Raith Overlay Alignment e-Beam over Optical Patterns
Roger Robbins 3/7/2017
Several students from the UTD Cleanroom have started an attempt to match a delicate e-beam pattern to a previously patterned optical layer using the manual P1 P2 P3 alignment scheme on the Raith 150TWO. This has proven to be unsuccessful by constantly missing the layer one pattern by a half micron up to almost 4 microns. I have explored this method with two e-beam pattern layers using the same manual alignment technique and experienced this same offset. This document proposes a reason for the offset and describes a more successful and accurate method of achieving the alignment needed by the students.
Manual Alignment Process
There are normally three manual alignment operations required before printing an overlay pattern on the Raith 150TWO. The first (manual) alignment for the sample sets the linkage between the sample edges (u,v) and the (x, y) coordinate system of the laser stage. The second (manual) align operation (writefield alignment) is used to achieve a coordinate link between the (u,v) coordinates of the sample edges and the electron beam scanning coordinate system that will orient the pattern. For overlays, the third (manual) alignment is the P1 P2 P3 alignment to markers on the first level pattern that matches the second level pattern to the coordinate system of the first layer. From this information the electron beam can print a pattern in alignment with the sample edges as well as orient the beam scan in location and angular orientation to the same coordinate system as the (u,v) coordinates of the previous pattern. The final alignment step is to match the writefield size and orientation to the previous layer writefield size and angle. This is generally an auto alignment step which will enable both overlay and “stitching” accuracy for geometries that cross write-field boundaries.
Figure 1 shows the misalignment after purely manual P1 P2 P3 alignment. Errors close to 4 microns appeared repeatedly in multiple alignment attempts. The offset was similar in amplitude but varied somewhat in direction between samples. All of the patterns in an array would have the same offset. This implies that the alignment system was working properly, but just had the wrong data to achieve overlay accuracy. This data came from the P1 P2 P3 manual alignment attempt.
Figure 1. Errant Overlay results from P1 P2 P3 manual alignment. Note the second layer overlaid outer rectangles appear oriented at the same angle as the cross due to the initial manual write-field alignment performed before the P1 P2 P3 manual alignments. (Pattern design shown in the small image: grey cross is layer one, red box is layer 2)
Speculation on the Source of the Alignment Offset
Proposed misalignment mechanism: Charging from long term scanning for focus and cross-hair aligning builds up a strong electric field above the surface which deflects the electron beam which shifts the pattern location the scan sees and thus creates the shift that produces misalignment. Note the region of high exposure in Figure 2 which caused the e-beam resist to change colors and stick through the development step. Around this will be the lesser exposed resist that washed away during development. In any case, this combination of concentrated electron bombardment of the resist would be the region where charge build up would have deflected the electron beam attempting an alignment. If the auto-alignment scan using the outer markers feels a push from this electric field, the correction logic would cause a misplacement of the marker location depending on the amount of charge and thus the amount of deflection away from the actual marker location.
Figure 2. Photo of first P1 marker scanned by the electron beam to locate the center cross for manual cross-hair alignment in addition to the time spent attempting to focus the e-beam spot. Note the developed spots on a -45 degree line in the upper left portion of the image – This represents the upper left corner extent of the search scan to find the center of the big marker in the center of the field. These large area scans would also land charge on the smaller corner markers used in any Auto Align measurement and cause beam deflection; thus justifying the reasoning for making a separated P’1, P’2, P’3 marker set for auto-alignment.
Quelling the Offset with Auto Align Methodology
The students had been attempting to use manual alignment on 1st layer optical patterns and were having a misalignment of about 3 microns in x and about a half micron in y. I repeated the manual alignment method on a 1st layer pattern I wrote with the Raith and obtained the same ballpark errors (PMMA on Si). ( Figure 1) During the manual alignment coordinate adjustment (P1 P2 P3 markers) on the 1st layer pattern, I found that I had to make 3 to 5 rounds of alignment adjustments to have a repeatable alignment final check-up. I reasoned that the beam was being offset due to the charge buildup and the pattern was written with only one beam passage, thus not enough charge collected in the pattern features to create the same beam shift in the pattern as the alignment had and thus the overlay pattern was offset. So I created two (P1 P2 P3) markers sets; one for manual coarse alignment and the other for an automatic line-scan alignment series which would incur only one beam passage and thus not be shifted due to previous charge deposition. This turned out to work very nicely with proper overlay alignment on all patterns in the test array.
It could be argued that with a set of P1 P2 P3 manual align markers, the coarse alignment could be found using a center cross and the auto alignment could be performed on the smaller markers in the corners of the 100 um field of the manual marker set. I was suspicious of this because many times one has to search for the P1 P2 P3 manual align markers with a general imaging scan usually at lower magnification and thus would coat the entire 100 micron align marker field with charge and create auto align errors due to charging. Therefore I created the second set of markers out of caution for the auto-align sequence.
The .pls listing of commands to tell the Raith what to do is shown in Figures 3 and 4. Note the first .pls list is the first level patterning which includes writing the two sets of manual/auto align markers along with the first level pattern used as a reference for alignment of the second layer to the first. Thus the second .pls command list will only have the Level 61 auto align marker scan performed before writing the matching second level pattern.
Figure 3. Command list for printing the first layer
Figure 4. Command list for aligning and printing the second (overlay) level.
Results of Auto Align Method
I have tested two alternative methodologies of using secondary auto align marks, both worked for the overlay pattern that I created and wrote with the Raith system. The first trial used a second set of P’1 P’2 P’3 markers to avoid the manual align offset errors. In examining the images of the auto align scans I found that the first (P’1) scans of the 4 small corner markers were off by micron sized errors, but still crossed the extended limbs of the marker crosses – thus obtaining the proper auto align data. (See Figure 5) But the second and third (P’2, P’3) markers were scanned with the measuring scan in the center of the marker limbs. This implies that the final alignment is determined in one auto align scan field inside the 100 um writefield – not a combination or average of the three auto align marker locations. This then implies that only one auto-align measurement is needed to align the entire overlay pattern to the previous level. However, if the alignment spec is extremely rigorous then the single field auto alignment might be required to be repeated for subsections of the large layer pattern.
This single P’3 auto align data was tested in a second overlay scheme in which I placed the “P’3” marker in the center of the first level pattern and just aligned on it for the auto align data collection. This method produced overlay alignment accuracy as good as the first tried (P’1 P’2 P’3) alignment. (Figure 8)
Figure 5. Obtaining proper alignment on P’1 even with manual offset.
Figure 6. Better alignment on P’2 after first try.
Figure 7. Duplicated accuracy on P’3. Note that it appears that alignment data from this single P’3 alignment was used for the entire overlay level placement.
The alignment data collected by the manual P1 P2 P3 alignment can only collect location and orientation data for the overall pattern array. If this data is correct, patterns will be placed in the proper location and array orientation, but there is no data from the manual alignment that calibrates the magnification or angular orientation of the write field. This means that the geometry pattern inside any of the write fields could still have errors (resulting from charging during the standard manual write field alignment procedure at the corner of the chip) which could spoil alignment across write field boundaries. It is the auto-align procedure using alignment data from markers at the corners of the write field that provide this local high accuracy data for proper final alignment. Figure 8 shows an alignment example from the 4 mm array of overlay test geometries using the auto align methodology discussed in this document.
Figure 8. Accurate Auto alignment over 4 mm array of alignment test geometries – Gap is one micron on all four wings of central cross.