Extreme ultraviolet lithography

Blank photomasks are mainly made by two companies: AGC Inc. and Hoya Corporation.[10][11] A blank photomask is covered with photoresist, which is then exposed using a laser, using maskless lithography.[12] The exposed photoresist is developed (removed) and the unprotected areas are etched. The remaining photoresist is then removed. Masks are then inspected and later repaired using an electron beam.[13] Etching must be done to a very specific depth thus making etching difficult when compared with conventional photomask manufacturing.[14]

Source: Gigaphoton, Sematech Symposium Japan, September 15, 2010

The required utility resources are significantly larger for EUV compared to 193 nm immersion, even with two exposures using the latter. Hynix reported at the 2009 EUV Symposium that the wall plug efficiency was ~0.02% for EUV, i.e., to get 200-watts at intermediate focus for 100 wafers-per-hour, one would require 1-megawatt of input power, compared to 165-kilowatts for an ArF immersion scanner, and that even at the same throughput, the footprint of the EUV scanner was ~3x the footprint of an ArF immersion scanner, resulting in productivity loss.[19] Additionally, to confine ion debris, a superconducting magnet may be required.[20]

A typical EUV tool weighs 180 tons.[21]

The following table summarizes key differences between EUV systems in development and ArF immersion systems which are widely used in production today:

The different degrees of resolution among the 0.33 NA tools are due to the different illumination options. Despite the potential of the optics to reach sub-20 nm resolution, secondary electrons in resist practically limit the resolution to around 20 nm.[31]

The EUV light source limits tool uptime besides throughput. In a two-week period, for example, over seven hours downtime may be scheduled, while total actual downtime including unscheduled issues could easily exceed a day.[41] A dose error over 2% warrants tool downtime.[41]

While state-of-the-art 193 nm ArF excimer lasers offer intensities of 200 W/cm²,[43] lasers for producing EUV-generating plasmas need to be much more intense, on the order of 10¹¹ W/cm².[44] A state-of-the-art ArF immersion lithography 120 W light source requires no more than 40 kW[45] while EUV sources are targeted to exceed 40 kW.[46]

The power target for EUV lithography is at least 250 W, while for other conventional lithography sources, it is much less.[41] For example, immersion lithography light sources target 90 W, dry ArF sources 45 W, and KrF sources 40 W. High-NA EUV sources are expected to require at least 500 W.[41]

GlobalFoundries and Lawrence Berkeley Labs carried out a Monte Carlo study to simulate the effects of intermixing between the molybdenum (Mo) and silicon (Si) layes in the multilayer that is used to reflect EUV light from the EUV mask.[59] The results indicated high sensitivity to the atomic-scale variations of layer thickness. Such variations could not be detected by wide-area reflectivity measurements, but would be significant on the scale of the critical dimension (CD).[59]

Wavelength dependence of reflectance angular spectrum. The reflectance as a function of incident angle depends significantly on the wavelength. Inset: Apodization due to angular dependence causes asymmetric distribution across the pupil for different wavelengths.[60]

Image shift due to defocus depends on wavelength. The angular dependence of multilayer reflectance of the object (mask) is different for different wavelengths, leading to different shifts when defocused.

Unlike DUV lithography sources, based on excimer lasers, EUV plasma sources produce light across a broad range of wavelengths.[61] Though the EUV spectrum is not completely monochromatic, nor even as spectrally pure as DUV laser sources, the working wavelength has generally been taken to be 13.5 nm. In actuality, the reflected power is distributed mostly in the 13.3-13.7 nm range.[62] The bandwidth of EUV light reflected by a multilayer mirror used for EUV lithography is over +/-2% (>270 pm);[63] the phase changes due to wavelength changes at a given illumination angle may be calculated[64] and compared to the aberration budget.[65] Wavelength dependence of reflectance[64][62] also affects the apodization, or illumination distribution across the pupil (for different angles); different wavelengths effectively 'see' different illuminations as they are reflected differently by the multilayer of the mask.[66][62] This effective source illumination tilt can lead to large image shifts due to defocus.[67] Conversely, the peak reflected wavelength varies across the pupil due to different incident angles.[62][68] This is aggravated when the angles span a wide radius, e.g., annular illumination. The peak reflectance wavelength increases for smaller incident angles.[69] Aperiodic multilayers have been proposed to reduce the sensitivity at the cost of lower reflectivity but are too sensitive to random fluctuations of layer thicknesses, such as from thickness control imprecision or interdiffusion.[70] In particular, defocused dense lines at pitches up to twice the minimum resolvable pitch suffer wavelength-dependent edge shifts.[71]

A narrower bandwidth would increase sensitivity to mask absorber and buffer thickness on the 1 nm scale.[72][73]

EUV mask phase shift effect. A phase shift on an EUV mask, even from a faintly bright region, shifts the feature position, and also changes its printed width.

Image changes through focus. The image contrast varies through focus as expected, but the best focus position is very different depending on a variety of factors, such as illumination angle in this case.

