Full Wafer Scale Manufacturing of Directly Printed TiO₂ Metalenses at Visible Wavelengths with Outstanding Focusing Efficiencies

Abstract

Highly efficient metalens arrays designed for 550 nm are directly printed using UV‐assisted nanoimprint lithography (UV‐NIL) and a TiO₂ nanoparticle (NP)‐based ink on 8″ optical wafers with imprint times less than 5 min. Approximately one‐thousand 4‐mm metalenses are fabricated per wafer with uniform optical performance using a reusable PDMS‐based elastomeric stamp. The absolute and relative focusing efficiencies are as high as 81.2% and 90.4%, respectively, matching closely with the simulated maximum efficiencies of 83% and 91% achievable with the given master design, indicating that future improvements are possible, and efficiencies are not limited by materials or process. The imprinted metalenses are free from organics due to a post‐imprint calcination step and exhibit outstanding dimensional and optical stabilities. The highest efficiencies are attained using imprint formulations comprised of mixtures of 10 and 20 nm TiO₂ NPs, whose denser packing not only increases the refractive index (RI) of the calcined lenses up to 2.0 but also reduces the feature shrinkage relative to the master. 25 cycles of atomic layer deposition of TiO₂ following imprinting increase the RI up to 2.3 without changing dimensions by uniform gap filling between NPs. This work opens a path for true, full‐scale additive manufacturing of metaoptics.

The highest experimentally determined focusing efficiency is reported here by using the direct imprinting of all‐inorganic TiO₂ metalens arrays. Absolute efficiency greater than 80% and relative efficiency greater than 90% are achieved by optimization of all fabrication parameters controllable in the additive manufacturing process. The highest refractive index of 2.3 is demonstrated by a short post‐imprint ALD process.

1. Introduction

Metasurfaces comprised of sub‐wavelength features and sub‐micron thickness are capable of tailoring the properties of light and enabling flat, multi‐functional optical components.^{[ 1} , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ^{8 ]} Fabrication platforms for metalenses operating at infrared wavelengths are typically based on the etching of Si using subtractive methods, common to well‐established semiconductor process flows, but are tool, time, and cost‐intensive.^{[ 9} , ¹⁰ , ^{11 ]} High‐performance metalenses operating in the visible range, however, cannot be fabricated out of Si due to the material's inherent absorption loss. There are viable candidates of loss‐less dielectrics in the visible range including titanium dioxide (TiO₂),^{[ 12} , ¹³ , ¹⁴ , ^{15 ]} silicon nitride,^{[ 16} , ¹⁷ , ^{18 ]} and niobium oxide.^{[ 19 ]} Fabrication of metalenses using such materials has often relied on subtractive manufacturing. Moreover, unlike the well‐known Si‐based etching technology, it is more challenging to etch some of these candidates with high precision at high aspect ratios to produce high‐efficiency metalenses at visible wavelengths. Fabrication issues associated with the dry‐etching of these dielectrics include the inevitable tapering, undercutting, and roughening of the sidewalls at the sub‐wavelength scale. Such deviations from the design in the shape, geometry, and surface roughness of these meta‐atoms introduce unwanted phase and amplitude modulation, which in turn degrades the performance of metalenses significantly. The fabrication constraints become even more demanding at shorter visible wavelengths.

Recently, the Capasso group developed a novel fabrication method that avoided the dry‐etching to produce TiO₂‐based metalenses designed for visible wavelengths.^{[ 12} , ²⁰ , ^{21 ]} They used a process similar to lift‐off using electron beam lithography (EBL) and atomic layer deposition (ALD). The geometry of the sub‐wavelength structures was achieved in an EBL resist and conformal coating of ALD followed to fill the large nano‐feature gaps defined by the resist, leading to the final metalens structures with high precision and smooth sidewalls. Despite the successful prototype visible metalenses using TiO₂, this fabrication strategy still raises some concerns when it comes to mass production level due to the low EBL throughput and because ALD was used for the deposition of the entire active lens material, requiring thousands of ALD cycles to reach a sub‐micron thickness for every device. In terms of time and cost, ALD alone is not a viable stand‐alone solution for high‐volume manufacturing. As the demand for visible metalenses is growing in many applications including AR/VR,^{[ 22} , ^{23 ]} sensing,^{[ 24} , ^{25 ]} and imaging^{[ 26} , ^{27 ]} markets, a more cost‐effective manufacturing process is essential for fabricating metalenses and other optical components at scale.

Remarkable progress has recently been made by additive manufacturing in fabricating optical structures including photonic crystals and metalenses in the visible wavelengths using a TiO₂ nanoparticle (NP)‐based structure and UV nanoimprint lithography (UV‐NIL).^{[ 13} , ¹⁴ , ¹⁵ , ^{28 ]} It was demonstrated that the direct and single‐step UV curing of a TiO₂ NP‐based ink with a patterned PDMS‐based elastomeric stamp is a promising pathway for the high‐throughput and low‐cost manufacturing of metalenses with a cycle time as fast as 2 min per wafer and at least 10 imprints per stamp.^{[ 14 ]} The optical performance of the directly printed metalenses was excellent compared to the state‐of‐the‐art at that time, as indicated by the absolute focusing efficiency of the metalenses, centered at ≈50% at 543 nm. More recently, simple post‐imprint processes such as calcination followed by a few cycles of ALD were shown to further improve the absolute focusing efficiencies of the metalenses, up to 75%, but only a few lenses achieved that efficiency.^{[ 15 ]} The optical performance of metalenses fabricated in the cost‐effective, scalable additive manufacturing platform is improving rapidly and is now competitive with those fabricated using subtractive platforms.

