28 January 2008—A technology that would allow a computer chip to change the electrical resistance of some of its own wiring could lead to more-powerful reconfigurable microchips that can quickly adapt themselves to new tasks, researchers at IBM say.

Engineers at IBM’s T.J. Watson Research Center in Yorktown Heights, N.Y., produced a prototype device based on a type of material found in experimental memory chips. Normally, a “via,” a hole leading from one layer of a chip’s wiring to another, is filled with a metal such as tungsten to provide an electrical connection between the layers. But in this case, the researchers used a phase-change material, a substance whose conductance can be switched between two states by briefly melting it. “By changing the state of the phase-change material, you create an on-off switch,” says Kuan-Neng Chen, the research staff member who led the project.

The team filled vias with an alloy of germanium, antimony, and tellurium (GST)—an often-experimented-with phase-change material, known as a chalcogenide. Beneath a GST-filled via, they deposited a strip of conductor. They then applied a series of electrical pulses to heat up the strip. They selected pulses that were of just the right amperage, shape, and duration to melt the GST. When the GST cooled, it changed from its original conductive crystalline state to a disordered, amorphous state. In that amorphous state, the material’s resistance is more than 2 megohms, effectively shutting off any circuit connected to the via.

A different pulse reverses the effect, melting the amorphous material again but letting it cool into a crystalline, conductive form. The reverse process is “kind of like an anneal,” Chen explains. “You cook it at a lower temperature but cook it longer.” When the GST is in the crystalline state, electrical resistance drops to less than 60 kilohms, more resistive than a tungsten via but conductive enough to turn the circuit back on.

Switching the via from off to on takes a little over 1 microsecond. The reverse process takes about 50 nanoseconds. The IBM group published its results in this month’s IEEE Electron Device Letters.

Building a computer chip with millions of those vias would create a new kind of field-programmable gate array (FPGA), a circuit whose design can be reconfigured depending on its application. One area that might benefit, Chen says, is high-speed routers for telecommunications, where the systems might have to change on the fly depending on the ebb and flow of data. FPGAs generally have a faster time to market than their counterparts, application-specific integrated circuits (ASICs), because they do not have to be fully designed before being manufactured. Today’s FPGAs, however, generally rely on flash memory cells controlling a transistor that acts as a switch for each interconnection. Unfortunately, that setup takes up real estate on the silicon that crowds out logic circuits. It also requires a higher voltage than the phase-change scheme, Chen says. Because his setup is built in the layers of wiring above the silicon, it would allow more logic in the same chip area.

Chen thinks the phase-change via is also superior to another type of reconfiguration technology, this one an invention of his own company. IBM’s eFuse uses a high electrical current to essentially blow fuses on the chip. But eFuse, too, requires a high voltage to operate and works only once. The phase-change vias, on the other hand, can be reconfigured repeatedly.

Stefan Lai, vice president of business development at Ovonyx, a Michigan company working to commercialize nonvolatile semiconductor memory based on phase-change materials, finds the IBM work interesting. “I see this as a very innovative approach to creating an electronic switch with a high on-off resistance ratio,” Lai says.

IBM’s Chen emphasizes that the work is in an early research stage, and that a proof of concept on a single via is very different from building a whole reconfigurable chip. “Right now, the major thing we are dealing with is how to integrate the materials in the process,” he says. He would like to find a phase-change material that not only has the optimal properties for switching—the best resistance-to-conductivity ratio and the best voltage requirements—but one that’s also compatible with the standard chip-manufacturing process. For instance, the team is testing a germanium-antimony alloy without the tellurium, because they don’t know whether tellurium will prove compatible, and they believe the material will change phases more quickly. They will likely look at other materials as well. It could be three to five years, he predicts, before a phase-change FPGA makes it to market.