The semiconductor industry is racing to develop a lithographic tool that can pattern IC features smaller than 100 nm—45 percent smaller than the 180-nm features on today’s lCs. There is no time to lose. The Semiconductor Industry Association, San Jose, Calif., predicts lCs with 70-nm features by 2008. But refractive optical tools, which have been the mainstay of chip manufacture to date, are not up to the task of imaging such small dimensions.
A leading candidate for the job is projection e-beam lithography. Recently, its standing in the running was buoyed as a result of two key developments by researchers at IBM Microelectronics, in East Fishkill, N.Y.: a set of magnetic lenses for the electron optics whose axes can be moved along with the scanning beam, and a high-emittance electron gun.
E-beam lithography has much in its favor. Electron wavelengths at the energies typically used—between 50 and 100 keV—are shorter than 0.1 nm, promising practically limitless resolution.
A type of e-beam lithography, called direct-write is a standard tool for building prototypes of micro- and nano-structures. In this approach a narrow e-beam scans a small area of photoresist and rapidly turns on and off to create the pattern. Once the small area is exposed, the wafer is shifted to another location under the beam and the process is repeated. But this technique is not practical for manufacturing because wafer throughput is low: direct-write systems do not use a mask, but rather fill in the wafer one feature at a time.
Projection e-beam systems use a wider beam of about 1 mm and do not write it directly on a wafer. Instead, it is projected through a reticle (sometimes called a mask) containing the pattern to be exposed, just as light is projected through a reticle in optical systems [see L.R. Harriott, "A new role for e-beam: elec-tron projection," IEEE Spectrum, July 1999, pp. 4 1—45.]
Although the throughput of projection e-beam lithography is much greater than the direct-write approach, it still falls short of the capabilities of optical tools. "Electron optical systems are limited to project in a single flash fields that are smaller than I mm2," explained Hans Pfeiffer, manager of electron-beam technology at IBM Microelectronics. The entire chip is exposed sequentially by accurately stitching the small subfields together. A strictly mechanical approach of moving wafer and mask under a stationary electron beam is still much too slow for commercial applications.
What is needed, according to Pfeiffer, is a tool that combines the 1-mm2 beam with the scanning capabilities of the direct-write instruments, as well as scanning electron microscopes. So he and his team developed Prevail, which stands for projection reduction exposure with variable axis immersion lenses [see figure-sorry!].
Shifting the axis
In creating a system capable of scanning such an area as wide as 20 mm, a big obstacle to overcome was the effect of off-axis aberrations. When the beam is scanned, it deteriorates as soon as it leaves the magnetic axis, because radial components of the magnetic field not present along the field axis cause electrons to be deflected from their paths. "So we came up with a technique to shift the axis of the magnetic lens in synchronism with the shifted beam. We are fooling the electrons to believe that the beam is still on axis," explained Pfeiffer.
The shifting of the beam and the field is done electronically at very high speed by the variable axis lenses which superimpose the appropriate magnetic fields. Prevail has several deflection systems in the illuminator and collimator lenses:
some shift the beam and others shift the symmetry of the magnetic field. "That is a complex undertaking," said Pfeiffer, "but it is the core of our Prevail system. The ability to shift the axis of the magnetic field allowed us to reach a platform from which we can project wafer throughputs of up to 30 wafers per hour." (Today’s optical tools can produce 100 wafers per hour.)
Using the variable axis lenses, the researchers successfully eliminated blurring and distortion of the outer subfields. The 80-nm lines and spaces produced with the beam deflected 10 mm at the mask showed no difference in resolution from those made by an undeflected beam.
Another factor that limits the through-put of projection e-beam systems is the Coulomb interaction among the electrons, which blurs and distorts the image, the higher the beam current, the greater the blur. But a large beam current is needed for high throughput. So the IBM team developed a high-emittance gun that projects a large uniform beam onto the sub-field of the reticle.
Any constriction of the electron beam increases the Coulomb interaction, so the gun has a 10-mm surface of emission and a large angle of emission—both needed to uniformly illuminate large areas. To achieve a uniform beam, Prevail uses only about one-third of the electrons from the emitting source and focuses them down to about 1 mm2 at the reticle.
IBM is partnering with Nikon Corp., Tokyo, to develop a commercial system based on Prevail. The first tools should be available by year end 2003.
LINDA GEPPERT, Editor, IEEE Spectrum August 2000
Prevail uses a series of magnetic lenses[green-in diagram] to focus a 1 mmxmm electron bean onto a 1-mm2 subfield of a reticle, deflecting the over a total range beam of 20 mm so as to rapidly image 20 subfields in sequence. The image of each subfield is then reduced by a factor of four and projected onto the wafer. |