The present study illustrates an unprecedented improvement in absolute focusing efficiencies up to 81.2% and relative efficiencies up to 90.4% using the NIL‐based direct imprinting method. Furthermore, we fabricate approximately one‐thousand 4‐mm‐sized metalenses on a full 8″ wafer with high uniformity across the wafer, which is different from other studies reporting efficiencies of a few lenses. That is, this study successfully demonstrates both the high performance and scalability of the process. More interestingly, the absolute and relative focusing efficiencies of 81.2% and 90.4% approach the same as the maximum efficiencies based on simulation (83% and 91%) with the given master mold design (numerical aperture ≈0.2 and design wavelength ≈550 nm), indicating that materials and processes for the NIL‐based direct imprinting method are not limiting further improvement. The lenses exhibit the highest experimentally determined efficiencies reported in the NIL‐based approaches and even exceed the focusing efficiency of most of the subtractively fabricated metalenses. There are studies that use NIL to imprint hard masks and create optical structures by etching, which still carry the aforementioned limitation of subtractive processing, but our approach is a direct, single‐step method. Moreover, other directly printed NIL structures are primarily derived from inorganic‐polymer composites but our final structures are fully inorganic. Our post‐imprint calcination removes all organics, including ligands used to stabilize the NP dispersion inks and any polymeric matrix used to bind NPs together in the imprinted structures. Removal of organics is essential for achieving optical and dimensional stability, especially for imprint materials containing TiO₂, which is an efficient photooxidation catalyst that promotes degradation and yellowing in polymer composites. The all‐inorganic structures achieved using this approach, therefore, offer the stability of optical components fabricated by etch‐based subtractive approaches, unlike those obtained from NIL‐based approaches that include a polymer‐based matrix in their metalens structures.^{[ 13} , ^{29 ]} In principle, all inorganic metalenses could be prepared by imprinting sol–gel formulations of metal oxide precursors directly followed by condensation and crystallization,^{[ 30} , ³¹ , ³² , ^{33 ]} however large contraction of the materials during imprinting and subsequent processing (greater than 50%), extended imprint times, and feature coarsening due to crystallization at high temperatures up to 800 °C constrain the ability to consistently achieve high aspect ratios, high fidelity reproduction of the nanostructures, and smooth sidewalls necessary for achieving high optical efficiencies.

Since meta‐optics is still an emerging technology, care must be taken when interpreting the results, especially efficiencies across multiple studies. Absolute efficiency (focused power/incident power) is a useful and reliable metric for the characterization of optical performance since it includes all losses and deviations taking place between incident power and focused power. Some studies, however, report only relative efficiencies (focused power/transmitted power) or only simulated efficiencies and sometimes arbitrarily define other measures of efficiencies that inflate performance.

In this study, the maximum focusing efficiencies were achieved by systematically optimizing every step of the NIL process. Adjustments in formulation especially by mixing two different‐sized TiO₂ NPs were essential for reaching more than 95% of the simulated efficiencies. Denser packing of the mixtures relative to that obtained using particles of a single size played a key role since the mixtures showed a synergistic effect in maximizing refractive index (RI) and minimizing feature shrinkage relative to the master during the imprinting process. Around 10 and 20 nm‐sized TiO₂ NPs were used and blended in a systematic approach to maximize the particle packing density and resultant optical density. The close relationship between the formulation of NP‐based inks and the optical performance of the printed metalenses from the inks was investigated.

2. Results and Discussion

2.1. A Synergistic Effect of Mixing Two Different‐Sized TiO₂ Particles on Optical Density

The efficiency of directly printed imprints based on NIL depends on the optimization of the RI of the imprint materials, the retention of feature fidelity during the imprint process, the maximum theoretical efficiency based on the design, and the agreement between the design specifications (dimensions and RI) and the actual dimensions and RI of the printed lens. While “master correction” can be used to offset feature shrinkages, the abilities to tune RI and control/minimize the offset in imprinted feature dimensions are essential to maximize the efficiency of additively manufactured optical components. Here we demonstrate that, in the case of TiO₂ NP‐based inks, controlling the size and distribution of NPs and their concentration enables fine‐tuning of material properties such as RI and feature shrinkages. In addition, post‐imprint processes such as calcination and ALD can be utilized for further adjustments.^{[ 15 ]} These modifications provide important handles of optimization that can minimize or obviate expensive, time‐consuming iteration cycles of mastering.

Figure 1a shows the variation of the RI in planar films prepared from TiO₂ NP‐based inks with various mixing ratios of two different‐sized particles (10 and 20 nm). The as‐deposited planar films composed of purely small and large NPs have RIs of 1.93 and 1.90, respectively. A binary mixture of the two results in the increase of the RI, yielding a maximum of 1.96 at the 4:6 weight ratio of 10 to 20 nm particles. This synergistic effect is attributed to a higher packing density of the film when two different‐sized particles are properly blended. For two sets of particles with an appropriate size difference, the smaller particles can fill gaps remaining between larger particles, improving the packing density of the film. In the case of monodispersed spherical particles, the maximum packing volume is always limited, but widening the distribution of particle sizes or using a mixture of different‐sized particles can bring about denser packing. This concept is well known and has been applied in many areas,^{[ 34} , ³⁵ , ^{36 ]} but to the best of our knowledge, this is the first application of it in the field of meta‐optics, especially for the direct‐NIL printed meta‐optics based on inorganic NPs. In Figure 1a, the trend of the RI change is retained after the calcination at 450 °C but the overall values are shifted to higher RIs due to the densification of the structure by removal of a small amount of organic residues between TiO₂ NPs. This observation would be different from other NP‐embedded resins where NPs are dispersed in a polymeric matrix, typical of most direct NIL approaches.^{[ 13} , ^{29 ]} In such cases, a significant RI drop is expected upon calcination because of the high porosity left by the burning off of organics. Unlike polymer/nanoparticle composites, our final structures include no organics, which is essential for high mechanical, optical, and thermal stability throughout extended service time term and/or under harsh conditions. The figure also shows the vertical film shrinkage during the calcination, whose trend is nearly opposite to that of RI, indicating the lowest film shrinkages at the 4:6 and 3:7 ratios. This implies that these specific ratios provide a more densely packed film prior to the calcination, reducing the degree of shrinkage during the calcination and more effectively retaining both the high density and structural stability.