The EUV mask absorber, due to partial transmission, generates a phase difference between the 0th and 1st diffraction orders of a line-space pattern, resulting in image shifts (at a given illumination angle) as well as changes in peak intensity (leading to linewidth changes)[74] which are further enhanced due to defocus.[75][76] Ultimately, this results in different positions of best focus for different pitches and different illumination angles. Generally, the image shift is balanced out due to illumination source points being paired (each on opposite sides of the optical axis). However, the separate images are superposed and the resulting image contrast is degraded when the individual source image shifts are large enough. The phase difference ultimately also determines the best focus position.

Pitches and orientations in metal layouts. The variation in pitches and orientations in metal layouts will cause complications for EUV lithography due to the use of reflective optics.

A fundamental aspect of EUVL tools, resulting from the use of reflective optics, is the off-axis illumination (at an angle of 6 degrees, in different direction at different positions within the illumination slit)[77] on a multilayer mask. This leads to shadowing effects resulting in asymmetry in the diffraction pattern that degrade pattern fidelity in various ways as described below.[78]

Shadowing effect on an EUV mask. When the feature size on the mask is comparable to the thickness, the shadowing effect is more significant to the formed image, since the shadow clearance distance is not negligible compared to the feature size.

The oblique incidence in a reflective optical system leads to shadowing effects in the presence of a mask absorber. For example, one side (behind the shadow) would appear brighter than the other (within the shadow).[79]

EUV H-V difference of focus. Horizontal and vertical features are focused differently in EUV optical systems. The numerical aperture (NA) also makes a difference.

EUV non-telecentricity. Left: Due to large multilayer reflection angle differences, one side of the illumination pupil results in more reflected light. Right: Consequently, illumination from one side will be dominant. This results in an optical path difference between diffraction orders with respect to defocus, leading to a tendency for the pattern to shift.

Most fundamentally, the behavior of light rays within the plane of reflection (affecting horizontal lines) is different from the behavior of light rays out of the plane of reflection (affecting vertical lines).[80] Most conspicuously, identically sized horizontal and vertical lines on the EUV mask are printed at different sizes on the wafer.

2-bar CD difference vs. focus. The difference between the widths of two adjacent horizontal lines varies as a function of focus.

The combination of the off-axis asymmetry and the mask shadowing effect leads to a fundamental inability of two identical features even in close proximity to be in focus simultaneously.[81] One of EUVL's key issues is the asymmetry between the top and bottom line of a pair of horizontal lines (the so-called "two-bar"). Some ways to partly compensate are the use of assist features as well as asymmetric illumination.[82]

An extension of the two-bar case to a grating consisting of many horizontal lines shows similar sensitivity to defocus.[83] It is manifest in the CD difference between the top and bottom edge lines of the set of 11 horizontal lines. The table below lists the CD difference over 100 nm focus range under quasar illumination (quasar illumination will be described in the section on optimum illumination vs. pitch).

For pitches of 40 nm or below, the linewidths are 20 nm or less, while the CD difference is at least 2.5 nm, resulting in at least 12.5% difference.

Two-bar occurrence in irregular layout. An irregular layout can include two-bar locations which are prone to asymmetric imaging.

The use of reflection causes wafer exposure position to be extremely sensitive to the reticle flatness and the reticle clamp. Reticle clamp cleanliness is therefore required to be maintained. Small (mrad-scale) deviations in mask flatness in the local slope, coupled with wafer defocus.[84] More significantly, mask defocus has been found to result in large overlay errors.[85][86] In particular, for a 10 nm node metal 1 layer (including 48 nm, 64 nm, 70 nm pitches, isolated, and power lines), the uncorrectable pattern placement error was 1 nm for 40 nm mask z-position shift.[87] This is a global pattern shift of the layer with respect to previously defined layers. However, features at different locations will also shift differently due to different local deviations from mask flatness, e.g., from defects buried under the multilayer. It can be estimated that the contribution of mask non-flatness to overlay error is roughly 1/40 times the peak-to-valley thickness variation.[88] With the blank peak-to-valley spec of 50 nm, ~1.25 nm image placement error is possible. Blank thickness variations up to 80 nm also contribute, which lead to up to 2 nm image shift.[88]

10nm node metal layer pattern shift vs defocus. Different patterns in the 10nm node metal layer (24 nm hp) are shifted differently through focus, depending on orientation and slit position as well as arrangement.

The off-axis illumination of the reticle is also the cause of non-telecentricity in wafer defocus, which consumes most of the 1.4 nm overlay budget of the NXE:3400 EUV scanner[89] even for design rules as loose as 100 nm pitch.[90] The worst uncorrectable pattern placement error for a 24 nm line was about 1.1 nm, relative to an adjacent 72 nm power line, per 80 nm wafer focus position shift at a single slit position; when across-slit performance is included, the worst error is over 1.5 nm in the wafer defocus window[87] In 2017, an actinic microscope mimicking a 0.33 NA EUV lithography system with 0.2/0.9 quasar 45 illumination showed that an 80 nm pitch contact array shifted -0.6 to 1.0 nm while a 56 nm pitch contact array shifted -1.7 to 1.0 nm relative to a horizontal reference line, within a +/- 50 nm defocus window.[91]

Wafer defocus also leads to image placement errors due to deviations from local mask flatness. If the local slope is indicated by an angle α, the image is projected to be shifted in a 4x projection tool by 8 α x (DOF/2) = 4 α DOF, where DOF is the depth of focus.[92] For a depth of focus of 100 nm, a small local deviation from flatness of 2.5 mrad (0.14°) can lead to a pattern shift of 1 nm.

