Intelligent Machines
Chips Go 3-D
Hardware
It’s an axiom in real estate: when land gets expensive, build up. For 30 years, chip designers have considered whether building integrated circuits with multiple layers might create cheaper, more powerful chips. Previous attempts to build such three-dimensional chips have failed or proved too expensive, but Santa Clara, CA-based startup Matrix Semiconductor plans to bring the first one to market in just a few months. While Matrix’s techniques won’t likely result in more computing power, they will produce cheaper chips for certain applications, like memory.
Matrix has adapted technology developed for making flat-panel liquid-crystal displays to build chips with multiple layers of circuitry. The company-founded in 1997 by Stanford University electrical engineer Tom Lee, Mike Farmwald, cofounder of chip connection technology company Rambus, and others-starts with a standard silicon chip. It paves the chip with glass, adds a new layer of silicon on top and starts the process over again. Matrix connects the layers by etching holes through the glass and filling them with silicon. So far, the company has built chips with eight layers on top of the base chip, and it believes it can go higher.
Matrix’s first products will be memory chips for consumer electronics like digital cameras and audio players. Current flash memory cards for such devices are rewritable but expensive; Matrix’s chips will cost ten times less, about as much as an audiotape or a roll of film, but will only record information once, says Dan Steere, Matrix’s chief marketing officer. The cost is so low largely because the stacked chips contain the same amount of circuitry as flash cards but use a much smaller area of the extremely expensive silicon wafers that form the bases for all silicon chips. The chips will also offer a permanent record of the images and sounds users record.
The amount of computing power the company can ultimately build into its chips could be limited, though. “You may never build a supercomputer out of this,” Lee says, “but we’ve got an awful lot of room to grow.” He hopes to eventually build chips for cell phones, or low-performance microprocessors like those found in appliances; such chips would be about one-tenth as expensive as current ones.
Most other teams working on three-dimensional chips, including groups at MIT, IBM, Rensselaer Polytechnic Institute and SUNY Albany, are betting on techniques for bonding conventional chips together to form multiple layers. Whichever approach ultimately wins out, though, multilayer-chip-building technology “opens up a whole new world of design,” says MIT electrical engineer L. Rafael Reif. And like a city skyline transformed by skyscrapers, the world of chips may never look the same.
