Research Article Current Issue Versions 1 Vol 4 (2) : 21040202 2021
Patterning with Organized Molecules
: 2021 - 06 - 01
1373 24 0
Abstract & Keywords
Abstract: Decades of progress in the semiconductor industry has led to lithographically printed dimensions that are small enough that the positions of individual molecules and the stochastic variation in the number of photons have a significant effect on the quality of photoresist patterns. These effects scale badly and will be more important as feature sizes continue to shrink. Self-organizing materials can provide regular patterns of molecules that have the potential to minimize stochastic effects. Some such reported materials are block copolymers, bottle brush polymers and DNA, all of which have been used as part of lithographic patterning. A key challenge for self-organizing materials is defect levels. The energy to rearrange has to be high enough that random defects aren’t created thermally but low enough that rearrangement into preferred domains can occur. All of the methods can generate accurate CDs based on the chemical composition of the material, but they all need some way to control the positions of the feature edges. There are methods for guiding the self-organization, but the final position is the sum of the guide pattern misalignment and the intrinsic alignment error of the self-organizing materials. Thus it can be worse than the positioning of the guide structures. Alignment and defect levels are thus two big challenges for manufacturing introduction of self-organizing materials.
Keywords: Stochastics; Self-assembly; overlay; edge placement error; self-organizing; DNA origami; bottle brush polymers
1.   Introduction
For many decades the semiconductor industry has relied on better and better patterning methods to improve resolution and make smaller patterns that support the production of smaller circuits and thus provide more functionality per unit area. The progress has occurred over decades and has enabled orders of magnitude improvement in the size and density of patterns on computer chips. It has been so successful that now the sizes of the smallest features on leading edge devices are comparable to that of polymer molecules. This success has brought challenges along with it. Once desired patterns are the size of molecules, then the positions of molecules and the number of molecules in a given area are random variables, and the imaging process can no longer be thought of as changing the solubility of a continuous material. Instead, the imaging process has to be modeled in molecular terms. Small-scale molecular randomness in a photo-imaging material will result in patterns that do not exactly match the original image that was projected on the photoresist. Instead, they have rough line edges, variable dimensions and other types of defects, even going so far as to have missing features or features with gaps or bridges. Of course, much work is underway to improve resists. But, long term, new classes of resists may be needed. One possibility is to use patterning materials that have self-organizing properties. “Self-organizing” refers to molecules where either intramolecular or intermolecular interactions drive molecular organization that creates organized structures of use in providing molecular scale patterning. This paper reviews some approaches that have been published that take advantage of self-organization and discusses some of the tradeoffs for each of the approaches.
2.   Challenge of Stochastics
The smallest pattern dimensions in leading edge chips are shown in the latest lithography roadmap [1] Critical dimensions are expected to continue to shrink for the next ten years. As described in the roadmap, a key challenge for future patterning is stochastics 2. Stochastics are the random variations in exposure photons, the positions and numbers of molecules in the exposed areas of resist, the generation and propagation of secondary electrons and random variations in light absorption. Stochastic effects can affect the size of features such as contact holes, giving random variations in pattern size; they can affect the edges of patterned features, for example creating unacceptable random roughness along the edge of a line feature; they can affect the printed position of features, for example the center of a contact hole could be in a different location than the designed location; and they can create defects such as missing contact holes or line breaks.
Stochastic effects scale badly. As critical dimensions shrink, stochastic effects become a larger fraction of the critical dimension. Consider photon shot noise as an example. This is the random variation in the number of photons in an exposure dose. The number of photons will have a Poisson distribution with a standard deviation that is the square root of the average number of photons. If the dose to print for a hole pattern currently in volume production requires 10,000 photons per hole, then the standard deviation from shot noise will 100 photons, or 1% of the imaging dose. If the next generation of patterns has a 30% reduction in printed feature size, then there will be 7000 photons per hole (assuming the same dose to print for resist), a standard deviation of 84 photons per hole, which is 1.2% of the imaging dose. The required tolerance on hole size shrinks proportionally to the hole size, but the magnitude of stochastic effects increases instead of decreasing. The proportion of stochastic effects due to other factors, such as random positions of molecules will also increase since there will be fewer molecules per contact hole. Thus, stochastic effects get worse with smaller feature sizes instead of better.