Ring-field for uniform illumination. A ring field refers to the arc-shaped field used to achieve uniform illumination from a point that is off-axis, as in reflective optical systems.

Illumination rotation across ring-field slit. Light reflected from curved optical surfaces will generate arc segments.[93] The illumination angles are rotated azimuthally across the arc-shaped slit (right), due to the reflection of an arc-shaped image from each pupil position as a point source (left).[94][95] The angle-dependent and wavelength-dependent multilayer reflectance distribution pattern is rotated accordingly.

The direction of illumination is also highly dependent on slit position, essentially rotated azimuthally.[96][97][28][98][99][100] The reason for this is a mirror is used to transform straight rectangular fields into arc-shaped fields.[101][102] More generally, so-called "ring-field" systems reduce aberrations by relying on the radial symmetry of an arc-shaped field derived from an off-axis annulus.[103] This is preferred, as reflective systems must use off-axis paths, which aggravate aberrations. Hence identical die patterns within different halves of the arc-shaped slit would require different OPC. This renders them uninspectable by die-to-die comparison, as they are no longer truly identical dies. For pitches requiring dipole, quadrupole, or hexapole illumination, the rotation also causes mismatch with the same pattern layout at a different slit position, i.e., edge vs. center. Even with annular or circular illumination, the rotational symmetry is destroyed by the angle-dependent multilayer reflectance described above. Although the azimuthal angle range is +/- ~20° on 0.33 NA scanners,[28][98][104][105] at 7nm design rules (36-40 nm pitch), the tolerance for (quasar) illumination can be +/-15°,[106][107] or even less.[108][109] Annular illumination nonuniformity and asymmetry also significantly impact the imaging.[110]

Rotated source illumination at different EUV slit locations. The quasar illumination is rotated across slit, requiring compensating counter-rotation at different locations at different times.

The slit position dependence is particularly difficult for the tilted patterns encountered in DRAM.[99] Besides the more complicated effects due to shadowing and pupil rotation, tilted edges are converted to stair shape, which may be distorted by OPC.

Rotated dipole illumination. Rotated dipole illumination (red) causes pattern degradation as only light from the overlap between the original (blue) and rotated locations can contribute to imaging. The remaining light reduces contrast by acting as unpatterned background light.

Aberrations, originating from deviations of optical surfaces from subatomic (<0.1 nm) specifications[111] as well as thermal deformations[112][113] and possibly including polarized reflectance effects,[114] are also dependent on slit position,[115][113] as will be further discussed below, with regard to source-mask optimization (SMO). The thermally induced aberrations are expected to exhibit differences among different positions across the slit, corresponding to different field positions, as each position encounters different parts of the deformed mirrors.[116] Ironically, the use of substrate materials with high thermal and mechanical stability make it more difficult to compensate wavefront errors[117]

Flare is the presence of background light originating from scattering off of surface features which are not resolved by the light. In EUV systems, this light can be EUV or out-of-band (OoB) light that is also produced by the EUV source. The OoB light adds the complication of affecting the resist exposure in ways other than accounted for by the EUV exposure. OoB light exposure may be alleviated by a layer coated above the resist, as well as 'black border' features on the EUV mask.[118] However, the layer coating inevitably absorbs EUV light, and the black border adds EUV mask processing cost.

Tip-to-tip difficulty for dense line patterns. The gap between line ends is difficult to optimize simultaneously with dense lines.

Tip-to-side gap printing difficulty. The tip-to-side gap is one of the hardest features to print in a bidirectional pattern. The dipole-like illumination favors the minimum pitch but not other pitches.

A key challenge for EUV is the counter-scaling behavior of the line tip-to-tip (T2T) distance as half-pitch (hp) is scaled down.[108] This is in part due to lower image contrast for the binary masks used in EUV lithography, which is not encountered with the use of phase shift masks in immersion lithography.[119][120] The rounding of the corners of the line end leads to line end shortening,[121] and this is worse for binary masks.[122] The use of phase-shift masks in EUV lithography has been studied but encounters difficulties from phase control in thin layers[123] as well as the bandwidth of the EUV light itself.[124] More conventionally, optical proximity correction (OPC) is used to address the corner rounding and line-end shortening. In spite of this, it has been shown that the tip-to-tip resolution and the line tip printability are traded off against each other, being effectively CDs of opposite polarity.[125] In addition, the effectiveness of optical corrections depends on other non-optical causes such as resist blur and diffusion effects, which may also include secondary electron blur (discussed in the section on photoresist exposure).[126] Also, larger molecular weights and sizes appear to reduce corner rounding.[127]

