Strong production demand for 0.33NA EUV Lithography has enabled investments in further tool development, resulting in a steady roadmap of 0.33NA tools with improved imaging, overlay, defectivity and productivity, as well as a robust ecosystem supporting use of these tools in HVM. In parallel, co-investments by end users in development and initial production of 0.55NA tools has kept this technology on track for production in 2025, only six years after 0.33NA tools entered production. Continued innovation in semiconductor device structures and processes is on track to support use of the designed resolution and EPE of the 0.55NA tools to continue scaling.

Development and manufacturing techniques applied in mask making and wafer lithography have many similarities. Moreover, these disciplines have jointly progressed enabling generations of IC performance advancement. Despite the similarities, there are also fundamental differences in how these two fields have evolved including the driving forces and constraints inherent in each. It is perhaps difficult to fully appreciate these differences unless one has walked a mile in the others shoes so to speak. From how yield and cycle time are characterized to equipment infrastructure and the evolution of performance parameters, we will open a curtain on certain aspects of mask making from an erstwhile lithographer's perspective in hope this can lead to new avenues for mask and lithography co-optimization. In addition, as the industry ramps EUV alongside an expansion in global design activity across all nodes, it is timely to revisit these unique challenges in mask making to assess whether new frameworks might further advance the capability in service to the design and lithography communities.

ASML NXE:3400 scanners are now commonly used for High Volume Manufacturing (HVM) of 7nm and 5nm logic devices as well as D1z memory devices. In 2021, ASML has introduced the NXE:3600D scanner to the market, targeting 3nm logic and D1a and D1b nodes. In this paper we will share the latest performance of these systems, including excellent overlay, critical dimension (CD) control, stability, reliability, and high productivity. Furthermore, we will address the ASML roadmap for meeting the requirements for the 2 nm node and beyond.

With the introduction of High NA EUV technology the continuation of scaling semiconductor devices in a cost efficient manner will be secured for several technology nodes to come. In this paper we will discuss the benefits of High NA technology from a process complexity reduction point of view as well as the positive impact that this new lithography platform will bring to the industry in terms of cost of technology reduction.

High-NA EUV lithography will improve resolution by increasing the EUV scanner NA from 0.33 to 0.55. To fully benefit from the resolution gain offered by the better scanner lens, it is key to develop and improve the EUV ecosystem. The role of the ecosystem is to ensure timely availability of the advanced resist materials, photomasks, metrology techniques, OPC/imaging strategies, and patterning techniques. In this context, in parallel to the EXE:5000 0.55 NA EUV scanner manufacturing, imec and ASML, together with our partners, are addressing the main challenges and needs towards High-NA ecosystem readiness. In this paper, we will discuss the key findings from simulations and experimental work to develop the high-NA lithography ecosystem (resist and patterning, mask technology) and highlight the key areas where development is needed.

ILT was developed by the team led by the author starting from 2003; Simulation and wafer printing demonstrated its superiority over traditional Manhattan OPC; But its adoption in HVM was hinder by long run time and long write-time; Almost 20 decades later, with the rise of GPU and ML, and the successful development of MBMW, all the issues have a solution. The author reviewed the history of ILT, survey the current status and provide some personal predictions for ILT in HVM.

