A new way to build computer memory chips takes its inspiration from the Manhattan skyline: when space is at a premium, build up rather than out.
A Silicon Valley start-up combats semiconductor sprawl by figuring out a way to build its circuits up rather than out.
There's a reason that Manhattan's skyline reaches high into the heavens -- and it's not just the view. When real estate is pricey, the smart money thinks tall, creating skyscrapers that fit as many people onto as little land as possible.
Semiconductors, which are built on silicon wafers that cost some $1 billion per acre, ought to follow the same principle. But traditionally their circuits have been spread horizontally, making their structure more akin to the sprawl of Los Angeles than the verticality of New York City.
Matrix Semiconductor intends to change that by introducing the high-rise memory chip, which will cost about one-tenth the price of its horizontally inclined competitors.
In a standard memory chip, circuits are etched into a base layer of pure monocrystalline silicon. To add memory, new circuits are spread across more silicon, increasing the cost and the size of the chip; it's like building more houses on an L.A. street. Only wiring and insulation are layered on top of the silicon.
Matrix Semiconductor's breakthrough, says CEO Dennis Segers, is that it has discovered a way to stack functional layers on top of the base layer using less refined, less expensive polysilicon. "We're able to pack a whole lot more memory bits in a smaller area of silicon real estate," Segers says. "Ultimately, that means that the bits cost us a lot less."
The approach has been tried before, in the 1980s. It failed, says Segers, in part because the materials that the engineers were then using for stacking would have required making costly changes in the semiconductor-manufacturing process. Fabrication facilities cost billions of dollars, and their operation relies on carefully controlled processing of pure silicon and polysilicon. To introduce a new material would require expensive testing of the material itself -- to ensure that it wouldn't contaminate the silicon -- and perhaps even new equipment. When two of Matrix Semiconductor's cofounders -- Mike Farmwald and Stanford professor Tom Lee -- revisited the existing research, in 1997, they discovered that advances in polysilicon layering and flattening techniques during the past 10 years had finally made the material suitable for stacking.
"We're able to pack a whole lot more memory bits in a smaller area of silicon real estate."
The company, which has raised about $80 million in financing from big-name investors like Microsoft, Benchmark Capital, Sony, Eastman Kodak, and Thomson Multimedia, plans to market its new chip to electronics manufacturers of portable consumer devices such as digital cameras.
Typically, digital photographers pay $35 or more for a single reusable memory card and then delete old photos when they want to make room for new ones. Memory cards containing the Matrix 3-D Memory chip, which can be used just once, will cost about the same as a roll of film -- inexpensive enough to enable users to purchase multiple cards. "Cards using our product plug in to the same kind of camera and operate in the same way as the old memory cards, but they're more like traditional film in that you can buy several off the rack at any retailer," says Segers.
Other possible applications for what Segers calls Matrix Semiconductor's "disposable memory" are digital music storage for MP3 players, computer-game cartridges, and PDA add-ons like prerecorded maps. Thomson Multimedia plans to offer the first digital memory cards using the company's 3-D Memory chip by the second half of the year.