3. Demonstrated Classes of Self-Organizing Materials for Semiconductor Patterning
The industry is working to accommodate or minimize these effects, but the possible solutions have significant costs. For example, a paper presented in 2020 predicted substantial increases in dose to print for EUV resists [3]. Since EUV tool throughput is slower for higher dose to print, this will increase patterning cost substantially. Using double patterning techniques to print larger features than will be required in the final device will minimize some stochastic effects but will also add significant cost and complexity. There is a need for patterning methods where stochastic effects are minimized or eliminated. One possibility is to use self-organizing materials. These are materials where favorable intra or inter molecular interaction energies are larger than entropic effects, giving structures that self-assemble. By definition, these structures will have less randomness than simple mixtures of molecules. Some such materials have already been demonstrated. The sections below discuss some such materials already demonstrated, their attributes related to semiconductor patterning, and their advantages and limitations. The best-known class of self-organizing materials used for patterning is block co-polymers. These materials consist of two or more polymer chains connected covalently [4,5]. The different blocks of polymers self-assemble into distinct domains of different phase whose dimensions are related to the size of the polymer. They have a characteristic length, L0 which is related to the size of the polymer. So far, these materials have been used to create simple repeating patterns, such as arrays of lamellae which can become lines and spaces after etch or arrays of cylinders that can become arrays of holes or pillars after etch. Guide structures are used to line up or “direct” the domains to put them in the right place. This gives rise to the general term “Directed Self-Assembly” (DSA) for this process. Another approach was reported in 2013 [6]. It used what are called “bottle brush polymers” which are large rigid polymers that have a cylindrical shape. The cylinders were 3 to 10nm in diameter and 30 to 100nm long. One end of the cylinder was functionalized to reduce surface energy and the other end was functionalized to adhere to a silicon oxide substrate. These materials were then able to give an array of vertical cylinders on the substrate if the spin coating thickness was set at the length of the polymer rods. By including photosensitive cross-linking chemistry, the polymer rods could be cross linked in exposed areas. Since a rod either stays or develops away if not cross linked, this basically gives a pixelated resist, where the size of the pixel is the diameter of the rods. Any feature imaged in this resist can only have a dimension that is a multiple of the cylinder to cylinder spacing along the direction of measurement. The cylinders are not guided into place, so the edge of a feature will depend both on where the edge of the aerial image is and on where the edges of the cylinders happen to fall. This method does not show any reduction of feature size below the aerial image size, but it changes the stochastics considerably. A third approach is called DNA origami [7]. This approach takes advantage of the ability to nanoengineer DNA to make two- and three-dimensional objects. In the cited paper, flat rectangles of DNA were designed that each had two small square holes and one larger square holes. The origami was aligned onto a wafer with a patterned hexamethyldisilazane coating (HMDS). An etch process was used to transfer the DNA pattern into the silicon, resulting in 25nm and 12nm holes in the silicon. In all the self-organizing molecule approaches described above, the actual patterned dimension is a function of the molecular size or of the molecular design. This means that stochastic variations in the critical dimension are suppressed. For DSA, the thickness of lamellae or cylinders will determine the dimension of the patterned line or hole, respectively. Random variations in the size of the guide pattern due to stochastic effects will have little influence on the critical dimension. For the bottle brush polymer approach, the width of a patterned feature has to be a multiple of the pitch of the polymeric cylinders. A small variation in the imaging dose will not change the pattern width. For DNA origami, the dimensions of the patterned holes are fixed by the molecular structure. The only lithography involved was patterning guide patterns for the larger DNA rectangles to adhere to. Variations in the pattern size of the guide patterns will not affect the etched hole size. This is different from conventional lithography where the feature size is a function of the imaging dose and is a major benefit for self-assembly methods of patterning On a smaller scale, the roughness at the edge of the patterns will also be different. For DSA, the edge of patterned features will have roughness caused by variations in the locations of boundary molecules. The variations will be determined by the c value of the blocks in the block co-polymer, not by anything in the patterning process for the guide patterns. The roughness of pattern will not get worse as feature sizes shrink, and in fact it may get somewhat smaller in absolute terms if higher c block copolymers are used to pattern the smaller features. For the bottle brush polymers, printed features will have designed in roughness that has the characteristic pitch of the pixels, but smaller scale and larger scale roughness will be absent. The effects of self-assembly on roughness and CD control show promise, but the effect of self-assembly on printed feature position has to be considered. For directed self-assembly, typical processes for reducing critical dimension use pitch multiplication, where some of the DSA blocks are aligned to a guide feature but others fall in between the guide feature. The position of the in-between blocks can vary a little and this adds to any positional errors in the guide features. For the bottle brush polymer approach, the edge of a printed feature will fall along a line of pixel edges closest to the edge of the aerial image, so the printed feature could be as much as half the diameter of a cylinder away from the actual placement of the aerial image. For the DNA approach, the alignment will be dependent on how well the DNA actually aligns to the pattern feature. There should be no further alignment variation due to variations in the DNA since the patterned feature placement with the DNA template is controlled by the molecular structure of the larger template.
4.   Defect Levels
A key question for any new technique is defect levels. With self-assembled methods, there is a reduction in free energy gained by the organizing of the molecules; but, after initial application of the material, the molecules are usually not in the minimum energy arrangement. Some sort of annealing step is needed to enable the molecules to arrange themselves in a way that minimizes free energy. If the free energy reduction by reaching the optimum arrangement is small, then thermal energy from the environment, as measured by kT, can cause random rearrangements away from the desired configuration. In this case, defects can spontaneously form or cannot be annealed out because further annealing starts creating new defects at the same time as other defects anneal away. For DSA, it has been shown that the free energy penalty of creating a defect is large enough that annealing can produce defect free structures [8]. However, for DNA origami, this is reported not to be the case [9]. For the bottle brush polymers, no such data is available.