Masks have always played an important role in lithography to improve resolution and achieve lower k1 values [1]. With EUV lithography now being in High Volume Manufacturing, many parties are looking at the various aspects of the EUV mask. However, the EUV mask as a system is much more complex than the ArFi mask. The EUV mask system roughly consists out of 5 parts which all have multiple functions and therefore interfaces. At the top we have the top-layer which serves to protect during cleaning, but also can help inspection of the absorber after patterning. Then we have the patterned absorber, whose primary function is to steer the diffraction and imaging. Below we have the cap layer that serves to protect the multilayer from EUV photons and environmental gases. Then comes the Multilayer that serves to reflect the maximum amount of EUV light possible. Last but not least we have the core material (ULE) and the back side coating needed to clamp the reticle on the reticle chuck. The EUV mask as such is much more complex than the ArFi mask, also depicted in Figure 1. Two main differences to be noted: - The attention to productivity in EUV as it can be steered with many of the mask properties [2] , - The presence of a cap and top layer is important for mask making and conditioning. For EUV the layers also affect imaging and need to be co-optimized. The cap layer serves many functions. It has also been shown that the cap layer can change its composition through oxidation [3], which can eventually lead to blister formation [4]. We will show experimental data that will enable visualization of the cap and multilayer contribution to state of the art EUV imaging. These experiments comprise contrast evaluation for monopole/dipole exposures and measurement of pattern placement through focus, There is quite some progress in demonstrating how absorber tuning can be used to steer the imaging parameters (contrast [NILS] and dose to size). For the absorber recent work has highlighted the physics and demonstrated that an alternative absorber can have improved imaging properties. We will give an update from our experimental findings using a low n mask ([2]). This might fuel even more appetite to optimize the other parts of the mask system as well. Before we do so we need to make sure all the necessary cross-links between functions and contributing parties are clear, without compromising the ability for an end user to differentiate. We will discuss tuning opportunities beyond absorber from imaging perspective, with focus on the cross-link with other aspects. We will also discuss how to interface the different functions by means of “system” documentation and human interaction (“eco-system workshop”) [1] Jo Finders and Christian Wagner "Imaging enhancement (low k1 imaging) in EUV lithography: current status and future resolution enhancement techniques", Proc. SPIE 11609, Extreme Ultraviolet (EUV) Lithography XII, 1160909 (22 February 2021) [2] Claire van Lare, Frank Timmermans, Jo Finders, Olena Romanets, Cheuk-Wah Man, Paul van Adrichem, Yohei Ikebe, Takeshi Aizawa, and Takahiro Onoue: "Investigation into a prototype extreme ultraviolet low-n attenuated phase-shift mask," Journal of Micro/Nanopatterning, Materials, and Metrology 20(2), 021006 (21 May 202 [3] “The refined EUV mask model”, I.A. MAKHOTKIN, M. WU, V. SOLTWISCH, F. SCHOLZE, V. PHILIPSEN, Journal of Applied Physics, 2019 [4] Jetske Stortelder, Arnold, Storm, Veronique de Rooij-Lohmann, Chien-Ching Wu, and Willem van Schaik "Compatibility assessment of novel reticle absorber materials for use in EUV lithography systems.", Proc. SPIE 10957, Extreme Ultraviolet (EUV) Lithography X, 1095713 (26 March 2019);

We investigate the induced best focus shifts by the mask absorber. The effect of n, k, bias and target size on BF shifts is studied. We consider lines and spaces with pitch = 5× target size. We present a correlation between the BF shifts and Zernike phase offset and the fourth-order Zernike coefficient that represents defocus. When no mitigation strategies are applied, low n absorber materials can show stronger BF shifts and stronger phase variation versus target size. The knowledge gained from this study will help to identify combinations of absorber properties (n, k, thickness) and biasing strategies, which provide high NILS, and threshold to size and enable proper focus control.

With the coming introduction of new EUV absorbers, there is an increasing need for actinic phase metrology, which is sensitive to both thin-film reflectance as well as 3D scattering effects. One promising measurement is actinic scatterometry, which can be carried out with relatively low hardware complexity while still being sensitive to the wavelength of interest. However, a drawback of scatterometry is the difficulty of computationally inverting the scattering process to interpret the measurements. We present a computational framework for linearizing the relationship between the measured intensity and the unknown phase using a set of rigorous scattering simulations of geometries sampled from a known random distribution. Once the scattering dataset has been generated no further heavy computations are required, and at runtime the phase can be extracted using only linear operations. We perform simulations to quantify the accuracy and precision of such a measurement, and to assess requirements on noise, coherence, and target size.

Previous stochastic simulation studies described the impact of EUV mask deficiencies upon imaging lines and spaces, in both 0.33 NA and high NA extreme UV lithography, using respectively isomorphic and anamorphic imaging. Specifically multilayer ripple and absorber line-edge roughness are considered to require adjusted mask specifications in high NA lithography, as they cause increased defectivity and deteriorated CD uniformity on wafer. Recent work demonstrates that for high NA at 22 nm pitch the spread of printed CD due to ML ripple is found to saturate at unacceptable values already below a simulated 50 pm RMS value, that is currently considered as state-of-the-art. Also, it has been shown that there are measurable differences between horizontal and vertical feature defectivity for lines and spaces. Two dimensional line features govern line-end to line-end spacing -- the so called tip-to-tip (T2T) metal lines gap. This spacing in turn impacts contact pitch, making both significant influencers on cell layout area. High NA is anticipated to allow reduction in line-end to line-end gap sizes in a single patterning situation. Figure 1 demonstrates that a 16 nm line-end to line-end gap in a 22 nm pitch feature is feasible. However, the stochastic simulation illustrates approximately 2 nm of T2T variation on wafer and significant variation in the simulated failure ratio with a 50 pm RMS ML-ripple EUV mask. Such mask defectivity impact further interacts with the photonic stochastics of the high NA scanner and the photoresist stochastics. The stochastic model used is explained in Ref. JM3i. This paper will explore the impact of mask ML ripple and absorber line edge roughness on high-NA imaging of T2T features, including their sensitivity to orientation relative to the slit/scan directions of the scanner, in view of anamorphicity. A proposal for more precise definitions of the concept of horizontal and vertical will be made, by which, in the author’s opinions, future coverage of high NA will benefit. These findings will improve insights on any potential additional mask specifications required for high NA single patterning.

