The ability to understand and control materials, and to manufacture them at the atomic level, will be essential if the computer industry is to continue on its historical productivity growth curve.
In a typical day, most people will interact with about 300 semiconductors. Semiconductors are all around us. They are in cars, radios, stereos, cell phones, pagers, microwave ovens, kitchen appliances, and medical equipment, to name just a few applications. They are everywhere. The question is, Why is that?
One of the big reasons is that the semiconductor industry has been able to improve the productivity of the integrated circuit (IC) by between 25 and 30 percent every year for more than 30 years. What does that mean for the consumer, for you and me?
Consider the power of computers. My car has more computing power than the first Apollo spacecraft that went to the moon. In 1970, a computer that could execute 1 million instructions per second cost over $1 million. This cost had dropped to well below $100,000 by 1980. Last week, it was less than $2. This exponential growth in performance is why the silicon-based IC is so pervasive. Indeed, the silicon IC is the engine that drives the information age. By concentrating on silicon, I don't want to downplay other important base technologies of the information age. Fiber optics, for example, is another key base technology. If silicon is the engine that drives the information age, then optical fiber is the highway of the information age.
There are other types of semiconductors that are important today. Compound semiconductors, for instance, provide the photons that are the vehicles that carry the information along optical-fiber highways. Compound semiconductors also make possible the very-high-frequency devices - cell phones and pagers - used in today's wireless communications.
Other key materials-intensive technologies for the information age include high-speed, high-density storage, both magnetic and optical, and displays, which are the interface between computers and humans. So, it's really the convergence of these technologies that is driving the information age. In this talk, however, I will concentrate on semiconductor materials research, and specifically on silicon and the silicon IC.
If one examines the economics of the semiconductor industry, it all starts with the materials (Figure 1). Worldwide in 1997, about $22 billion worth of materials were used in the manufacture of semiconductors. The United States has about 10 percent of the revenue from those materials. The revenue from the semiconductor manufacturing equipment was about $36 billion last year, about half of which went to U.S. companies. Semiconductors were a $147 billion business in 1997, and the electronics industry was approaching a nearly $1 trillion enterprise. The United States captured almost 40 percent of the electronics market. All of this rests on the basic sciences.
I'd like to focus now on the role and costs of materials, the base of the pyramid in Figure 1. In 1995, about 20 percent of the cost of a processed wafer was represented by materials. At that time, the design rule, or the feature size, was 0.35 mm. By 1999, when feature size will be 0.18 mm, materials will account for about 30 percent of wafer cost. Materials content and costs are increasing with time, and they are increasing significantly. One reason for this increase is that chips are made with multiple layers of electrical interconnections, which means more and more materials and more processing steps. Materials also account for greater than 50 percent of IC packaging costs.
The semiconductor industry is very conservative when it comes to introducing new materials. The same basic materials used in the 1960s are still in use today. In the 1970s, when the design rule was 3 mm, integrated circuits were made using silicon doped with phosphorous, boron, and arsenic to control the electrical properties. The insulators were silicon dioxide and silicon nitride, and interconnects were aluminum.
At 2 mm feature sizes, molybdenum was introduced, at 1.2 mm, titanium silicide was added. At 0.8 mm, the aluminum interconnects were doped with copper to improve resistance to electromigration. When feature sizes decreased to 0.35 mm, tungsten plugs were used to connect the insulating layers.
As feature sizes became even smaller, the dielectric constant in the insulator had to be reduced, and some companies have added fluorine to the silicon dioxide. Now, we are facing massive introduction - by this industry's standards - of new materials as we move forward. Silicon dioxide is going to be fluorinated; the industry will introduce families of low-dielectric-constant materials in the interlayers; there will also be high-dielectric-constant materials, deposited copper, barrier materials, and other materials.
Why do we need these new materials? One important reason relates to gate-delay performance. As feature size continues to shrink, switching speed continues to get faster. To take advantage of that speed, one has to connect these transistors in the ICs. If the industry stayed with aluminum silicon dioxide, circuit performance would be significantly degraded at very small feature sizes.
Let me try to make that a little bit more concrete. Today's high-end microprocessor, 0.25 mm technology, has six levels of aluminum conductor separated by silicon dioxide dielectrics and connected tungsten plugs (Figure 2). In the future, say with 0.1 mm devices, the conductors will be copper and the insulators will be some low-dielectric-constant material. This will allow IC manufacture with about eight levels of interconnect.
There are two reasons for making the switch from aluminum to copper. The first is that copper gives superior performance in the circuits. The second is that if the industry were to stay with aluminum and silicon dioxide, we would require up to 14 levels of interconnect, and it is not clear that such circuits would be manufacturable at an affordable cost.
Not all companies are waiting until 0.1 mm to introduce copper. IBM, as you may have seen in the press, is introducing copper in its current technology, or certainly in their next generation. They are doing so because copper is a better conductor than aluminum. It reduces resistance and therefore improves device performance, even if one does not replace the insulator.
