Scientists from IBM, Macronix, and Qimonda recently announced the results from a collaborative research project that may give a major boost to a new type of computer memory with the potential to be the successor to flash memory chips.
Working together at IBM Research labs on both U.S. coasts, the scientists designed, built, and demonstrated a prototype phase-change memory device that switched more than 500 times faster than flash while using less than one-half the power to write data into a cell. The device's cross section is a minuscule 3 by 20 nm, far smaller than flash can be built today and equivalent to the industry's chip-making capabilities targeted for 2015. This new result shows that unlike flash, phase-change memory technology can improve as it gets smaller with Moore's Law advancements.
"These results dramatically demonstrate that phase-change memory has a very bright future," said T.C. Chen, vice president, Science & Technology, IBM Research. "Many expect flash memory to encounter significant scaling limitations in the near future. Today we unveil a new phase-change memory material that has high performance even in an extremely small volume. This should ultimately lead to phase-change memories that will be very attractive for many applications."
The advance heralds future success for "phase-change" memory, which appears to be much faster and can be scaled to dimensions smaller than flash—enabling future generations of high-density “nonvolatile” memory devices, as well as more powerful electronics. Nonvolatile memories do not require electrical power to retain their information. By combining nonvolatility with good performance and reliability, this phase-change technology may also enable a path toward a universal memory for mobile applications. The new material is a complex semiconductor alloy created in an exhaustive search conducted at IBM's Almaden Research Center in San Jose, Calif. It was designed with the help of mathematical simulations specifically for use in phase-change memory cells.
A computer memory cell stores information—a digital "zero" or "one"—in a structure that can be rapidly switched between two readily discernible states. Most memories today are based on the presence or absence of electrical charge contained in a tiny confined region of the cell. The fastest and most economical memory designs—SRAM and DRAM, respectively—use inherently leaky memory cells, so they must be powered continuously and, in the case of DRAM, refreshed frequently as well. These "volatile" memories lose their stored information whenever their power supply is interrupted.
Most flash memory used today has a "floating gate" charge-storing cell that is designed not to leak. As a result, flash retains its stored data and requires power only to read, write, or erase information. This "nonvolatile" characteristic makes flash memory popular in battery-powered portable electronics. Nonvolatile data retention would also be a big advantage in general computer applications, but writing data onto flash memory is thousands of times slower than DRAM or SRAM. Also, flash memory cells degrade and become unreliable after being rewritten about 100,000 times. This is not a problem in many consumer uses, but is another show-stopper for using flash in applications that must be frequently rewritten, such as computer main memories or the buffer memories in networks or storage systems. A third concern for flash's future is that it may become extremely difficult to keep its current cell design nonvolatile as Moore's Law shrinks its minimum feature sizes below 45 nm.
The IBM/Macronix/Qimonda joint project's phase-change memory achievement is important because it demonstrates a new nonvolatile phase-change material that can switch more than 500 times faster than flash memory, with less than one-half the power consumption, and, most significantly, achieves these desirable properties when scaled down to at least the 22-nm node, two chip-processing generations beyond floating-gate flash's predicted brick wall.
At the heart of phase-change memory is a tiny chunk of a semiconductor alloy that can be changed rapidly between an ordered, crystalline phase having lower electrical resistance to a disordered, amorphous phase with much higher electrical resistance. Because no electrical power is required to maintain either phase of the material, phase-change memory is nonvolatile.
The material's phase is set by the amplitude and duration of an electrical pulse that heats the material. When heated to a temperature just above melting, the alloy's energized atoms move around into random arrangements. Suddenly stopping the electrical pulse freezes the atoms into a random, amorphous phase. Turning the pulse off more gradually—over about 10 ns—allows enough time for the atoms to rearrange themselves back into the well-ordered crystalline phase they prefer.
The new memory material is a germanium-antimony alloy (GeSb) to which small amounts of other elements have been added (doped) to enhance its properties. Simulation studies enabled the researchers to fine tune and optimize the material's properties and to study the details of its crystallization behavior. A patent has been filed covering the composition of the new material.