to hit physics wall
'There's been a resurgence in terms of pessimism ... in the last few months.'
By Scott Hillis, Reuters via Computerworld | Techworld | Published: 08:00, 23 March 2007
Hundreds of rectangles, each the size of a grain of rice, cover a shiny platter of silicon at a research facility belonging to Micron Technology Inc., a maker of semiconductor products for memory, storage and imaging. These cells contain circuits etched at a width of 50 nanometers -- 2,000 times thinner than a human hair. They represent the leading edge in a shrinking process in which a single memory chip can now hold hours of music or hundreds of digital pictures.
But makers of memory chips are looking ahead to a day, not too far off, when technology based on silicon bumps up against the laws of physics and memory can't be made any smaller. That development will have implications for gadgets like MP3 players and digital cameras.
"You get into the 25nm regime, and there may need to be a new structure for nonvolatile memory," said Mike Splinter, CEO of Santa Clara, Calif.-based Applied Materials Inc., the world's biggest supplier of tools for making microchips. "I'm quite nervous about this, because 25nm is not that far away, and if you have to change a process in a couple generations, then that is really challenging," Splinter said in a recent interview.
That would slow the development of things like digital music players and cameras, for which current flash memory -- used to store music and images -- will not suffice beyond the next couple of years. Until now, this shrinking of memory and processors has been governed by an industry maxim known as Moore's Law, formulated by Gordon Moore in 1965, three years before he helped found chip maker Intel Corp.
Moore stated that the number of transistors that could be housed on a given area of silicon would double every two years. He later reduced the time period to 18 months. The end of Moore's Law is expected to come more quickly for memory chips than processors because of the different ways in which they work. Whereas processors have circuits that act as pipes that guide streams of electrons, memory chips use pools of charged electrons to store data, and it gets harder to read the data as the number of electrons in each pool shrinks.
Such concerns aren't far from the mind of Tom Trill, a marketing director at South Korea's Samsung Semiconductor Inc., the biggest memory chip company in the world. "It's a question we've had forever, and we've always had an answer," Trill told Reuters. "There's been a resurgence in terms of pessimism ... in the last few months."
The concerns have major memory makers pouring hundreds of millions of dollars into perfecting the next big technology. The possible alternatives sound like science fiction: M-RAM, P-RAM, molecular memory and carbon nanotubes.
"In the next decade, we're going to need some significant new technologies," Mark Durcan, Micron's chief operating officer, said at the company's Boise, Idaho, headquarters.
Other major chip firms working on new technologies include Intel Corp., South Korea's Hynix Semiconductor Inc., Germany's Infineon Technologies AG and Japan's Toshiba Corp., Hitachi Ltd. and Fujitsu Ltd. One of the most promising new technologies is P-RAM, or phase-change memory, in which the physical state of a germanium alloy is changed between crystalline and amorphous to store data, rather than a change in electrical charge. The same principle stores music on a CD.
In what analysts said was a major step forward for phase-change memory, IBM said in December that it had developed a prototype chip that performed 500 times faster than current flash memory while using less than half the power.
Importantly, IBM researchers showed the technology can be used to create circuits as small as 20nm, less than half the size of current cutting-edge flash technology. "It's like a green light to the industry to say, 'OK, let's invest in this technology going forward,'" said Spike Narayan, IBM's senior manager for nanoscale science.
Other promising new technologies include magnetic memories that use magnetic fields instead of electrical charge, polymers or custom-designed molecules whose electrons can be easily manipulated, and carbon nanotubes. The challenge researchers face with most of those technologies is finding a way to make them cheaply in large quantities.
Just like in today's memory industry -- where each of four different technologies has its own multibillion-dollar market -- future technologies will probably find their own niches. New technologies will likely be cross-licensed throughout the industry, which is now dominated by five companies that account for more than 80 percent of the total output. The five next-largest vendors account for all but 1 percent of the remaining output.
"There's probably an opportunity for some of those smaller five to be melted into the larger five, but nothing that would shake up the whole industry," said Doug Freedman, an analyst in the San Francisco office of American Technology Research Inc. New memory technologies will all have to meet several key requirements. For example, they will have to be able to store a lot of data, read and write that data quickly, and be able to retain that data when the power is switched off.