Hardware: Researchers pursue a cheaper way of designing and fabricating computer chips.
As consumers come to expect that everything from cell phones to stuffed animals will pack significant computing power, manufacturers are under pressure to churn out ever faster and cheaper microchips. But making computer chips using photolithography-the standard manufacturing technique-is wildly expensive. A significant part of that cost is the stencil-like “masks” that filter the light beam used to pattern millions of transistors onto a chip. Indeed, making a single silicon chip can require as many as 30 masks costing more than a million dollars-and as the transistors on a chip continue to shrink, the cost of the masks only grows.
No wonder, then, researchers are racing to develop ways to do away with masks entirely. One of the most promising efforts, led by Henry Smith, director of MIT’s NanoStructures Laboratory, uses an array of tiny mirrors, each just 16 micrometers across, to direct light through microscopic lenses; each lens focuses a beam of light to a spot on the silicon wafer, and the more powerful the lens, the smaller the spot. By tilting individual mirrors back and forth, a computer can turn individual beams on and off as the whole setup scans across the wafer. With as many as a million mirrors, the system could create the same complex pattern on the silicon chip that would normally require a series of masks.
So far, Smith’s group has used the system to pattern chip features 350 nanometers wide-40 percent wider than those on today’s best chips. But computer simulations predict the MIT technology can generate features as small or even smaller than those derived using conventional lithography by switching to shorter wavelengths of light.
At the University of California, Berkeley, a group led by electrical engineer William Oldham is taking a similar approach; but where the MIT group has focused on increasing the power of the lenses to make smaller features, the Berkeley researchers are reducing the size of the mirrors. Without increasing the power of the lenses, “to get smaller patterns you need smaller mirrors,” says Yashesh Shroff, a graduate student in Oldham’s lab.
Currently, Oldham’s team is making mirrors measuring only one micrometer across. “Nobody has ever made such tiny mirrors before,” says Shroff. Within five years, he adds, the researchers hope to have a complete system that can etch features 50 nanometers wide or less into silicon chips.
Maskless technologies could give chip designers unheard-of flexibility. “If you want to test a design a day, you can’t afford to build a million-dollar mask set a day,” says Dan Herr, director of materials and process science research at the industry-backed Semiconductor Research Corporation in Research Triangle Park, NC. With micromirrors, on the other hand, a designer could simply reprogram the array. And the technique could make the fabrication of customized chips for things like synthesizing speech in toys or playing MP3s in handheld computers-chips manufactured in much smaller quantities than, say, Pentium processors-much more cost effective. “Say I want to make a chip for a talking teddy bear, but I only expect to sell 2,000 of them,” says engineer David Carter, a member of the MIT group. “Now, with mask costs at a million dollars, who’s going to pay $500 for a teddy bear?”
Greater flexibility and lower cost could also be a boon for other industries pursuing emerging applications of lithography. Smith, for example, thinks his technology will be well suited for patterning the chambers and channels that help process biological samples in microfluidic chips, which could be used for drug discovery or in handheld diagnostic devices.
Observers suggest that the MIT team is the closest to a product that would replace masks; the researchers hope to have a commercial mirror-and-lens device for chip prototyping on the market in a year or two. Still, chip makers are also taking notice of the lithography efforts at Berkeley, Stanford University and the University of Texas at Austin. “Until about two years ago, all of this maskless technology was seen as very blue sky,” says Herr. But advances in computer software as well as technologies for fabricating things like micromirrors-coupled with the growing cost of existing production methods-could bring maskless lithography out of the lab and into fabrication plants within four to five years, Herr says.
If and when that happens, it will bring down one more barrier to computing innovation.
AI is here. Will you lead or follow?
Join us at EmTech Digital 2019.