In unidirectional metal layers, tip-to-tip spacing is one of the more severe issues for single exposure patterning. For the 40 nm pitch vertical lines, an 18 nm nominal tip-to-tip drawn gap resulted in an actual tip-to-tip distance of 29 nm with OPC (optical proximity correction),[108] while for 32 nm pitch horizontal lines, the tip-to-tip distance with a 14 nm nominal gap went to 31 nm with OPC.[128] These actual tip-to-tip distances define a lower limit of the half-pitch of the metal running in the direction perpendicular to the tip. In this case, the lower limit is around 30 nm. With further optimization of the illumination (discussed in the section on source-mask optimization), the lower limit can be further reduced to around 25 nm.[129]

For larger pitches, where conventional illumination can be used, the line tip-to-tip distance is generally larger. For the 24 nm half-pitch lines, with a 20 nm nominally drawn gap, the distance was actually 45 nm, while for 32 nm half-pitch lines, the same nominal gap resulted in a tip-to-tip distance of 34 nm.[128] With OPC, these become 39 nm and 28 nm for 24 nm half-pitch and 32 nm half-pitch, respectively.[130]

The printed space between a line tip and a perpendicular line it faces is 25-35 nm for 22 nm half-pitch lines (with a 20 nm nominally drawn gap).[130] For a 22 nm line-space pattern with 22 nm nominal gap, conventional illumination yields a 38 nm tip-to-line distance, while quasar illumination yields a 28 nm distance.[131] The tip-to-side gap is one of the hardest features to print in a bidirectional pattern.[130]

Summary of EUV line tip and corner effects:[132]

Source: Semicon West 2017, IBM

The line end clearance distance of 28 nm essentially forces pitches to be at least 56 nm for EUV single exposure bi-directional patterning. 7 nm node minimum metal pitch is already at 40 nm or below, while gate pitch can also be below 56 nm,[133][134] so this is an indication that multiple patterning would be needed even for EUV at 7 nm.[135]

Assist feature OPC. Assist features help improve the image of isolated features (blue) to be more like dense features (gray). However, the more effective they are, the greater the risk that the assist feature will print (orange).

Pattern effect on SMO. A mere widening of the power rail (top and bottom of sample pattern) leads to significant changes in the optimized pupil as well as reduced focus window, even with the use of assist features.

Assist features are often used to help balance asymmetry from non-telecentricity at different slit positions, due to different illumination angles, starting at the 7 nm node,[136][137] where the pitch is ~ 41 nm for a wavelength ~13.5 nm and NA=0.33, corresponding to k1 ~ 0.5.[138] However, the asymmetry is reduced but not completely eliminated, since the assist features mainly enhance the highest spatial frequencies, whereas intermediate spatial frequencies, which also affect feature focus and position, are not much affected. The coupling between the primary image and the self images is too strong for the asymmetry to be completely eliminated by assist features; only asymmetric illumination can achieve this.[82] Assist features may also get in the way of access to power/ground rails. Power rails are expected to be wider, which also limits the effectiveness of using assist features, by constraining the local pitch. Local pitches between 1x and 2x the minimum pitch forbid assist feature placement, as there is simply no room to preserve the local pitch symmetry. In fact, for the application to the two-bar asymmetry case, the optimum assist feature placement may be less than or exceed the two-bar pitch.[137] Depending on the parameter to be optimized (process window area, depth of focus, exposure latitude), the optimum assist feature configuration can be very different, e.g., pitch between assist feature and bar being different from two-bar pitch, symmetric or asymmetric, etc..

At pitches smaller than 58 nm, there is a tradeoff between depth of focus enhancement and contrast loss by assist feature placement.[137] Generally, there is still a focus-exposure tradeoff as the dose window is constrained by the need to have the assist features not print accidentally.

An additional concern comes from shot noise;[139] sub-resolution assist features (SRAFs) cause the required dose to be lower, so as not to print the assist features accidentally.[140] This results in fewer photons defining smaller features (see discussion in section on shot noise).

It is now known that the underlying mechanism for the asymmetry is different shadowing from different angles of incidence. Hence, reducing absorber thickness would be the most direct way to resolve the issue.[141]

Stochastic printing of sub-resolution assist features. SRAFs receive low enough doses which are close enough to printing that they will have more significant stochastic impact on printing. Here the SRAF printing error occurs at the far right.

As SRAFs are smaller features than primary features and are not supposed to receive doses high enough to print, they are more susceptible to stochastic dose variations causing printing errors.[142] This is particularly prohibitive for EUV, where even when the primary feature is printed at 80 mJ/cm², the SRAF suffers from stochastic printing.

Source-mask optimization (SMO) is used to reduce pattern shift for different features in a metal layer (targeted for 16 nm CD with 32 nm anchor pitch) in a single exposure, but cannot satisfy every possible case.