In this work, the authors will review the laser operation and imaging fundamentals behind a new method of alternating wavelengths in a single exposure and the application to Customer use cases. Two focal positions are created that are averaged over the exposure field, which can be turned on and off, thus eliminating any potential scanner calibration issues. The main focus of this work is the application of this imaging method (single exposure with two focus positions) to significantly improve the sidewall angle linearity of features in extremely thick photoresist applications. This novel technique, called MFI (multi-focal imaging), can be tuned specifically to provide the required amount of focus offset for a specific thick photoresist application. There are several Customer use cases that have been evaluated in simulation and demonstrated on wafer.

Since the beginning of lithography using a reticle, defects have been one of the key factors in determining reticle quality and wafer yields. With the rapid adoption to EUV Lithography and the high cost of existing EUV mask defect review systems there is an industry need for an alternative defect review and dispositioning system. To meet this need Micron Mask Technology Center, the supplier of state-of-the-art mask technology, has partnered with EUV Tech, the world’s leading supplier of EUV Metrology equipment, to develop a standalone Zoneplate-based EUV wavelength microscope. EUV Tech has introduced and delivered the AIRES (Actinic Image REview System), an at-wavelength EUV Mask Defect Review Tool for EUV photomasks. The tool features EUV Tech’s patented zone plate imaging system equipped with the EUV Tech’s automated mask transfer system to the process chamber. AIRES is based on the successful Zoneplate EUV Microscope at Lawrence Berkeley Lab called SHARP which has been used to support semiconductor development for over a decade. AIRES is the first commercial standalone EUV projection imaging zoneplate microscope supporting photomask defect review and dispositioning to support EUV high volume manufacturing. In this paper, we introduce the tool and its performance with defect cases and pattern imaging application on EUV production reticles.

Pellicles for EUV lithography manufactured using carbon nanotubes (CNT) films were characterized for EUV transmission, scattering, reflectivity, mechanical properties, and capability to stand high intensity (20 W/cm2) EUV radiation in environmental conditions similar to a EUV scanner. Full size (110 x 140 mm), free-standing pellicles less than 20 nm thick can be routinely manufactured using this type of CNT film. Concentration of contaminants such as catalyst particles is very small (< 0.5 wt.%). Several types of coated CNT films were also tested to increase the lifetime of the pellicles. Coating layers didn’t degrade mechanical strength significantly and even an improvement in rupture pressure and reduction of deflection under pressure was observed for some coatings. Coated films were capable of surviving a EUV dose equivalent to 32,000 wafers in a 600 W scanner. The EUV transmissions ranged from 97% for uncoated samples to 93% for coated. Scattering of uncoated samples was of only 0.15% a practically zero reflectance. The most promising coating increased scattering by a factor of 2.

The potential of deep learning as a supplement for image simulations to allow for more efficient modeling of new lithographic configurations has been explored in recent years. A routine challenge with deep learning solutions is their inherent data inefficiency. This work details a deep learning model capable of predicting aerial images for different mask absorbers and illumination settings. We expand on this model by investigating its accuracy potential and data efficiency. This investigation provides insights into the amounts of training data required to achieve the optimum accuracy for different absorber stacks. A significant variance in data requirements and achievable accuracy across different absorbers is observed. The observed trends indicate that the amount of training data required to train the model is directly correlated to the severity of the mask-3D effects of the absorber. This work presents a method that can improve the data efficiency of this predictive model without compromising the accuracy for novel absorbers or new lithographic configurations.

We introduce and demonstrate a new simulator that utilizes resist-specific chemical formulations to drive a Monte Carlo multi-physics and chemistry modeling flow for NA 0.33 and NA 0.55 EUV lithography. With this model stochastic variations are a natural consequence of the chemical and atomic composition of the material. Fully predictive EUV absorption coefficients and exposure latitudes have less than 5% error. LCDU values are predicted with errors smaller than 15% and LWR/LER values are predicted with errors smaller than 7%. We believe chemistry-focused simulations will be key to the development of new EUV photoresists tailored to high-NA imaging.

Spatial fluctuation in reactions throughout exposed resist films is analyzed by principal component analysis (PCA) or singular value decomposition (SVD), which is known as generalization of Fourie transform. In optical simulations, image is expressed by a wave function obtained from SVD of matrix representation of optical images. We attempt to extend this to density/probability distributions of resist reactions for analyzing probabilistic behavior of resists such as LER/LCDU and stochastic defects. To show its effectiveness, PCA is applied to reaction distributions calculated by the fully coupled Monte-Carlo simulation, which visualizes correlated reaction influences on spatial feature and probability of pattern anomalies.

