SPIE Handbook of Microlithography, Micromachining and Microfabrication, Volume 1: Microlithography

Section 2.5 Systems: 2.5.4 Gaussian vector scan systems

2.5.1 Environment
2.5.2 SEM and STEM Conversions
2.5.3 Commercial SEM Conversion Systems
2.5.4 Gaussian vector scan systems
2.5.5 Gaussian Spot Mask Writers
2.5.6 Shaped Spot and Cell Projection Systems
2.5.7 SCALPEL
2.5.8 Other E-Beam System Research
2.5.9 Electron Beam Fabrication Services

Table of Contents

Like the converted SEMs, Gaussian vector scan systems use the writing strategy of stopping in each field, deflecting the beam from shape to shape, and filling in the shapes with a raster pattern. Large commercial systems, however, break the deflection into two (or more) sections, usually making use of a 16-bit DAC for "subfield" placement, and a faster 12-bit DAC for deflection inside the subfield (see Fig. 2.15). This is the scheme used in systems from JEOL, and some of the systems from Leica. Leica's EBPG series, and the Vector Scan (VS) tools built by IBM use an alternative technique: the slower DACs are used for placing the origin of each primitive shape and the faster DACs are used for filling in the shape. In addition to deflecting the beam with separate DACs, systems from Hitachi and Leica use these separate DACs to drive physically separate deflectors (magnetic or electrostatic). JEOL systems, in contrast, use a single stage electrostatic deflector. Single stage deflectors have fewer problems with matching deflections of the "fast" and "slow" electronics, but sacrifice some speed.

The largest distinction of these commercial Gaussian spot systems (and in fact all commercial e-beam systems) is the use of high precision laser-controlled stages. Stage controllers from Hewlett-Packard or Zygo use the Zeeman effect to split the line of a He-Ne laser. The split-frequency laser beam is reflected off a mirror attached to the stage, and the beat frequency from the two lines is measured by high speed electronics. When the stage moves, the beat frequency shifts according to the Doppler effect, and the stage position is calculated by integrating the beat counts. While often referred to as "interferometers," these stages actually have more in common with radar speed guns.

Analysis of multiple points on the stage mirror allows the measurement of X, Y, and rotation about Z (yaw). Stage precision is often given in terms of a fraction of the laser's wavelength; a precision of /128 = 5 nm is commonly used in commercial systems, and the best stages now use /1024 = 0.6 nm. Even though the controller reports the stage location to this precision, the accuracy of the stage is limited by unmeasured rotations about the X and Y axes, and by bow in the mirrors. These nonlinearities, called "runout", limit the absolute placement accuracy to the order of 0.1 um over 5 cm of stage travel.

The high precision in reading the stage position means that the stage motors and drive do not have to be highly refined. In fact, simple capstan motors and push rods have been used at IBM. [52-53] The stage controller receives a target location from a computer, drives the motors to a point close to this location, then sends an interrupt back to the computer and corrects the field position by applying an electronic shift. This shift is applied continuously, in real time, to compensate also for stage drift and low frequency vibration. In comparison, the laser stage built by Raith for SEM conversions applies corrections to relatively slow piezoelectric translators on the stage itself. By moving and measuring an alignment mark at various locations in the writing field, laser stages are used to calibrate the deflection gain, deflection linearity, and field distortion; that is, the stage is used as an absolute reference, and the deflection amplifiers are calibrated using the stage controller.

Other common features of commercial systems include a flat stage, a fixed working distance (contrasting with a SEM), and automated substrate handling. A flat stage keeps the sample in focus but requires the use of a detector either on or above the objective pole-piece. Most commonly, a microchannel plate or a set of silicon diodes is mounted on the pole-piece.

The market niche for commercial Gaussian spot high resolution e-beam tools has been primarily in research, and to a lesser extent for small-scale production of MMICs, high-speed T-gate transistors, and integrated optics.

Table 2.1. Characteristics of SEM-based lithography systems. In all cases the resolution is high, depending (for Nabity and Raith) on the chosen SEM. All of these systems have relatively small stage motion, ~ 2 in. The Nabity and Raith devices are add-on products, while the Leica Nanowriter is an integrated system.

