A New Lease On Life For Old Fashioned Chips

Ran-Hong Yan won't forget the day after Christmas in 1990. He had seen snow atop distant mountains in his native Taiwan, but he still wasn't prepared for the powdery flakes drifting down outside his Aberdeen (N.J.) apartment. He and his wife, Jennifer, dashed outside to marvel at their designs. "It was enchanting," he recalls. So much so that Yan, then 28 and a rookie on the semiconductor research staff at AT&T Bell Laboratories, called in late so he could help his daughter, Rose, build her first snowman.

After hanging up, he watched snow blanketing the tops of cars without sticking to their sides, and something clicked: Yan found himself envisioning a miniature snowfall of boron atoms piling up inside silicon wafers. It was an idea that would dramatically alter the outlook for silicon microchips.

QUICK FREEZE. On Dec. 6, one of Yan's co-workers, Kwing F. Lee, explained exactly how to the International Electron Device Meeting (IEDM) in Washington. Lee reported that the "snowfall" concept transforms complementary metal-on-silicon (CMOS) transistors--the most common switch, and heretofore a relative slowpoke--into the sleekest silicon speedsters ever. In fact, Yan's technique triples the speed of CMOS devices to at least 116 gigahertz, or billions of on/off cycles per second. Even at room temperature, such transistors are 30% faster than similar devices cooled with liquid nitrogen, a turbocharger for chips.

Group these transistors in an integrated circuit, and the gains can be even more startling. For example, Bell Labs has made a so-called frequency-divider chip similar to those used to change channels in communications gear. Before Yan's work, "the best number for a frequency divider was 2.6 gigahertz," says Abbas Ourmazd, head of microphysics research at the lab. "We've measured ours at nearly 14 gigahertz." Bell Labs won't speculate on how much the new technique, called vertical doping engineering (VDE), will speed up microprocessors. But it seems to confirm predictions by Intel Corp. and Digital Equipment Corp. that personal computer speeds will triple by the year 2000.

More important, silicon chips will drive ever faster electronics systems long past the turn of the century. Until now, indications were that silicon would hit redline speeds by 2000. To tweak more performance from chips without installing costly refrigeration even in PCs, chip companies might have had to make an expensive switch to a faster material, such as gallium arsenide. But VDE postpones that. "We see clear sailing for at least 20 years," asserts Mark Melliar-Smith, chief technical officer of AT&T Microelectronics, American Telephone & Telegraph Co.'s chipmaking arm.

He isn't the only optimist. In early 1991, when Bell Labs turned a crew of 16 scientists loose on Yan's idea, the news spread to the tight circle of researchers who hone the cutting edge of semiconductor technology. That prodded similar work among Japanese chipmakers such as NEC, Hitachi, and Fujitsu, and some of this also popped up at the Washington meeting. Perhaps the most impressive is from Toshiba Corp. Its scientists have made the world's tiniest CMOS transistor, an experimental switch 0.04 micron wide--one-tenth the size of the smallest transistors now heading into volume production.

EVER TINIER. Further advances may be possible. At last year's IEDM, IBM researchers said they had wrapped up a probe to the frontiers of transistor physics. Using Damocles, a computer program that Thomas N. Theis, manager of semiconductor physics research, describes as "the world's most sophisticated model" of solid-state microdevices, IBM found what may be the ultimate size for silicon transistors. "Before we did this study," says Theis, "0.1 micron was widely accepted as the physical limit. Now, it's 300 angstroms"--just 0.03 micron. IBM's ultratiny models aren't VDE designs. But computer simulations at Bell Labs suggest that VDE, which has already helped create chips with the 0.1-micron transistors that the industry expects to produce commercially around 2010, might get down to 0.03 micron, too. It would take 3,000 such devices, laid side by side, to equal the diameter of a human hair.

The innovation that boosts silicon's longevity is an elegantly simple addition to the structure of CMOS transistors (diagram). What Yan envisioned that snowy day is a plug that fills up part of a transistor's so-called channel, the conduit through which electrons zip when an "on" signal arrives at the gate. The shorter this path, the quicker the electrons make the jaunt.

Trouble is, shortening the channel increases the chance that electrons will leak across when they aren't supposed to--and wreak havoc with computing. To prevent this, as the channel is truncated it is "doped" with more of some impurity, such as boron atoms. This boosts the channel's electrical resistance and keeps the electrons bottled up. But when the channel shrinks to 0.1 micron, its resistance must climb so high that heat, generated by electrons pushing through the boron, roasts an uncooled chip. Vertical doping engineering dams up part of the channel, constricting the movement of electrons. Because electron leakage is then less likely, less resistance is needed to keep the transistor turned off--and a shorter channel is possible.

The secret to VDE is driving boron atoms to precise depths. This is done by hammering the boron into the silicon with an ion beam, whose energy level determines how deep the boron penetrates. In a 0.1-micron transistor, the dam goes 60 nanometers deep and rises to a height of 40 nanometers, or billionths of a meter--leaving a 20-nanometer-deep path for the channel.

PLANE TRUTH. Why has such a common-sense solution gone overlooked? Yan attributes that to mind-set. Silicon transistors have always been a planar technology--that is, semiconductor designers were preoccupied with the horizontal size of components.

By contrast, Yan's work for his 1990 PhD in electrical engineering from the University of California at Santa Barbara focused on gallium arsenide and related materials. Crafting transistors with these can involve stacking materials in layers. Ourmazd thought Yan might bring a fresh perspective to silicon's dilemma. But he never dreamed his hunch would pay off so handsomely so soon.

    Before it's here, it's on the Bloomberg Terminal. LEARN MORE