Figure 1. This source of extreme ultraviolet (EUV) light is based on a plasma created when a laser is focused on a beam of xenon gas clusters expanding at supersonic speeds (in addition to EUV light, some visible light is created, as seen in the blue glow). Patented at Sandia National Labs in 1996, the light source is used in an experimental lithography tool for patterning microchips with features one-third the size of current, commercially available chips. Smaller features of about 0.1 µm should enable a 10-fold increase in computer chip speed and 1,000-fold increase in memory.
Accelerated development of a new extreme ultraviolet (EUV) lithography system, along with proof of concept demonstrations around the world, prompted the microchip industry in December to put their full support behind optical lithography as the next generation lithography system. EUV is expected to take silicon microchips into the third decade of the new millennium when, it is expected, a completely new kind of microchip will need to be created.
For the past two decades, Moore's law has been gospel to the microchip industry. Simply stated, the dictum from Intel cofounder and chairman emeritus, Gordon Moore, states that the number of transistors on a microchip should double approximately every two years. This has been the driver behind the success of the personal computer and the key ingredient to growth for microchip producers around the world. Yet, despite the advances of the past twenty years, the future of microchip manufacture has had some of the world's largest companies scrambling to find a way to maintain the necessary exponential growth of microchip technology defined by Moore.
Until recently, the industry watched as four different technologies struggled to become the preferred next-generation lithography system. In December of 1998, however, SEMATECH (SPIE Corporate Member; Austin, TX) brought together industry leaders for a strategic meeting. By 11 December, optical lithography had proven to those assembled that it has the ability to keep creating faster microchips well into the second decade of the 21st century.
Critical decisions about critical dimensions
Figure 2. Atomic force microscope image of transistor gate printed in photo resist using EUV lithography. The active region is shown in red.
Within the microchip industry, speed and efficiency are largely governed by the all-important "critical dimension" (CD). Rather than referring to the latest musical recording technology, this CD refers to the size of the smallest feature on a microchip. These features make up the transistors, electronic gates, and beltways that funnel information in the form of electrons through a computer's brain.
Today, cutting-edge lithography techniques use KrF excimer lasers at 248 nm to create CDs of 0.25 µm (see sidebar). Optical techniques, such as phase sensitive masks, will allow CDs smaller than the wavelength of the source; however, these techniques add considerably to the cost of the process. Manufacturers, such as SPIE Corporate Members Cymer (San Diego, CA) and Lambda Physik (Ft. Lauderdale, FL) are putting every effort into ruggedizing the ArF excimer, which emits at 193 nm, for the next generation of optical lithography.
Excimer manufacturers have gone to solid state excitation sources to offer better pulse control and to reduce the chances of misfiring. In addition to these controls, source providers have worked hard to increase the per-pulse power and repetition rates of ArF lasers in order to reduce exposure times and increase the throughput of a wafer fabrication system.
Outliving the excimer
As lithography ventures further into the UV, reduction optics become problematic for lithographic stepper manufacturers. Fused silica, the material of choice for today's lithographic steppers, can develop chromatic aberrations with continuous exposure to UV light. CaF optics resist UV damage, but even the synthetic version of the naturally occurring material is hygroscopic. This water-absorbing property makes grinding and polishing the lenses a difficult proposition. Although CaF and its counterpart, MgF (which is used when polarization is not an issue), have been used in microscope and military imaging systems for many years, lenses made from these materials have been low-volume, customized items.
Figure 3. This wafer was patterned on an integrated laboratory research system capable of printing proof-of-principle, functioning microelectronic devices using extreme ultraviolet lithography (EUVL). The EUV lithography research tool was assembled at Sandia National Laboratories, which has joined with two other Department of Energy labs -- Lawrence Livermore National Lab. and Lawrence Berkeley National Lab. -- creating a Virtual National Laboratory to help develop EUV lithography for commercial use.
In the past, optics industry experts have expressed concerns that the semiconductor industry may not give the optics industry enough time to build furnaces and other tools to prepare for large-volume orders of large-aperture CaF optics, creating an optics bottleneck for the deployment of ArF and F2 lithographic systems. One compensatory fact is that lithographic systems with ArF excimer laser light sources may be able to use a combination of fused-silica and CaF lenses, depending on the beam's fluence at a particular spot in the optical train. However, future excimer laser sources, such as the F2 at 157 nm, would require CaF exclusively.