The self-organizing systems described above are of different types. The bottle brush polymers basically provide imaging pixels. In the described work, the individual pixels aren’t aligned to any guide structures. The pixels are circular rather than square or some other shape, and the resolution is not as small as the pixels because cross linking of the cylinders is used to create the pattern after development. In theory, the cylinders could be aligned to guide structures and the pixels could have a different shapy using some clever (but as yet unknown) chemistry. Assuming this could be done, there would still be some limitations. Printed CDs would have to be an integer fraction of the feature size, and there would be a tradeoff of pixel size and feature size. Large pixels would give less roughness, but constrain the dimensions you can print. Small pixels would enable a wider range of dimensions but provide more random feature edges. What is more, the mechanism doesn’t enable smaller features than the aerial image provides. You may get cleaner patterns than with ordinary photoresist, but it this may not be valuable enough to make this effort worthwhile.
In DSA there also defined domains, and in some cases, such as vertical domains that are cylindrical, they can be thought of as pixels. In other cases, such as with lamellae or with horizontal cylinders, the domains are close to infinite in length and are not typical pixels at all. But a printed pattern is one domain wide, so the edge of the pattern is determined by molecular effects at the edges of domains, and the edge roughness is not directly related to the domain size. In DNA origami, the patterns have boundaries that are related to the DNA molecular structure and should not suffer from patterning noise. However, there may be some pattern transfer noise. In the cited example, even though the opening in the DNA were square the transferred patterns were more circular.
In all of these examples, the position of the final pattern is a challenge. The bottle brush polymers do not line up on a preferred grid and this adds variation in the positions of the edges of a printed pattern. The other two methods use a guide pattern to control placement, but there is noise in the alignment of the organized molecules to the guide pattern which adds to the alignment error already present due to the overlay variation in the printing process. Overlay is already a key challenge in moving to smaller critical dimensions. Future nodes have projected overlay requirements for three sigma overlay of less than 2nm. Meeting this target is a challenge for all organized molecule methods. It’s not enough that molecules self-organize into useful patterns. They have to self-organize in the right places.
5.   Conclusion
Molecular size structures are now part of leading-edge semiconductor devices. Organized molecules have shown great promise for making features of this size. There are several patterning methods using organized molecules that have been demonstrated. DSA is much closer to possible implementation than the others. Effective use of these methods requires not only achieving structures of the correct shape and size, but also putting those structures in the correct location. Ways of ensuring molecular alignment are needed if these methods are to reach their full potential.
[1] See and links therein to the lithography and More Moore roadmaps.
[2] Ibid.
[3] Mark Neisser, Harry J. Levinson, "Projecting EUV photo-speeds for future logic nodes," Proc. SPIE 11323, Extreme Ultraviolet (EUV) Lithography XI, 113231N (23 March 2020).
[4] Christian Pinto-Gómez, Francesc Pérez-Murano, Joan Bausells, Luis Guillermo Villanueva, and Marta Fernández-Regúlez, “Directed Self-Assembly of Block Copolymers for the Fabrication of Functional Devices”, Polymers 2020, 12, 2432.
[5] Hanqiong Hu, Manesh Gopinadhan† and Chinedum O. Osuji, “Directed self-assembly of block copolymers: a tutorial review of strategies for enabling nanotechnology with soft matter”, Soft Matter, 2014, 10, 3867.
[6] Trefonas et al., “Bottom-up/top-down, high-resolution, high-throughput lithography using vertically assembled block bottle brush polymers”, Proc. SPIE, Vol. 8682 (2013).
[7] Marie Marmiesse, Raluca Tiron, Guillaume Thomas, Shimon Levi, Xavier Baillin, "Nanoengineering DNA origami for lithography," Proc. SPIE 11324, Novel Patterning Technologies for Semiconductors, MEMS/NEMS and MOEMS 2020, 1132411 (27 March 2020).
[8] Nabil Laachi, Hassei Takahashi, Kris T. Delaney, Su-Mi Hur, David Shykind, Corey J. Weinheimer, Glenn H. Fredrickson, "Self-consistent field theory of directed self-assembly in laterally confined lamellae-forming diblock copolymers," Proc. SPIE 8323, Alternative Lithographic Technologies IV, 83230K (21 March 2012).
[9] J. Alexander Liddle, Jacob M. Majikes, "Nucleic acid nanofabrication: existential angst and killer applications," Proc. SPIE 11610, Novel Patterning Technologies 2021, 1161015 (22 February 2021).
Article and author information
Mark Neisser
Publication records
Published: Aug. 9, 2021 (Versions1
Journal of Microelectronic Manufacturing