Making a computer chip is, in many ways, like taking a photograph. And one of the challenges in creating the next generation of chips lies in sharpening the images.
For the past six years, Andy Neureuther, an electrical engineering and computer science professor at the University of California, Berkeley, and his graduate students have been working on TEMPEST, a computer program that examines how light bounces through quartz, around chrome, and onto silicon. The stormy acronym stands for an ungainly mouthful-Time-domain Electromagnetic Massively Parallel Evaluation of Scattering from Topography-but the program's calculations have aided engineers at such companies as AT&T, IBM, and Motorola in designing computer chips and other integrated circuits.
To understand what bouncing light has to do with circuit design, you need to know how integrated circuits are usually made: the negative image of the circuit-called the mask-is laid out in chrome on a plate of quartz glass. Like projecting a slide, a beam of light shines through the mask and projects an image of the circuit onto a silicon wafer. The light exposes a photosensitive coating on the wafer, which is then washed away in the exposed areas. The remaining coating serves as a pattern for etching the circuit into the silicon.
In practice, however, the process is not black and white. Look at any shadow: the edges are not sharp. Light leaks from bright regions into the dark. This fuzziness of the image limits how small one can make integrated circuits. Try to make them smaller, and the shadows blur into a hazy gray that is useless.
If someone figures out how to create a sharper picture of the circuit, that immediately opens the door to faster computer chips and memory chips as well as hard disks that hold more data. That's what a technique called "phase-shifted masks" is designed to do. The phase-shift in the masks adjusts light waves so that the blur into the dark region between two features cancels each other out giving true dark regions. Consider two neighboring holes in the chrome mask. Because light travels slower in glass than in air, creating a trench in one of the holes causes the light waves to exit this hole with less phase delay than the other.
Light waves consist of oscillating magnetic and electric fields. Choose the thicknesses just right-so the path length is one half of a wavelength of the light-and the light waves from one hole will be exactly 180 degrees out of phase with waves out of the other hole. When one light wave's electric field is up, the other is down. When the blur contributions meet in the middle region, they add up to zero. The result is a darker region to separate the features.
Of course, it's not quite that simple in the real world. The Berkeley group first turned to the gigaflop calculating speed of the CM-2 and then to NCSA's CM-5 that they use now. They solve in detail how light moves in and about the phase-shifted masks. Back in 1989, "parallel computers were the only game in town for complex typographies and inhomogeneous materials," Neureuther says. Happily, the problem-solving Maxwell's equations on a grid representing the mask-also maps into neat, separate parts, which take full advantage of the parallel processing abilities of the Connection Machine.
"What TEMPEST does is allow you to look at the nonideal physical condition," Neureuther says. The light leaving the bottom of the trench and entering the air travels faster than light still inside the glass sidewalls. As a result, the light that ideally propagates vertically begins to diverge with light near the edges of the trench curving back toward the glass sidewalls of the trench. Some of this light does not make it out of the trench, which makes the trench region not quite as bright as the light from an opening in the chrome that has not been etched.
TEMPEST allows researchers to investigate technology solutions for overcoming these effects in the space of a few minutes on a CM- rather than the months needed for experiments. Once they find something promising, then they can test it out for real. "It definitely does not replace experiments," Neureuther says. "It's just more cost-effective." That could prove a pretty picture both for chip makers and chip users-which, in today's world, is almost everyone.
Kenneth Chang , formerly a research programmer at NCSA, graduated in June from the Science Writing Program at the University of California, Santa Cruz, and is currently working on the Los Angeles Times.
