The two standard reticle defect inspection methods are die-to-die and die-to-database. The die-to-die inspection method compares images from the two dice on the same reticle to identify any defect. However, the die-to-database inspection method compares images from the reticle with the design data (CAD). The previous year, we built an SEM-based VSB writer classification system for die-to-die inspection that used state-of-the-art deep learning models to identify errors such as shape, position, and dose [1]. Using the deep neural networks and DL-based SEM digital twins [2], we showed better accuracy than the average human expert in classifying SEM-based defects. However, a limitation remained that the DL model wasn’t aware of chrome and glass regions, just from the input SEM. This information is helpful to make better decisions in classifying some typical errors achieving higher accuracy. In the current paper, we improve the accuracy of the existing classifier by enhancing the underlying deep learning model and supplementing it with the recognition of chrome and glass (exposed and unexposed) regions further. We make it possible with yet another DL-based SEM2CAD digital twin to automatically identify exposed/unexposed areas from the SEM and augment manual input by the expert to it. We feed this new information into the SEM classifier that currently takes a reference and error SEM image for more accurate results. In addition, we also built an SEM-based defect classification system for the die-to-database inspection to categorize various types of VSB mask writer defects, which requires defect SEM images and the reference CAD. Using several deep neural network models and digital twins, in this paper, we provide a production-grade system for the VSB writer’s SEM-based defect classification that works for both die-to-die and die-to-database inspection methods.
Sub-nanometer accuracy attainable with electron micrograph SEM images is the only way to “see” well enough for the mask analysis needed in EUV mask production. Because SEM images are pixel dose maps, deep learning (DL) offers an attractive alternative to the tedious and error-prone mask analysis performed by the operators and expert field application engineers in today’s mask shops. However, production demands preclude collecting a large enough variety and number of real SEM images to effectively train deep learning models. We have found that digital twins that can mimic the SEM images derived from CAD data provide an exceptional way to synthesize ample data to train effective DL models. Previous studies [1, 2, 3, 4] have shown how deep learning can be used to create digital twins. However, it was unclear if SEM images generated with digital twins would have sufficient quality to train a deep learning network to classify real SEM images. This paper shows how we built three DL tools for SEM-based mask analysis. The first tool automatically filters good quality SEM images, particularly for test chips, using a DL-based binary classifier. A second tool uses another DL model to align CAD and SEM images for applications where it is important that features on both the images are properly aligned. A third tool uses a DL multi-class classifier to categorize various types of VSB mask writer defects. In developing the three tools, we trained state-of-the-art deep neural networks on SEM images generated using digital twins to achieve accurate results on real SEM images. Furthermore, we validated the results of trained deep learning models through model visualization and accuracy-metric evaluation.
Sub-nanometer accuracy attainable with electron micrograph SEM images is the only way to “see” well enough for the mask analysis needed in EUV mask production. Because SEM images are pixel dose maps, deep learning (DL) offers an attractive alternative to the tedious and error-prone mask analysis performed by the operators and expert field application engineers in today’s mask shops. However, production demands preclude collecting a large enough variety and number of real SEM images to effectively train deep learning models. We have found that digital twins that can mimic the SEM images derived from CAD data provide an exceptional way to synthesize ample data to train effective DL models. Previous studies [1, 2, 3, 4] have shown how deep learning can be used to create digital twins. However, it was unclear if SEM images generated with digital twins would have sufficient quality to train a deep learning network to classify real SEM images. This paper shows how we built three DL tools for SEM-based mask analysis. The first tool automatically filters good quality SEM images, particularly for test chips, using a DL-based binary classifier. A second tool uses another DL model to align CAD and SEM images for applications where it is important that features on both the images are properly aligned. A third tool uses a DL multi-class classifier to categorize various types of VSB mask writer defects. In developing the three tools, we trained state-of-the-art deep neural networks on SEM images generated using digital twins to achieve accurate results on real SEM images. Furthermore, we validated the results of trained deep learning models through model visualization and accuracy-metric evaluation.