Until recently, most U.S. manufacturers did not expect the F2 laser to develop into a lithographic source because of the remaining development required for both the source and the optics, combined with the fact that it would represent only a half-generation improvement in CD size reduction. At the recent SEMATECH meeting, however, some of the world's largest U.S. stepper manufacturers came forward to announce that they would rather build upon their optical lithography expertise than attempt to interject a new technology. According to Richard Stulen, the EUV project leader at Sandia National Labs, adapting to the 157-nm F2 laser means giving the industry the tools to continue improving microchip performance, while giving a next generation lithography system, currently under development at three U.S. national laboratories, the chance to mature.
Big business still supports lasers
Although using lasers as a direct exposure source will soon reach the physical limits, the microchip industry now believes that optical lithography will be able to continually produce smaller and smaller features until silicon chips are no longer a viable material. In 1996, Researchers at Sandia National Labs developed an EUV source that uses a high-power laser beam to excite a supersonic stream of xenon gas. Once excited, the gas emits peak radiation at 13 nm, straddling the line between UV and x rays. "With EUV there is no doubt that it can go all the way down to [feature sizes of ] 30 nm without having throughput problems," explained a triumphant Stulen following the December SEMATECH meeting. "With SCALPEL, there are questions of beam blur [at that feature size]. So while it might be possible to generate [30 nm] feature sizes, it would pay a stiff penalty in throughput."
Optics have posed the largest problem for this system, although recent results from around the world have quelled many doubts. Conventional refractive optics are useless at such short wavelengths. Instead, scientists at Lawrence Livermore National Laboratory developed reflective optics made from multiple layers of molybdenum and silicon, with roughness factors controlled to within a few atoms. According to Stulen, Zeiss (Oberkochen, FRG) has shown extraordinary progress in manufacturing these special mirrors, showing the world that "this is something that you can do not only in a national lab, but broadly, around the world," Stulen explained. Zeiss's results were as good or better than those of SVG-Tinsley (Richmond, CA), the U.S. contractor for the EUV project.
Though improved, these mirrors still offer many challenges. They only reflect 70 percent of EUV light, resulting in a commensurate need for highly sensitive photoresists and masking technologies. Livermore leads the development of the optics, coating, and masks, while Sandia continues to develop the source and new photoresists.
According to Stulen, two other significant factors prompted the industry to get behind the EUV project. At the SEMATECH meeting, wafer manufacturers revealed that the electron beam from SCALPEL -- the front runner before December -- deposited a great deal of energy at the wafer, expanding the material. New techniques would be needed to compensate for this expansion if it were to go into mass production.
Perhaps the greatest show of support, however, came when a major manufacturer said it would be ready to start taking orders for EUV lithographic systems in the fourth quarter of this year.
Despite its success, Stulen cautions that EUV still has many significant challenges. The difficulties in developing a next-generation lithography system are matched only by the cost. Bill O'Leary of IBM expects that bringing a new system to fruition will cost on the order of $500 million. EUV has half this amount thanks to a private-public consortium formed last year. The consortium, called the EUV Limited Liability Corp. (EUV LLC), is led by Intel Corp., Motorola Corp., and Advanced Micro Devices, Inc. It was formed last year to finance the development of EUV for the next 3 years. In the past year, Japan has initiated its own EUV program financed with $85 million, and a European consortium has a similar, but smaller, investment. Now that the industry is behind EUV development, it stands to reason that the system's already accelerated development will leap forward again as new corporate members strive to develop their own expertise in this exciting field.
Lithography in a nutshell
In many ways, printing microchips is just like any other kind of printing. Lithographic processes are those which copy a master design onto a secondary material. Microchip lithography, although much more technically involved than conventional printing processes, follows many of the same general principles.
Optical lithography has dominated microchip manufacturing since the earliest microchips. The general process involves shining light through a mask or stencil, reducing that transient image through optics, and then imaging the design on a light-sensitive material called a photoresist. Subsequent processes bathe the wafer in chemicals to remove the unimaged material, and then the process is repeated. A single microchip may have as many as 20 or more lithographic layers, with each layer gaining in complexity.
Standard mercury lamps are still used for some of the early layers with the largest features. Excimer laser technology is beginning to take over the industry as the preferred source for the newest generation of microchips with critical dimensions of 0.25 µm. The excimer laser is an excellent lithographic source because, while it is wavelength-stable and offers good powers at uv wavelengths, the light is not too coherent. Wavefronts that are too flat cause problems in the lithographic imaging process, which is one of the major reasons why solid state lasers, although capable of reaching the proper wavelengths with reduced power and cost, have not made significant inroads into the microchip manufacturing industry.
R. Winn Hardin
R. Winn Hardin is a science and technology writer based in Fairbury, NE.