Figure 1.

a) Refractive indices (RIs) at 543 nm of the TiO₂ nanoparticle‐based films before and after calcination and their % shrinkages during the calcination as a function of mixing ratios of the two different‐sized TiO₂ nanoparticles. b) Dependence of the RIs of the calcined films on the number of cycles of TiO₂ ALD at 250 °C for the 10:0 (purely small nanoparticles) and 0:10 ratios (purely large nanoparticles) and c) measurement of wavelength‐dependent RI before and after the porosity filling by ALD. d) Dependence of the porosity (%) of the calcined films on the number of ALD cycles comparing the 10 and 20 nm particles. e) Schematic description of the ALD backfill process.

Figure 1b shows that ALD backfilling of the calcined structures enables further tuning of RI. In our recent work, it was demonstrated that a few cycles of TiO₂ ALD onto and within the calcined films or imprints can uniformly backfill a few nanometer‐sized interstitial gaps between NPs.^{[ 15 ]} That is, this short ALD process results in an abrupt increase of RI by replacing air gaps (RI of 1.0) between NPs with high index ALD‐deposited TiO₂ layers (RI of 2.5–2.6) as schematically illustrated in Figure 1e.^{[ 37} , ^{38 ]} The calcined film prepared from the purely small NPs shows a continuous RI change from 2.0 to 2.15 through 10 cycles of ALD (Figure 1b) while the RI of the film from the purely large NPs changes from 1.89 to 2.27 over 25 cycles, which takes just a few minutes. The RI increases level off in both cases when all the interstitial gaps are filled and closed by ALD. The large NP‐based film initially has a lower RI, but the RI increases gradually over a wider range of ALD cycles, resulting in a higher RI at the end. This greater RI change of the larger NP‐based films would indicate more accessible porosity by ALD. The wavelength‐dependent RI values shown in Figure 1c confirm such changes at the testing wavelength of our metalens, 543 nm. The ability to tune RI provides a useful handle to optimize the performance of imprinted optics, including metalenses.

The porosity of the TiO₂ NP‐based films is quantitatively evaluated by ellipsometric porosimetry (EP) analysis using the adsorption–desorption behavior of water within the nanometer‐scale porosity.^{[ 39} , ⁴⁰ , ⁴¹ , ^{42 ]} Figure 1d compares the variations of porosity throughout the ALD backfill process of the films composed of purely small (10:0 ratio) and large NPs (0:10 ratio). The porosity gradually decreases in both cases as expected. Before ALD, the larger NP‐based film had a slightly higher porosity (≈32%) than the small NP‐based film (≈30%) but its rate of porosity reduction is much faster. The trend in porosity mirrors that observed in RI (Figure 1b). The complete ALD backfill process reduced the porosity to 12.3% and 5.3% for the small and large NP‐based films, respectively, which corresponds to the RIs of 2.15 and 2.27.

The maximum RI achievable by post‐process ALD is observed in films with mixed particles as shown in Figure 2a. Overall, RI increases as the ratio of the large particles increases, yielding the highest RI of 2.30 at the 2:8 mixing ratio of 10 to 20 nm particles. In the figure, it is interesting to see the changing tendency of ALD‐driven RI increments in terms of mixing ratio. The greatest RI change by ALD is observed at the 0:10 ratio but there is a critical point where the slope changes dramatically at around the 4:6 ratio. This critical ratio corresponds to the ratio yielding the maximum RI for the as‐deposited and calcined films (Figure 1a). This would suggest two different roles of adding large particles to small particles in the perspectives of packing density and ALD‐accessible porosity, as schematically illustrated in Figure 2c. In the first range (from 10:0 to 5:5 ratios), where smaller NPs are dominant, adding larger particles increases the overall packing density (based on the observation of RI increase in Figure 1a) but has a relatively smaller impact on ALD‐accessible porosity (based on the lower slope in Figure 2a), meaning that ALD‐accessible pores are predominantly limited by small gaps between smaller 10 nm particles. In the second regime (from 4:6 to 0:10 ratios) when large NPs become dominant, adding more large NPs seems to move away from the maximum packing efficiency, lowering the RI again (Figure 1a). This less perfect packing in return means that more ALD‐accessible, large pores are created by larger 20 nm particles, which is more beneficial from the perspective of the ALD‐driven RI increment in this second regime (Figure 2a).

Figure 2.

a) The effects of mixing the two different‐sized TiO₂ nanoparticles on the degree of the RI change by 25 cycles of ALD relative to the RIs of the calcined films given in Figure 1a. b) Pore size distribution graphs were obtained during the water adsorption onto the calcined films and their ALD‐backfilled films (25 cycles of ALD) comparing three different ratios of the small and large nanoparticles (10:0, 4:6, and 0:10). c) Schematic illustration comparing the effects of TiO₂ nanoparticle size and distribution on the particle packing density and ALD‐accessible pores between the particles.

