Specific Process Knowledge/Lithography/Aligners/Aligner: Maskless 01 processing: Difference between revisions
Line 275: | Line 275: | ||
==Writing speed== | ==Writing speed== | ||
The site acceptance test performed during the installation of the Aligner: Maskless 01 showed an exposure time/speed of 0.025 min/ | The site acceptance test performed during the installation of the Aligner: Maskless 01 showed an exposure time/speed of 0.025 min/mm<sup>2</sup>, which is slightly lower than the 0.02 min/mm<sup>2</sup> given in the specifications. At such speeds, a full 4" design would take 2.6-3.3 hours to expose. In practice, we observe exposure times in excess of 2 hours for a full 4" design. | ||
At such exposure speeds, we would quickly run into a bottleneck as more and more users expose more and more wafers, if full 4" designs are used all the time. Another concern is that the exposure dose seems to be unstable over long exposure times, as consecutive exposures using the same parameters have been observed to yield different lithographic results. The conclusion is that exposure using the maskless aligner requires a different design philosophy than when a mask aligner is used for the exposure. | At such exposure speeds, we would quickly run into a bottleneck as more and more users expose more and more wafers, if full 4" designs are used all the time. Another concern is that the exposure dose seems to be unstable over long exposure times, as consecutive exposures using the same parameters have been observed to yield different lithographic results. The conclusion is that exposure using the maskless aligner requires a different design philosophy than when a mask aligner is used for the exposure. |
Revision as of 10:48, 14 February 2018
Feedback to this page: click here
Process Parameters
Resolution
The pixel-size of the DMM is 0.5µmX0.5µm (at the sample surface), so this is the ultimate resolution of the Aligner: Maskless 01, and also the smallest step that will produce features of different size. The lithographic resolution of the machine is 1µm on paper, which was demonstrated in the acceptance test after installation. The resist thickness needed for this resolution is 0.5µm. In 1.5µm thick resist, the resolution is around 2-3µm.
Exposure dose
The exposure dose needed in the Aligner: Maskless 01 seems to follow the dose needed to process the same substrate in Aligner: MA6-2. As doses get higher, there is a tendency for the dose needed in the Aligner: Maskless 01 to exceed the dose needed in Aligner: MA6-2.
Defocus
The optimal defocus setting is probably a function of the resist thickness, but for 1.5µm resist, a defocus of -4 seems to be optimal.
Substrate positioning
During load, the machine will focus on the surface of the sample. Then, using the pneumatic focusing mechanism, it will detect the edges of the sample (this function dependes on the substrate template used) in order to determine the center of the sample. The following results rapport findings using the "4 inch wafer" template on a standard 100mm Si substrate.
Substrate centering
During (4") substrate detection, the sample is scanned along the X- and Y-axes, as well as diagonally. From these measurements, the diameter of the substrate is calculated, as well as the stage position matching the center of the substrate. This stage position will be the default origin for the subsequent exposure.
Unfortunately, the centering does not compensate for major or minor flats. The center position will therefore typically be displaced several hundred µm from the center of the substrate along the Y-axis due to the major flat, and possibly also shifted due to any minor flats. These shifts have successfully been compensated during file conversion, which yielded a centering accuracy of ±200µm. The table below lists the corrections needed during file conversion in order to compensate for shifts due to major and minor flat, dependent on the position of the minor flat. Keep in mind that the tolerances of substrate diameter and flat lengths in the SEMI wafer standard introduce uncertainties to these numbers in the order of 0.4-0.9mm. Accurate positioning of the design relative to the substrate center would require measurement of diameter and flat lengths of the individual substrate, and subsequent recalculation of the correction values.
Minor flat position | X correction [mm] | Y correction [mm] |
---|---|---|
none | 1.35 | 0 |
left | 1.35 | 0.4 |
up | 0.95 | 0 |
Flat alignment
At the end of (4") substrate detection, the sample is scanned twice along the flat, in order to determine the substrate rotation. This angle will be presented in the exposure panel along with the option to expose the design rotated in order to compensate for this angle, i.e. aligned to the flat.