Pitch effect on SMO. SMO carried out targeted for one pitch (32 nm in this case) may have varying performance for other pitches. Here 36 nm has best performance, but barely exceeds the lower NILS limit of 2.0

Due to the effects of non-telecentricity, standard illumination pupil shapes, such as disc or annular, are not sufficient to be used for feature sizes of ~20 nm or below (10 nm node and beyond).[90] Instead certain parts of the pupil (often over 50%) must be asymmetrically excluded. The parts to be excluded depend on the pattern. In particular, the densest allowed lines need to be aligned along one direction and prefer a dipole shape. For this situation, double exposure lithography would be required for 2D patterns, due to the presence of both X- and Y-oriented patterns, each requiring its own 1D pattern mask and dipole orientation.[143][144] There may be 200–400 illuminating points, each contributing its weight of the dose to balance the overall image through focus. Thus the shot noise effect (to be discussed later) critically affects the image position through focus, in a large population of features.

Double or multiple patterning would also be required if a pattern consists of sub-patterns which require significantly different optimized illuminations, due to different pitches, orientations, shapes, and sizes.

Impact of different wavelengths. Different wavelengths effectively have different pupils, resulting in different results of source-mask optimization.

Largely due to the slit shape,[104] and the presence of residual aberrations,[145] the effectiveness of SMO varies across slit position.[146] At each slit position, there are different aberrations[115] and different azimuthal angles of incidence leading to different shadowing.[28] Consequently, there could be uncorrected variations across slit for aberration-sensitive features, which may not be obviously seen with regular line-space patterns.[137] At each slit position, although optical proximity correction (OPC), including the assist features mentioned above, may also be applied to address the aberrations,[147][148] they also feedback into the illumination specification,[149][146][150][151] since the benefits differ for different illumination conditions.[147] This would necessitate the use of different source-mask combinations at each slit position, i.e., multiple mask exposures per layer.[115]

The above-mentioned chromatic aberrations, due to mask-induced apodization,[66] also lead to inconsistent source-mask optimizations for different wavelengths.

Degradation of image quality with pitch reduction under conventional EUV illumination. This has motivated greater consideration of other illumination modes such as QUASAR.

The optimum illumination, taking into account both exposure dose and focus windows, is a strong function of pitch in the range between 32 nm and 48 nm (relevant to 7 nm and 10 nm foundry nodes), which is where most of the work on EUV application has been focused. For pitches larger than 44 nm, the illumination pupil shape is preferably conventional, which is a circular disc, possibly including a central obscuration to provide an annular appearance.[128] For pitches in the range 44 nm down to 34 nm, the optimum shape is no longer conventional or annular but more shaped like the "quasar" (Quadrupole-shaped annular)[152] source, i.e., an arc within each quadrant of the pupil.[128] For pitches of 32 nm and below, the optimum illumination becomes more dipole like, i.e., concentrated toward the top and bottom or the left and right ends of the pupil.[108] When source-mask optimization is performed, the resulting shape will resemble the closest of the standard set (conventional, annular, quasar, dipole). For pitches less than 41 nm, the central portion of the pupil must be excluded for a tool with NA=0.33, as 13.5 nm EUV light going through that portion would only contribute the zeroth diffraction order (unscattered light), effectively adding flare.[153]

The best focus for a given feature size varies as a strong function of pitch, polarity, and orientation under a given illumination.[154] At 36 nm pitch, horizontal and vertical darkfield features have more than 30 nm difference of focus. The 34 nm pitch and 48 nm pitch features have the largest difference of best focus regardless of feature type. In the 48-64 nm pitch range, the best focus position shifts roughly linearly as a function of pitch, by as much as 10-20 nm.[155] For the 34-48 nm pitch range, the best focus position shifts roughly linearly in the opposite direction as a function of pitch. This can be correlated with the phase difference between the zero and first diffraction orders.[156] Assist features, if they can fit within the pitch, were found not to reduce this tendency much, for a range of intermediate pitches,[157] or even worsened it for the case of 18-27 nm and quasar illumination.[158] 50 nm contact holes on 100 nm and 150 pitches had best focus positions separated by roughly 25 nm; smaller features are expected to be worse.[159] Contact holes in the 48-100 nm pitch range showed a 37 nm best focus range.[160] The best focus position vs. pitch is also dependent on resist.[161] Critical layers often contain lines at one minimum pitch of one polarity, e.g., darkfield trenches, in one orientation, e.g., vertical, mixed with spaces of the other polarity of the other orientation. This often magnifies the best focus differences, and challenges the tip-to-tip and tip-to-line imaging.[162]

For the foundry "5nm" node, the minimum metal pitch for horizontal lines is taken to be around 32 nm,[163] for which dipole-like illumination is preferred, but the minimum metal pitch for vertical lines (parallel to the gates) is taken to be around 40 nm,[163] for which quasar-like illumination is preferred. Moreover, for the foundry "7nm" node, the minimum metal pitch for horizontal lines is taken to be around 40 nm,[163] for which quasar-like illumination is expected, while the minimum metal pitch for vertical lines can be taken to be around 50 nm,[163] for which conventional or annular illumination is preferred. For the quasar illumination, the best focus position varies strongly as a function of pitch, particularly 36-40 nm vs. 48-60 nm, as well as over the 48-72 nm range.[164] For these nodes, it is impossible to have a single EUV exposure illumination setting that fits both metal line directions at different respective pitches. Unidirectional metal layers are expected for these nodes anyway.[163] The line tip-to-tip gap in this case is expected to be kept small enough by using cut exposures in a multiple patterning scenario.[108]

Phase profile of attenuated phase shift mask for EUV. The phase profile (red) for an attenuated phase shift mask used with a partially transmitting EUV absorber does not match the ideal profile design (dotted), due to oblique incidence illumination and absorber edge scattering.