Figure 2b shows the pore size distributions of calcined films and fully ALD‐backfilled films for three different mixing ratios (10:0, 4:6, and 0:10) analyzed by the EP method. Prior to ALD, the pore sizes are primarily centered ≈2–3 nm. After 25 cycles of TiO₂ ALD, all the data indicate a shift to smaller pore diameters. It is difficult to identify clear peaks in the range of pore sizes examined after ALD, but it is clear that the peak population is shifted toward 1.5 nm for the 10:0 and 4:6 ratios. The pore size reduction confirms that ALD is an efficient way to fill the interstitial gaps between TiO₂ NPs with high uniformity throughout the film. Prior to ALD, the maximum in the pore size distribution is observed at 2.7 and 3.0 nm for the 10 and 20 nm particles, respectively. Their 4:6 ratio blend gives a relatively broad peak over the range of 2.2–2.7 nm. This peak position of the blend is smaller than the peak positions of the other two, which is indicative of the denser packing of binary particle sizes resulting in a reduced pore size. One may observe a weak bimodal‐like distribution with the bimodal blend (4:6 ratio) in Figure 2b, unlike the sharp, single peak shown by the other two unimodal systems (10:0 and 0:10 ratios). This broad or bimodal‐like distribution would also suggest the presence of two different pore sizes when two particles are almost equally mixed (4:6 ratio) as illustrated in Figure 2c.^{[ 43} , ⁴⁴ , ^{45 ]} Figure 2b also shows that after the ALD backfill, the large NP‐based film seems to exhibit the most complete backfill of pores, which results from a combination of the larger porosity (30% vs 32% in Figure 1d) and the larger pores (2.7 nm vs 3.0 nm in Figure 2b). Although the differences in porosity and pore size may look somewhat small, this is a meaningful increase in the accessibility and diffusion of the ALD precursor, TDMAT, considering its kinetic diameter of ≈0.7 nm.^{[ 46} , ⁴⁷ , ^{48 ]} The ALD precursor would experience an increase in pore tortuosity more quickly in the presence of any small pores in its diffusion path, which is expected for the 10:0 and 4:6 ratios.

2.2. Effects of the Particle Packing Density on the Optical Performance of Metalenses

Metalens imprints are prepared by a direct, single‐step fabrication process using a patterned elastomeric stamp onto the spin‐coated film of a TiO₂ NP‐based ink followed by UV curing. The Si‐based master mold used for patterning the stamp is shown in Figure 3a and consists of more than a hundred 4‐mm‐sized metalenses. In the master, five different metalens designs are included to systematically bias the nanoposts’ width by ± 20 nm with the constant post height across the designs, but all are designed for focusing 550 nm light. The width variation is designed to elucidate the sensitivity of device performance to variation in post width, as a consequence of design and process variances. As shown in Figure 3b, a set of 30 lenses is the basic repeating unit of the reticle used for the master fabrication. The diameter of the circular nanoposts is biased from −20 to +20 nm in steps of 10 nm from the nominal design (+20, +10, 0, −10, −20 nm) with the constant height across the designs. In every repeating unit of the 30 lens array, each design has six identical lenses (6 × 5 = 30). More details are described in the Experimental Section.

Figure 3.

a) The Si‐based master mold for the metalens arrays consisting of 121 4 mm‐sized metalenses designed for focusing 550 nm light. b) The basic repeat unit of the master includes five different designs whose post diameters are uniformly biased from +20 nm to −20 nm in steps of 10 nm from the nominal design (+20, +10, 0, −10, and −20 nm) with the fixed posts’ height of 820 nm. c,d) SEM images of directly printed metalens replicated from the master mold. Dependence of the absolute focusing efficiencies e) on the mixing ratio of two different‐sized TiO₂ nanoparticles and f) on the design variations. The absolute focusing efficiency is defined by the ratio of the optical power focused by the lens to the incident power corrected for Fresnel reflectance loss at the backside of the substrate. g) Schematic illustration summarizing differences in dimensions and refractive indices between designed and fabricated structures.

Figure 3c,d show SEM images of a directly printed metalens where nanoposts with varying diameters and high aspect ratios are successfully replicated from the master mold with high feature fidelity. The optical performance of the fabricated metalens arrays is present in Figure 3e. It is revealed that the average and the maximum values of the absolute focusing efficiencies of the as‐imprinted metalenses are dependent on the mixing ratio of the small to the large TiO₂ particles. Four different ratios (10:0, 7:3, 4:6, and 0:10) are representatively investigated based on the planar film studies. 20 lenses (4 lenses for each design) are characterized for each ratio to get the average and standard deviation. In Figure 3e, the 4:6 weight ratio ink yields the highest focusing efficiencies. Figure 3f deconvolutes the total average of the 20 lenses into the average of each design. In terms of design variations, efficiency increases as the biased post diameter increases from −20 to +20 nm.

The reason that the highest focusing efficiencies are obtained at the 4:6 ratio (Figure 3e) and with the widest feature dimensions at the given formulation of the ink (+10 and +20 nm biased) (Figure 3f) primarily arise from dimensional mismatches between designed master and fabricated master. As compared in Figure 3g, the nanoposts in the fabricated master are shorter in height and much smaller in width than the parent design. The master was designed to have a uniform post height of 840 nm and a post width of 329 nm (the widest post in the 0 nm‐biased design) but in the fabricated master, the actual dimensions were determined to be 820 and 275 nm, respectively. Such deviations resulted in lower efficiencies when we first fabricated lenses with purely small NPs (10:0 ratio in Figure 2e). As explained above, one advantage of this NIL‐based process lies in that adjustments in formulation provide important handles for optimizing the efficiencies of printed optical elements. As an illustration, we blended two different‐sized particles to offset the fabrication errors in mastering. As shown by planar films in Figure 1, a mixture of the two particles not only led to a higher RI but also reduced shrinkage. These two factors played key roles in counterbalancing the shortages in dimensions, most efficiently at the 4:6 mixing ratio. Having a higher RI (n = 2.01 for the 4:6 ratio) than the design RI (n = 1.90) increases the “optical height” of the imprinted structures although their physical height is less than that of the design. In Figure S1 (Supporting Information), the dimensional shrinkages during the imprinting and the post‐imprint calcination processes relative to the master were determined using SEM images.

