{
  "id": "extreme-ultraviolet-lithography",
  "title": "Extreme Ultraviolet Lithography",
  "lang": "en",
  "source_lang": "en",
  "summary": "Extreme ultraviolet lithography uses 13.5-nanometre radiation, reflective optics, and photosensitive materials to manufacture exceptionally small semiconductor circuit patterns.",
  "category": "technology",
  "subcategory": "engineering",
  "type": "concept",
  "aliases": [
    "EUV lithography",
    "EUVL"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/extreme-ultraviolet-lithography/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/extreme-ultraviolet-lithography/",
      "json": "https://aiwiki.page/zh/extreme-ultraviolet-lithography.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Extreme ultraviolet lithography (EUV lithography or EUVL) is a form of photolithography that uses radiation with a wavelength near 13.5 nanometres to transfer patterns onto a photosensitive layer on a semiconductor wafer. It is a central manufacturing technology for advanced integrated circuits, including logic processors and high-density memory. Its short wavelength permits smaller features to be printed with fewer patterning operations than would otherwise be required with longer-wavelength optical systems.\n\nEUV lithography differs fundamentally from conventional optical projection lithography in its hardware. Air and most materials strongly absorb extreme ultraviolet radiation, so the optical path operates under vacuum and uses multilayer mirrors rather than transmissive lenses. A commercial exposure system combines a plasma light source, precision reflective optics, a reflective patterned mask, rapidly moving stages, and tightly controlled resist chemistry. Its performance depends not only on optical resolution but also on defect control, statistical fluctuations in exposure, alignment, and compatibility with subsequent manufacturing steps.\n\nThe technology reached high-volume semiconductor manufacturing in the late 2010s after several decades of research. Its development connects short-wavelength optics, plasma physics, surface science, chemical engineering, and precision instrumentation. The transition from conventional EUV systems with a numerical aperture of 0.33 to high-numerical-aperture systems with a numerical aperture of 0.55 extends its resolving power while introducing new constraints on masks, materials, and process integration."
    },
    {
      "heading": "Imaging principles and resolution",
      "text": "Lithography produces a patterned chemical image that is subsequently converted into a physical structure. A wafer is coated with photoresist, exposed through a projection system, and developed so that selected regions dissolve. The remaining resist can protect underlying material during etching or serve as part of a more elaborate pattern-transfer stack. In this context, lithography is related to historical lithography and printing, but the image dimensions, optical requirements, and substrate-processing methods are radically different.\n\nThe ability of a projection system to resolve small features is commonly described by a Rayleigh-type relationship:\n\n\\[R=k_1\\frac{\\lambda}{\\mathrm{NA}},\\]\n\nwhere \\(R\\) is a characteristic resolvable dimension, \\(\\lambda\\) is wavelength, \\(\\mathrm{NA}\\) is numerical aperture, and \\(k_1\\) represents the combined effects of illumination, mask design, resist response, and processing. This expression is a practical scaling rule rather than a universal boundary for every pattern. Isolated lines, dense periodic lines, and contact holes have different imaging requirements. The approximately fourteenfold wavelength reduction from 193-nanometre argon-fluoride exposure to 13.5-nanometre EUV greatly improves the available resolution, although system architecture and process effects prevent a simple fourteenfold reduction in manufactured dimensions.\n\nNumerical aperture measures the angular range of light accepted by the projection optics. In vacuum it is \\(\\mathrm{NA}=\\sin\\theta\\), where \\(\\theta\\) is the relevant maximum ray angle at the image. Increasing numerical aperture captures more of the light diffracted by a mask pattern and enables finer detail to be reconstructed. Diffraction therefore remains central even when the wavelength is much smaller than visible light. For illustration, \\(k_1=0.4\\), \\(\\lambda=13.5\\) nanometres, and \\(\\mathrm{NA}=0.33\\) give a characteristic resolution of about 16 nanometres; this is not a universal specification for a complete manufacturing process.\n\nResolution is only one dimension of performance. Depth of focus scales approximately as\n\n\\[\\mathrm{DOF}\\propto\\frac{\\lambda}{\\mathrm{NA}^{2}},\\]\n\nso an increase in numerical aperture narrows the permissible range of wafer height and focus error. Exposure latitude, image contrast, pattern orientation, and resist thickness also matter. Moreover, commercial process labels such as “7 nm” or “3 nm” are technology-generation names, not direct statements of EUV wavelength, transistor gate length, or the minimum printable dimension. A circuit fabricated at one named node contains many different pitches and feature sizes, often produced using several lithographic technologies."