A commonly touted advantage of EUV has been the relative ease of lithography, as indicated by the ratio of feature size to the wavelength multiplied by the numerical aperture, also known as the k1 ratio. An 18 nm metal linewidth has a k1 of 0.44 for 13.5 nm wavelength, 0.33 NA, for example. For the k1 approaching 0.5, some weak resolution enhancement including attenuated phase shift masks has been used as essential to production with the ArF laser wavelength (193 nm),[165][166][167][168][169][170] whereas this resolution enhancement is not available for EUV.[171][172] In particular, 3D mask effects including scattering at the absorber edges distort the desired phase profile.[172] Also, the phase profile is effectively derived from the plane wave spectrum reflected from the multilayer through the absorber rather than the incident plane wave.[173] Without absorbers, near-field distortion also occurs at an etched multilayer sidewall due to the oblique incidence illumination;[174] some light traverses only a limited number of bilayers near the sidewall.[79] Additionally, the different polarizations (TE and TM) have different phase shifts.[79]

Low energy electron blur alters the image by reducing the local dose gradient.

Low energy electron migration. Low energy electron travel distance (r) can reach at least several nanometers, even though the distance between electron collisions (scattering) is 1 nm.

A study by the College of Nanoscale Science and Engineering (CNSE) presented at the 2013 EUVL Workshop indicated that, as a measure of EUV photoelectron and secondary electron blur, 50–100 eV electrons easily penetrated beyond 15 nm of resist thickness (PMMA or commercial resist), indicating more than 30 nm range of resist affected centered on the EUV point of absorption, for doses exceeding 200–300 uC/cm².[182] This can be compared with the image contrast degradation reported for sub-40 nm pitches later in 2015.[55]

The process of electron penetration through a resist is essentially a stochastic process; there is a finite probability that resist exposure by released electrons can occur quite far from the point of photon absorption.[183][184] Increasing the dose increases the number of far-reaching electrons, resulting in more extended resist loss. A leading EUV chemically amplified resist exposed to 80 eV electrons at a dose up to 80 uc/cm² showed up to 7.5 nm resist thickness loss.[185] For an open-source resist exposed near 200 uC/cm² by 80 eV electrons, the resist thickness lost after post-exposure bake and development was around 13 nm, while doubling the dose resulted in increasing the loss to 15 nm.[52] On the other hand, for doses >500 uC/cm², the resist begins to thicken due to crosslinking.[185]

The degree of photoelectron emission from the layer underlying the EUV photoresist has been shown to affect the depth of focus.[186] Unfortunately, hardmask layers tend to increase photoelectron emission, degrading the depth of focus.

Secondary electron blur vs. dose. The secondary electron blur can increase significantly with increasing dose.

Secondary electron spread vs. dose. The secondary electron radial spread from a point source increases as a function of dose.

Direct photocurrent measurements have been used to obtain secondary electron yields in response to EUV radiation. Simulations roughly calibrated to this yield show that the secondary electron blur increases with dose.[187] This is associated with fewer trapping sites as dose is increased and the reactions depleting the trapping sites proceed.

Low-energy electron-induced events also increase in number at a given distance from the photon absorption site, as dose is increased. This has been confirmed by resist thickness loss measurements as a function of low-energy electron dose.[188] The dose-dependent spread of secondary electrons was also known before from electron beam lithography.[189]

The increased secondary electron blur with increased dose makes control of stochastic defects more difficult.[190]

Due to the production of secondary electrons of various energies, the charge of the resist may locally fluctuate.[191] An EUV exposure with less blur leads to more pronounced charge differences at the feature edge,[192] which can lead to larger electric fields.[193] Such large electric fields have been observed to lead to dielectric breakdown.[191] The trapping of secondary electrons leads to a reduction of secondary electrons emitted from the film;[191] however, the trap sites may themselves be depleted, resulting in effectively extended secondary electron blur for larger doses.[187] Electron trapping is predicted to occur as part of polaronic behavior,[194][195] which confines the electron's final energy deposition to trap site locations. The polaron size can be quite large in resists, e.g., 46 nm in PMMA.[195]

EUV resists are also exposable by wavelengths longer than EUV, particular VUV and DUV wavelengths in the 150–250 nm range.[196]

Outgassing contamination vs. EUV dose: The increase of dose to size (E_size) to reduce shot noise and roughness comes at price of increased contamination from outgassing. The contamination thickness shown here is relative to a reference resist.

