May 2002  

Back to the future

Eye-popping products are on the horizon, as new semiconductor materials prepare to mix it up with silicon

By Stephen Lawton

Electronic paper. Sure. Flexible computer screens. Sounds good. Internet-enabled cereal boxes. Say what? These are just a few of the devices that could hit the market in years-rather than decades-as some of the world's largest chip companies develop new semiconductor materials to enhance or replace silicon.

But as eye-popping, and in some cases mind-boggling, as these products are, they're not the only driving force behind cutting-edge materials R&D. In some cases, they're more like desirable byproducts of the drive to keep up with Moore's Law in an economically feasible way.

Existing silicon-based technology likely will reach the limits of our ability to build smaller and faster chips in 10 to 20 years, warn scientists and industry analysts. Potential limitations of silicon chips include heat dissipation, power consumption and signal noise from current bleeding through the ultrathin insulation layers of transistors. Simply put, when you challenge the laws of physics, physics will win.

That's why nanotechnology, plastic semiconductors and hybrid chips-created by marrying silicon with a compound material such as gallium arsenide-are poised to capture a share of what has been silicon's domain. How much of a share is still open to speculation by industry analysts. Such chips generally are easier and less expensive to manufacture, require less power and perform well enough to displace-and in some cases, replace-silicon chips, manufacturers say. Plus, they promise, these chips will be fully compatible with existing technologies.

As semiconductors become more integrated, the greater number of transistors requires more power, and that translates into more heat. Helping chips keep their cool and maintaining low power levels while increasing speed and integration are the technological barriers chip makers will face during this decade, says Intel Corp.'s Gerald Marcyk, director of components research at Intel Labs in Santa Clara, CA: "Smaller and faster just isn't good enough anymore."

Still, Intel isn't exactly tracking the rest of the industry in its pursuit of next-generation chips. Instead, the semiconductor behemoth is using a new transistor design to further integrate circuits using existing silicon technology (see sidebar, below).

Like any good corporate gambler, however, Intel is hedging its bets, developing technology that allows silicon to serve as a substrate to other materials-an approach that's being duplicated by other chip companies as a way of leveraging existing technology while exploiting advanced new materials.

The hybrid approach

The use of compound materials isn't new. These inorganic materials, which include gallium arsenide (GaAs), silicon germanium (SiGa), indium phosphide (InP) and gallium nitride (GaN), offer capabilities that silicon alone can't provide, such as the ability to transmit light.

While they can offer exceptional performance, chips built exclusively from compound materials tend to be more expensive and more brittle than silicon. Indium phosphide, for example, "will crack if you look at it the wrong way," says Bill Ooms, vice president and director of material devices and energy research at Motorola Labs in Tempe, AZ. As a result, use of these materials has been limited to niche applications.

But all that could change with the impending introduction of hybrid chips, which marry silicon with another material, or compound. For example, growing a gallium arsenide crystal on top of a silicon base creates a chip that has higher performance than silicon, plus the ability to transmit light. Another method of building a hybrid chip is to incorporate nanotechnology on top of silicon. In each case, building on a silicon substrate means that the new chips would be compatible with existing CMOS technology and wouldn't require special conversion technology to be used with existing semiconductors.

Motorola Inc., Schaumburg, IL, last year announced it had successfully grown a compound material-a gallium arsenide crystal-on top of silicon. At the time, Motorola's Senior Vice President and Chief Technology Officer Dennis Roberson compared the scope of the accomplishment to the transition from discrete semiconductors to integrated circuits.

The challenge to growing a gallium arsenide crystal on silicon is that the two materials inherently are incompatible at the atomic level, says Motorola's Ooms. Motorola was able to overcome this by adding a "bonding" layer-in this case a single crystal of strontium titanate the size of the silicon substrate-between the silicon and gallium arsenide layers. This layer closely matches the molecular properties of the silicon, but is flexible enough to expand and match the atomic density of the gallium arsenide, he says.

