Thursday, October 04, 2007

IBM Attempts to Reinvent Memory

A new type of memory using nanowires could be simpler, cheaper, denser, faster, and more reliable.

By Kevin Bullis

Nanowire memory: A magnetic nanowire (thin line at top) made of iron and nickel is stretched between electrical contacts to test properties that could lead to better memory devices. The device stores bits in the form of domain walls (shown at bottom), which are areas where two magnetic regions in the nanowire meet. Credit: Stuart Parkin, IBM Research

Multimedia •  See two possible configurations of a novel type of memory.

A new nanowire-based memory device being developed by researchers at IBM could combine the best qualities of the various types of memory used today, driving down costs and improving performance. If the experimental memory pans out--and the work is still in the very early stages of development--it could serve as a universal memory, substituting for the different types now used.

Stuart Parkin, an experimental physicist at IBM's Almaden Research Center, in San Jose, CA, says that the memory, which would pack a hundred bits of data on a single nanowire, could potentially store 10 to 100 times more data than flash--the type of memory used in digital cameras and other small portable devices--while operating at much faster speeds. And because it's solid-state memory, it would be far sturdier than magnetic hard drives, which require mechanical devices to read and write data. "In principle, we could be cheaper than flash would be, denser than flash would be, and orders of magnitude faster," Parkin says. "And there's no wear-out mechanism, so it's totally reliable."

All of this should be possible, Parkin says, as a result of applying new insights into the nanoscale behavior of magnetic materials and the electronic currents in these materials that open the way to storing many bits of data on a single nanowire. Parkin has demonstrated the basic elements of the new type of memory, but has not yet built a complete prototype.

Although it is early in development, the research has attracted attention because of Parkin's track record for making major breakthroughs in magnetic memory. His earlier discoveries and inventions have led to a thousand-fold increase in the storage density of magnetic hard drives, paving the way for the massive data-storage centers crucial to today's Internet, as well as making possible the immense storage capacity of portable devices such as iPods. The new memory devices would combine advantages of the three types of memory used widely today--hard drives, flash drives, and dynamic random access memory (DRAM)--while avoiding many of their disadvantages. Like hard drives, which are the cheapest form of memory, Parkin's proposed devices would store bits of data in a magnetic medium. But unlike hard drives, they would not require a mobile head and spinning disks to read and write these bits. Indeed, there would be no mechanical parts, making Parkin's memory much more robust than a hard drive: there would be no danger of the read-write head crashing into the magnetic media and destroying data.

Parkin's memory would also have advantages over conventional solid-state memory such as DRAM and flash. Unlike DRAM, the new memory would not require a continuous supply of energy to store data. Flash memory also has this advantage over DRAM: it can store data without power. But it's slow. The new memory could be orders of magnitude faster than flash, even rivaling the speed of DRAM, Parkin says.

The devices could also be more compact and cheaper than conventional solid-state memory. They would resemble such memory in that they would use millions of tightly packed read-write devices arrayed in a grid on a memory chip, rather than the few read-write heads used in hard drives. But unlike conventional solid-state memory, in which each read-write device can store between one and four bits, each would be paired with a nanowire that can store between 10 and 100 bits. These bits would be quickly shuttled along the length of the nanowire, propelled by electronic pulses, then read or written at one point along the nanowire.