Due to the high efficiency of absorption of EUV by photoresists, heating and outgassing become primary concerns. Organic photoresists outgas hydrocarbons[197] while metal oxide photoresists outgas water and oxygen[198] and metal (in a hydrogen ambient); the last is uncleanable.[37] The carbon contamination is known to affect multilayer reflectivity[199] while the oxygen is particularly harmful for the ruthenium capping layers on the EUV multilayer optics.[200]

Atomic hydrogen in the tool chambers is used to clean tin and carbon which deposit on the EUV optical surfaces.[202] The reaction with tin in the light source or resist or on an optical surface to form volatile SnH4 proceeds via the reaction:

Sn(s) + 4H(g) => SnH4(g).[202]

The SnH₄ can reach the coatings of other EUV optical surfaces, where it redeposits Sn via the reaction:

SnH4 => Sn(s) + 2H2(g).[202]

Redeposition may also occur by other intermediate reactions.[203]

The redeposited Sn[36][37] might be subsequently removed by atomic hydrogen exposure. However, overall, the tin cleaning efficiency is less than 0.01%, due to both redeposition and hydrogen desorption, leading to formation of hydrogen molecules at the expense of atomic hydrogen.[202]

The removal of carbon proceeds by methane formation at lower temperatures or acetyelene formation at higher temperatures:[202]

H(ads) + C => CH(ads)
Methane formation:
CH(ads) + H(ads) => CH2(ads)
CH2(ads) + H(ads) => CH3(ads)
CH3(ads) + H(ads) => CH4(g)
Acetylene formation:
CH(ads) + CH(ads) => C2H2(g)

Atomic hydrogen is produced by EUV light directly photoionizing H₂:

hν + H2 => H+ + H + e−[204]

Electrons generated in the above reaction may also dissociate H₂ to form atomic hydrogen:

e− + H2 => H+ + H + 2e−[204]

Hydrogen-induced blistering defects. Atomic hydrogen (red dots) used for cleaning surfaces can penetrate underneath the surface. In the Mo/Si multilayers, H₂ (paired red dots) is formed and trapped, resulting in blister (white region) formation.

Hydrogen also reacts with metal-containing compounds to reduce them to metal,[205] and diffuses through the silicon[206] and molybdenum[207] in the multilayer, eventually causing blistering.[208][209] Capping layers that mitigate hydrogen-related damage often reduce reflectivity to well below 70%.[208] Capping layers are known to be permeable to ambient gases including oxygen[210] and hydrogen,[211][212][213][214] as well as susceptible to the hydrogen-induced blistering defects.[215] Hydrogen may also react with the capping layer, resulting in its removal.[216] Hydrogen also reacts with resists to etch[217][218] or decompose[219] them. Besides photoresist, hydrogen plasmas can also etch silicon, albeit very slowly.[220]

To help mitigate the above effects, the latest EUV tool introduced in 2017, the NXE:3400B, features a membrane that separates the wafer from the projection optics of the tool, protecting the latter from outgassing from the resist on the wafer.[39] The membrane contains layers which absorb DUV and IR radiation, and transmits 85-90% of the incident EUV radiation. There is of course, accumulated contamination from wafer outgassing as well as particles in general (although the latter are out of focus, they may still obstruct light).

Multiple EUV pulses at less than 10 mJ/cm² could accumulate damage to a Ru-capped Mo/Si multilayer mirror optic element.[227] The angle of incidence was 16° or 0.28 rads, which is within the range of angles for a 0.33 NA optical system.

Production EUV tools need a pellicle to protect the mask from contamination. Currently, the pellicle is not yet guaranteed to withstand 250 W power necessary for high volume manufacturing; the specification is 40 W.[228]

Pellicles are normally expected to protect the mask from particles during transport, entry into or exit from the exposure chamber, as well as the exposure itself. Without pellicles, particle adders would reduce yield, which has not been an issue for conventional optical lithography with 193 nm light and pellicles. However, for EUV, the feasibility of pellicle use is severely challenged, due to the required thinness of the shielding films to prevent excessive EUV absorption. Particle contamination would be prohibitive if pellicles were not stable above 200 W, i.e., the targeted power for manufacturing.[229]

Heating of the EUV mask pellicle (film temperature up to 750 K for 80 W incident power) is a significant concern, due to the resulting deformation and transmission decrease.[230] ASML developed a 70 nm thick polysilicon pellicle membrane, which allows EUV transmission of 82%; however, less than half of the membranes survived expected EUV power levels.[231] SiNx pellicle membranes also failed at 82 W equivalent EUV source power levels.[232] At target 250 W levels, the pellicle is expected to reach 686 degrees Celsius,[233] well over the melting point of aluminum. Alternative materials need to allow sufficient transmission as well as maintain mechanical and thermal stability. However, graphite, graphene or other carbon nanomaterials (nanosheets, nanotubes) are damaged by EUV due to the release of electrons[234] and also too easily etched in the hydrogen cleaning plasma expected to be deployed in EUV scanners.[235] Hydrogen plasmas can also etch silicon as well.[236][237] A coating helps improve hydrogen resistance, but this reduces transmission and/or emissivity, and may also affect mechanical stability (e.g., bulging).[238] The current lack of any suitable pellicle material, aggravated by the use of hydrogen plasma cleaning in the EUV scanner,[239][240] presents an obstacle to volume production.[241]

ASML, the sole EUV tool supplier, reported in June 2019 that pellicles required for critical layers still required improvements.[242]

In the absence of pellicles, EUV mask cleanliness would have to be checked before actual product wafers are exposed, using wafers specially prepared for defect inspection.[243] These wafers are inspected after printing for repeating defects indicating a dirty mask; if any are found, the mask must be cleaned and another set of inspection wafers are exposed, repeating the flow until the mask is clean. Any affected product wafers must be reworked.