While inorganic compounds may offer greater performance, the economics to replace silicon on a wide scale simply don't exist, save for some specific applications, Ooms says. Wafers made exclusively from these materials are rather small-typically 4 inches to 6 inches in diameter, but sometimes smaller-expensive to manufacture and not economical to build in small lots, he notes. An indium phosphide wafer, for example, is only 3 inches wide and costs $1,200 to manufacture, he says. Silicon wafers, by contrast, cost pennies for an 8-inch wafer. Hybrid chips will be more expensive than silicon wafers, Ooms says, but they should be significantly less expensive than a pure gallium arsenide or indium phosphide wafer. Hybrids also should provide higher yields because they use 8-inch wafers and are expected to have fewer damaged wafers than pure compound material chips.

By combining traditional CMOS with gallium arsenide, Ooms believes Motorola will be able to build faster radio frequency chips that will be cost-competitive with silicon alone. Motorola has licensed the gallium arsenide-on-silicon technology to IQE Ltd. of Cardiff, Wales, for cell phone applications. Ooms anticipates the first engineering prototypes to be released in the second quarter of this year.

Theoretically, Ooms adds, combining indium phosphide with silicon should allow Motorola to build better collision-avoidance radar using its improved infrared capabilities, while gallium nitride on silicon could be used for blue lasers for storage applications. But these two hybrid chips have yet to reach even the design stage, so he has no estimate of when commercial implementation could begin.

One word: plastics

One of the more exotic approaches to chip manufacturing today is the use of organic thin-film transistors (OTFTs). Also called plastic semiconductors, they're "manufactured" by printing the chip directly onto a thin film, using inkjet printers. The film then is bonded to a substrate that can be rigid or flexible; as a result, the semiconductor itself can be rigid or flexible. Another benefit: light-emitting properties that allow OTFTs to be used for some applications not possible with silicon.

(One might think that calling the same chip both "organic" and "plastic" is contradictory. It's not. OTFTs are built using polymers that can be processed as liquids and deposited on a substrate to make a semiconductor. A polymer, by definition, is an organic material due to its molecular makeup; thus, plastic semiconductors are organic.)

John Rogers, director of nanotechnology research at Lucent Technologies Inc.'s Bell Labs, characterizes plastic semiconductors as a "most disruptive technology" that potentially could replace the world's No. 1 reflective display technology used today: paper.

"Electronic paper is right around the corner," Rogers says, predicting that in five to 10 years, we'll have commercial enabling technology to build a display as thin, flexible and light as paper with color pixels that can change dynamically from on to off. By connecting a wireless adapter to the electronic paper, he says, the unit could access the Internet and change the images being shown.

In the movie Back to the Future II, a futuristic video wall changes from displaying an outdoor scene to becoming a large-screen videophone. Rogers predicts that such a wall, made up of multiple sheets of plastic semiconductors used as a display, could be available sooner than 2015-the year in which the movie was set.

Among the potential commercial applications for electronic paper are computer and advertising displays, labels and smart bar codes. In fact, Rogers says, "Plastic circuits could revolutionize big segments of consumer devices." It's conceivable that we could see bumper stickers that change messages or cereal boxes that connect to the Internet and provide personalized content based on the breakfast-eater's preference, he notes.

Lucent's partner in its electronic paper endeavor, E Ink Corp. of Cambridge, MA, currently laminates its monochrome electronic display to a rigid substrate. Next year, says Michael McCreary, E Ink's vice president of R&D, the display will be manufactured on a flexible film coming off large rollers and laminated to a glass substrate. By 2004, the company will laminate a full-color display to a flexible substrate, which could be the basis for a flexible computer monitor, or a video screen that conceivably would wrap around a pole for 360-degree viewing, he projects.

Once an image is created using E Ink's technology today, no additional power is required to maintain the image. Power is only required to change a pixel from on to off-black to white. Advertising displays will be able to run for months on two AA batteries, he anticipates.

IBM Corp., Armonk, NY, also sees advances for lightweight, flexible and rugged displays. Christos Dimitrakopoulos, organic electronics researcher at IBM, says he expects to see wearable displays and computers within five years.