In NIL‐based printed optics, the presence of a residual film under the nanoposts and its thickness should be precisely controlled to maximize efficiencies. The optimization of the formulation was further conducted to reduce etalon absorption by the residual film. Instead of adding any post‐imprint etch process, a more cost‐effective approach is to modify the NP concentration in the formulation to control the thickness of a spin‐coated film. Optical simulations revealed that a residual layer thickness having an even multiple of wavelength/(4^*RI) produced the highest efficiency and offered comparable efficiency to a residual layer‐free structure. In the case of the 4:6 ratio blend, this number is calculated to be ≈69 nm, and the first even multiple is 138 nm. Figure 4a shows the relationship between the absolute focusing efficiencies and the residual layer thickness varied by TiO₂ NP concentrations in the ink. Figure 4b confirms that residual thicknesses can be properly adjusted while retaining high imprint fidelity. The 25 wt.% formulation yields the best performance with a residual layer thickness of 124 nm, which is close to the first even multiple (138 nm). The lower efficiencies at 20 and 32 wt.% result from the residual thicknesses close to the odd multiples (69 and 207 nm).

Figure 4.

a) Dependence of the residual layer thickness on the absolute focusing efficiencies of the as‐imprinted metalenses verified by b) SEM images. c) The change of the absolute focusing efficiencies throughout the post‐imprint processes including the calcination and 5 cycles of TiO₂ ALD. d) The relative (the ratio of the optical power focused by the lens to the power transmitted through the lens) and the absolute focusing efficiencies measured at different wavelengths (520, 543, and 560 nm) after 5 cycles of ALD and e) their focal spot images. f) Recent progress in efficiencies of TiO₂ metalenses at visible wavelengths using the NIL‐based direct printing. Yoon et al.^{[ 13 ]} used a TiO₂/polymer composite while our structures (Einck et al.,^{[ 14 ]} Jung et al.,^{[ 15 ]} and this paper) are all‐inorganic.

Figure 4c shows the effects of the post‐imprint processes on the focusing efficiencies of the metalenses prepared from the 25 wt.% ink (4:6 ratio) that is the best formulation optimized so far. After calcining the lenses, the efficiencies slightly increased by about 1%. The calcination is expected to make feature dimensions slightly smaller (Figure S2, Supporting Information) but increases RI (Figure 1a). This trend may indicate that the effect of the higher RI contributes more to the performance than that of the reduced feature dimensions. 5 cycles of ALD following the calcination decreases the efficiencies. This is because even a short ALD results in a significant increase of RI (Figure 2a), leading to a deviation from the design RI (≈1.90), which negatively impacts the performance of the device originally designed for the lower RI. In future master designs, a new RI limit of 2.3 can be applied. With this ALD‐backfilled metalens sample, we tested the lenses at three different wavelengths (520, 543, and 560 nm) and compared absolute (focused power/incident power) and relative (focused power/transmitted power) focusing efficiencies. Figure 4d shows the effects of varying wavelengths on absolute and relative efficiencies. Over the wavelength regime examined, the relative and absolute efficiencies are reduced only by 5–10%. The lenses still yield good performance at these other wavelengths, as is also confirmed by the focal spot images showing good symmetry and diffraction‐limited focusing (Figure 4e). The highest relative efficiency is 90.4% at 543 nm.

Figure 4f summarizes the recent progress in the efficiencies of metalenses fabricated by the NIL‐based direct imprinting method. Yoon et al. reported an absolute efficiency of 33% using a TiO₂/polymer nanocomposite. Our all‐inorganic TiO₂ metalenses have shown a steep increase in efficiencies over the last few years and the current results indicate a very small offset between the maximum theoretical efficiencies based on the dimensions and RI of the lenses and the actual realized efficiencies of the imprinted lenses. The near theoretical performance of the lenses suggests that this approach is not limited by materials and process in the NIL step and that further improvement can be realized by optimizing design, the production of the master, and master correction to enable printing of lenses that closely match the dimensions of the optimized designs.

Figure 5 demonstrates the uniform optical performance of the lenses across a large device area footprint (2″ × 2″) on a single wafer. 81 lenses were tested to show the scalability of the process with the best‐performing design. One of them (lens C7R9) exhibits an efficiency beyond 80% and many of the lenses are found to have efficiencies above 75%, mostly populated in the +10 and +20 nm designs as expected (Figure 3f). The C7R9 lens had a relative efficiency of 90.4%. It is also important to note in Figure 5b that the efficiency deviations are small within the same design (diameter bias). These are the highest experimentally determined efficiencies reported in the field of NIL‐based meta‐optics. There is a recent report that also achieved efficiency as high as 77% through patterning of sol–gel TiO₂ followed by high‐temperature crystallization. It was, however, revealed that obtaining high‐efficiency consistently and reliably remained challenging due to some difficulties of sol–gel approaches such as large feature contraction, particle coarsening at crystallization and annealing temperatures up to 800 °C, and rough sidewalls, all of which make it difficult to precisely control the quality of high aspect ratio meta‐atoms and to scale for manufacturing.^{[ 31 ]} In Table S4 (Supporting Information), our efficiency value is compared with other manufacturing approaches that reported efficiencies in a similar visible wavelength range. Figure 5d also verifies the uniformity of focal spot images across the large device area footprint, except for only two lenses (C1R1 and C9R1) close to the edge. The same 81 lenses were tested prior to calcination and after 5 cycles of ALD, respectively available in Tables S1 and S2 (Supporting Information). The efficiencies of the ALD‐deposited sample at different wavelengths are also available in Table S3 (Supporting Information). It should be pointed out that the highest absolute and relative efficiencies (81.2% and 90.4% of the C7R9 lens) closely match the simulated efficiencies (83% and 91%). Figure S2 (Supporting Information) shows the line cuts of the measured focal spot intensities of the C7R9 metalens and the on‐axis modulation transfer functions (MTF) with the corresponding curves for an ideal lens (Airy function), further demonstrating that the imprinted metalenses are nearly diffraction limited. Further advances in achieving higher NAs and higher efficiencies are possible through the redesign of metasurfaces. The redesign requires taller meta‐atom features, higher aspect ratios, and higher refractive indices. We are currently optimizing the fabrication of such lenses.

