[ Return to Special Projects ] A. O'Malley's Quote "They are the hidden engines of the information age, the driving forces behind today's computers and tomorrow's digital televisions, stereos, appliances, and unimagined gadgets. Improvements in their technology occur with such breakneck regularity that a generation may last only a year or two. In the 1990's, they're speeding along even faster than the experts had expected. And they may be headed straight for a wall." This is the statement Chris O'Malley used to explain the current situation for microprocessors in his Popular Science Magazine article. B. Moore's Law Also, in 1965, Gordon Moore, an Intel Corporation co-founder, made a long-range prediction that the number of transistors on computer chips would double every 18 months. This prediction, later dubbed Moore's Law, has been proved true for the past 35 years thanks, in part, to continuous advances in lithography, which is the process of etching features on a silicon wafer. Many semiconductor experts have begun to fear that Moore's Law could soon be broken. The laws of physics have begun to limit how small the features of the processor can get, and the laws of economics have begun to limit how much companies are going to be willing to upgrade the technology. II. Short History of Microprocessors The era of microprocessors began in 1971 when a small company named Intel Corporation dared to enter a struggling to survive computer industry and introducted its 4004 processor. This 4-bit processor had 2300 transistors, and it was intended for use in calculators. Then in 1975, Intel introduced the 8086, which had 29,000 transistors, and pc computing was born. Microprocessors continued evolving, and new processors were introduced like the 4.77MHz Intel 8088 in 1979, the 80286 in 1982, the 16MHz 80386 which had 250,000 transistors, the Intel 80486 which had 1.2 million transistors, and today's Pentium II's which have 7.5 million transistors. Each processor's features get smaller and smaller as processors evolve. The features in current Pentium II processors are as small as .25 microns, or approximately 1000 atoms wide. These smaller features are possible through the advances in Optical Lithography processes. III. Overview of Optical Lithography Optical lithography is the process that is currently used to etch processor features onto a silicon wafer. The process was originally supposed to run out of steam when processor features reached 1.25 microns wide; however, new techniques and equipment have allowed for the .25 micron Pentium II features to be possible. It is also predicted that .13 micron features could be possible. The process of optical lithography consists of three stages. The first stage holds the light source that is shined through the second stage, the pattern mask/lens stage. When the light passes this stage, it enters the third stage where the silicon wafer is etched with the processor features. The light source stage currently uses ultraviolet light that has a wavelength between 193 and 248 nanometers, which is located near the top end of the ultraviolet spectrum. The most important part of the system is the pattern mask/lens stage. This pattern mask is made out of a sheet of glass topped with a very thin coating of an opaque material, usually chrome, that is laid out in the pattern of the processor features. Each of the different processor stages, such as the transistor gate regions, has a different mask designed for it. In order to minimize pattern mask defects being transferred onto the wafer, the pattern on the mask is usually five times larger than what is patterened on the chip. Once the light passes through the pattern mask it goes through a series of lenses that shrink the pattern down to the size that is desired for the features on the chip. The third stage of the system consists of a silicon wafer with a thin layer of a photoresist chemical on its top. Wherever the light passes through the mask, it contacts the chemical and causes the photoresists solubility to either increase or decrease, depending on whether the wafer has a positive or negative photoresist layer. The wafer is then put through a solvent that washes away the unwanted photoresist. This produces either a positive or negative image of the pattern mask on the wafer. The resolution of this system can be found by using Rayleigh's equation, R equals KL over NA, from basic physics. The goals of microprocessor fabrication companies are to lower the value of R so that more transistors can be put on the processor increasing the processor's speed. The two main ways that the companies lower the value of R is decreasing the light source's wavelength (the value of L) or by increasing the lens' NA value. Increasing the value of NA is the less successful way to decrease the value of R. Minor imperfections in the silicon wafer cause hills and valleys to be formed in the wafer, which means that the image is projected