The flat alignment accuracy has been measured to be 0±0.1° (1.7mRad) quite consistently. Out of a total of 15 exposures, 13 showed misalignment of 0.1° or better, despite initial sample rotations exceeding 5°.
Alignment
The alignment accuracy of the Aligner: Maskless 01 is a combination of the position accuracy of the stage, the accuracy of the alignment mark detection, and the accuracy of the pattern already on the wafer (first print).
By measuring the stitching accuracy between two layers printed on the same substrate (without unloading the substrate), we can assess the stage accuracy. By aligning to a pattern previously exposed by the Aligner: Maskless 01, we can assess the mark detection accuracy. And finally, by aligning to a pattern exposed on a mask aligner, we can assess the mask-less aligner's ability to compensate for any scaling and orthogonality errors between the two machines.
The results reported here use printed verniers to assess the misalignment along the two axes at different points on the wafer using an optical microscope. Two different designs were used; a ±5µm vernier and a ±1µm vernier. Both consist of a scale of 5µm lines with 10µm pitch, and a vernier scale to enable subdivision of the 5µm or 1µm scale into tenths, i.e. 0.5µm or 0.1µm. During inspection, observation of the symmetry of neighboring lines enables the observer to read the shifts with ±0.25µm or ±0.05µm accuracy.
The measurements are used to calculate the misalignment of the second layer with respect to the first print: The shift [µm] is the median of all measurement points in X or Y; the misplacement [µm] is the amount by which the image is shifted; the rotation [ppm] is the angle by which the image is rotated; and the run-out [ppm] is the amount of gain in the image. The unit of ppm (parts per million) is used as the rotation and run-out are generally small. A rotation of 1ppm corresponds to an angle of 0.2" (arcseconds) or a shift of 100nm across an entire 4" wafer, while a run-out of 1ppm corresponds to a shift of 50nm at the edge of a 4" wafer compared to the center. For comparison, the pixel size at the wafer surface is 500nmX500nm.
The deviations (±) given for the results here are calculated as half the range of measurements. If the range is small, the measurement uncertainty is used in stead.
The samples used for these tests are 100mm Si wafers coated with a 1.5µm layer of the positive tone resist AZ 5214E.
The conclusion to the tests are that the stitching accuracy of the Aligner: Maskless 01 is ±0.1µm. The machine-to-self overlay accuracy is ±0.5µm (an alignment error of -2µm in X must be compensated, but this error may be subject to machine calibration). The machine-to-machine overlay accuracy could not be determined.
Stitching
In the stitching test, the design consists of ±5µm and ±1µm verniers along the X and Y axis placed in a 3 by 3 matrix covering a 60mm by 60mm area centered on the wafer. The sample is loaded, and the first layer (the linear scales) is printed. Without unloading, the second layer (the vernier scales) is printed on top of the first, and then the sample is developed.
The results in the table below show that the errors are at or below the measurement uncertainty for the stitching tests using no flat alignment. In other words, the stage is accurate to within 100nm.
When flat alignment is used, a rotation error of ~1.5ppm appears, along with a ~0.3µm misalignment and a -6ppm scaling of the Y axis. This level of accuracy (which corresponds to approximately half a pixel) is what we can expect when the rotation compensation is applied, i.e. when aligning to a previously printed layer.
Shift (median) [µm] | Misplacement [µm] | Run-out (gain) [ppm] | Rotation [ppm] | ||
---|---|---|---|---|---|
No flat alignment
(two samples) |
X | -0.03±0.08 | 0.00±0.05 | -0.97±1.67 | 0.42±1.18 |
Y | -0.05±0.05 | -0.07±0.05 | -0.42±1.67 | 0.00±1.18 | |
With flat alignment | X | 0.05±0.10 | 0.05±0.05 | 0.56±1.67 | -1.67±1.18 |
Y | -0.25±0.35 | -0.29±0.16 | -6.39±1.67 | -1.39±1.18 |
Overlay
In the overlay test, two alignment accuracies are assessed: The machine-to-self overlay accuracy (MLA-MLA), and the machine-to-machine (MA6-MLA) overlay accuracy. Because alignment is possible using two marks or four marks, both are tested in each case. Exposing the first print with or without flat alignment was also tested, but no significant effect was observed.
