Button: Contact CNFButton: MultimediaButton: About CNF
Button: Getting StartedButton: PublicationButton: REU ProgramButton: Events & SeminarsButton: Education OutreachButton: TechnologiesButton: Lab Equipment

Button: Lab User

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

Section 2.5 Systems: 2.5.2 SEM and STEM Conversions

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
2.5.8 Other E-Beam System Research
2.5.9 Electron Beam Fabrication Services
Table of Contents

2.5.2 SEM and STEM Conversions

Any tool for microscopy - optical, electron, or scanning probe - may be adapted to work in reverse; that is, for writing instead of reading. Converted electron microscopes suffer the same limitations as light microscopes used for photolithography, namely, a small field of view and low throughput. Nevertheless, for a subset of research and R&D applications, converted SEMs offer a relatively inexpensive solution.

Of the many custom designed SEM conversions, most use a single set of digital-to-analog converters (DACs), from 12 to 16 bits wide, to drive the scan coils of the microscope. The beam is modulated with an electrostatic or magnetic beam blanker, which is usually located near a crossover of the beam. Alternatively, the beam can be blanked magnetically by biasing the gun alignment coils or not blanked at all. In the later case, the beam must be "dumped" to unused sections of the pattern. Figure 2.15 illustrates the "vector scan" method, in which shapes are filled with a raster pattern and the beam jumps from one shape to the next via a direct vector. By taking over the scan coils and beam blanking, a SEM can be used as a simple but high resolution lithography tool.

SEM conversions have evolved greatly in the past twenty years, primarily due to improvements in small computers and commercially available DAC boards. Early designs used relatively slow computers that sent primitive shapes (rectangles, trapezoids, and lines) to custom hardware. The custom pattern generator filled in the shapes by calculating coordinates inside the shapes and feeding these numbers to the DACs. While this approach is still the best way to avoid data transmission bottlenecks (and is used in commercial systems), inexpensive SEM conversions can now rely on the CPU to generate the shape filling data. A typical configuration uses an Intel CPU based PC, with a DAC card plugged into an ISA bus. In this case, the CPU can generate data much faster than it can be transmitted over an ISA bus.

  FIGURE 2.15 The vector-scan writing strategy. (a) Patterns are split into "fields". The stage moves from field to field, as shown by the arrows. Full patterns are stitched together from multiple fields. (b) In many vector-scan systems the fields are further tiled into subfields. A major DAC (16 bits) deflects the beam (a small "Gaussian" spot) to a subfield boundary, and a faster DAC (12 bits) deflects the beam within a subfield. SEM conversion kits typically do not include the faster 12-bit DAC. (c) The primitive shape is filled in by rastering the spot. Between shapes the beam is turned off ("blanked") and is deflected in a direct vector to the next shape. An alternative deflection strategy (not shown) is to use the major DAC to deflect the beam to the origin of each primitive shape

The bus limits the deflection speed to around 100 kHz, that is, to a dwell time per point of 10 us.

What dwell time is required? With a 16-bit DAC and a SEM viewing field of 100 um, the size of a pixel (the smallest logically addressable element of an exposure field) is 100 um/216=1.5 nm, and its area A is the square of this. The charge delivered to this pixel in a time t is It, where I is the beam current. This must equal the dose times the pixel area. Given a beam current I on the order of 50 pA and a required dose D around 200 uC/cm2 (typical for PMMA), we have a pixel dwell time

t = DA / I = 910-8 s, (2.1)

or a deflection speed of 11 MHz. This being impossible with an ISA bus, we must either space out the exposure points, apply a short strobe to the beam blanker, or use a combination of the two. When the exposure points are spaced every n pixels (that is, when the 216 available exposure points are reduced by a factor of n) then the "pixel area" and thus the dwell time is increased by a factor of n2. Note that the placement of features can still be specified to a precision of 216 within the writing field, while the shapes are filled in with a more coarse grid.

In the above example, we can set n to 11 so that the dwell time is increased to 1.110-5 s (91 kHz), increasing the pitch of exposure points to 16.5 nm. This spacing is a good match to the resolution of PMMA, and allows fine lines to be defined without any bumps due to pixelization. However, when we require 100 times the current (5000 pA in this example), the exposure point spacing must be increased by a factor of 10, possibly leading to rough edges. Some pattern generators (see Sect. avoid this problem by allowing different exposure point spacings in the X and Y (or in the r and theta) directions, thereby allowing a larger exposure point spacing in the less critical dimension.

To use a SEM without a beam blanker, one must consider the large exposure point spacing required for common resists. Lack of a beam blanker leads to the additional problem of artifacts from the settling of scan coils and exposure at beam dump sites. Many SEM manufacturers offer factory-installed beam blankers. Retrofitted blankers are also sold by Raith GmbH. [47]

The scan coils of a SEM are designed for imaging in a raster pattern and so are not commonly optimized for the random placements of a vector scan pattern generator. Settling times are typically around 10 us for a JEOL 840 to as long as 1 ms for the Hitachi S800, where the bandwidth of the scan coils has been purposely limited to reduce noise in the imaging system. Thus, it is important to consider the bandwidth of the deflection system when purchasing a SEM for beamwriting.

The other major limitation of a SEM is its stage. Being designed for flexible imaging applications, SEM stages are not flat, and even when equipped with stepper motor control are no more accurate than ~1 to 5 um. Periodic alignment marks can be used to stitch fields accurately, but this requires extra processing as well as the use of photolithography for printing alignment marks. The mark mask would presumably be fabricated on a commercial system with a laser-controlled stage. Fortunately, alignment with a converted SEM can be quite accurate, especially when using Moiré patterns for manual alignment. Automated alignment in the center of a SEM writing field is at least as good as in large commercial systems. Alignment at the edges of a SEM field will be compromised by distortions, which are typically much larger than in dedicated e-beam systems. Laser-controlled stages can be purchased for SEMs, but these are usually beyond the budgets of small research groups.

Electron beam lithography requires a flat sample close to the objective lens, making secondary electron imaging difficult with an ordinary Everhart-Thornley detector (a scintillator-photomultiplier in the chamber). A few high end SEMs are equipped with a detector above the objective lens or can be equipped with a microchannel plate on the pole-piece. These types of detectors are a great advantage for lithography since they allow the operator to decrease the working distance, and thus the spot size, while keeping the sample flat and in focus.

With patterning speed limited by beam settling and bus speed, it is clear that inexpensive SEM conversions cannot match the high speed writing of dedicated e-beam systems. However, a SEM based lithography system can provide adequate results for a wide variety of applications, at a small fraction of the cost of a dedicated system. The number of applications is limited by stitching, alignment, and automation. Practical applications include small numbers of quantum devices (metal lines, junctions, SQUIDs, split gates), small numbers of transistors, small area gratings, small masks, tests of resists, and direct deposition. The main limitations with SEM lithography are observed with writing over large areas, or when deflection speed and throughput are critical. Specifically, difficulties with stitching and/or distortions due to the electron optics of the microscope can become significant. SEMs are not practical for most mask making, integration of many devices over many fields, large area gratings, multifield optical devices, or any application requiring a large substrate.

Next Sub-Section: 2.5.3 Commercial SEM Conversion Systems

Table of Contents

Back to Top

Button: Search Button: Search Keywords
Cornell University