In the last years, the continuous efforts on the development of extreme ultraviolet lithography (EUVL) have allowed to push the lithographic performance of the EUV photoresists on the ASML NXE:3400 full field exposure tool, however, stochastic resist roughness, local critical dimension uniformity (LCDU) and pattern defectivity at nano-scale are still the major limiting factors of the lithographic process window of EUV resist when looking at sub-40nm pitches for both linespace (LS) and contact hole (CH) applications, especially in the low exposure dose regime [1]. To be effective during the lithographic EUV resist screening evaluation phase for such tight pitches, imec has implemented since 2018 [2] additional metrology analysis after resist development inspection (ADI) to further quick feedback on the quantification of nano-failures (nano-bridges, broken lines, merging or missing contacts) induced by a stochastic EUV patterning regime, and thus to improve the resist design at lithographic step in a faster manner. In this work, we have further extended the examination of the resist performance introducing additional metrology analysis after pattern transfer in a silicon nitride (SiN) substrate. We present the characterization results on 40nm and 36nm pitch staggered dense contact holes looking at both lithographic and etching knobs to mitigate the patterning process stochastic issues, confirming that the holistic litho-etch approach is an important and necessary step in the development path of EUV advanced patterning applications towards high volume manufacturing and high-NA EUV lithography.

Extreme Ultraviolet (EUV) lithography has gained maturity and is now the de facto leading edge patterning technology for advanced nodes beyond 7nm. Looking forward, the industry is exploring how to push the resolution even further. A high NA EUV scanner promises to increase the numerical aperture to capture larger diffraction orders enabling resolution improvement. The desire to use existing mask infrastructure with the anamorphic high NA projection lens project results in a half-field on wafer as compared to the currently available low NA EUV scanner. Once the high NA scanner is available, integrated devices will be manufactured with mix and match full-field and half-field lithography. This will bring its own set of challenges from an overlay error minimization standpoint, as half-field and full-field will not share the same center of gravity. Given the fact that overlay models are well known and that the high NA EUV scanner is not yet commercially available we took the approach to emulate the process through computer Monte Carlo simulations. This paper will explain the methods and assumptions used for the Monte Carlo simulation and explore how the mix and match affect the overlay correctable and none-correctable errors

An ERL-based EUV-FEL can provide EUV power of more than 1 kW for multiple scanners to overcome stochastic effects with a higher throughput. An IR-FEL project started at the KEK cERL as a NEDO project in order to develop high-power IR lasers for high-efficiency laser processing, and it can demonstrate proof of concept of the EUV-FEL for future lithography. The IR-FEL was constructed in May 2020 and commissioned in June to July 2020 and in February to March 2021. We will briefly review the EUV-FEL and present the construction and commissioning of the cERL IR-FEL for realizing the EUV-FEL for future lithography.

We improved EUV mask imaging performance through phase and amplitude optimization between 0th and 1st diffraction orders. A specific phase difference between the 0th and 1st diffraction orders can reduce the aerial image split and then improve the imaging performance. The optimum phase difference is 27 degrees in the 14 nm HP L/S pattern and 36 degrees in the 12 nm HP L/S pattern. This effect becomes more evident with the amplitude balancing between the two diffraction orders.

The attenuation of a tin ion beam was studied as it traveled through hydrogen gas and hydrogen plasma of varying pressures. The results of the experiment were compared against predicted results obtained using the collisional code RustBCA. The results of this experiment will provide greater insight into ion behavior from generation in EUV vessels to wall contamination.

Tin contamination of the collector mirror surface remains one of the main issues of EUV sources, directly impacting the availability of the tool. Hydrogen plasma based tin removal processes employ hydrogen radicals and ions to interact with tin deposits to form gaseous tin hydride (SnH 4 ), which can be removed through pumping. An annular surface wave plasma (SWP) source developed at the University of Illinois – Urbana Champaign is integrated into the cone and perimeter of the collection mirror for in-situ tin removal. The SWP is characterized by high ion and radical densities, low electron temperature, and local generation where etching is needed. This method has the potential to significantly reduce downtime and increase mirror lifetime. Radical probe measurements show hydrogen radical densities in the order of 10 13 cm -3 , while Langmuir probe measurements show electron temperatures of up to 4 eV and plasma densities on the order of 10 11-12 cm -3 . The generated ions are essential to the tin cleaning and have sufficiently low energy to cause no damage to the collector capping layer. Tin etch rates of up to 250 nm/min were observed in a variety of experimental conditions, including various powers, pressures, flowrates and temperatures. The high etch rates demonstrated in this study exceed the expected contamination rate of the EUV source.