2.7.3 Electron-Beam

Up to now all presented lithography techniques have been based on photon exposure radiation. However, particle beams like accelerated electrons can also be used for lithographic tasks. To compare the different methods, the De Broglie wavelength of the electrons is approximately calculated from^c $\lambda$ $\cong$ 1.23/$\sqrt{V}$, whereby $\lambda$ is given in nm and the accelerating voltage V in V. For example, electrons accelerated at 10-100 keV would have a wavelength of 0.0123-0.0039 nm, which is one to two orders of magnitude smaller than the wavelengths used with X-ray exposure. Hence, the resolution to be achieved theoretically is extremely high, 50 nm and below are possible. There are two major problems involved in electron-beam lithography, namely throughput and particle-to-particle interaction.

The throughput is too low for manufacturing floor usage due to the scanning exposure mode that writes a pattern sequentially into the resist. Direct-write tools sweep a finely focused Gaussian beam spot of fixed diameter across the wafer. The beam is either blanked or unblanked at the exposure sites commonly called excel. An excel is typically one-fourth or one-fifth of the minimum feature size, e.g., 10 nm for 50 nm linewidth. Two different motion types are used, raster-scan and vector-scan. Variably shaped beams are also used to increase throughput. However, for wafer-writing applications the throughput is too low and will remain too low for the foreseeable future. Therefore, direct-write systems are primarily used for the fabrication of UV and X-ray masks, research and development, and other applications, where a high resolution and short turn-around-time are more important than throughput.

An alternative approach is based on a multibeam exposure system realized by arrays of some kind of field-emitting sources or locally addressed photocathodes. This would solve the throughput problem for direct-write tools, but due to the high current densities stochastic particle-to-particle interactions blur the final image. The beam ``thermalizes'' and the forward momentum converts to a transverse momentum, a phenomenon also known as Boersch effect [27,28]. Hence, there is a fundamental tradeoff between beam number versus blurring. Further problems arise from the heating and charging of the resist. Temperature changes between 10-20 ^oC can occur and cause uncontrollable linewidth variations. Inhomogeneous charging might also lead to significant feature displacement.

Alignment is especially important to fully utilize the high resolution capability of electron beam lithography. Reference based alignment schemes with alignment fiducial well away from the writing field lack of calibration accuracy. A spatial phase-lock approach with marks placed inside the mask enables local alignment, and accuracies better than 5 nm have been reported. However, the ultimate resolution of electron-beam lithography is limited by beam-to-matter interactions. While beam diameters in the order of 1 nm are possible, the beam splays out once it has struck the solid. This problem can be analyzed by considering the point spread function of the electrons, which is typically composed of two parts, namely a relatively narrow peak caused by forward scattering, and a broad background due to backscatter and second order effects. The resolution of isolated features is limited by the width of the narrow peak, e.g., 10 nm for 100 eV-electrons. For ``real world'' patterns the broad background component of even far away excels starts to sum up to significant amounts of deposited energy. This phenomenon is called proximity effect and seriously reduces the achievable resolution. Possible solutions include a thin interlayer between resist and substrate to filter out backscattered electrons, a scaling of the mask patterns, as well as the so-called GHOST technique [29,30,31,32]. In this technique also nominally unexposed regions are weakly irradiated to homogenize the background and thereby equalize the energy within the desired excels. However, for sub-quarter-micron dimensions also the GHOST approach fails and has to be replaced by ``dose modulation.'' Here, individual excel doses are varied in such a way to compensate forward and backscattering distortions.

A totally different approach tries to implement high-throughput electron projection systems. This technology was pioneered at the AT&T Bell Laboratories at Lucent Technologies, Murray Hill, NJ, under the acronym SCALPEL [33].^d A broad beam of electrons is collimated and incident upon a mask made from an approximately 100 nm thick, low-stress silicon nitride film on which a thin metal pattern is placed, e.g., 50 nm tungsten or gold. The pattern does not absorb the beam but rather creates enough angular deflection to prevent the electrons from being brought to focus in the image plane. Such systems typically work with a demagnification of 4x. Absorbing stencil masks are avoided, which is a big advantage since disturbing heating effects do not occur. Full-field optics that do not scale with increasing die sizes and decreasing minimum feature sizes are not required. The numerical aperture can thus be larger and the beam currents can be smaller for an acceptable throughput. The resolution is therefore not destroyed by space-charge effects. Problems primarily arise from mask stability and robustness. The mask delicacy requires rib supports to the membrane that must be accounted for as the electron-beam is swept over the mask. These massive support struts between the cells carrying the features are not imaged onto the wafer since the patterns are shifted into place as they are illuminated. While the mask structure is similar to that used for X-ray lithography, the support struts provide greater dimensional stability [24], and the use of the reduction optics makes the mask fabrication simpler. Another serious problem is due to the proximity effect. A ``dark field'' effect can be employed by providing a GHOST image correction [34]. A portion of the scattered beam is admitted to the image plane through lateral apertures that are cut into the final SCALPEL aperture. This in effect creates a dark field image that mimics to some extent the GHOST background dose and provides some degree of contrast equalization. The throughput of a fully-developed SCALPEL tool is expected to be comparable to that of an optical stepper, while delivering resolution on the scale of 100 nm. However, several questions remain concerning its practical use: At energies in the 100 keV range resists are proportionally less sensitive, and the energy delivered to the substrate will be larger than in other electron-beam systems. The effects this may have on transistor thresholds and mobility are still unknown.

Another emerging technique with potential applicability for high-resolution lithography on an industrially competitive scale is based on proximal probes, such as scanning tunneling microscope (STM) [35] and atomic force microscope (AFM) [36]. The dramatic advances in micro-electromechanical technology has made the fabrication of massively parallel arrays consisting of metal-coated tips and piezoceramic height sensors feasible.

Footnotes

... from^c

The exact expression for the De Broglie wavelength of an electron is given by $\lambda$ = h/p = h/$\sqrt{2m_0qV}$, where h = 6.626  10^-34 J/Hz is the Planck constant, p = $\sqrt{2m_0qV}$ the electron impulse, m₀ = 9.1095  10^-31 kg the electron rest mass, q = 1.602  10^-19 C the elementary charge, and V the applied acceleration voltage in V.

...[33].^d

The acronym SCALPEL stands for Scattering with Angular Limitation for Projection Electron Lithography. For detailed information on SCALPEL see the WWW at http://www.bell-labs.com/project/SCALPEL/.