    },
    {
      "heading": "Research origins and industrial development",
      "text": "The scientific foundations of EUV lithography emerged from work on soft-X-ray imaging, short-wavelength sources, and multilayer reflectors. During the 1980s, researchers demonstrated that alternating thin films could provide useful reflectivity at near-normal incidence in a spectral region where ordinary optical materials were unsuitable. Hiroo Kinoshita’s work in Japan was among the early demonstrations of projection imaging with multilayer-coated reflective optics. Parallel investigations in the United States and Europe explored whether such systems could become a practical successor to longer-wavelength lithography.\n\nIn the 1990s, EUV became a major coordinated research undertaking rather than an isolated optical experiment. The United States’ EUV Limited Liability Company, formed in 1997, supported development with national laboratories, notably Lawrence Livermore, Sandia, and Lawrence Berkeley. European programs brought together lithography-equipment manufacturers, optical specialists, research institutes, and semiconductor producers. These efforts addressed different but interdependent problems: an optical demonstration was insufficient without a reliable source, defect-controlled masks, durable mirrors, sensitive resists, and a system capable of processing wafers at economically useful rates.\n\nASML subsequently became the supplier of commercial EUV projection scanners, with ZEISS providing the projection optics and Cymer contributing source technology. ASML acquired Cymer in 2013. Early full-field tools and preproduction systems established wafer-scale imaging and integration experience, but the move to mass production took considerably longer than early roadmaps anticipated. Source output, collector lifetime, machine availability, mask infrastructure, and resist performance all required sustained improvement. The history illustrates why lithography transitions cannot be dated solely by the first successful exposure: laboratory resolution and factory productivity are separate milestones.\n\nCommercial adoption accelerated in the late 2010s. Samsung announced production using its 7-nanometre EUV process in 2018, and TSMC introduced its N7+ EUV process into volume production in 2019. Early adoption typically applied EUV to a limited number of especially demanding layers rather than every layer of a chip. Subsequent process generations expanded its use in logic and introduced it into advanced dynamic random-access memory manufacturing. Semiconductor foundries and integrated device manufacturers nevertheless continued to use deep-ultraviolet tools extensively, because many layers did not require EUV resolution. Industrial development therefore produced a mixed lithography environment rather than a wholesale replacement of older exposure systems."
    },
    {
      "heading": "Plasma light sources and radiation delivery",
      "text": "Commercial EUV scanners generate radiation using a laser-produced plasma. A stream of microscopic tin droplets passes through an interaction region, where carefully timed laser pulses convert each droplet into a hot, highly ionized plasma. The resulting emission contains a strong band around 13.5 nanometres, associated with transitions in multiply charged tin ions. This wavelength is particularly useful because it coincides with the operating band of molybdenum–silicon multilayer mirrors. Source design and mirror design are consequently linked: a bright plasma at an unsuitable wavelength would not produce an efficient exposure system.\n\nThe droplet is generally conditioned by a preparatory pulse before receiving the main pulse from a high-power carbon-dioxide laser. The preparatory pulse reshapes or expands the tin target, improving its interaction with the main beam. Systems repeat this process tens of thousands of times per second. Stable operation requires precise synchronization of droplet generation, laser timing, laser focus, and target position. Small disturbances can reduce the useful EUV output or increase debris. Only part of the energy delivered to the target emerges as radiation within the narrow wavelength band accepted by the downstream optics.\n\nA collector mirror gathers a portion of the plasma emission and directs it toward an intermediate focus, which serves as a defined interface between the source and scanner illumination system. Source power is often specified at this intermediate focus rather than at the wafer. The distinction is essential because each subsequent mirror reflects only a fraction of the incident radiation. Source power, useful in-band power, wafer-plane power, and electrical input power are therefore different quantities. Increasing source output can raise wafer throughput, but the benefit depends on optical transmission, exposure dose, and the time required for stage motion and wafer handling.\n\nThe source environment is hostile to optical surfaces. Tin ions, neutral particles, droplets, and deposited material can damage or obscure the collector. Debris mitigation combines source geometry, gas flows, protective arrangements, and cleaning methods. Hydrogen is used in EUV systems partly because reactive hydrogen species can assist removal of certain contaminants, although the detailed surface chemistry must be controlled to avoid damaging coatings. The source also presents substantial thermal and engineering loads. Cooling systems remove heat from lasers and plasma-facing components, while monitoring and feedback maintain stable radiation delivery. Practical source performance is thus measured not merely by peak brightness but by sustained output, component lifetime, maintenance frequency, and availability."
