Breaking up is hard to do—especially after 40 years together. It’s even tougher when those 40 years were spent constantly side by side, fostering an upstart technology that has changed the world.
Little noticed outside the semiconductor industry, silicon dioxide has supported and protected silicon, as well as facilitated the element’s special electronic properties.
“The guys like us who work with the stuff every day consider silicon dioxide the greatest gift from God,” says John S. Suehle of the National Institute of Standards and Technology (NIST) in Gaithersburg, Md.
Today, the disaffection is spreading to computing circuits. Circuit features are becoming so small that chip makers are battling on many fronts to keep up with the pace of change (SN: 11/8/97, p. 302: https://www.sciencenews.org/sn_arc97/11_8_97/bob1.htm). In the confines of smaller circuits, silicon dioxide’s even-handed skill at managing passels of unruly charges has become a flaw.
Subscribe to Science News
Get great science journalism, from the most trusted source, delivered to your doorstep.
Consequently, research groups in the industry and elsewhere are searching for alternate materials. Although they don’t know what will successfully replace silicon dioxide, manufacturers are specifying new machinery and otherwise making accommodations to handle likely alternatives in their production lines. Chip makers are already using some substitute materials on a limited scale in their products.
“If we can’t replace silicon dioxide, it’s a showstopper for device scaling”—the main process by which the industry has made circuit elements smaller and smaller, says James H. Stathis of the IBM Thomas J. Watson Research Center in Yorktown Heights, N.Y.
The semiconductor industry will part with silicon dioxide only reluctantly. Chip makers are chafing at the uncertainty and expense of switching to what may prove to be a less-than-perfect substitute, electronics specialists say.
Circuit manufacturers would do much worse, however, not to change, they add. Unless the divorce takes place—and soon—the astounding pace of innovation on which the semiconductor industry thrives could slacken for the first time.
The density of components on microcircuits has grown exponentially since the early days of the industry, doubling roughly every 18 months. It’s a grueling rate of change that the industry has come to expect. For anyone who can’t keep up, “the penalty is death,” Robert D. Miller of the IBM Almaden Research Center in San Jose, Calif., grimly jokes.
When fashioning an integrated circuit on a chip, manufacturers use light to project patterns onto surfaces of silicon or other semiconductors. Each pattern defines the contours of a layer of material deposited on the surface. By repeatedly applying patterns and adding to or carving away those layers, chip builders create and wire together millions of circuit components on one chip.
To achieve greater component density, circuit makers use optical methods, such as lenses, to shrink the patterns—a strategy called scaling. They must then make other adjustments, such as altering the recipes and thicknesses of deposited materials, in order for the diminished devices to work properly.
From the start, silicon dioxide has been one of the stars of this scaling process. However, circuits have become so small that the material can no longer keep up, transforming it from God’s gift to semiconductors to silicon’s soon-to-be ex.
Paradoxically, silicon dioxide is losing its attractiveness in different microcircuit applications for opposite reasons. In its role in the transistors of integrated circuits, it no longer promotes accumulation of electric charge strongly enough. In engineering parlance, it has too low a value of a property called the dielectric constant, or k.
At the same time, silicon dioxide’s value of k is turning out to be too high for its other main function, electrically isolating wiring between devices.
To continue making faster circuits, chip makers have recently switched from aluminum wiring to copper, which has less resistance and therefore increases signal speed (SN: 9/27/97, p. 196). But that’s not enough.
For insulation, circuit fabricators need a low-k material because charge storage, or capacitance, between wires slows signals. It also encourages signals to bleed from one wire to another.
To protect wires in some commercial products, manufacturers already spike the silicon dioxide with a little fluorine to drive down the mixture’s dielectric constant, Miller says. In a few years, however, the industry will have to break completely with silicon dioxide as the wiring insulator, he says.
This problem with silicon dioxide as a wiring insulator cropped up a little sooner than the transistor woes, so chip makers have already gone much further in dealing with it. Research into low-k materials is making progress, Miller says. “There are people very far along on that, far enough along that we know it can be done,” he says. Some researchers have even explored replacing silicon dioxide insulation with wispy, highly porous aerogels (SN:12/14/96, p. 383) or simply with air (SN: 7/18/98, p. 37).
Replacing silicon dioxide
The challenge of replacing silicon dioxide in transistors looms ahead. Suehle says that it may be “the most serious challenge” among many that the industry faces.
Transistors are the workhorses of integrated circuits. In essence, a transistor is like a valve: It modifies flow. Applying voltage or current to one of the three terminals of a transistor controls the current moving between the other two.
Most commonly used in integrated circuits is the complementary metal oxide semiconductor, or CMOS, transistor. There, a layer of silicon dioxide serves as an electrical insulator, or dielectric, preventing unwanted vertical current flow between the control electrode, known as the gate, and the underlying silicon.
Because of its high capacitance, the gate-silicon dioxide-silicon sandwich can store substantial amounts of electric charge. That charge accumulation controls the flow of electricity that traverses a thin skin of silicon just beneath the oxide.
As circuit devices have shrunk, so too has the area of the gate. However, the area in part determines how much charge the gate can hold. To keep the capacitance up, designers have had to make the gate oxide thinner with each scaling.
Materials scientists figured out how to grow silicon dioxide films so free of defects, such as pinholes and impurities, that the gate oxides insulated well even when pared down from hundreds of nanometers thick in the 1970s to only about 2.5 nm today.