Another reason for introducing new materials relates to the devices themselves. As the device size gets smaller, the junctions must become shallower. The insulators in the gate must become thinner. Everything has to scale on all dimensions. That means that more and more control will be required in manufacturing. More characterization will be required, leading to in situ monitoring and on-line control.
Lucent Technologies is a leader in research on thin oxides and new materials in this area. The company has been able to apply a layer of oxide that is about 15 ?, or 5 atoms, thick. The oxide has to be put down uniformly, with perfect interfaces across the entire wafer.
The Problem of Tunneling Current
One of the big limitations of this technology as it is scaled to smaller dimensions is electron tunneling current, which travels through the oxide between the substrate and the gate as voltage is applied to the gate. This gate leakage is especially a problem if the thin oxide layers have any nonuniformities, pin holes, or surface imperfections. By understanding the properties of the materials, how to clean the surfaces, and how to process and deposit or grow oxides with good uniformity, Lucent has been able to reduce the tunneling current by a factor of 100 compared with the best previous results.
But even this may not be good enough for high-volume manufacturing, because of the uniformity and control required. Keeping in mind our desire to retain the excellent electrical properties of the oxide-silicon interface, the solution may be in building sandwich structures (e.g., putting a high-dielectric-constant material on top of a silicon dioxide-silicon interface). This, in fact, is where current R&D seems to be leading.
If we take this technology to its limit, we have essentially atomic-scale transistors. At Lucent, researchers have created a transistor with a 36 nm gate consisting of a 182-monolayer channel and an oxide layer of 1.2 nm, or about four atoms thick. By 2010, this technology will allow 64- or 256-gigabit memory, 200-GHz transistor speeds, a 10-GHz processor clock, and major reductions in power. This may be about as far as one can go by just simply continuing to decrease the feature size in silicon.
So, what's beyond silicon? I used to answer that question by saying, silicon. In fact, a couple of areas are being looked at seriously, but they are silicon based. One area is the so-called silicon on insulator (SOI). SOI has been around for a long time. It was pioneered for radiation-resistant and military applications.
IBM has adapted the technology for high-performance microprocessor manufacturing. IBM implants the silicon with oxygen and anneals the implantation-induced damage, creating an oxide layer that isolates the SOI layer. Transistors are then built in the silicon on the top layer. This approach gives a 20 to 30 percent increase in performance compared with bulk silicon transistors and also can reduce some of the "body effects" and junction capacity.
Another approach that is being looked at is building transistors out of silicon-germanium alloys. A silicon-germanium heterojunction bipolar transistor under development at IBM, for example, is highly planar and operates at a much higher speed than the comparable silicon transistor. The advantage of SOI and silicon-germanium is that they are compatible with conventional silicon processing and manufacturing technologies. So, silicon ICs and the silicon-germanium ICs can enter the same processing flow, thus taking advantage of the enormous worldwide manufacturing infrastructure that the semiconductor industry has built.
An interesting question is, How will these new transistors be manufactured in high volume? Fifty nanometers, after all, is pretty small compared with the wavelength of light that the industry has used to etch its circuits. Optical lithography, using steppers and scanners, has been the approach up to this point. We started with visible light, have now moved to the deep ultraviolet (248 nm lasers), and will eventually move to 193 nm. The practical limit for conventional through-the-lens lithography might be around 100 nm, or 0.1 um.
Several approaches are being studied to develop the industry's next-generation lithography. That's the technology that will print the smaller feature sizes. One of these, pioneered and led by Bell Laboratories, uses electron scattering, or electron printing, of feature sizes. Instead of trying to absorb the electrons in the metal, which would require thick masks, or reflect the electrons as we do with photons, for the mask Bell researchers simply put some heavy-metal scattering features where you do not want features printed. This scatters the electrons. These scattered electrons are stopped by an aperture, while electrons that come through the membrane go through the aperture and print the desired features.
Another approach is called extreme ultraviolet lithography. This technology is being developed by Sandia National Laboratories, Lawrence Livermore National Laboratories, and Lawrence Berkeley National Laboratories. The work is funded by industry in a consortium led by Intel. The idea behind this technology is that a laser pulse is used to vaporize a material, creating a plasma that radiates at all wavelengths, including soft X-rays. Optics collect the 13-nm, 134-? light, which then goes to a reflective mask through reduction optics to print the pattern on the wafer. Metal is deposited in the patterned regions.
I mention these technologies for two reasons. First, if we want to continue to reduce feature size and increase processing speeds, a new lithography technology will be needed. Second, creating the new technology will require an enormous amount of materials research. A lot of materials science and atomic-level engineering is required to create the equipment that will be used to manufacture the ICs as we continue to push to smaller and smaller feature sizes.
Semiconductors are enabling the convergence of computing, communication, and consumer electronics. The result is shaping the way we live and work. It is having a huge effect not only in our country, but also around the world.
The question is, how far will it go? How long will this technology revolution continue? I don't know the answer to that. But, for it to continue, advances in materials are required. Our ability to understand and control materials, and to manufacture them at the atomic level, will be essential if the computer industry is to continue on its historical productivity growth curve.