A new kind of flash memory technology with potentially greater capacity and durability, lower power requirements, and the same design as flash NAND is primed to challenge today's solid-state disk products.
Fremont, Calif.-based Nanochip said it has made breakthroughs in its array-based memory research that will enable it to deliver working prototypes to potential manufacturing partners next year. Three investors, including Intel Capital, recently put $14 million into the company, which has been developing the technology since its founding in 1996.
"It's a technology that doesn't depend on Moore's Law," says Gordon Knight, CEO of Nanochip. "This technology should go at least 10 generations."
Knight was alluding to the decades-long trend in which the number of transistors that can be placed on an integrated circuit roughly doubles every two years. Current thinking is that flash memory could hit its limit at around 32 to 45 nanometers. That describes the smallest possible width of a metal line on the circuit or the amount of space between that line and the next line. The capacity of an IC is restricted by the ability to "print" to a smaller and smaller two-dimensional plane, otherwise known as the lithography.
And that, according to Stefan Lai, is where Nanochip's technology shines. "Moore's Law is driven by lithography," says Lai, a member of Nanochip's technical advisory board, as well as vice president of business development at Ovonyx and former vice president of Intel's flash memory group. "Every two years, you need to buy this new machine that allows you to print something that's smaller and finer."
Array-based memory uses a grid of microscopic probes to read and write to a storage material. The storage area isn't defined by the lithography but by the movement of the probes. "If [Nanochip] can move the probes one-tenth the distance, for example, they can get 100 times the density with no change in the lithography," says Lai. "You don't have to buy all these new machines."
Lai said that in principle, Nanochip could develop the ability to move the probe a single atom at a time. The company said its current generation of probes has a radius smaller than 25nm, but it projects that eventually the probes could be shrunk to two or three nanometers apiece. That scale, said Knight will enable development in 10 to 12 years of a memory chip greater than 1TB. For a first generation, anticipated in 2010, Knight says he expects a small number of chips to be in excess of 100GB, but a more realistic number is "tens of gigabytes" per integrated circuit, a capacity comparable to the current generation of flash devices.
Reusing the old equipment for new chips
Knight sees a market for Nanochip's technology in USB drives, solid-state disk drives, and even enterprise servers. In each case, he believes, there are advantages with array-based memory.
Unlike flash NAND, where the frequently changing lithography requires construction of ever-pricier manufacturing plants, Nanochip can manufacture its chips on existing low-cost semiconductor equipment, according to Marlene Bourne, head of analyst firm The Bourne Report. "They're using used equipment [and] adapting to their needs," she says. "Same machinery, same equipment, same materials, same basic processing steps. You're just creating three-dimensional objects instead of a flat [integrated circuit]." That will hold true, she says, even as the company increases the density of its chips. That could provide a cost advantage over solid-state drives, which are currently in the range of $15 to $18 per gigabyte.
Like solid-state drives, array-based memory requires no motor, which reduces its power consumption and heat output in comparison with spinning disk hard drives, says Lai. The mechanism used to move the probes is very low power, he says. Because they don't require "a hundred pieces to make the hard drive work," Lai says he believes Nanochip products will be more rugged.
Unlike traditional disk drives in servers, says Knight, his company's technology prevents the queuing problems that surface when multiple users try to access data. "When you have an array of these chips, you have many, many points of access," he says. An internal controller inside the Nanochip sends the tips down to locate specific data, which is returned in multiplex form and output in serial form, "just like the output of a NAND flash drive or disk drive -- but in fact, the data is spread out over a few hundred tips."
Array-based technology isn't something new and unique, says Bourne. Nanochip is simply applying it in a slightly different way. "The tips that form the core of this memory technology are what's being used in atomic force microscopes," she points out.
IBM's first attempt with Millipede
IBM first showed a similar technology in the late 1990s. The Millipede project, which is no longer in active research at IBM's Zurich Research Laboratory, used microelectrical mechanical system (MEMS) components. In MEMS, the electronics or "brains" of the chip are usually fabricated using integrated circuits, while the moving parts are microscopic components etched from silicon in a micromachining process. Millipede itself was based on nanoscale research in which individual iron atoms were arranged with atomic precision on a special copper surface. That research won two IBM scientists the 1986 Nobel Prize in physics.