KEYWORDS: Photomasks, Lithography, Electron beams, Logic, Electron beam melting, Line edge roughness, Electron beam lithography, Extreme ultraviolet, Optical lithography, LCDs
Semiconductor scaling is slowing down because of difficulties of device manufacturing below logic 7nm
node generation. Various lithography candidates which include ArF immersion with resolution enhancement
technology (like Inversed Lithography technology), Extreme Ultra Violet lithography and Nano Imprint
lithography are being developed to address the situation. In such advanced lithography, shot counts of mask
patterns are estimated to increase explosively in critical layers, and then it is hoped that multi beam mask
writer (MBMW) is released to handle them within realistic write time. However, ArF immersion technology
with multiple patterning will continue to be a mainstream lithography solution for most of the layers. Then,
the shot counts in less critical layers are estimated to be stable because of the limitation of resolution in ArF
immersion technology. Therefore, single beam mask writer (SBMW) can play an important role for mask
production still, relative to MBMW. Also the demand of SBMW seems actually strong for the logic 7nm
node. To realize this, we have developed a new SBMW, EBM-9500 for mask fabrication in this generation. A
newly introduced electron beam source enables higher current density of 1200A/cm2. Heating effect
correction function has also been newly introduced to satisfy the requirements for both pattern accuracy and
throughput. In this paper, we will report the configuration and performance of EBM-9500.
Resist heating effect which is caused in electron beam lithography by rise in substrate temperature of a few tens or hundreds of degrees changes resist sensitivity and leads to degradation of local critical dimension uniformity (LCDU). Increasing writing pass count and reducing dose per pass is one way to avoid the resist heating effect, but it worsens writing throughput. As an alternative way, NuFlare Technology is developing a heating effect correction system which corrects CD deviation induced by resist heating effect and mitigates LCDU degradation even in high dose per pass conditions. Our developing correction model is based on a dose modulation method. Therefore, a kind of conversion equation to modify the dose corresponding to CD change by temperature rise is necessary. For this purpose, a CD variation model depending on local pattern density was introduced and its validity was confirmed by experiments and temperature simulations. And then the dose modulation rate which is a parameter to be used in the heating effect correction system was defined as ideally irrelevant to the local pattern density, and the actual values were also determined with the experimental results for several resist types. The accuracy of the heating effect correction was also discussed. Even when deviations depending on the pattern density slightly remains in the dose modulation rates (i.e., not ideal in actual), the estimated residual errors in the correction are sufficiently small and acceptable for practical 2 pass writing with the constant dose modulation rates. In these results, it is demonstrated that the CD variation model is effective for the heating effect correction system.
In the half pitch (hp) 16nm generation, the shot count on a mask is expected to become bipolar. The multi-patterning
technology in lithography seems to maintain the shot count around 300G shots instead of increase in the number of
masks needed for one layer. However, as a result of mask multiplication, the better positional accuracy would be
required especially in Mask-to-Mask overlay. On the other hand, in complex OPC, the shot count on a mask is expected
to exceed 1T shots.
In addition, regardless of the shot count forecast, the resist sensitivity needs to be lower to reduce the shot noise effect so
as to get better LER. In other words, slow resist would appear on main stream, in near future. Hence, such trend would
result in longer write time than that of the previous generations. At the same time, most mask makers request masks to
be written within 24 hours. Thus, a faster mask writer with better writing accuracy than those of previous generations is
needed.
With this background, a new electron beam mask writing system, EBM- 9000, has been developed to satisfy such
requirements of the hp 16nm generation. The development of EBM-9000 has focused on improving throughput for
larger shot counts and improving the writing accuracy.
EBM-9000 equipped with new features such as new electron optics, high current density (800A/cm2) and high speed deflection control has been developed for the 11nm technology node(tn) (half pitch (hp) 16nm). Also in parallel of aggressive introduction of new technologies, EBM-9000 inherits the 50kV variable shaped electron beam / vector scan architecture, continuous stage motion and VSB-12 data format handling from the preceding EBM series to maintain high reliability accepted by many customers. This paper will report our technical challenges and results obtained through the development.
