Multiple patterning (or multi-patterning) is a class of technologies for manufacturing integrated circuits (ICs), developed for photolithography to enhance the feature density. It is expected to be necessary for the 10 nm and 7 nm node semiconductor processes and beyond. The premise is that a single lithographic exposure may not be enough to provide sufficient resolution. Hence additional exposures would be needed, or else positioning patterns using etched feature sidewalls (using spacers) would be necessary.
Even with single exposure having sufficient resolution, extra masks have been implemented for better patterning quality such as by Intel for line-cutting at its 45 nm nodeIntel 45nm HKMG or TSMC at its 28 nm node.TSMC 28nm cutpoly Even for electron-beam lithography, single exposure appears insufficient at ~10 nm half-pitch, hence requiring double patterning.
Double patterning lithography was first demonstrated in 1983 by D. C. Flanders and N. N. Efremow. Since then several double patterning techniques have been developed such as self alignment double patterning (SADP) and a litho-only approach to double patterning.
Pitch double-patterning was pioneered by Gurtej Singh Sandhu of Micron Technology during the 2000s, leading to the development of 30-nm class NAND flash memory. Multi-patterning has since been widely adopted by NAND flash and random-access memory manufacturers worldwide.
Situations requiring multiple patterning There are a number of situations which lead to multiple patterning being required.
Sub-resolution pitch
The most obvious case requiring multiple patterning is when the feature pitch is below the resolution limit of the optical projection system. For a system with numerical aperture NA and wavelength λ, any pitch below 0.5 λ/NA would not be resolvable in a single wafer exposure. The resolution limit may also originate from stochastic effects, as in the case of EUV. Consequently, 20 nm linewidth still requires EUV double patterning, due to larger defectivity at larger pitches.
Two-dimensional pattern rounding
It is well-established that dense two-dimensional patterns, which are formed from the interference of two or three beams along one direction, as in quadrupole or QUASAR illumination, are subject to significant rounding, particularly at bends and corners. The corner rounding radius is larger than the minimum pitch (~0.7 λ/NA).R. L. Jones and J. D. Byers, Proc. SPIE 5040, 1035 (2003). This also contributes to hot spots for feature sizes of ~0.4 λ/NA or smaller.S. Kobayashi et al., Proc. SPIE 6521, 65210B (2007). For this reason, it is advantageous to first define line patterns, then cut segments from such lines accordingly.R. Kotb et al., Proc. SPIE 10583, 1058321 (2018). This requires additional exposures. The cut shapes themselves may also be round, which requires tight placement accuracy.L. T.-N. Wang et al., Proc. SPIE 9781, 97810B (2016).
Line tip vs. linewidth tradeoff The rounding of line tips naturally leads to a tradeoff between shrinking the line width (i.e., the width of the line tip) and shrinking the gap between opposite facing tips. As the line width shrinks, the tip radius shrinks. When the line tip is already less than the point spread function (k1~0.6–0.7), the line tip naturally pulls back,C. A. Mack, Proc. SPIE 4226, 83 (2000). increasing the gap between opposite facing tips. The point spread function likewise limits the resolvable distance between the centers of the line tips (modeled as circles). This leads in turn to a tradeoff between reducing cell width and reducing cell height. The tradeoff is avoided by adding a cut/trim mask (see discussion below). Hence, for the EUV-targeted 7 nm node, with an 18 nm metal linewidth (k1=0.44 for λ=13.5 nm, NA=0.33), the line tip gap of less than 25 nm (k1=0.61) entails EUV single patterning is not sufficient; a second cut exposure is necessary.
Different parts of layout requiring different illuminations
When patterns include feature sizes near the resolution limit, it is common that different arrangements of such features will require specific illuminations for them to be printed.C. T. Bodendorf, Proc. SPIE 5992, 599224 (2005).
The most basic example is horizontal dense lines vs. vertical lines (half-pitch < 0.35 λ/NA), where the former requires a North-South dipole illumination while the latter requires an East-West dipole illumination. If both types are used (also known as cross-quadrupole C-Quad), the inappropriate dipole degrades the image of the respective line orientation.M. Burkhardt et al., Proc. SPIE 6520, 65200K (2007). Larger pitches up to λ/NA can have both horizontal and vertical lines accommodated by quadrupole or QUASAR illumination, but diagonally spaced features and elbow features are degraded.E. Hendrickx et al., Proc. SPIE 6924, 69240L (2008).W. N. Partlo et al., Prof. SPIE 1927, 137 (1993).