Manufacturability is key, McCreary notes. Semiconductor vendors will be able to create the next generation of OTFTs using much of today's silicon manufacturing technology, but it will be easier, cheaper and require less complex manufacturing lines, he says.

Manufacture of silicon semiconductors requires a clean room, the use of toxic chemicals and extremely high heat. It requires costly photolithography steps or vacuum deposition-both of which can be eliminated with plastic semiconductors, says Tracey Stephens, marketing manager and co-founder of Plastic Logic Ltd. of Cambridge, England.

Plastic semiconductors can be inkjet printed outside a clean room at room temperature and require less manufacturing equipment, Dimitrakopoulos adds. Existing manufacturing lines could be used, so manufacturers won't need to build costly new fabs. Additionally, the technology should result in higher yields, Stephens and Dimitrakopoulos agree.

"Since we are inkjet-printing circuits," Stephens notes, "lower batch sizes are much more economical than with conventional silicon." Stephens sees additional applications for inkjet-printed circuits in electronic labels, disposable electronics and "novel packaging." She anticipates shipping commercial versions of the circuits next year.

The technology is application agnostic, says Dan Gamota, department manager for Printing Organic Electronics at Motorola; it could replace both digital and analog silicon chips, depending on the application. However, these circuits can't replace microprocessors that require much higher performance.

Nanotech: The Next Generation

The silicon squeeze is on, with plastic chips at the low end and nanotechnology targeting high-performance applications. Hewlett-Packard Co., Palo Alto, CA, is one of the champions of the nanotechnology-on-silicon hybrid semiconductors.

HP, in conjunction with UCLA, Westwood, CA, recently patented a grid design for a one-atom-high chip. Functioning like a field-programmable gate array (FPGA), the chip can be programmed electrically and is more tolerant of defects than traditional silicon, says Philip Kuekes, a computer architect in the Quantum Science Research department at HP Laboratories. If a defect on the nanochip is encountered, the engineer simply reroutes the circuit's signals to bypass the problem. Not only are these circuits smaller than those in traditional silicon, but a single molecule of a nanochip could replace six or seven transistors in a silicon chip, he says.

Kuekes expects memory chips using the grid design to be available by 2005, and other chips using the nanochip-on-silicon design to reach the market by 2007.

As with organic chips, nanochips are less expensive to build than silicon chips, Kuekes says. The equipment for manufacturing semiconductors at the quantum level is less expensive because, in part, a nanochip can be built at room temperature, doesn't need the same level of clean room and current manufacturing lines can be used. Again, some modifications of equipment would be required, but it would be far less expensive to modify an existing fab than to build a new fab to make silicon chips, he says.

Nanochips also require very little power to switch a microcircuit, so these chips will meet the low power requirement, Kuekes adds.

HP isn't the only company bullish on nanotechnology for memory chips. Nantero Inc., Woburn, MA, is using carbon nanotubes to create a new type of memory chip transistor that the company says will carry it through the rest of the decade. The memory is expected to be compatible with today's CMOS chips and could be manufactured in today's fabrication facilities. He expects to have commercial prototypes in one to two years.

Unlike HP's technology, in which semiconductors are built using a layer of active material just one atom high, Nantero's nanotubes could range from a few nanometers to as much as a micron, says CEO Greg Schmergel. Nantero's approach differs significantly from other technological advances; it might be considered a throwback to computer equipment of decades ago. The company uses an electromechanical design, meaning that the nanotubes must physically move and touch to switch from a zero to a one. However, Schmergel says, the nanotubes, which have walls one atom thick, are "many times stronger than steel-as strong as a diamond-a better conductor [of electrical signals] than copper, and [they're] flexible."

Schmergel says the technology virtually could eliminate a computer's boot time. Because carbon nanotubes are nonvolatile, computers could boot instantly, regardless of the operating system, he says.

