The active storage medium is a thin polymer film on the surface of the chip that represents bits in the form of 10-nanometer-diameter holes. Called the Millipede, the system features both high density in a small area and nonvolatile, erasable capability, suiting it for portable digital systems, IBM researchers said.

"One of the key distinctions with this technology is that it is just at the beginning of its lifetime in terms of density, while most of the current storage system technologies are at the end of their lifetimes [and thus are] facing more and more problems to make further improvements," said project leader Peter Vettiger.

He stressed that the Millipede is only a prototype and will need at least two more years of basic research before even being considered for manufacturing. Nevertheless, the research group is already targeting its development toward specific applications areas, such as personal digital assistants, cellular phones and multifunctional watches.

"Our next prototype will be demonstrated next year on a printed-circuit board, and it will take at least another year to do the engineering for the package form factor of a flash memory chip," said Vettiger. "We want it to be compatible with existing flash memories but at considerably higher capacities.

"After that, it may take a few more years to go to production, even if everything goes well. But it is important to realize that we have not made a decision at IBM yet that this is going to be a product at all."

Vettiger said the Millipede device has natural advantages — such as nonvolatility, low power and large capacity storage — for mobile applications where flash memories are so important today. If those advantages can be capitalized on successfully for flash memory replacement, he said, then the burgeoning cell phone market alone could bootstrap the Millipede project into a full-blown production technology, after which other applications could be developed.

"The concept is of a general nature and in the distant future could be used in other areas," said Vettiger, citing such applications as large-area microscopic imaging, nanoscale lithography and atomic-scale molecular manipulation.

Previous Millipede prototypes could only bulk-erase the entire chip. But the current prototype erases with a heated tip by coaxing the polymer to flow back into the hole, erasing the bit.

"One of the breakthroughs we have made for this announcement is that you can erase individual indentations in the polymer. We discovered that if you make a series of holes closer and closer to the original, surface tension will force the heated polymer to flow back into the original hole, in effect erasing it. This works because the polymer is not destroyed when we make the holes. We have written and erased successfully over 100,000 times with the current prototype," said Vettiger.

The nanoscale read/write tips are kept under static tension by the architecture of the silicon from which they are sculpted. The slight amount of static tension is not enough to cause wear between the polymer and the tips moving over its surface. When heated to 400°C, the tension presses the tip into the polymer to write or erase. To read out, the same tips are heated to just 300°C to prevent damaging the polymer. When the tip drops into a hole marking a bit, it is abruptly cooled by the better heat transport, and a measurable change in its resistance can be detected that is enough to distinguish "1" from "0."

Cantilever configuration

"We are also experimenting with electrostatic activation of each cantilever. It basically has the same effect, which is to push the hot tip into the polymer," said Vettiger. "In addition to the heating pulse, you apply a voltage to the media, which generates the electrostatic force."

The 1,024 cantilevers for the Millipede prototype were arrayed in a 32 x 32 configuration over a 3 x 3-mm polymer-coated silicon platform. The two-dimensional array of v-shaped silicon cantilevers is 0.5 micron thick and 70 microns long, ending in a sharply pointed tip less than 2 microns long. The terabit per square inch was measured by only using one tip to see how close together the holes could be spaced, but the density (200 Gbits/square inch) of the 32 x 32 array was limited by the pitch of the cantilevers.

Since the cantilevers are spaced at a 100-micron pitch, each cantilever's tip is responsible for only a 100 x 100-micron area. Consequently, the silicon platform only needs to be moved, at most, by 100 microns in the X direction and 100 microns in the Y direction. All 1,024 read/write heads can then work in parallel to simultaneously address 1,024 bits on the chip (one from each 100 x 100-micron area.

"Each cantilever is responsible for its area in a two-dimensional array, depending on the pitch, which is now 100 microns," Vettiger said. "The polymer-coated substrate is being shuttled 100 microns in two dimensions by electromagnetic forces; a [microelectromechanical system] coil and a magnet drive the little silicon platform back and forth."

The next phase of research, which has already been under way for a year, is to increase the number of read/write tips to 4,096 over a 7-mm² area. Eventually, Vettiger predicts, capacities could hit 15 Gbytes per Millipede flash-memory replacement chip — a tenfold increase over current predictions for flash memory capacities.

In addition to density, however, the Millipede needs to speed its data transfer rate if it is going to go into mobile systems. Each tip operates in the kHz range now, meaning many must be used in parallel to achieve the kind of data rates that flash memory chips typically exhibit. Using all 1,024 tips in parallel would raise the data rate to the MHz range, but simulations indicate that individual tips will eventually be able to read at MHz rates anyway. Combining parallel MHz reading with higher speed MHz tips could result in GHz data rates in the long run, albeit at higher power consumption levels.

Mobile obstacle

The complex erasing procedure, though an improvement over previous bulk erase methods, is still the biggest obstacle to the Millipede's going mobile. There is no theoretical IBM research to indicate that erasing will get much faster in the future. And erasing takes much more power than the 100 milliwatts required for reading.

"Erasing as well as the writing process takes place in the order of microseconds, but you can use as many of the tips as you want in parallel to make them faster. To save power, you could also program your mobile device to do erasures when it is plugged in to its charger," Vettiger said.

The researchers plan to transfer their lab bench prototype into a printed-circuit board system that's closer to what will be required if the Millipede ever becomes an IBM product — a real memory storage unit that can be attached to a host processor. Besides refinement and optimization of materials and design, IBM plans to have an all-in-one pc board-level working prototype of a full-blown Millipede memory subsystem ready by early next year.

If all goes well, it will take until at least 2004 for Vettiger's team to shoe-horn its pc board prototype into a flash-replacement-sized package.