Figure 5.

a) The absolute focusing efficiencies of 81 lenses across 5 different designs prepared from the 25 wt.% ink are shown in Figure 5c and b) their average and standard deviation for each design in c) one calcined metalens imprint. d) Focal spot images in two diagonal directions of the 81 lens array showing the uniformity of the focusing performance of the metalens over a large device area footprint (2″ × 2″) and e) the magnified focal spot image of the best‐performing lens (C7R9) giving the relative efficiency of 90.4% and the absolute efficiency of 81.2%.

The optimized process was scalable up to the full 8″ wafer imprinting, which is the most common wafer size in commercial imprinting platforms available. Figure 6 shows that approximately one‐thousand 4‐mm‐sized metalenses can be fabricated on a single 8″ wafer. As demonstrated in Figure 6e, the maximum, average, and standard deviation of the absolute efficiency is almost the same as the coupon scale testing result, which truly demonstrates the scalability of the entire process on the full wafer scale manufacturing platform. This would be the very first demonstration of additive manufacturing of meta‐optics showing theory‐approaching high performance, full‐scale manufacturing, wafer‐scale uniformity, and organic‐free structure, all at the same time. Moreover, Myrias Optics, Inc. has recently demonstrated a successful transfer of the materials and processes to the EVG manufacturing platform as shown in Figure S3 (Supporting Information).

Figure 6.

Images of a) an 8″ master mold, b) an 8″ stamp, c) a cassette of 8″ calcined imprints on fused silica that were patterned manually, each of which contains approximately one‐thousand 4‐mm‐sized focusing metalenses per wafer. d) An SEM image is taken from a comparable imprint fabricated on a Si wafer and e) the absolute focusing efficiencies of 120 lenses across the 8″ imprint focusing on the two designs. The formulation used for this full wafer imprinting was solely composed of 10 nm particles without 20 nm particles, which turned out to closely match the designs due to a little oversized post in the master mold. The focal spot images, line‐cut intensities, and on‐axis MTF curves are available in Figure S2 (Supporting Information). The images of imprints manufactured on the EVG imprinting tool are available in Figure S3 (Supporting Information).

3. Conclusion

Direct imprint using NIL, an additive manufacturing approach, yields high‐performance metalenses with absolute focusing efficiencies of greater than 80%, which are close to theoretical efficiencies (83%) of the designs employed. Directly imprinted lenses are now comparable to or more efficient than the metalenses fabricated using subtractive manufacturing. The close agreement between observed and theoretical efficiency in this study suggests that the material and NIL process used are not limiting, and further improvements can be realized by optimizing the design and mastering process. Moreover, deviations resulting from mastering and process errors can be offset by tuning formulations and post‐treatments to adjust RI and feature dimensions in the imprinted lenses. RI as high as 2.3 can be achieved by using formulations containing two different‐sized TiO₂ NPs and a few cycles of the post‐imprint ALD, which means that higher NAs and even higher efficiencies are accessible in future master designs, broadening the design window. The cost‐effectiveness and scalability of this additive manufacturing method is more attractive for mass production of high‐efficiency optical devices, not limited to metalenses, but including waveguides, holographic elements, and diffractive optics, to meet the volumes required to match the fast‐growing demand of markets across widespread applications in high‐precision optics.

4. Experimental Section

Materials

All materials were purchased from Sigma–Aldrich and used as received without further purification unless specified. 50 wt.% titanium dioxide (TiO₂) NPs dispersed in propylene glycol monomethyl ether acetate (PGMEA) were purchased from Pixelligent Technologies, LLC and two different sizes of TiO₂ NPs (10  and 20 nm) were provided. A binary mixture of the two different‐sized particles was formulated by blending the two commercial products dispersed in the same solvent, PGMEA to a desired weight ratio.

Preparation of TiO₂ NP‐Based Inks

The as‐received 50 wt.% TiO₂ dispersions were diluted into a desired concentration ranging from 20 to 32 wt.% to target a certain thickness for a spin‐coated film and the two different dispersions containing TiO₂ NPs of different sizes were mixed into various weight ratios. A solvent mixture of ethyl lactate (EL) and 2‐(2‐butoxyethoxy)ethyl acetate (which contains a 0.15 wt.% of an anionic surfactant relative to EL) was used as a diluent and a silane coupling agent was then added to the diluted dispersion at the concentration of 15 wt.% relative to the total mass of two different‐sized TiO₂ NPs. After mixing all the components, the TiO₂ NP‐based ink was sonicated for at least 10 min and used for imprinting as it was. The best formulation used for the small coupon scale imprinting was the 4:6 ratio of small to large particles while that used for the full 8″ scale imprinting was solely composed of small particles.