    },
    {
      "heading": "Reflective optics, vacuum, and precision motion",
      "text": "At 13.5 nanometres, ordinary optical glass is effectively opaque over useful lens thicknesses. EUV projection systems therefore use mirrors throughout the main imaging path. Their reflectivity comes from alternating layers of materials, commonly molybdenum and silicon, whose interfaces reinforce reflection within a narrow wavelength and angular range. This interference mechanism is related to Bragg diffraction. A typical multilayer contains several dozen layer pairs, with individual layers only a few nanometres thick. The exact stack design balances reflectivity, bandwidth, interface quality, stress, and long-term stability.\n\nEven a highly optimized EUV mirror reflects substantially less than all the incident radiation. Peak reflectivity near 70 percent is representative of high-quality molybdenum–silicon multilayers under suitable conditions, not a constant for every surface and angle. Successive reflections multiply the losses: ten hypothetical reflections at 70 percent each would transmit only about \\(0.7^{10}\\), or 2.8 percent, of the starting power. Real system efficiency also includes bandwidth selection, illumination configuration, the mask, and other losses. This cumulative attenuation explains the extraordinary demands placed on the source and the importance of maintaining clean optical surfaces.\n\nMirror substrates and multilayers must preserve the wavefront with exceptionally small errors. Surface figure, roughness, coating uniformity, and mounting distortion all affect image quality. EUV mirrors are often aspheric, and their shapes must be measured and corrected using specialized metrology. Absorbed radiation can cause thermal distortion even when only a small fraction of the source power reaches a particular optic. The optical assembly therefore requires controlled thermal conditions, mechanical isolation, and sophisticated calibration. Contamination is equally consequential: thin carbon deposits or oxidation can change reflectivity and imaging performance.\n\nThe optical path operates under vacuum because gases absorb EUV radiation. Differentially controlled regions, pumping systems, and material-compatibility rules allow the source, optics, mask, and wafer environments to coexist without imposing identical conditions everywhere. Meanwhile, the mask and wafer move synchronously through a narrow illuminated field during scanning exposure. High-speed stages combine actuators, precision position measurement, and control theory to maintain the required trajectory. Overlay—the placement of a new pattern relative to previously fabricated layers—depends on stage accuracy, optical distortion correction, wafer deformation, and alignment measurements. The scanner is consequently both an optical instrument and a high-dynamic-performance positioning machine, with errors in either subsystem appearing in the printed circuit pattern."
