Usually the easiest way to generate the CAD for a clear field mask is to drawy your pattern as if it were a dark field mask and then invert the pattern afterword using a logical operation. CNF has CAD software that provides this functionality should it be needed.
The tone of the mask you produce should also consider the tone of the resist that you will use to image your pattern. A dark field mask printed into negative tone resist will look the same as a clear field mask imaged into a positive tone resist. The various outcomes of combining mask tones with different resist tones are shown in the following figure.
The easiest way to select the right combination is to consider the steps following your lithographic process. Are you going to etch the wafer? Are you going to perform liftoff? This will guide you to the best resist use and, ultimately, to the tone of the mask you need to make. Here are some general examples:
- Lift off of metal electrodes using image reversal: use a clear field mask with a positive tone resist.
- Etching trenches into a Si wafer: darkfield mask with a positive tone resist.
- Etching mesas into a Si wafer: clearfield mask with a positive tone resist.
- Fluidic channels in SU-8: clearfield mask with SU-8 resist (negative tone).
Click here to return to the top of this page.
Click here to go back to the CNF Process Technologies page.
Guidelines for GCA 3600F Pattern Generator
The PG exposes variably sized rotated rectangles onto a mask blank. These rectangles are defined by a set of precision controlled moveable aperture blades. CAD pattern data is fractured into a format such that the entire pattern can be described by these rotated rectangles. Consequently, pattern write time is entirely dependent on the both the complexity and amount of exposed areas. The rate of exposure is usually in the range of 4000 - 5000 exposures/hour where each exposure--commonly referred to as a flash--is one rotated rectangle.
The geometries used in your pattern can have a significant impact on the number of flashes it takes to create your pattern. Here are some considerations to make when laying out your pattern.
Angles
Shapes with 90 and 45 degree angles can be written quite efficiently in the PG. Shapes with obtuse angles can require more flashes than those composed of shapes with ony 90 and 45 degree angles. This stems from the fact that it takes at least two rotated rectangles to define an obtuse angle. This is demonstrated in the following figure where the upper shape contains onle 45 and 90 degree angles and the lower contains 90 degree angles and one obtuse angle.
Shapes with acute angles can require significantly more flashes compared to shapes with only 45, 90 and obtuse angles. Consider the example shown in the following figure where the shape is composed of 90 degree angles and one acute angle. The data increases by a factor of 5 compared to the data required to print a shape with only 90 and 45 degree angles shown above.
Sides
Shapes with more sides require more flashes. Going from a rectangle to a pentagon increases the amount of flashes by a factor of 5 as shown in the following figure.
Circles
While some CAD tools will allow the definition of a true circle (infinite number of sides) GDSII does not. A circle must ultimately be defined as an n-sided polygon where n is some user defined number. For circles printed at the diffraction limit of a given exposure tool n=4 will print an adequate looking circle. This limit in feature size is given as follows:
- GCA AutoStep 200 and 6300 10x: 500 nm
- GCA 6300 5x: 800 nm
- Contact aligners: 2 microns
For features larger than this, n=4 will not be adequate. As the features increase n=4 will result in the printing of a square. Using an octagon to approximate a circle can produce good looking features up to 2-3 times the size of the diffraction limit. Octagons require 4 times the amount of flashes as a square, increasing the size of the data and, consequently, the write time.
The increase of data continues as n is increased. For a circle with n=80 and a radius of 100 microns 62 flashes will be required as shown below.
The maximum size of a single flash is fixed at 1.5 mm. For shapes larger than this, multiple flashes will need to be stitched together. This is demonstrated in the following figure. A 4.5 mm square requires 9 flashes instead of a single flash even though it is composed strictly of shapes with 90 degree angles.
As the diameter of the circle increases beyond the size of the maximum flash size the data experiences an enormous increase. A circle with a radius of 3 mm and n=80 will require over 1000 flashes as shown below.