Carbon nanotube chips also would have significantly higher storage density than existing memory chips, he claims; whereas today's DRAM stick might offer 512MB of memory, carbon nanotubes could provide many gigabytes of memory in that same physical space. And if the technology were packaged to fit in the same space as a 5.25-inch disk drive, the user conceivably could obtain as much storage capacity as in a large array of disks, he argues.

Still, it's not likely that the memory industry will be dominated by carbon nanotubes in the future, says Steve Cullen, director and principal analyst for semiconductor research at Scottsdale, AZ-based Reed In-Stat Group, which is owned by Reed Elsevier Plc, ELECTRONIC BUSINESS ' parent company. For any new memory technology to succeed, he says, the manufacturer will have to create a chip that is extremely inexpensive to build. As prices continue to fall and capacities increase, memory has become a loss leader for some companies, he notes. A pure memory chip company could have a hard time making a profit in that kind of market.

Undaunted, Schmergel notes that every 1% of the memory market captured by Nantero chips would translate to $250 million, based on In-Stat's estimate of $25 billion in worldwide memory revenue in 2001.

Despite advances in nanotechnology, plastic semiconductors and hybrid chips, analysts and chipmakers alike see no end to the dominance of silicon for the next five to 10 years. Silicon still is inexpensive to manufacture, and a lot of companies have experience in silicon semiconductors, notes Elliot Grant, an engagement manager at McKinsey & Co., a San Francisco-based management consultancy. But while silicon semiconductors in general are safe, technologies like OTFTs "can be category killers within two years," he says.

Meanwhile, scientists at HP, IBM, Lucent and other companies agree that silicon is about to feel pressure from emerging technologies and materials in vertical markets. At least in the storage business, Schmergel says, carbon nanotubes could be a very disruptive technology within two years. "There are no significant physics issues to hold back production," he says. "The laws of physics are on our side."

Intel plays silicon card

It might be hard to picture Intel as a contrarian. But while some chip makers look to organic compound materials or nanotechnology to take their chips to the next level, scientists at this PC industry monolith are focusing on new designs to squeeze more life out of silicon technologies.

The company's goal is to build microprocessors that have 25 times the number of transistors as today's Pentium 4, with 10 times the speed and no increase in power consumption, all by the end of the decade. To put that in perspective, today's Pentium 4 has approximately 42 million transistors, the next generation Itanium processor will have 220 million transistors-and Intel believes it can build a silicon processor that can scale up to 1 billion transistors.

As for speed, Intel expects to be able to deliver what Star Trek only promises-real-time speech recognition and translation, as well as graphics-intensive programs and proactive computing, says Gerald Marcyk, director of Intel's Components Research Lab in Santa Clara, CA.

To achieve this goal, Marcyk says, Intel developed 15-nanometer transistors that can operate in the terahertz range. But first, Intel had to overcome the barriers of heat, power and electron drift by redesigning the most basic component in the semiconductor: the transistor.

These new, low-power transistors, which should be available in five years, will operate at approximately 0.75 volts, according to Intel's Web site. Today's "low-power" chips need 1.3 volts, almost twice the power.

Exotic technologies aren't necessarily required to maintain the technology advances in Moore's Law, maintains Marcyk, but then, he doesn't rule them out either. "Moore's Law is about economics," he says, "not science." Chips today are very different from those of 1965, he notes, and those of 2010 might well use new technologies that are not solely silicon. As the economics of semiconductor manufacturing and design change, so might the semiconductors themselves. Moore's Law "will be extended by new materials and new structures."

Intel is addressing the power problem associated with highly integrated chips by making chips that can run on lower voltage per transistor. Lower voltage means less heat and less power, but since the new chips have many more transistors than existing chips, the result is a wash-essentially, the chips use the same amount of power and create the same amount of heat as today's chips do, even though they're more complex.


E Ink Corp.

Hewlett-Packard Labs

IBM Journal of Research and Development, January 2001

IBM Microelectronics

Intel Labs

Motorola Inc.

Nantero Inc.

Plastic Logic Ltd.

Semiconductors Information Web site

Copyright 2001-2002
All trademarks are the property of their respective companies.