As discussed above, with regard to contamination removal, hydrogen used in recent EUV systems can penetrate into the EUV mask layers. Once trapped, bulge defects were produced.[215] These are essentially the blister defects which arise after a sufficient number of EUV mask exposures in the hydrogen environment.

Reduction of field size by demagnification. Increasing the demagnification from 4X to 8X in one dimension would split the original full imaging field into two parts to preserve the same die area (26 mm × 33 mm).

Field stitching. Stitching together exposure fields is a concern where critical features cross a field boundary (red dotted line).

Preparation of an anamorphic lens with an NA between 0.5 and 0.6 is underway as of 2016. The demagnification will be 8X in one dimension and 4X in the other, and the angle of reflection will increase.[244]

Higher demagnification will increase the mask size or reduce the size of the printed field. Reduced field size would divide full-size chip patterns (normally taking up 26 mm × 33 mm) among two or more conventional 6-inch EUV masks. Large (approaching or exceeding 500 mm²) chips, typically used for GPUs[245] or servers,[246] would have to be stitched together from two or more sub-patterns from different masks.[247] Without field stitching, die size would be limited. With field stitching, features that cross field boundaries would have alignment errors, and the extra time required to change masks would reduce the throughput of the EUV system.[248]

Shot noise causing significant CD variations. A set of holes patterned at 64 nm pitch shows significant effects of shot noise with an EUV dose of just over 10 mJ/cm². A larger dose would result in the same range of variation over a larger sample size.

With the natural Poisson distribution due to the random arrival and absorption times of the photons,[249][250] there is an expected natural dose (photon number) variation of at least several percent 3 sigma, making the exposure process susceptible to stochastic variations. The dose variation leads to a variation of the feature edge position, effectively becoming a blur component. Unlike the hard resolution limit imposed by diffraction, shot noise imposes a softer limit, with the main guideline being the ITRS line width roughness (LWR) spec of 8% (3s) of linewidth.[251] Increasing the dose will reduce the shot noise,[252] but this also requires higher source power.

A 10 nm wide, 10 nm long assist feature region, at a target non-printing dose of 15 mJ/cm², with 10% absorption, is defined by just over 100 photons, which leads to a 6s noise of 59%, corresponding to a stochastic dose range of 6 to 24 mJ/cm², which could affect the printability.

A 2017 study by Intel showed that for semi-isolated vias (whose Airy disk can be approximated by a Gaussian), the sensitivity of CD to dose was particularly strong,[253] strong enough that a reduction of dose could nonlinearly lead to failure to print the via.

Via printing failure from noise-induced dose reduction. Shot noise-induced dose reduction could in extreme cases lead to via printing failure (CD->0).

Minimum dose to restrain shot noise for process variation tolerance areas:

+ A process variation tolerance area is the largest region over which process variation is allowed.

The two issues of shot noise and EUV-released electrons point out two constraining factors: 1) keeping dose high enough to reduce shot noise to tolerable levels, but also 2) avoiding too high a dose due to the increased contribution of EUV-released photoelectrons and secondary electrons to the resist exposure process, increasing the edge blur and thereby limiting the resolution. Aside from the resolution impact, higher dose also increases outgassing[254] and limits throughput, and crosslinking[255] occurs at very high dose levels. For chemically amplified resists, higher dose exposure also increases line edge roughness due to acid generator decomposition.[256]

As mentioned earlier, a more absorbing resist actually leads to less vertical dose uniformity. This also means shot noise is worse toward the bottom of a highly absorbing EUV resist layer.

Line end position affected by shot noise. The natural variation of EUV photon number can cause the position of the line end to shift.

Even with higher absorption, EUV has a larger shot noise concern than the ArF (193 nm) wavelength, mainly because it is applied to smaller dimensions and current dose targets are lower due to currently available source power levels.

As can be seen above, at the target incident dose levels, significantly fewer EUV photons are absorbed in EUV resists compared to ArF photons in ArF resists. Despite greater transparency of the resist, the incident photon flux is about 14 times larger (193/13.5) for the same energy dose per unit area. The resist thickness is limited by transparency as well as resist collapse[261] and resist strip[262] considerations.

The NXE:3400C was announced to be introduced in 2019, including features that focused on improving uptime significantly, such as a modular design for faster changing, continuous tin supply, and better collector degradation control.[300] However, aberration improvements have yet to be implemented, as aberrations have to be measured directly in-situ first.[301]

ASML plans to introduce an improved EUV system late 2021.[302] It will be called NXE:3600[303] and based on previous roadmaps it should improve throughput to over 185 wafers per hour and have an overlay of 1.1 nm.