Stamp Making

A two‐layered stamp composed of hard PDMS (h‐PDMS) and soft PDMS (s‐PDMS) was fabricated. To make the first h‐PDMS layer, 1.7 g of 7–8% vinyl methylsiloxane‐dimethylsiloxane (Gelest, Inc.), 5 µL of 2,4,6,8‐tetramethyl‐2,4,6,8 tetravinylcyclotetrasiloxane, 9 µL of platinum‐divinyltetramethyl disiloxane (Gelest, Inc.), and 3 g of toluene were vigorously hand‐mixed and then 0.5 mL of 25–30% methylhydrosiloxane‐dimethylsiloxane (Gelest, Inc.) was added and mixed again. The mixture was spin‐coated onto the Si‐based metalens master mold (3000 rpm, 40 s, 1500 rpm s⁻¹ ramp rate). The spin coating was done twice to ensure the full coverage of the high‐density metalens array without any curing step between the two spins. Then, the spin‐coated master was cured at 100 °C for 45 min in an oven. The master was removed from the oven and cooled to room temperature before depositing the second s‐PDMS layer, Sylgard 184 (Dow Corning). The base and the curing agent of Sylgard 184 were vigorously hand‐mixed at the weight ratio of 10 to 1 and degassed under vacuum prior to the lamination with the pre‐cured h‐PDMS layer. The second s‐PDMS layer was manually poured onto a 0.3 mm‐thickness glass backing (Howard glass) whose surface was pre‐treated with 1,1,1,3,3,3‐hexamethyldisilazane (HMDS, Acros Organics) to make a better stamp adhesion to the glass. The stamp thickness of the poured s‐PDMS layer was controlled to be ≈0.7 mm by placing four footers of that thickness at each corner of the rectangular‐shaped metalens master. The h‐PDMS coated Si master was slowly placed onto the drop‐cast Sylgard with the pattern facing down to the glass backing and the backside of the master was gently pressed until the desired thickness limited by the footers was obtained. The entire assembly on the glass backing was cured in a 75 °C oven overnight. Then, the master was demolded from the PDMS‐based stamp, and isopropyl alcohol (IPA) was utilized to help alleviate the release tension between the master mold and the stamp. Once the master was removed from the stamp, the stamp was rinsed immediately with IPA and water. The master was also rinsed with IPA and the whole stamp‐making process can be repeated.

Fabrication of TiO₂ NP‐Based Planar Films and Imprints

All the fabrication steps described here were done in general lab space, except for the post‐imprint ALD process carried out in a cleanroom. TiO₂ NP‐based planar films and imprints were fabricated on Si (University Wafers) or fused silica (Precision Micro‐Optics) substrates. The substrates were cleaned and activated by UV‐Ozone treatment (UVO 342 from Jelight Company Inc.) for 10 min immediately before being used for imprinting. The diluted TiO₂ NP‐based ink (20–32 wt.%) was deposited onto a substrate and spun at 2000 rpm for 30 s with a ramp rate of 1500 rpm s⁻¹. A stamp was placed down onto the spin‐coated film right after spinning. The assembly was transferred to the high‐intensity UV tool and cured. A custom‐made UV tool (Carpe Diem Technologies) was used and designed to generate a pulsed UV source at 365 nm. The UV curing conditions were as follows: 35 V, pulse on for 5 milliseconds, pulse off for 15 milliseconds, and 4000 repetitions of the on/off pulses. To make unpatterned planar films which were used for the film characterization, the films were pre‐baked at 100 °C for 30 s to remove some excess solvents prior to the transfer to the UV curing chamber. 2000 repetitions of the pulses were employed for the planar films. The overall curing time of the pulsed UV took 40 and 80 s for planar films and imprints, respectively. When the UV curing was done and the imprint was fully cured and solidified, the stamp can be removed from the imprint very slowly from one edge to the other edge in a single motion to avoid any re‐contact between the imprint and the stamp. The released stamp can be re‐used for the next imprinting runs without any cleaning process or any pause. Some planar films or imprints went through some post‐imprint processes including calcination and atomic layer deposition (ALD). The calcination was made at 450 °C for 2 h in an air atmosphere with a heating rate of 3.3 °C min⁻¹. All the ALD processes were made on the calcined samples.

Atomic Layer Deposition

The post‐imprint ALD process was done at the chamber temperature of 250 °C using a Cambridge NanoTech Savannah 90 Atomic Layer Deposition system. Tetrakis(dimethylamido)titanium (IV) (TDMAT) was used and heated to 75 °C for the deposition of TiO₂ onto the calcined TiO₂ NP‐based planar films or imprints. The repeating cycle of the ALD deposition was as follows: 1) the TDMAT pulse (0.1 s), 2) purge (5 s), 3) the H₂O pulse (0.015 s), and 4) purge (5 s). A slightly different procedure was adjusted for the patterned metalens structures having a higher surface area as well as high aspect ratio features. The precursor was allowed to diffuse into the patterns for an extended time by closing the stop valve connected to the vacuum line. The modified recipe was as follows: 1) the TDMAT pulse (0.1 s), 2) wait (1 s) with the stop valve closed, 3) purge (5 s) with the stop valve open, 4) the H₂O pulse (0.015 s), 5) wait (1 s) with the stop valve closed, and 6) purge (5 s) with the stop valve open.

Master Design and Fabrication

The metalens was designed to have circular nanoposts with varying widths across the lens ranging from 100 to 280 nm arranged on a triangular lattice (430 nm lattice constant) with a constant height of 700 nm. The metalens was designed to perform best on fused silica substrates with a refractive index (RI) of 1.46. A target RI of the imprinted metalens structure sitting on fused silica was 1.9. The metalens had a diameter of 4 mm and a focal length of 9.8 mm, that was, numerical aperture (NA) of 0.2, and the design wavelength ws 550 nm.

The basic repeating unit of the metalens master comprised five different columns and six identical rows and the diameter of nanoposts was uniformly biased from −20 to +20 nm in steps of 10 nm from the nominal design (0 nm) for the five different columns. This variation in nanoposts’ width was made to account for any dimensional mismatch between the original design of the lens and the imprinted metalens structures. The Si‐based master was fabricated by NYCreates (Albany, NY) using DUV photolithography and plasma etching. For the 8″ master mold fabrication, a silicon‐on‐insulator (SOI) substrate was used to have a relatively flat etched baseline utilizing the etch selectivity of the insulator layer as an etch stop. The etch depth was targeted to be 840 nm and the post width was to be 329 nm for the biggest post in the nominal design (the 0 nm‐biased) based on an estimation of 17% shrinkage in the post height (15% for lateral shrinkage) during the stamp making and imprinting step relative to the master's feature dimension. The feature dimensions of the fabricated master were measured using SEM and the nanoposts were found to have a height of 800–840 nm (840 nm target) and the biggest post width found in the nominal design was 275 nm (329 nm target).