2.5.4.1 JEOL systems

JEOL's popular JBX-5DII Gaussian vector scan system uses a LaB₆ emitter running at either 25 or 50 kV. Figure 2.17 shows the 5DII with two condenser lenses and two objective lenses. Only one of the objectives is used at a time; the operator has the choice of using the long working distance lens for a field size of 800 um, or the short working distance lens, for an 80 um field at 50 kV. (The fields are twice as large at 25 kV.) The pattern generator runs at 6 MHz (> 0.167 us per exposure point) and the stage has a precision of /1024 = 0.6nm. As with all commercial systems, alignment, field stitching, and sample handling are fully automated. In fact, one drawback for research purposes is that there is no manual mode of operation. The system is capable of aligning to within 40 nm (2) and writing 30 nm wide features over an entire 5 in. wafer or mask plate. JEOL systems are known for their simple, high quality sample holders. The 5DII is one of the highest resolution (though not one of the fastest) e-beam tools in the LaB₆ class.

FIGURE 2.17 Schematic of the JEOL JBX-5DII system with LaB6 emitter. The system features two objective lenses for two different working distances (courtesy of JEOL Ltd.).

JEOL's JBX-6000 implements a number of improvements on the 5DII. The LaB₆ emitter is replaced with a thermal field emitter, eliminating the need for one of the condenser lenses. The pattern generator speed is increased to 12 Mhz, and the PDP-11 controller is replaced with a VAX. The system uses the same set of two objective lenses, and for a given objective lens the magnification is fixed (that is, the DAC's deflection is not scaled with the field size). As can be seen in the graph of figure 2.18, the ultimate spot size is somewhat improved over that of the LaB₆ machine, but more importantly, the current density at smaller spot sizes is greatly improved. The JBX-6000 runs at 25 kV or 50 kV.

With higher current density comes the property that the probe size is sometimes smaller than a pixel. For example, consider a pixel grid of spacing 0.0025 um. If the rastering beam skips every n grid points, then the pixel area is (n 0.0025 um)². With a current of 10 nA and a dose of 200 uC/cm², we must have (n 0.0025 um)² 200 uC/cm² = 10 nA (exposure time for one pixel), and since the minimum exposure time is 1/(12 MHz) = 0.08 us, the smallest value of n is 9. In this case the pixel spacing is 22.5 nm and the spot size, according to Fig. 2.18 is 12 nm. In this example the pixel spacing is larger than the spot size, and the exposed features may develop as a lumpy set of connected dots. The problem will be even more pronounced when using high speed resists, large field sizes, and larger currents. One solution would be to implement a faster pattern generator; however, JEOL's approach is to retain the superior noise immunity of the 12 MHz deflector and instead to use less current when necessary, or to increase the spot size by using a larger aperture. Alternatively, one can purposely defocus the beam. The NPGS system (see Sect. 2.5.3.1) attacks the problem by allowing different pixel spacings in X and Y (or in r and ).

It is interesting to note that future high resolution systems under development at Hitachi [54] are likely to resemble the JEOL Gaussian-spot tools, with field sizes >= 500 um and a single stage electrostatic deflector. Small fields avoid the complexities of dynamic focus and astigmatism corrections, and allow the short working distance needed to reduce the spot size. Single stage deflectors limit the bandwidth (speed) of the system, but improve intrafield stitching between deflections of coarse and fine DACs. The design tradeoff is clearly between high speed and high accuracy.

2.5.4.2 Leica Lithography Systems

Electron beam systems from Leica Lithography Systems Ltd. (LLS) are a combination of products previously manufactured by Cambridge Instruments, the electron beam lithography division of Philips, and most recently products from the former Jenoptik Microlit Division. Leica sells eight different models of Gaussian spot vector scan machines (the EBL Nanowriter has been described above). Systems in the mid-range of resolution include the EBML-300, a LaB₆ tool directly evolved from the Cambridge line, and the EBPG-5, a LaB₆ machine evolved from the Philips line. The EBPG-5 is comparable to the JEOL JBX-5DII in resolution but has accelerating voltage up to 100 kV. The EBMLand EBPG are both known for their versatile control software. On Leica's high end is the VectorBeam, with optics evolved from the Philips EBPG line and control electronics and software evolved from the Cambridge EBML line. The VectorBeam (Fig. 2.19) has a thermal field emission electron source running at 100 kV and a 6 in. stage motion with up to /1024 = 0.6 nm precision. The 25 MHz pattern generator has the useful feature that it is able to hold a small pattern in a buffer, so that repeated patterns do not have to be retransmitted to the pattern generator. This can significantly decrease the transmission overhead time when writing a large array of simple figures.

