Fracturing


For rectangular patterns, fracturing is pretty simple - the pattern is broken into simple rectangles.

Before, with polygons:


After fracturing into J52 (.v30) format, the pattern consists only of rectangles:





A slightly more complex pattern involves shapes with angled edges:


You can see these fracture into trapezoids. All patterns must be fractured into simple shapes: rectangles and trapezoids.




Curved structures are really just as simple -- they must be fractured into simple rectangles and trapezoids. So this segment showing two curved waveguides:



will get fractured into this:


A few more points about these structures. As mentioned in the CAD section, GDSII doesn't have the ability to store curved structures -- the best you can get are polygon approximations to curves. You can see this in the next figure, which shows the source waveguides. In this case, I'm showing a very useful capability of the LayoutBEAMER viewer, the "Pick". By right-double-clicking on a shape, that shape is identified, by showing its outline in black and highlighting its vertices. You'll also note some identifying text which showings the GDS-II cell, layer and datatype numbers, as well as computed exposure energy (1.0 in this case, but this is used in an advanced function called Proximity Correction.)
'



Finally, let's look at some simple circle structures, as this is a pretty common structure in the physics and materials world.
Here's the original structure, showing polygons of various numbers of vertices, from 4 to 100.


Now, LayoutBEAMER offers a choice of Fracture Mode, in the Export JEOL module, General panel, under Fracture Mode. The three choices, and sample outputs for these circles are:

Conventional



LRFT (Large Rectangle, Fine Trapezoid)



Curved




Which of these options is best? For mechanical reasons within the machine, the LRFT usually produces the best result, and this is the fracture mode I typically use.
Curved mode is a special capability of LayoutBEAMER that is still being developed.


The output examples are also labeled with the number of exposed shapes in each output file. Is this important? Well, that depends, If your pattern has a modest number of circular shapes, no, it probably doesn't matter. But if you have many curved shapes in a large pattern, it could make a difference, particularly in exposure time. Each shape has some finite overhead time during the exposure, so exposing, for example 65 shapes compared to 13 won't make much difference, but say you want a square millimeter of these circles, on a 130 nm center-to-center spacing. This would be 59,171,597 circles in a single square millimeter. In this case, the difference between a 13-shape and 65-shape pattern would be more than 3 Billion shapes. That would be a significant amount of overhead time, and if the circles are small enough and exposure conditions chosen properly, you wouldn't be able to tell any difference between a 12-sided and a 100-sided polygon approximation. In fact, at small enough dimensions, even the single shape square approximation is rounded sufficiently to look like a circle. In the SEM shown here, each circular dot was exposed as a single square shape.

]


In these photos, you can see side-by-side comparison of a 4-sided (square) shape on the left, and an 8-sided (octagon) shape on the right. No discernible difference in "roundness", but the right-hand, 8-sided pattern takes about 30% longer to expose. For a small enough array, this difference is possibly insignificant. For a large array, this could be hundreds of dollars in difference in price for machine time. Something to consider.