The fabricated metalens master was further cleaned with a hot RCA‐1 procedure to ensure the complete removal of any residues. A cleaning mixture with the 5:1:1 volume ratios of deionized water (DI H₂O), ammonium hydroxide (28–30% NH₄OH), and hydrogen peroxide (30% H₂O₂) was used. DI H₂O and NH₄OH were first mixed and heated on a hotplate to a temperature of 70±5 °C and removed from the hotplate at that temperature range, and H₂O₂ was added. Once vigorous bubble formation was observed in 1–2 min, the master was soaked into the mixture for 15 min. The master was rinsed with overflowing water, washed with IPA, and dried with compressed nitrogen gas.

Master Fluorination

The surface of the fabricated Si master was activated by UV/Ozone for 30 min prior to a vapor phase fluorination. The master was placed in a vacuum desiccator with a vial containing a 0.1 mL of heptadecafluoro‐1,1,2,2‐tetrahydrodecyltrichlorosilane (Gelest, Inc.) and the desiccator was vacuumed for 30 s, sealed, and left overnight at room temperature. The fluorinated master was removed from the desiccator, rinsed with IPA and water, and dried with compressed nitrogen gas.

Characterization of Films and Imprints

The RI measurement of planar films on Si wafers was taken on a JA Woollam RC2 variable angle spectroscopic ellipsometer (VASE) at five different angles with an interval of 5° (45–65°) over the wavelength range from 450 to 1690 nm. The obtained data were fit to the Cauchy transparent thin film model available in the CompleteEase software. An advanced Cauchy model that accounted for any RI gradient in the film was also utilized to verify the uniformity of the RI across the film even after the post‐imprint ALD process. The RI reported in the present study was extracted at 543 nm unless specified which was the wavelength of the power source of the optical testing setup used for the characterization of metalens performance.

The SEM images of the master and imprints were taken with a FEI Magellan 400 FESEM and their feature dimensions were measured using ImageJ software.

Environmental Ellipsometric Porosimetry Analysis

The environmental ellipsometric porosimetry (EP) investigation was carried out using a spectroscopic ellipsometry system (Woollam M‐2000 V) combined with an atmospheric control chamber with water as the adsorbate. The volume of adsorbed (and capillary condensed) water into the pores of the TiO₂ NP‐based films was measured through a time‐resolved measurement of the RI variation as a function of P/P₀. The RI was calculated using the Cauchy transparent model, while the volume of adsorbed water (porosity) was then deduced from the RI using the Bruggeman effective medium approximation. The pore size distribution was obtained by using the Kelvin equation and assuming cylindrical pores.

Optical Testing of Metalenses

The focusing efficiencies and focal spot images of the printed metalenses were taken using a Uniphase JDSU 1676 HeNe laser source with an unpolarized collimated beam operating at the wavelength of 543 nm.^{[ 14} , ^{15 ]} A laser beam expander (Melles Griot 10X) was used to provide a large incident beam with a planar phase front and an aperture with a diameter of 3.8 mm was placed between the beam expander and a lens. Focal spot images were captured by a CCD with a 50x objective lens and a tube lens (focal length: 20 cm).

Both absolute and relative focusing efficiencies were measured. The absolute efficiency of metalens was defined by the ratio (P_f/P₀) of the optical power focused by the metalens (P_f) to the optical power incident on it (P₀). The powers of the incident light and the focused light were measured by a power meter (ILX Lightwave OMM‐6810B with an OMH‐6722 power head). For measuring the focused power (P_f), a 40‐µm‐diameter pinhole was placed at the focal plane and the power passing through the pinhole was measured.^{[ 17 ]} The collected incident power was corrected for Fresnel reflection losses at the backside of the substrate. The loss expected at the air‐substrate interface was 3.5% considering the RI contrast of air (n = 1.0) and fused silica substrate (n = 1.46). The relative efficiency was defined by the ratio (P_f/P_t) of the same optical power focused by the metalens (P_f) to the optical power transmitted through the metalens device (P_t). The transmitted power was measured by placing the power meter as close as possible to the surface of the metalens at least less than a distance of 1–2 mm to collect all the transmitted light. There was some light diffracted at extremely large angles that cannot be collected by this method, which would have a negligible impact on the reported relative efficiencies.

The full 8″ wafer imprint was characterized in the same way but after dicing the imprint into a coupon size. Resist S1813 was coated onto the calcined imprint by spin coating and baking at 115 °C for 1 min, followed by mechanical dicing and resist removal process.

Metalens Simulations

The focusing efficiencies of the metalenses were estimated using the grating averaging technique.^{[ 49 ]} The highest efficiency was found to be 83% with the given target RI of 1.9 and the post height of 700 nm and post diameters ranging from 100 to 280 nm. The simulated absolute/relative efficiencies at a few different wavelengths (543, 550, 560) were 91%/83%, 90%/83%, and 88%/82.5%, respectively.

Conflict of Interest

The authors declare the following competing financial interest(s): D.E.J, V.J.E, A.A., and J.J.W. have a financial interest in Myrias Optics, Inc., which has licensed technology associated with imprinted optical components.

Supporting information

Supporting Information

Jung D. E., Einck V. J., Dawicki A., Malgras V., Verrastro L. D., Grosso D., Arbabi A., Watkins J. J., Full Wafer Scale Manufacturing of Directly Printed TiO₂ Metalenses at Visible Wavelengths with Outstanding Focusing Efficiencies. Adv. Mater. 2025, 37, 2500327. 10.1002/adma.202500327

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.