    },
    {
      "heading": "Reflective masks, defects, and protective membranes",
      "text": "An EUV photomask, often called a reticle, is a patterned reflective object rather than a transparent plate carrying opaque features. Its substrate supports a multilayer reflector, protective layers, and a patterned absorber. Light incident on exposed reflective regions is returned toward the projection optics, while absorber-covered regions suppress or modify the reflected field. Conventional 0.33-NA EUV scanners use fourfold reduction between the mask and wafer, so a nominal mask feature is four times its corresponding image dimension. Nevertheless, mask errors remain important because the optical and resist processes do not simply reproduce every edge with a fixed geometric scale.\n\nDefects can originate in the substrate, within the multilayer stack, or in the patterned absorber. A buried bump or depression may disturb the phase of reflected light even if the top surface appears relatively unobstructed. Such phase defects are distinctive challenges for reflective EUV masks. Absorber defects can instead create missing or unwanted pattern elements. Inspection therefore needs to distinguish geometrical imperfections from their actual imaging consequences. Measurements made at the EUV exposure wavelength are called actinic measurements; they can reveal effects that are difficult to infer from inspection at other wavelengths.\n\nA protective membrane, or pellicle, intercepts particles before they reach the mask surface. Because it sits away from the mask’s image plane, a small particle on the pellicle is less likely to print as a sharply resolved defect. EUV pellicles must be exceptionally thin and sufficiently transparent at 13.5 nanometres while remaining mechanically robust. The light crosses the membrane on the way to the mask and again after reflection, so even modest absorption reduces exposure efficiency. Absorbed power also heats the membrane. Pellicle development consequently involves a difficult combination of transmission, thermal endurance, chemical compatibility, handling strength, and particle protection.\n\nThe mask is illuminated obliquely so that incoming and outgoing radiation can be separated. Finite absorber thickness and oblique illumination generate three-dimensional mask effects, including shadowing, orientation-dependent imaging, phase shifts, and changes in the best-focus position. These effects become more consequential as dimensions shrink and numerical aperture rises. Mask design therefore includes electromagnetic modeling rather than treating the absorber as an ideal two-dimensional stencil. Alternative absorber materials and stack designs seek to improve contrast and reduce unwanted phase effects. Mask fabrication, cleaning, inspection, repair, and storage all form part of the manufacturing infrastructure, because an otherwise capable scanner cannot compensate for a mask that repeatedly prints defects across every exposed wafer."
    },
    {
      "heading": "Resist chemistry and stochastic printing failures",
      "text": "An EUV photon carries much more energy than a deep-ultraviolet photon. Its energy is\n\n\\[E=\\frac{hc}{\\lambda},\\]\n\nwhere \\(h\\) is the Planck constant and \\(c\\) is the speed of light. At 13.5 nanometres, the energy is approximately 92 electronvolts. Absorption commonly ejects an electron, which can initiate further electronic excitations and chemical reactions. The spatial distribution of these secondary processes affects the eventual pattern. EUV exposure is therefore not adequately described as a purely local optical event in which each absorbed photon changes only the material directly beneath its point of entry.\n\nChemically amplified resists use radiation-sensitive compounds to generate acid, followed by a post-exposure bake that allows acid-catalyzed reactions to alter solubility. The matrix often contains a polymer, and the chemistry can be configured for positive- or negative-tone processing. Amplification increases sensitivity because one initiating event can produce multiple chemical transformations, but acid movement and reaction variability can blur edges. Metal-containing resists, including metal-oxide-based materials, offer alternative combinations of EUV absorption, etch resistance, and chemical response. Their practical value depends on the complete process, including coating, development, contamination management, and pattern transfer.\n\nSmall printed features receive a finite number of photons. For an incident exposure dose of 20 millijoules per square centimetre, a square area measuring 10 nanometres on each side receives approximately 1,400 photons before accounting for reflection and incomplete absorption. The number actually absorbed in the relevant volume is smaller. Under an ideal Poisson distribution, fluctuations scale approximately as \\(\\sqrt{N}\\), so the relative fluctuation is \\(1/\\sqrt{N}\\). This shot noise combines with variations in molecular composition, electron transport, acid generation, and development. Statistical fluctuations can produce rough edges or rare failures even when the average printed dimension is correct.\n\nCharacteristic failures include missing contact holes, merged holes, breaks in narrow lines, and bridges between adjacent features. Their probabilities may vary steeply with dose, feature size, and local pattern environment. This creates a sensitivity–resolution–roughness trade-off: increasing dose generally improves photon statistics but reduces throughput; stronger chemical amplification can improve sensitivity while increasing blur; thinner resists can improve imaging and mechanical stability while making subsequent etching more difficult. A large chip contains enormous numbers of opportunities for rare defects, so average linewidth alone is an inadequate qualification metric. Resist development increasingly emphasizes failure probability across extensive pattern populations, together with line-edge roughness, critical-dimension uniformity, and compatibility with the underlying material stack."
