Throughput--the number of wafers that a lithographic tool can process in an hour--determines how fast a semiconductor manufacturer can build ICs. So it's directly connected to profits. But electron projection lithography performs its magic slowly; its throughput is low, and the likelihood is that it will be unable to achieve the dominant position once expected of it. At most, semiconductor manufacturers expect it to process 30 wafers per hour.
But Bernie Roman, manager of advanced lithography development, Motorola Semiconductor Products Sector, in Austin, Texas, thinks that to meet the needs of industry, throughput should be much higher. "We would be looking for throughput of 50 to 80 wafers per hour on any 45-nm production lithography tool that uses a mask," he says. "I'm not sure that it's possible with EPL."
Nevertheless, Nikon Corp. is pushing ahead with its development of the one and only electron projection system slated for commercialization. (The company declined to be quoted in this article.)
Electron Projection Lithography
Design and build a tool to print very small structures--65 nm and below--on semiconductor wafers
Why it's a Loser: Although electron projection lithography is capable of producing 45-nm structures and even smaller ones, it can't process enough wafers in an hour to satisfy industry expectations
Organization: Nikon Corp.
Center of Activity: Tokyo, Japan
Number of People on the Project: Confidential
Electron projection lithography works by passing a beam of energetic electrons through a mask containing holes that define the circuit pattern for one layer of the IC. The electrons print the layer's pattern on a photoresist--a film of photosensitive material--that covers the wafer. But since electrons are charged particles, they repel each other. This phenomenon limits the strength of the beam current that can be used to expose the photoresist. The smaller the current, the longer the exposure time needed to create the pattern. Unfortunately, as feature sizes shrink in future semiconductor generations, the problem gets worse, slowing throughput even more.
Another factor that limits throughput is that the diameter of the electron beam is slightly larger than a millimeter. As a result, the mask is divided into 1-mm2 subfields and the beam illuminates the subfields one at a time.
Nikon's electron optics, designed by IBM Corp. scientists in the company's Microelectronics Division, in Wappinger's Falls, N.Y., can steer the beam quickly over 20 1-mm2 subfields in succession without the need to move the wafer. But then the wafer must be shifted to a new position so that the beam can expose a different area. (IBM declined repeated requests for comment.)
Thus, to expose a typical 100-mm2 chip, the wafer, which currently can be as large as 300 mm in diameter, must be shifted many times. And repeatedly shifting the wafer adds a good deal of time to the exposure process. Today's optical lithographic systems, in contrast, can expose an entire chip in a fraction of a second.
Beyond its weakness of low throughput, however, EPL has several strengths. For one thing, electron-beam optics similar to EPL's has been around for decades and has reached a fine art.
One type of system, called direct-write, uses an extremely narrow beam of electrons to create the patterns on the chrome masks used for optical lithography. In fact, many development laboratories already use such a system to develop advanced devices.
Another strength is the electron projection system's incredible depth of focus. There's an inverse relationship between the numerical aperture of an optical system--a number analogous to a camera's f-stop--and the depth of focus. Because the wavelength of the electrons is only a fraction of a nanometer, the numerical aperture of the electron optics can be extremely small. Nikon's tool has a numerical aperture of less than 0.02, giving it a depth of focus of micrometers.
Compare that result with today's optical tools, which have pushed numerical apertures close to the fundamental limit of 1.00 to obtain good resolution "A Little Light Magic," IEEE Spectrum , September, pp. 3439]. "With optical tools," says Obert Wood, a senior technologist at International Sematech in Austin, Texas, "there's no depth of focus left. You have to be within a tenth of a micrometer or so of the optimum focus or the [chip] features blur and enlarge."
Electron projection's large depth of focus is a strong advantage in printing transistor contacts and vias, which join two different levels of metal wiring.
Both require deep, narrow holes in the photoresist. Indeed, if the technology is accepted by industry, it will be used first to form these levels on the wafer.
Still, the large depth of focus may not be enough of a motivation for companies to accept electron projection lithography. "The industry has been able to deal with shrinking depth of focus in optical lithography pretty much since its inception," says Motorola's Roman. "And I firmly believe that we will be able to continue to do that."
While throughput is an important factor for semiconductor manufacturers, a high yield may be more important. After all, what good is a high throughput if the yield of good chips is poor?
A third strength of electron projection is the relative simplicity of its masks. That simplicity should make them much cheaper than optical masks, which now require expensive correction techniques like phase-shift masks and double exposures [see, again, "A Little Light Magic"]. The projected cost of an optical mask set for next-generation ICs, which is expected to exceed US $1 million, may not be worth it for those ASIC manufacturers who need only a few wafers at a time for their product. These manufacturers may care less about throughput than they do about mask costs.
Besides electron projection, several other contenders are vying for the top spot in next-generation lithography, the most popular of which is the extreme-ultraviolet approach, now being strongly promoted by Intel Corp., Santa Clara, Calif. An extreme-ultraviolet tool uses light--actually soft X-rays--with a wavelength of only 13.5 nm.
The system employs reflective optics rather than the refractive optics of today's tools, and the mirrors and masks are complicated structures with many thin, atomically smooth alternating layers of molybdenum and silicon. In addition, the throughput of these tools so far is also limited, not by the basic physics but because the ultraviolet beam isn't strong enough for fast exposure.
Developers are hoping that extreme-ultraviolet tools and the infrastructure needed to support them will be in place in 2009, when the semiconductor industry is scheduled to begin producing ICs with feature sizes of 45 nm.
Optimistically, electron projection lithography may not have to compete with extreme-ultraviolet. In fact they could peacefully coexist in a wafer fabrication facility, with electron projection used to make the contacts and vias and extreme ultraviolet used for the other critical levels.
Nikon has already built two experimental systems and has shipped one of them to Selete, short for Semiconductor Leading Edge Technologies Inc., a consortium of Japanese companies developing the infrastructure--the mask, photoresist, and other technology--needed for electron projection.
All told, it would be an advantage for electron projection lithography to gain a foothold in the industry before extreme ultraviolet comes on-line in 2009. Otherwise, if extreme ultraviolet can make contacts and vias well enough to satisfy the requirements of semiconductor manufacturing, it may not matter that electron projection can make them better.