We report our development of fogging effect correction method aimed for EBM-8000, our newest series of EB mask
writers for mask production of 22nm half-pitch generation and for mask development of 16nm half-pitch generation. We
refined the method of fogging effect correction by taking account of dose modulation for proximity effects correction
and loading effect correction into fogging effect correction, greatly reducing theoretical error. Writing experiment has
shown that our method based on the threshold dose model is effective, though deviation from the model is observed.
We previously proposed a new method to correct critical dimension (CD) errors appearing in large-scale integrated circuit (LSI) fabrication processes, such as long range loading effect, local flare, and micro loading effect. The method provides high accuracy correction dimensions when using the pattern modulation method (method correcting CD errors by controlling figure sizes of LSI patterns). Now the case that several processes cause CD errors when a layer of an LSI pattern is fabricated on a wafer is discussed. These CD errors are corrected by generalizing the method proposed previously and taking the sequence of processes into account. It is shown from numerical calculation that the method can suppress the CD error to less than 0.01 nm with three iterations, under the condition that the maximum CD errors by micro loading effect and flare are 10 nm and 20 nm, respectively. It is strongly suggested that our methods will provide the necessary CD accuracies in the future.
In our previous paper, we proposed a new method to obtain
accurate pattern dimensions for correcting global critical dimension CD
errors, which are defined as errors of CD uniformity in a region of several
millimeters to several centimeters. The method is based on the pattern
modulation method a method of correcting CD errors by controlling figure
sizes of large-scale integration LSI patterns. An essential point of
our method is to take into account the difference between the pattern
density of the original LSI pattern and that of the corrected pattern which
has a pattern dimension different from original one after pattern modulation
for correction to provide an accurate correction. In this paper, we
apply the proposed method to correct CD errors caused by flare and
microloading effects. It is shown from numerical calculations that our
method can suppress the correction error to less than 0.1 nm for both
cases by three iterations. It is strongly recommended that our method be
used for a wide range of applications to provide the necessary CD accuracies
of the future.
Image placement (IP) errors caused by electro-static chuck (ESC) and non-flatness of mask are additional factors in
writing extreme ultra-violet (EUV) mask, and minimizing their influences is being fervently addressed. New correction
technique of EBM-6000 has been developed for EUV mask writing based on the conventional grid matching correction
(GMC) without ESC to obtain good reproducibility to satisfy user's requirement to develop EUV mask at an early stage.
Heating effect was evaluated for EBM-6000 which is operated at high current density of 70A/cm2 and acceleration
voltage of 50kV. FEP171 as widely used for current productions and lower sensitivity resists are tested. Lower
sensitivity resist is one of key items to achieve highly accurate Local critical dimension uniformity (LCDU) because of
shot noise reduction.
CD variations in experiment are compared with simulated temperature changes induced by heating effect. Then, the
ratio, ΔCD/ΔT, is found mostly constant for every resist, 0.1 nm/C°.
Writing conditions are estimated to meet CDU spec of hp45 generation for a worst case pattern, i.e. 100% density
pattern. For FEP171, the maximum shot size of 0.85 μm shot size at 2pass writing mode is sufficient. It should be
reduced to 0.5 μm at 2pass writing mode for every lower sensitivity resist. When 4pass writing mode is used, the
maximum shot size of 0.85 μm is available. Writing conditions and writing time for realistic patterns are also discussed.
In order to comply with the demanding technology requirements for 45 nm half pitch (HP) node (32 nm technology
node), Nuflare Technology Inc. (NFT) has developed Electron-beam mask writing equipment, EBM-6000, with
increased current density (70A/cm2), while its other primary features basically remain unchanged, namely 50 kV
acceleration voltage, Variable Shaped Beam (VSB)/vector scan, like its predecessors [1-5]. In addition, new
functionalities and capabilities such as astigmatism correction in subfield, optimized variable stage speed control,
electron gun with multiple cathodes (Turret electron gun), and optimized data handling system have been
employed to improve writing accuracy, throughput, and up-time. VSB-12 is the standard input data format for
EBM-6000, and as optional features to be selected by users, direct input function for VSB-11 and CREF-flatpoly
are offered as well.
In this paper, the new features and capabilities of EBM-6000 together with supporting technologies are reported to
solidly prove the viability of EBM-6000 for 45 nm HP node.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.