{
  "id": "multiple-patterning",
  "title": "Multiple Patterning",
  "lang": "en",
  "source_lang": "en",
  "summary": "Multiple patterning is a family of semiconductor fabrication techniques that combines several patterning steps to produce features too closely spaced for a single lithographic exposure.",
  "category": "computer-science",
  "subcategory": "hardware",
  "type": "concept",
  "aliases": [
    "Multiple-patterning lithography",
    "Multipatterning"
  ],
  "generated_at": "2026-11-06",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/multiple-patterning/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/multiple-patterning/",
      "json": "https://aiwiki.page/zh/multiple-patterning.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Multiple patterning is a family of manufacturing techniques used to create densely spaced structures in integrated circuits by combining two or more pattern-definition steps. Rather than printing an entire dense layout in a single photolithographic exposure, a process divides the layout among several exposures or generates additional features from previously patterned structures. The resulting pattern can have a smaller pitch—the distance between corresponding points on adjacent repeating features—than a single exposure can reliably produce. Multiple patterning is used for structures such as transistor fins, interconnect lines, and contacts, and includes both repeated lithography-and-etch processes and self-aligned spacer techniques."
    },
    {
      "heading": "Physical basis",
      "text": "Photolithography projects a pattern from a photomask onto a light-sensitive material on a semiconductor wafer. Optical resolution depends on the exposure wavelength, the numerical aperture of the projection system, and the imaging and processing methods employed.\n\nA common expression for the characteristic printable feature dimension is\n\n\\[CD = k_1\\frac{\\lambda}{NA},\\]\n\nwhere \\(CD\\) is the critical dimension, \\(\\lambda\\) is the exposure wavelength, \\(NA\\) is numerical aperture, and \\(k_1\\) represents process-dependent factors. This expression summarizes resolution scaling rather than defining a universal minimum dimension for every possible layout.\n\nAs patterns become denser, neighboring features can no longer be separated reliably in a single exposure. Multiple patterning addresses this problem by ensuring that each exposure contains a less densely packed subset of the final pattern, or by using deposited films to generate additional features without directly imaging every one of them.\n\nFor an ideal regular array with final pitch \\(p\\), splitting alternating lines between two exposures gives each exposure a pitch of \\(2p\\). Combining the two patterns restores the desired pitch \\(p\\). Real processes require additional allowances for alignment, feature shape, etching, and variation.\n\nMultiple patterning therefore improves the density of the combined pattern, not the intrinsic resolving power of an individual exposure. It also distinguishes pitch reduction from linewidth reduction: making isolated lines narrower does not, by itself, make a densely spaced array printable."
    },
    {
      "heading": "Principal process families",
      "text": "Lithography–etch repetition\n\nIn lithography–etch–lithography–etch, usually abbreviated LELE, a target layout is divided into two masks. The first pattern is exposed, developed, and etched into a suitable transfer layer. A second lithographic sequence defines the complementary pattern, which is combined with the first through subsequent pattern transfer.\n\nFor a regular line array, one mask may contain the odd-numbered lines and the other the even-numbered lines. Each mask therefore contains features spaced farther apart than in the finished structure.\n\nThe same principle can be extended to three or more lithographic patterning cycles. Triple patterning is often described as LELELE, although actual process sequences can contain additional deposition, cleaning, hard-mask, and transfer operations.\n\nAn advantage of exposure-based decomposition is its ability to accommodate a comparatively broad range of two-dimensional shapes. Its major challenge is overlay: independently exposed patterns must be aligned accurately enough that the combined geometry remains within specification.\n\nLithography–lithography–etch approaches\n\nSome methods form two lithographic patterns before transferring their combination in a principal etch step. These are often grouped under lithography–lithography–etch, or LLE.\n\nThe first pattern must survive the materials and processing used for the second. Techniques can include chemically stabilizing the first resist pattern or using material systems that allow successive patterning steps without destroying earlier features.\n\nTwo exposures of an ordinary resist do not automatically provide the same capability. If their optical images merely add together, closely spaced features may still fail to resolve. Effective pitch splitting generally requires a material or process mechanism that preserves the separately defined patterns.