In the MLA-MLA overlay test, the design is the same as for stitching; ±5µm and ±1µm verniers along the X and Y axis placed in a 3 by 3 matrix covering a 60mm by 60mm area centered on the wafer. The sample is loaded, and the first layer (the linear scales) is printed. The sample is unloaded and developed. The second layer (the vernier scales) is aligned to marks contained in the first layer, and then the sample is developed again. The alignment marks used for 2 mark alignment are placed 60mm apart on the X axis, while the marks used for 4 mark alignment are placed at the corners of a 60mm by 30mm rectangle.
The results in the table below show an alignment error of -2µm in X. They also show that including the scaling and shearing of the axes calculated by the machine from alignment to four marks in the exposure only seems to introduce a gain error in Y. However, if the wafer has significant bow or other distortions due to processing of the first layer, scaling and/or shearing correction may be necessary.
Since the alignment error in X seems so consistent, the machine was calibrated by correcting the X-axis beam off-set 2µm in the machine configuration files. As seen below, this removed the alignment error, and suggests that the machine-to-self overlay accuracy of the Aligner: Maskless 01 is ±0.5µm or better.
MLA-MLA | Scaling [ppm] | Shearing [mRad] | Shift (median) [µm] | Misplacement [µm] | Run-out (gain) [ppm] | Rotation [ppm] | |
---|---|---|---|---|---|---|---|
2 alignment marks
(three samples) |
X | NA | NA | -2.08±0.25 | -2.07±0.25 | -2.31±8.33 | 1.02±1.67 |
Y | NA | -0.33±0.18 | -0.34±0.19 | 1.20±1.67 | -0.46±5.89 | ||
4 alignment marks
(three samples) |
X | 1±2 | 0.002±0.001 | -2.08±0.25 | -2.08±0.25 | -3.70±8.33 | 0.56±1.67 |
Y | -12±1 | -0.22±0.35 | -0.21±0.08 | -6.30±1.67 | -0.93±5.89 | ||
2 alignment marks
after calibration (three samples) |
X | NA | NA | -0.22±0.25 | -0.19±0.19 | 1.85±2.08 | -0.09±2.92 |
Y | NA | -0.28±0.38 | -0.28±0.13 | 4.35±5.97 | -0.09±2.64 |
In the MA6-MLA overlay test, a mask from an existing design (GreenBelt METAL v2, dark field) is reused to print the first layer using the Aligner: MA6-2, and the sample is developed. The second layer is aligned and printed in Aligner: Maskless 01, using a design containing four sets of ±5 vernier scales located at the corners of a ~90mm by ~3mm rectangle, before being developed again. For 2 mark alignment, the original alignment marks at X±~43mm are used. Since only two alignment marks exist in the original design, four corners (positioned in a ~60mm by ~60mm square pattern) are used for manual alignment in the 4 mark test. Using corners rather than crosses reduces the accuracy of the alignment, as the positions of corners are subject to shifts due to bias in the first print lithography.
The results in the table below reproduce the alignment error in X seen in the machine-to-self overlay tests (should be better now that the machine has been calibrated). Due to the close proximity of the verniers in Y, run-out and rotation have not been calculated for that axis. But the X axis shows some run-out, consistent with alignment between two stages (the mask writer and the maskless aligner) without scaling compensation. The fact that using scaling and shearing compensation doesn't remove this gain, and seems to add a significant rotation error, is probably more due to the alignment "marks" used than to the machine itself. More tests using a purpose-designed mask would have to be conducted in order to estimate the machine-to-machine overlay accuracy of the Aligner: Maskless 01.