FIGURE 2.18 Probe beam diameter versus current for (a) a LaB6 cathode with a 120 um objective aperture, (b) a thermal field-emission (TFE) cathode with a 40 um objective aperture, and (c) a thermal field-emission cathode with a 100 um objective aperture. Data is from JEOL Gaussian-spot e-beam systems using 50 kV acceleration and a short working distance objective ("5th lens") (courtesy of JEOL Ltd.).

Leica e-beam tools are also distinguished from those of JEOL by their use of a single objective lens (one working distance), and scaleable writing fields with 2¹⁵=32768 or 2¹⁶ =65536 pixels across the field. In the case of the EBML-300, field sizes up to 3.2 mm may be used, although the benefit of using such a large field is debatable.

FIGURE 2.19 Schematic of the Leica VectorBeam 100 kV column with a thermally assisted field emission electron source (courtesy of Leica Lithography Systems Ltd.)

The largest systems from Leica are also equipped with 100 kV TFE emitters, and have stages with up to 8 in. travel. Additional features include a glancing-angle laser height sensor for dynamic field size corrections, and dynamic focus/astigmatism corrections -- features more commonly found on high speed maskmaking tools. Systems using large writing fields, with deflection angles exceeding 5 to 10 milliradians, make use of a number of higher order corrections including deflection linearization maps, field rotation maps, dynamic focus and stigmation tables, and even shift corrections for the dynamic focus coil.

2.5.4.3 Leica Lithographie Systeme Jena (Jenoptik) LION

One of the most unique Gaussian vector scan systems is the LION-LV1 from Leica Lithographie Systeme Jena GmbH,[55] a company better known for its large mask making machines (previously sold only in Eastern Bloc countries). The LION-LV1 combines a column designed by ICT GmbH (Heimstetten, Germany) with the pattern generator from Raith GmbH. This pattern generator has the unusual feature that it allows "continuous path control" of curves. In this mode the beam is held close to the center of the field while stage motion defines the shape of a Bezier curve. The ICT column is very similar to that used in the Leo 982 SEM, [51] except for the use of a beam blanker and higher bandwidth deflection coils (see Fig. 2.20). In this system, proximity effects are avoided by using beam energies as low as 1 to 2 keV. Although the voltage may be set as high as 20 kV, the system's selling point is low voltage -- avoiding both damage to the substrate and complications due to the proximity effect.

The column provides a spot size as small as 5 nm at 1 kV, through the use of an unusual compound objective lens. An electrostatic lens produces a diverging field, while the surrounding magnetic lens converges the beam. The complementary lenses reduce chromatic aberration, just as in a compound optical lens. A high resolution automated stage, substrate cassette loader, and substrate height measuring system complete the LION-LV1 as a full-featured system.

Low voltage operation avoids substrate damage and proximity effects, and offers the capability of three dimensional patterning by tailoring the electron penetration depth. However, the disadvantage is in greatly complicated resist processing. If the beam does not penetrate the resist, there will be significant effects from resist charging, [56] and placement errors due to charging may be dependent on the writing order and on the shape of the pattern itself. Charging may be avoided by using a resist trilayer with a conducting center (e.g., PMMA on Ti on polyimide), or by using a conducting overlayer (see sect. 2.7.1). Increased processing is required also for removing the resist layer over alignment marks. In a production environment this complexity adds significantly to the cost of ownership.

FIGURE 2.20 Low-voltage column developed by ICT GmbH, used in the LION e-beam system from Leica. The beam blanker is directly above the anode. The objective lens combines electrostatic and magnetic elements to reduce the net chromatic aberration. Beam diameter at 1kV is approximately 5nm. (Courtesy of LLS Jena GmbH.)

Table 2.2 Comparison of Gaussian-spot, vector-scan systems. All of these systems are equipped with thermally-assisted (Schottky) field emission electron sources.

Next Sub-Section: 2.5.5 Gaussian Spot Mask Writers

Table of Contents