In DRAM, the array and periphery are exposed at different illumination conditions. For example, the array could be exposed with dipole illumination while the periphery could use annular illumination.T. Winkler et al., Proc. SPIE 5754, 1169 (2005). This situation applies to any set of patterns (half-pitch < 0.5 λ/NA) with different pitches or different feature arrangements, e.g., rectangular arrays vs. staggered arrays.Y. Chen et al., J. Vac. Sci. & Tech. B 35, 06G601 (2017).K. Tsujita et al., Proc. SPIE 6520, 652036 (2007).K. Tian et al., Proc. SPIE 7274, 72740C (2009). Any of the individual patterns is resolvable, but a single illumination cannot be used simultaneously for all of them. A minimum pitch may require an illumination that is detrimental to twice the minimum pitch with defocus.DRAM Periphery Metal Patterning Pitch IncompatibilitiesA Forbidden Pitch Combination at Advanced Lithography Nodes
The inclusion of both isolated and dense features is a well-known instance of multi-pitch patterning. Subresolution assist features (SRAFs) have been designed to enable the patterning of isolated features when using illumination tailored for the dense features. However, not all pitch ranges can be covered. In particular, semi-dense features may not be easy to include.F. M. Schellenberg et al., DAC 2001, Adoption of OPC and the Impact on Design and Layout.V. Wiaux et al., Proc. SPIE 5040, 270 (2003).
Specific example: hole arrays
For the specific case of hole arrays (minimum half-pitch < 0.6 λ/NA), three well-known cases require three entirely different illuminations. A regular array generally requires Quasar illumination, while the same array rotated 45 degrees results in a checkerboard array that requires C-quad illumination. Different from both cases, an array with close to triangular or hexagonal symmetry requires hexapole illumination.B. Bilski et al., Proc. SPIE 10466, 1046605 (2017).
Multi-pitch patterns
Sometimes a feature pattern inherently contains more than one pitch, and furthermore, these pitches are incompatible to the extent that no illumination can simultaneously image both pitches satisfactorily. A common example, again from DRAM, is the brick pattern defining the active regions of the array.Samsung 80nm DRAMWhen Lithography Requires Multiple Patterning: Avoiding Analog Distortions in the Resist Layer In addition to the narrow pitch of the active regions, there is also the pitch between the active region separations or breaks, which is different from that of the narrow pitch in the same direction. When the narrow pitch is < λ/NA (but still > 0.5 λ/NA), it cannot be imaged simultaneously with the double pitch due to the focus limitations of the latter. Selective etching, along with SADP or SAQP (to be described below), is the current best approach to achieve the simultaneous patterning of both pitches.Feature-selective etching in SAQP for sub-20nm patterning
Small deviations from 2-beam interference A two-beam interference pattern (half-pitch <0.5 λ/NA) forms a set of regularly spaced lines. Breaks in such lines, e.g., brick patterns, are deviations from the interference pattern. Such breaks generally do not dominate the pattern, and are thus small deviations. These deviations are insufficient to completely offset the constructive or destructive interference of the underlying regular line pattern; sidelobes often result.N. Singh and M. Mukherjee-Roy, Proc. SPIE vol. 4691, 1054 (2002).J. Garofalo et al., J. Vac. Sci. & Tech. B 11, 2651 (1993). Line end gaps are easily bridged under dipole illumination.T. Matsuda et al., Proc. SPIE 7973, 797316 (2011). Another mask exposure (usually referred as a cut mask) is therefore necessary to break the line pattern more robustly.
Line cutting
The earliest implementation of multiple patterning involved line cutting. This first occurred for Intel's 45 nm node, for 160 nm gate pitch.Intel Technology Journal 45nm issue The use of a second mask to cut lines defined by a first mask does not help increase feature density directly. Instead it allows definition of features, e.g., brick patterns, which are based on lines spaced at a minimum pitch, in particular, when the lines are near the resolution limit and are generated by the two-beam interference mentioned above. The two-beam interference still dominates the diffraction pattern. In fact, in the absence of a separate cut exposure, the gap between the ends of the minimum pitch lines will be prohibitively large.How Line Cuts Became Necessarily Separate Steps in LithographyM. Eurlings et al., Proc. SPIE 4404, 266 (2001). This is due to rounding resulting from reduced spatial frequencies.Vanishing of Half the Fourier Coefficients in Staggered Arrays
The line cut shapes themselves are subject to rounding; this rounding can be minimized with optimized illumination,Optimized Illumination for Dense Metal Cut Patterns but cannot be eliminated completely.
When applying the second mask to cut lines, the overlay relative to the first mask needs to be considered; otherwise, edge placement errors (EPE) may result. If the line pitch is already near the resolution limit, the cut pattern itself may have imaging difficulty, from reduced dose or focus window. EUV stochastic variability causes random shaping of the cuts.EUV Stochastic Variability in Line Cuts In this case, more than one cut mask would have to be used, or else the cut has to extend over more than one line. Self-aligned line cutting (to be discussed below) may be a preferred option.
Pitch splitting
The earliest form of multiple patterning involved simply dividing a pattern into two or three parts, each of which may be processed conventionally, with the entire pattern combined at the end in the final layer. This is sometimes called pitch splitting, since two features separated by one pitch cannot be imaged, so only skipped features can be imaged at once. It is also named more directly as "LELE" (Litho-Etch-Litho-Etch). This approach has been used for the 20 nm and 14 nm nodes. The additional cost of extra exposures was tolerated since only a few critical layers would need them. A more serious concern was the effect of feature-to-feature positioning errors (overlay). Consequently, the self-aligned sidewall imaging approach (described below) has succeeded this approach.