MA6-MLA | Scaling [ppm] | Shearing [mRad] | Shift (median) [µm] | Misplacement [µm] | Run-out (gain) [ppm] | Rotation [ppm] | |
---|---|---|---|---|---|---|---|
2 alignment marks | X | NA | NA | -1.50±0.75 | -1.50±0.25 | 6.94±5.56 | 2.78±3.89 |
Y | NA | 0.00±0.25 | 0.00±0.25 | NA | NA | ||
4 alignment marks | X | -8 | 0.005 | -1.75±0.63 | -1.69±0.25 | 6.25±5.56 | -12.5±3.89 |
Y | 13 | 0.25±0.63 | 0.19±0.56 | NA | NA |
Optimal use of the maskless aligner![](/images/thumb/9/9a/Section_under_construction.jpg/70px-Section_under_construction.jpg)
This section is under construction![](/images/thumb/9/9a/Section_under_construction.jpg/70px-Section_under_construction.jpg)
Writing speed
The site acceptance test performed during the installation of the Aligner: Maskless 01 showed an exposure time/speed of 0.025 min/mm2, which is slightly lower than the 0.02 min/mm2 given in the specifications. At such speeds, a full 4" design would take 2.6-3.3 hours to expose. In practice, we observe exposure times in excess of 2 hours for a full 4" design.
At such exposure speeds, we would quickly run into a bottleneck as more and more users expose more and more wafers, if full 4" designs are used all the time. Another concern is that the exposure dose seems to be unstable over long exposure times, as consecutive exposures using the same parameters have been observed to yield different lithographic results. The conclusion is that exposure using the maskless aligner requires a different design philosophy than when a mask aligner is used for the exposure.
Design philosophy
In order to effectively use the Aligner: Maskless 01, we need to steer away from the "as many variations of the design as possible" approach typically used in mask design when the design of a device is in it's test phase. Instead, we should print only one or a few devices at a time, and iterate the design based on the result of that test, thus truly making use of the rapid prototyping capabilities of the maskless aligner.
Alternatively, we should split our design into two parts; one comprising of all the parts of the device that remain constant through the iterations, e.g. contacts, and the other part being the part of the device that is subject to design optimization. The first part should be printed using a mask aligner, while the other design can be printed much faster on the maskless aligner due to the significantly reduced exposure area. This method is especially useful if the two exposures can be done in the same layer of resist using mix-and-match resist/process.
Mix-and-match
In order to establish a mix-and-match process for Aligner: Maskless 01, different resists have been tested by exposing them in a mask aligner, and subsequently aligning and exposing using Aligner: Maskless 01. Screen dumps taken during the alignment procedure may be seen in the table below. All these tests were carried out on 4" Si wafers.
The easiest approach to mix-and-match is to use a (simple) positive tone resist, such as AZ 5214E, and develop the pattern after the first exposure. The remaining resist can be exposed using the same parameters as normal, and there is plenty of contrast for the alignment process. This process can also be used if the maskless aligner is used first.
Alternatively, the development after first exposure can be skipped. This enables the use of chemically amplified positive tone resists, such as AZ MiR 701, without the drop in sensitivity usually associated with the post-exposure bake. The contrast, however, is greatly reduced, making alignment more difficult. AZ MiR 701 shows more contrast than AZ 5214E, but in both cases the focus and brightness has to be optimized in order to make automatic alignment possible. The design may also have to be adapted, as the poor contrast makes navigation using the overview camera difficult. Also, a less reflective substrate than Si, e.g. oxidized Si, may complicate alignment further.
In the case of negative tone resist, such as AZ nLOF 2020, there is no contrast after exposure, making alignment impossible. Performing post-exposure bake before the second exposure yields a slight contrast, but in the case of nLOF the automatic alignment failed. Manual alignment was possible, but the accuracy suffers.
High resolution camera | Overview camera | |
---|---|---|
AZ 5214E
Exposed + developed |
![]() |
![]() |
AZ MiR 701
Exposed only |
![]() |
![]() |
AZ 5214E
Exposed only |
![]() |
![]() |
AZ nLOF 2020
Exposed + PEB |
![]() |
![]() |