Search

EP-4341750-B1 - MACHINE LEARNING FOR SELECTING INITIAL SOURCE SHAPES FOR SOURCE MASK OPTIMIZATION

EP4341750B1EP 4341750 B1EP4341750 B1EP 4341750B1EP-4341750-B1

Inventors

  • STANTON, WILLIAM, A.
  • BERTHIAUME, SYLVAIN
  • STOCK, HANS-JURGEN
  • HISEROTE, JAY, A.

Dates

Publication Date
20260513
Application Date
20220520

Claims (15)

  1. A method comprising: receiving a layout of a lithographic mask (210); applying, by a processor, a machine learning model (230) to infer source shapes from the layout of the lithographic mask; and providing the inferred source shapes as initial source shapes (245) to a source mask optimization process.
  2. The method of claim 1, wherein the machine learning model infers source seeds (235) from the layout of the lithographic mask; and using the inferred source shapes as initial source shapes comprises generating (240) the initial source shapes from the inferred source seeds.
  3. The method of claim 1, further comprising: selecting clips (220) from the layout of the lithographic mask based on a sensitivity metric to process variations in the clips, wherein the machine learning model infers source shapes from the selected clips (225).
  4. The method of claim 3, wherein the sensitivity metric is a sensitivity metric of aerial images (106) produced by the clips to process variations in the clips.
  5. The method of claim 1, further comprising: selecting clips (220) from the layout of the lithographic mask based on which clips are close to design rule limits for the layout of the lithographic mask, wherein the machine learning model infers source shapes from the selected clips (225).
  6. The method of claim 1, wherein the layout of the lithographic mask includes features for testing compliance with design rules, and the method further comprises selecting clips (220) from the layout of the lithographic mask that include the features, wherein the machine learning model infers source shapes from the selected clips (225).
  7. The method of claim 1, wherein an output of the machine learning model comprises images of Fourier Bessel coefficients for the inferred source shapes.
  8. The method of claim 1, wherein the machine learning model comprises a U-net.
  9. The method of claim 1, wherein the method is performed iteratively for different design iterations of the layout of the lithographic mask.
  10. A non-transitory computer readable medium (824) comprising stored instructions (826), which when executed by a processor (802), cause the processor to: receiving a layout of a lithographic mask (210); select clips (220) from the layout of the lithographic mask; select source shapes based on the selected clips (225); and provide the selected source shapes as initial source shapes (245) to a source mask optimization process.
  11. The non-transitory computer readable medium of claim 10, wherein selecting the clips is based on sensitivity to process variations in the clips.
  12. The non-transitory computer readable medium of claim 11, wherein - selecting clips is based on sensitivity of aerial images (106) produced by the clips to process variations in the clips, and/or - selecting clips is based on sensitivity of resist profiles (110) produced by the clips to process variations in the clips, and/or - selecting clips is based on which clips are close to design rule limits for the layout of the lithographic mask.
  13. The non-transitory computer readable medium of claim 10, wherein the layout of the lithographic mask includes features for testing compliance with design rules, and the selected clips include said features.
  14. A system (800) comprising: a memory (804) storing instructions (826); and a processor (802), coupled with the memory and to execute the instructions, the instructions when executed cause the processor to: access a training set comprising (a) inputs comprising clips taken from layouts of lithographic masks; and (b) outputs comprising corresponding source shapes; and train a machine learning model using the training set, wherein the machine learning model infers initial source shapes for a source mask optimization process from layouts of lithographic masks.
  15. The system of claim 14, wherein - the input clips in the training set comprise instances of parameterized clips for different parameter values, and/or - the input clips in the training set comprise clips based on design rules for the layout of the lithographic mask, and/or - the input clips in the training set comprise clips representing contacts, lines and spaces of different geometries, and/or - the input clips in the training set comprise input clips and corresponding source shapes from previously designed lithographic masks and source shapes.

Description

TECHNICAL FIELD The present disclosure generally relates to a mask synthesis simulation system. In particular, the present disclosure relates to using machine learning models to select initial source shapes. BACKGROUND One step in the manufacture of semiconductor wafers involves lithography. In a typical lithography process, an illumination source produces light that is collected and directed by collection/illumination optics to illuminate a lithographic mask. Projection optics relay the pattern produced by the illuminated mask onto a wafer, exposing resist on the wafer according to the illumination pattern. The patterned resist is then used in a process to fabricate structures on the wafer. Source mask optimization refers to the process of designing the source shape in conjunction with the lithographic mask. Source shape is a term that includes both the illumination source and the collection/illumination optics. In one approach to source mask optimization, various different starting points are used for the source shapes in the source mask optimization. These initial source shapes are then improved and combined to yield the final source shape, i.e., the final design of the illumination source and/or collection/illumination optics. Document US 2012/113404 A1 describes a method for optimizing a lithographic projection apparatus including optimizing projection optics therein including several flows including optimizing a source, a mask, and the projection optics as well as sequential and iterative optimization steps. SUMMARY The present invention is defined by the attached set of claims. In some aspects, initial source shapes are determined based on a layout of the lithographic mask, instead of being arbitrarily chosen. In one approach, a layout of a lithographic mask is received. Different sections of the lithographic mask, referred to as clips, are selected. These clips are applied to a machine learning model which infers source shapes from the clips. The inferred source shapes are used as the initial source shapes for the source mask optimization process. Other aspects include components, devices, systems, improvements, methods, processes, applications, computer readable mediums, and other technologies related to any of the above. BRIEF DESCRIPTION OF THE DRAWINGS The disclosure will be understood more fully from the detailed description given below and from the accompanying figures of embodiments of the disclosure. The figures are used to provide knowledge and understanding of embodiments of the disclosure and do not limit the scope of the disclosure to these specific embodiments. Furthermore, the figures are not necessarily drawn to scale. FIG. 1 is a flowchart of a source mask optimization flow in accordance with embodiments of the present disclosure.FIG. 2 is a flowchart illustrating use of a machine learning model to infer source seeds from an input mask layout in accordance with embodiments of the present disclosure.FIG. 3 shows an architecture of a neural network suitable for use in the process of FIG. 2.FIGS. 4A-4C show examples of parameterized clip patterns in accordance with embodiments of the present disclosure.FIG. 5 shows an example mask clip and corresponding source seed in accordance with embodiments of the present disclosure.FIG. 6 illustrates source mask optimization results comparing source seeds inferred by neural network versus a typical seed selection method.FIG. 7 depicts a flowchart of various processes used during the design and manufacture of an integrated circuit in accordance with some embodiments of the present disclosure.FIG. 8 depicts a diagram of an example computer system in which embodiments of the present disclosure may operate. DETAILED DESCRIPTION Aspects of the present disclosure relate to machine learning for selecting initial source shapes for source mask optimization. Source mask optimization (SMO) uses starting conditions which include initial source shape(s) for the SMO process. The source shapes may be represented in pixelated, parameterized or other forms. Typically, these initial source shape(s) are arbitrary. They are not related to features on the lithographic mask, even though the source will be used to illuminate the lithographic mask and the source and lithographic mask will be the subject of the SMO design process. Because the initial source shapes do not take into account information about the mask layout, this approach can lead to longer optimization runtimes and non-optimal solutions. In one aspect, a machine learning model is used to determine initial source shapes based on the actual mask layout. The machine learning model may be trained using results from prior mask designs. When a new design task is encountered, clips from the layout of the lithographic mask are selected. For example, these clips may be the sections of the mask layout that are most challenging to manufacture or most sensitive to variations in the lithography process. The selected cl