    },
    {
      "heading": "Computational patterning and manufacturing integration",
      "text": "The circuit designer’s intended geometry cannot ordinarily be placed unchanged on a mask and expected to print accurately. Diffraction, illumination, mask structure, resist chemistry, and etching all alter the pattern. Optical proximity correction modifies mask edges and adds carefully designed auxiliary features so that the resulting wafer image more closely matches the target. EUV uses these computational methods extensively, even though its shorter wavelength reduces some of the resolution pressure encountered with 193-nanometre exposure. Accurate models must account for the reflective mask and its three-dimensional electromagnetic behavior.\n\nComputational lithography links optical simulation to process measurements. Test patterns are exposed over combinations of dose and focus, then measured to calibrate models of printed dimensions and defect behavior. Optimization seeks patterns that work across a process window rather than only at one ideal setting. Source–mask optimization adjusts illumination and mask geometry together. More computationally intensive inverse methods search for mask shapes that produce a desired wafer pattern, sometimes yielding curvilinear geometries rather than conventional rectilinear edges. The resulting designs must still be writable, inspectable, and sufficiently robust for manufacturing.\n\nEUV often replaces several exposures and processing steps that would otherwise be required by multiple patterning. In a 193-nanometre immersion process, a dense target may be decomposed across multiple masks or formed through spacer-based techniques. Each additional operation can introduce alignment errors, process variability, and cycle time. A single EUV exposure can simplify selected layers, but it does not remove all complexity. At sufficiently tight pitches, EUV itself can be combined with multiple patterning, self-aligned processes, or specialized cut and block masks. The choice depends on the entire integration scheme rather than resolution alone.\n\nAfter development, the resist image usually passes through one or more hard-mask layers before reaching the final device material. Etch selectivity, sidewall profile, local loading, and material damage determine whether the optical image becomes a useful circuit structure. Measurements with an electron microscope, optical instruments, and electrical test structures reveal different aspects of performance. Overlay control must also accommodate wafer distortions caused by previous deposition, heating, and etching operations. An EUV layer may need to align with patterns made by a different scanner type, making cross-tool matching important. In production, success is measured by functioning devices and yield, not simply by attractive resist images. This requirement links lithography to process integration, defect inspection, design rules, and statistical manufacturing control."
    },
    {
      "heading": "High-numerical-aperture EUV lithography",
      "text": "High-numerical-aperture EUV lithography, generally shortened to high-NA EUV, increases numerical aperture from 0.33 to 0.55 while retaining the 13.5-nanometre wavelength. At equal \\(k_1\\), the Rayleigh relationship predicts that the characteristic resolvable dimension scales by \\(0.33/0.55=0.6\\). The resulting improvement can extend single-exposure patterning to smaller pitches and potentially avoid some multiple-patterning sequences. High-NA EUV is not a different spectral technology; it is a substantial redesign of the projection and imaging system within the same wavelength regime.\n\nThe larger angular range creates difficult optical and mask constraints. Commercial high-NA architecture uses anamorphic projection: the reduction factor differs between two perpendicular directions. The image reduction is fourfold in one direction and eightfold in the other. With a standard-sized reticle, this produces a maximum exposure field approximately half the area of the conventional full field. A commonly cited conventional field is about 26 by 33 millimetres, whereas the high-NA field is approximately 26 by 16.5 millimetres. The arrangement helps manage the ray geometry at the mask while preserving a practical reticle format.\n\nThe smaller field changes manufacturing logistics. More fields may be needed to expose a wafer, and a die larger than the available field may require an alternative layout or a stitching strategy. Stitching joins adjacent exposures and imposes demanding boundary-placement requirements. At the same time, depth of focus becomes more restrictive: with other factors held constant, changing from 0.33 to 0.55 NA reduces the Rayleigh depth-of-focus scale to roughly 36 percent of its former value. Wafer flatness, local topography, focus control, and resist thickness therefore become increasingly important. High-NA imaging also intensifies the need to understand mask-induced phase and polarization effects.\n\nASML’s EXE platform embodies the high-NA architecture, and delivery of the first EXE:5000 system to Intel began in late 2023. Early systems support process development, optical characterization, resist qualification, and the construction of manufacturing recipes. Delivery of development equipment is distinct from broad high-volume adoption, which depends on qualified processes and production economics. Research focuses on ultrathin resists, improved hard masks, low-defect printing, advanced mask materials, and metrology capable of resolving the relevant structures. High-NA and 0.33-NA systems can coexist within the same technology generation, with the newer architecture reserved for layers where its resolution advantage offsets its additional integration and equipment demands."