“Being able to make [the oxide] thinner and thinner is what has really allowed us to continue scaling,” says NIST’s Eric M. Vogel.
But even thinness and purity can’t keep a relationship going forever. As the oxide becomes thinner, a quantum-mechanical effect known as tunneling permits more and more electrons to slip like ghosts through the oxide wall. Although transistors can function despite tunneling leakage, they require more power. Circuit makers find this power loss unacceptable, especially as electronic products are increasingly becoming portable and battery-powered.
A new partner
The time has come for silicon to find a new partner, most semiconductor specialists agree. Circuit builders need a substitute that they can pile more thickly under the gate electrode, to prevent tunneling, without at the same time reducing capacitance to an unacceptable level.
To continue scaling, “the only option is to go to higher-k materials. There is no choice,” Miller says.
With the change to copper and to silicon dioxide substitutes, “the industry is going to be facing the most significant [materials] changes ever,” says Howard R. Huff of International Sematech, a research consortium based in Austin, Tex., and sponsored by 13 semiconductor companies.
The high-k imperative has sent academic and industry researchers racing through the periodic table. In particular, scientists are focusing on the so-called transition metals, such as titanium, tantalum, and zirconium. These elements form high-k compounds, some of which have long been used in capacitors. Investigators also became acquainted with them during initial attempts to replace silicon dioxide in memories.
Considering oxides and silicates of transition metals and more complex combinations of several metals and perhaps other substances, “we have hundreds of candidates,” says Tso-Ping Ma of Yale University.
Mr. or Ms. Right
Like someone searching for Mr. or Ms. Right, researchers trolling for high-k dielectrics can’t afford to focus on just one appealing trait. Making their task particularly daunting is the way solid-state physicists and engineers recite lists of good things about silicon dioxide. The tributes read like Elizabeth Barrett Browning’s poem “How Do I Love Thee?”
“If you want all those things, it’s hard to find another material like silicon dioxide,” Miller says.
At the top of all the lists: Silicon dioxide mates with the silicon surface in such a way that so-called dangling bonds become satisfied.
That’s not just idle pleasure. Dangling bonds stick up from the surface when sausage-like silicon crystals are sliced into wafers. Ordinarily bonded to four other silicon atoms, many silicon atoms at a cut surface find themselves bereft of a bond or two.
When the first CMOS transistors were invented, those hungry bonds spoiled the performance of the devices. Then, in 1958, a group of scientists at Bell Telephone Laboratories found that growing a thin layer of silicon dioxide on the surface fulfilled the bonds.
In the view of researchers at the time, noted Chih-Tang Sah of the University of Florida in Gainesville in a 1988 article about transistor history, that discovery was the “most important and significant technology advance” leading to silicon-integrated circuit technology.
“Next to silicon, [silicon dioxide] is really what ignited the industry,” adds Robert R. Doering of Texas Instruments in Dallas.
How good are the best prospects—mainly transition-metal oxides—on the high-k horizon? They quash tunneling currents admirably in laboratory experiments and also maintain a hefty gate capacitance thanks to dielectric constants 6 to 20 times that of silicon dioxide.
Ma’s Yale group and three other research teams reported such performance results last December at the 1999 Institute of Electrical and Electronics Engineers International Electron Devices Meeting in Washington, D.C.
Yet in some ways, those new dielectrics leave semiconductor specialists pining for silicon dioxide.
Troubles arise at the silicon-dielectric boundary, for instance. Researchers find lower-k compounds forming there, silicon dioxide itself or byproducts of reactions between silicon and the new partner materials.
Huff notes also that none of the leading high-k candidates can tolerate the heat of the current process for fabricating circuits. The industry should be cautious not to jump prematurely into major, costly changes to manufacturing to accommodate the materials, he says.
It may be wiser to cope with circuit leakage by counteracting it with power-conserving designs and more judicious placement of the thinnest silicon dioxide layers. Given uncertainty about the new materials, he asks, “Are you going to overthrow 40 years of industrial experience with silicon dioxide?”
Realizing the change may turn out to be necessary, however, Huff helps lead a Sematech team that is rushing to meet a June deadline. By then, the group must produce specifications for a machine capable of depositing on circuits whatever high-k materials will be ready for pilot studies 3 years from now. Very likely, those will be a stopgap measure—silicon-based compounds already familiar to the industry but only modestly higher k than silicon dioxide.
Cutting the knot
No one knows when silicon and its oxide will finally cut the knot. The Semiconductor Industry Association in San Jose, Calif., gives manufacturers no more than 5 years to take the high-k plunge. Otherwise, the industry’s growth rate will slow, the association says.
In its latest guide, or roadmap, to the industry’s future, the group dryly notes, however, that “history has shown that changes of this magnitude ordinarily require 10 years or more to implement.”
In 1998, Stathis and Donelli J. DiMaria, also of IBM’s Watson Center, sent tremors through the industry by predicting that gate oxides would start to fail at a thickness of about 2.2 nm, which at that time was only a year or so away.
A more optimistic prediction—1.6 nm—came out of the December 1999 electron devices meeting. The IBM researchers revised their estimate to 1.8 nm, Stathis says, although he warns that setting a hard-and-fast limit is difficult.
“Certainly by 2005, we’ll hit that point,” he predicts. “It could come a bit sooner.”
When it does, the microchip world in which silicon and silicon dioxide have been the perfect couple will slip away. While the semiconductor industry toasts silicon’s new companion materials, some technologists may quietly mourn the end of the good thing that silicon and its oxide had for so long.