For patterns with large circles, curves and arcs it may be faster to write your mask using the DWL 66. This tool employs a different writing strategy resulting in write times that are less sensitive to the complexity of the patterned data.
Intersecting Lines
Arrays of intersecting lines can require a surprising amount of data to print in the PG. Consider the following array of 9 intersecting lines. When this pattern is converted into PG data, it requires 27 flashes to print. This is largely because the algorithm used to translate CAD data into PG data does not take into account the fact that overlapping lines can sometimes reduce the flash count. This is illustrated in the figure below.
One way around this problem is to draw the pattern as two separate layers (shown below). By writing the mask as two separate layers sequentially the amount of data can be reduced in the most efficient manner: 9 lines converting to 9 flashes.
Click here to return to the top of this page.
Click here to go back to the CNF Process Technologies page.
Guidelines for Heidelberg DWL 66
The Heidelberg Instruments DWL 66 is a laser direct write patterning tool. The writing technique employs both a stage and a beam scan. That is, the stage is scanned in one direction while the beam is scanned in the transverse direction. This creates a "stripe" where the width of the stripe is determined by the objective lens. This is shown schematically in the following figure. The beam and stage are scanned over the entire pattern area regardless of the presence of pattern data. If the beam is scanning in a region of the substrate where no pattern data is present it is blanked (i.e. the beam path is blocked). The beam is unblanked in regions where pattern data is present. This means that write time is highly dependent on the extent of the pattern data, not on the pattern data itself.
CNF currently uses two different working distance lenses, a 10 mm and a 2 mm. For the 10 mm lens, a stripe of 50 microns is created while a 10 micron stripe is created by the 2 mm lens. This results in write times that are at least 5 times longer when using the 2 mm lens. However, the resolution of the 2 mm lens is significantly better than the 10 mm.
When CAD pattern data is converted into the DWL 66 format it is essentially broken up into a series of horizontal rectangles. This is depicted in the following figure. Each box represents an area where the laser is exposing the substrate. This results in a data stream that is basically an instruction to blank or unblank the beam.
The beam scan is always parallel to the X-axis of the stage motion and pattern data. If pattern data that requires a lot of beam modulation (i.e. bkanking and unblanking) is oriented perpedicular to the scan axis a significant amount of data results. For example, a grating oriented such that the lines are parallel to the Y-axis will generate more data than a grating oriented parallel to the X-axis.
Pattern data is stored in a 1 MB data buffer during the write. When the buffer is depleted the stage stops, the buffer is reloaded and patterning continues. Patterns that require intense beam modulation fill up the data buffer quickly. If the extent of the pattern is larger than the distance the stage can travel without having to reload the buffer write times can increase dramatically. The suggestion made in the above figure can reduce this problem for gratings. For large 2D arrays of repetitive shapes (circles, squares, triangles) this problem can be particularly vexing. Unfortunately, there is no known solution for these situations.
Contact Masks in the DWL 66
In general contact masks fall into two categories: (1) masks where the same pattern is repeated multiple times and (2) the mask contains no clearly repeatiing pattern. In the first case, converting a single pattern and arraying it in the DWL 66 control software can lead to faster write times. This prevents the beam from scanning in the areas between each pattern where no pattern data is present. This is shown int he following figure. For the second case, converting the entire pattern is the only option and the writing cannot be optimized further.
Stepper Masks in the DWL 66
The pattern data for stepper masks usually consists of the die area. For the AutoStep 200, special marks for aligning the reticle are usually included in the CAD because the PG cannot generate them automatically. However, these can be generated automatically using the DWL 66. This means that pattern data for stepper masks should only include the die area and NO reticle alignment marks. Including these marks will significantly increase the write time of most stepper masks
Click here to return to the top of this page.
Click here to go back to the CNF Process Technologies page.
Guidelines for EBL maskwriting
Pattern preparation for EBL mask writing is covered under the EBL pattern preparation page.