    },
    {
      "heading": "Productivity, supply chains, and practical limits",
      "text": "The economic value of EUV depends on cost per functioning chip, not on the exposure tool’s purchase price alone. Relevant variables include wafers processed per hour, resist dose, equipment availability, maintenance requirements, mask costs, and yield. A more expensive exposure step can reduce total manufacturing cost if it eliminates several masks, deposition operations, etches, and alignment-sensitive transfers. Conversely, an EUV process that requires an excessive dose or generates too many stochastic defects can lose that advantage. Comparisons must therefore consider complete patterning flows at equivalent design requirements and production quality.\n\nThroughput is constrained by several interacting subsystems. Greater source power can shorten exposure time, but stage acceleration, scanning speed, alignment, wafer exchange, and thermal limits eventually become important. Higher resist sensitivity may improve throughput while worsening roughness or defect probability. Protective pellicles reduce the risk of mask contamination but absorb useful radiation. Scheduled maintenance and unscheduled failures affect the fraction of time a scanner is available for production. Consequently, maximum demonstrated wafers per hour and sustained factory productivity are not interchangeable measures, and specifications require a stated dose and operating context.\n\nThe supply chain is unusually specialized. It includes precision mirror production, multilayer deposition, high-power lasers, tin-droplet generators, vacuum equipment, masks, pellicles, resists, metrology instruments, and contamination-controlled service operations. ASML integrates commercial EUV scanners, while suppliers and semiconductor manufacturers contribute essential subsystems, materials, and process knowledge. Replicating the capability requires more than reproducing a single machine design: manufacturing tolerances, measurement methods, accumulated operating data, and supplier coordination are deeply interdependent. The technology also consumes substantial electrical power and cooling capacity, particularly in the laser and source systems, although environmental comparisons depend on which alternative process steps EUV replaces.\n\nSeveral limits remain active research problems rather than fixed physical cutoffs. Rare stochastic defects must be characterized over sufficiently large areas to support reliable yield predictions. Buried mask defects must be detected and assessed for printability. Optics and pellicles must withstand increasing radiation loads without unacceptable degradation. Computational models must connect electromagnetic imaging, electron-driven chemistry, and pattern transfer at useful accuracy and speed. Moving to still shorter wavelengths has been investigated, but would require new combinations of sources, reflectors, masks, and resists rather than a simple adjustment to existing tools. Within present EUV systems, progress continues through coordinated changes to source stability, optical control, materials, design rules, and the distribution of patterning tasks among different exposure technologies."
    }
  ],
  "links": [
    "photolithography",
    "integrated-circuit",
    "photoresist",
    "lithography",
    "printing",
    "numerical-aperture",
    "diffraction",
    "semiconductor-foundry",
    "plasma",
    "laser",
    "control-theory",
    "photomask",
    "planck-constant",
    "speed-of-light",
    "electron",
    "photon",
    "polymer",
    "poisson-distribution",
    "multiple-patterning",
    "electron-microscope"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}