Millipede works by using a microscopic probe to make an indent in a polymer material. Each indent represents a single bit as part of the write operation. The indentations can then be removed from the material surface during an erase operation.
By using thousands of such probes in parallel, array-based memory achieves high data rates, with each probe able to read, write, and erase in its own data field.
Where Millipede puts "dents in plastic," Nanochip has found a better material for the read-write process to occur, according to Knight, though he declines say what that better material is. A year and a half ago, Knight says, the company made a breakthrough on a new media type that could be infinitely rewritable. "The media never wears out," Knight claims. "That's really what got the company rolling fast."
In-Stat analyst Steve Cullen believes Nanochip has licensed a material that uses chalcogenide glass from Ovonyx. Knight acknowledges that his company has worked with that kind of recording material but is unwilling to say more on the topic.
Lai, who works for Ovonyx, declines to comment on the material being used by Nanochip but points out that the phase-change semiconductor work being done by Ovonyx has more to do with reducing the size of current circuit technologies. "We will continue to follow Moore's Law."
A potential stumbling block for Nanochip's technology is that the tips on the probes, which have a radius smaller than 25nm, could wear out quickly.
Tip wear is particularly relevant if array-based probes are adopted as storage mechanisms in servers. "Obviously, you have a lot of tip wear that goes into an enterprise server that's operating 24/7, for five, six or seven years," says Knight.
Lai concurs. The tip is a problem, he says, because it touches the surface of the material.
Knight declines to specify how Nanochip has resolved the tip-wear dilemma, but he insists the company has had a breakthrough in its research that has addressed the problem.
Probe-based storage in the real world
In-Stat's Cullen claims the new technology will find a home as a replacement for hard drives in notebook computers. "The thing that strikes me about 100GB is that it's a nice size for something to replace a disk in a notebook PC," he says. "All they've got to do is come close to the price of a disk and then offer some other advantage. It may consume less power than a disk. It could be more rugged."
Nanochip is confident in its ability to produce a product with the same size as existing drives. "We'll make the interface so it'll just plug and play," says Knight. It's a new technology, but you want it to fit right in."
Lai believes that the new memory could herald breakthroughs in mobile devices and biotechnology. "You now need your whole life history stored in your mobile device," he says. "If you want something to store your genome in, it may take a lot of memory, and you'll want to carry it with you."
The big question that remains for Nanochip is whether the company can create working prototypes with the cost advantages that array-based technology is supposed to offer over conventional forms of memory. The fact that IBM appears to have moved on from its Millipede research doesn't alarm Bourne. In fact, she points out, several people from the IBM team have joined Nanochip's board of advisers. Knight said the company has 50 engineers and scientists working around the world on the prototypes, either as part of Nanochip itself or within the companies that his firm is partnering with.
IBM last publicly shared details of its probe-based storage research in a gathering of companies and organizations involved in a joint research project called ProTeM, for "probe-based terabit memory."
According to Evangelos Eleftheriou, an IBM fellow and manager of IBM Labs' storage technologies group, the company built a prototype that achieved a storage capacity of a terabyte per square inch. He says that research will be published in an article appearing in a couple of months in the IBM Journal of Research and Development. But the group doesn't have plans to develop any products. It will leave that to other companies that might choose to license the research, he says.
The challenge for adoption of any new type of memory, points out Eleftheriou, is that flash itself isn't standing still. "In 2010, it's going to be $1 per gigabyte ... so hopefully the cost per gigabyte [of probe-based arrays] is going to be low."
Now, he says, the areas of interest for probe-based technology at IBM have moved onto topics including archival storage and maskless lithography, a technique that separates individual molecules and places them precisely onto a surface.
"The focus of our research is [to] explore ways to enhance the speed in probe sensing and the way we modify the surface -- how fast we can do those things... There are many things that come together, from positioning control, to materials to micro-machining, micro-fabricating, so it's extremely fascinating altogether."
This story, "A storage technology that breaks Moore's Law" was originally published by Computerworld.