\n\nSelf-aligned double patterning\n\nSelf-aligned double patterning, or SADP, uses sidewall spacers to generate a denser pattern from a more widely spaced initial template.\n\nA typical sequence is:\n\n- Pattern relatively widely spaced mandrels, which serve as temporary cores.\n- Deposit a conformal film over the mandrels and surrounding surface.\n- Perform a directional etch that removes the film from horizontal surfaces while leaving material on the mandrel sidewalls.\n- Remove the mandrels.\n- Transfer the remaining spacer pattern into an underlying layer.\n\nBecause each mandrel has two sidewalls, a suitable geometry produces two spacer lines for each mandrel line. Properly chosen mandrel widths, spacing, and spacer thickness can yield an approximately uniform array at half the original pitch.\n\nSpacer films may be formed using processes such as atomic layer deposition or conformal chemical vapor deposition. Spacer width depends strongly on film thickness and etch behavior rather than solely on the linewidth printed in the initial exposure.\n\nThe term self-aligned refers to the geometric relationship between the spacers and the mandrels. It does not mean that the complete layer requires no alignment control.\n\nSelf-aligned quadruple patterning\n\nSelf-aligned quadruple patterning, or SAQP, performs two successive stages of spacer-based pitch multiplication. A pattern generated by the first spacer cycle becomes the basis for a second cycle, producing approximately four times the original line density in a suitable regular layout.\n\nSAQP can reach smaller pitches while avoiding the need to align four independent exposures that each define an interleaved line set. However, it requires more deposition, etching, material removal, and pattern-transfer operations than SADP. Variations introduced during one stage can influence subsequent stages.\n\nThe double and quadruple labels describe the nominal pattern-density multiplication. They do not specify the total number of masks, exposures, or processing operations needed to finish the layer.\n\nCut, block, and trim patterning\n\nSpacer methods naturally produce long lines or closed contours around mandrels, whereas functional circuits require particular line ends, breaks, and connections.\n\nAdditional masks therefore define cuts, blocks, or trim regions. Depending on the process, these masks remove unwanted portions of a line array or prevent selected regions from transferring into the final layer.\n\nA dense line grid may be generated by self-aligned patterning while its line ends are defined by a separate exposure. If the required cuts are themselves too closely spaced, their pattern may also need decomposition across multiple masks. A nominally self-aligned process can consequently remain limited by the lithography and overlay requirements of its cut patterns."
    },
    {
      "heading": "Layout decomposition and design automation",
      "text": "For exposure-based multiple patterning, the target layout must be assigned to masks so that each mask satisfies the applicable printing constraints. This is called layout decomposition.\n\nA common abstraction represents the problem as graph coloring:\n\n- Each relevant layout feature or feature segment becomes a vertex.\n- An edge connects features that cannot be printed together on the same mask.\n- Each color represents a different mask.\n\nA valid assignment places connected vertices on different masks. For a simple two-mask conflict graph, a valid assignment exists exactly when the graph is bipartite; an odd cycle indicates a conflict that cannot be resolved by assigning whole features to only two masks.\n\nOne remedy is a stitch, which divides a feature into overlapping segments assigned to different masks. Stitching can remove certain decomposition conflicts, but it creates a region where registration and pattern-transfer errors can affect continuity or width.\n\nWith three or more masks, general graph-coloring formulations can be NP-complete. Industrial decomposition also involves geometric constraints, stitch choices, mask-density balance, and process-specific rules. Electronic design automation tools therefore combine graph methods, mathematical optimization, and heuristics to find manufacturable assignments.\n\nSpacer-based processes require different decomposition models because many final features are generated from mandrel boundaries rather than independently assigned to exposures. Tools must determine mandrel placement, spacer topology, and cut locations together.\n\nEffects on circuit design\n\nMultiple patterning makes manufacturability dependent on relationships among nearby shapes, not only on each shape’s dimensions.\n\nRelevant design constraints can include:\n\n- Minimum spacing between features assigned to the same mask.\n- Restrictions on line-end positions and cut spacing.\n- Compatible mask assignments at standard-cell boundaries.\n- Limits on stitch placement.\n- Preferred routing directions and regular track arrangements.\n\nThese requirements influence cell-library design, placement, and interconnect routing. More regular layouts often simplify decomposition, although regularity can limit local layout flexibility or increase area."