onto an uneven surface. Higher NA values means greater "less-than-perfect" imaging on the wafer resulting on lower wafer yields. Higher NA lenses are also much more expensive. For the past 10 to 15 years, companies have been more successful decreasing the light source's wavelength. During this period of time the wavelengths of the light sources has gone from 436 nanometers (dubbed the g-line) to 365 nanometers (dubbed the i-line) and to 248 and 193 nanometer (dubbed deep or super-uv) wavelengths currently used today. Many universities and national laboratories are currently experimenting with 13 nanometer extreme-uv, or soft x-ray, systems, and some have been successful in using this system to etch lines that are .08 microns, or 250 atoms wide. IV. Alternatives to Optical Lithography Each reduction in the light source's wavelength means a large expense in new equipment and new materials used in the process, both for pattern masks and photoresists. When 193 nanometer uv light no longer is able to print processor features small enough, the jump to 13 nanometer uv will cost processor companies billions of dollars to update their equipment. This has prompted several companies to start developing alternatives to the current tecnhiques, and two techniques are serious contenders to becoming the next generation of lithography techniques. A. X-Ray The first of these currently under development is x-ray lithography, where 4 nanometer wavelength x-ray photons replace the uv light source. In order to generate x-rays, however, companies will require synchrotron x-ray generating machines that cost 25 million dollars each. Scientists at AT&T, IBM, Loral Federal Systems, and Motorola are working together to try to make this system commercially available by the start of the next century despite the large economic risk. Henry Smith and his partners at the Massachusetts Institute of Technology used an x-ray lithography system to carve features that were .017 microns, or approximately 10 atoms, wide. They later used the system to build working electronic devices with features of .005 microns. The main problem with x-ray lithography is that, over time, the x-rays distort materials used to make the pattern masks; however, program manager of microlithographic mask development, Steve Schnur, sees x-ray lithography as being the next generation of lithographic tools required to do advanced semiconductors. B. Electron Beam Lithography The seond alternative under development is electron beam lithography which fires high-energy electrons through a pattern mask rather than firing high-energy photons through the mask. The traditional way that electron beam lithography was used was through the use of direct-write electron "pencils," and features were literally drawn on the wafer. This process is much too slow for manufacturing purposes, however, so research companies, like the Bell-Labs division of Lucent Technologies, have been researching ways to shine showers of electrons through the pattern mask. the SCALPEL system is what they have come up with. The main problem that researchers have had with this system is that the high-energy electrons that get absorbed by traditional pattern mask materials heat the mask up and deform it. To minimize this problem, a mask made with chromium and tungsten has been created. These metals scatter rather than absorb electrons. A rigid layer of silicon nitride supports these metals, which is transparent to electrons. The major difference between the SCALPEL system and traditional lithography techniques is the molybdenum or metrology plate. Both the electrons that pass through the pattern's features and the scattered electrons pass through the silicon nitride and through a magnetic lens which focuses the unscattered electrons through a tiny hole in the molybdenum sheet. The scattered electrons miss the hole and are simply absorbed by the molybdenum. A second magnetic lens, below the molybdenum, then focuses the beam onto the wafer reducing the pattern to 1/4th its original size. Using this system, a Bell-Labs team reported in June 1996 that they were able to etch processor features .08 microns, or 250 atoms, wide. V. Conclusion Wired Magazine interviewed Gordon Moore for their May 1997 issue. In this interview Wired questioned Moore about Moore's Law and how long he felt it would hold up. Moore predicted that in about ten years, a distinct slowdown in the doubling rate would occur. He didn't extimate how much that slowdown would be, but it could be as slow as every five years instead of 18 months. The reason, the industry is reaching a point where the light wavelengths are so small that focusing lenses cannot be built anymore. Whatever appears to be the future of microprocessors, one fact has microprocessor companies shivering. Once optical lithography has reached its limit, companies will be looking at spending up to ten billion dollars to convert their plants over to new equipment. |