    },
    {
      "heading": "Process control and error mechanisms",
      "text": "Overlay and edge placement\n\nIn repeated lithographic patterning, an alignment error between masks changes the spacing between features printed in different exposures. Even if every line has the intended width, their relative positions can be incorrect.\n\nThe broader manufacturing quantity is edge-placement error: the displacement of an actual feature edge from its intended location. Its contributors include overlay, critical-dimension variation, mask error, resist behavior, and etch bias.\n\nSelf-aligned patterning reduces reliance on overlay for particular neighboring edges. Nevertheless, alignment remains important for cut masks and for registration to other device layers.\n\nPitch walking\n\nPitch walking is a systematic alternation or other repeating variation in the spacing of features that should form a uniform array.\n\nIn spacer processes, it can arise from mismatches among mandrel width, mandrel pitch, spacer thickness, and etch behavior. Different neighboring spaces may be determined by different combinations of these quantities. An error can therefore widen one class of spaces while narrowing another, even when the average pitch remains near its target.\n\nSuccessive multiplication stages can introduce more complex repeating patterns of dimensional variation.\n\nRoughness, defects, and transfer variation\n\nMultiple patterning can inherit or modify line-edge roughness from earlier steps. Deposition and etching can smooth some variations while introducing others. Additional operations also create opportunities for particles, incomplete removal, missing features, bridges, or pattern collapse.\n\nThese effects matter differently for different structures. Variation in a transistor fin can affect device behavior, while a misplaced interconnect cut can create an open circuit or an unintended connection. More patterning steps do not necessarily produce an unacceptable yield, but each step adds requirements for control and defect management."
    },
    {
      "heading": "Historical development and applications",
      "text": "Multiple patterning became increasingly important as semiconductor manufacturing extended 193-nanometer immersion lithography to pitches that were difficult or impossible to print in one exposure. Improvements in imaging, masks, resists, and computational correction extended single-exposure capability, while pitch splitting provided a further route to higher density.\n\nSpacer patterning became particularly useful for repetitive structures, including memory arrays and the fins of fin field-effect transistors. Exposure-based decomposition and combinations of spacer-generated lines with cut masks have also been used for dense interconnect layers.\n\nThe appropriate method depends on the geometry and requirements of each layer. A single chip can use single-exposure lithography, double patterning, spacer-based multiplication, and other patterning approaches on different layers. A commercial process-node label does not uniquely identify the pitch, number of masks, or patterning method used.\n\nMultiple patterning has supported the continued scaling associated with Moore’s law, but increasing feature density does not guarantee a proportional reduction in manufacturing cost."
    },
    {
      "heading": "Relationship to extreme ultraviolet lithography",
      "text": "Extreme ultraviolet lithography—EUV lithography—uses a much shorter exposure wavelength than 193-nanometer lithography. For suitable layers and pitches, it can replace a multiple-patterning sequence with a single exposure, reducing mask count and some overlay dependencies.\n\nEUV does not eliminate multiple patterning as a general technique. At sufficiently small pitches, EUV patterns can also require division among exposures or combination with self-aligned processes. The choice depends on resolution, defectivity, throughput, process margin, design constraints, and cost.\n\nHigher numerical aperture can further extend single-exposure resolution, but wavelength and aperture alone do not determine the preferred manufacturing sequence. Resist performance, stochastic defects, pattern transfer, and metrology also affect the result."
    },
    {
      "heading": "Economic and technical limitations",
      "text": "Multiple patterning exchanges additional process complexity for greater pattern density. Important costs include extra masks, scanner exposures, deposition and etch operations, inspection, metrology, and longer manufacturing cycle time.\n\nIts main trade-offs are:\n\n- Density versus process complexity: smaller pitch requires more elaborate pattern generation and transfer.\n- Geometric flexibility versus manufacturability: unrestricted shapes can be harder to decompose than regular arrays.\n- Overlay dependence versus spacer dependence: self-aligned techniques reduce some alignment sensitivities but introduce demanding film-thickness and etch controls.\n- Local resolution versus complete-layer feasibility: successfully forming a dense line array does not ensure that all required cuts, contacts, and connections can also be manufactured.\n- Process cost versus product benefit: additional density may justify more processing for some layers and products but not others.\n\nConsequently, multiple patterning is evaluated as a complete design-and-manufacturing system rather than solely as a method for obtaining the smallest possible pitch."
    }
  ],
  "links": [
    "integrated-circuit",
    "photolithography",
    "photomask",
    "semiconductor-wafer",
    "graph-coloring",
    "np-completeness",
    "electronic-design-automation",
    "mathematical-optimization",
    "heuristic",
    "semiconductor",
    "finfet",
    "moores-law",
    "extreme-ultraviolet-lithography"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}