A hybrid acrylate-epoxy monomer, (3,4-epoxycyclohexyl)methyl acrylate (ECA), was chosen for wavelength selective radical and cationic photocuring to create mechanically resolved multimaterial thermosets (Fig. 2A). Efficient Type I photoacid generation was optimized for epoxy curing with UV light exposure only, while a Type I radical photoinitiator was employed to induce acrylate curing upon exposure to either violet or UV light (optimization details provided subsequently). This cross-reactivity was viewed as favorable given that the persistent acrylate network would tie the two disparate domains together, strengthening the interfaces that are often points of failure in multimaterial structures.43 Additionally, dual curing of hybrid acrylate-epoxy resins allowed for the incorporation of acrylate diluents to tune resin viscosity, photocuring rate, and material properties of the cured parts without leading to high sol fractions. Acrylate diluents incorporated into the present resin were 2-hydroxyethyl acrylate (HEA) and tetra(ethylene glycol) diacrylate (TEGDA), as described in detail later.
Photosystem optimization to achieve spectral control within the constraints of our 3D printing system (≤ 15 mW/cm2 and ≤ 80 mW/cm2 from 365 nm and 405 nm LEDs, respectively) was accomplished using real-time Fourier transform infrared (RT-FTIR) spectroscopy to monitor monomer to polymer conversion (ρ) (Figs. S1 − S8). Seven industrial photoacid generators (triarylsulfonium and diphenyliodonium salts) were screened with UV (365 nm) light exposure (Scheme S2, Table S1), while phenylbis(2,4,6-trimethylbenzoyl)phosphine oxide (BAPO) was selected as the violet (405 nm) light reactive Type I radical photoinitiator (Fig. 2B). Photoacid generator testing was accomplished using 3,4-epoxycyclohexylmethyl 3,4-epoxycyclohexanecarboxylate (ECC) as a proxy monomer owing to the overlap in epoxy and acrylate IR absorption bands when using ECA (Fig. S1). Photoacid generators with ECC alone confirmed that the epoxide polymerization kinetics at ~ 10 mW/cm2 365 nm LED exposure was insufficient for printing (> 60 seconds to completion, Figs. S2 − S8). However, among those tested, bis[4-(diphenylsulfonio)phenyl]sulfide bis(hexafluoroantimonate) (THS) led to one of the fastest epoxide conversion rates and was thus used going forward. Quantifying the spectral overlap (Fig. 2B, shaded regions) between absorbance profiles of each compound and emission profiles of the 365 and 405 nm LEDs provided insight into their potential wavelength selective reactivity. The LED output profiles were collected after a 387 nm longpass dichroic beamsplitter (394 nm reflection cut-off and 375 nm transmission cut-on) to match the DLP setup described later, which reduces emission overlap to potentially improve spectral control (Figs. S9 − S12). Average molar absorptivity values across each complete emission profile for the 365 and 405 nm LEDs, respectively, were ~ 930 and 570 M− 1·cm− 1 for BAPO and ~ 150 and 12 M− 1·cm− 1 for THS (Figs. S13 − S17 and Table S2). Notably, the weak absorption of THS at 365 nm corresponded with the slow epoxide polymerizations observed (Fig. S8).
To accelerate the cationic polymerizations, we examined the use of photosensitizers exhibiting a stronger 365 nm absorption relative to THS. Specifically, 4-isopropylthioxanthone was considered first given its common utility as an efficient photosensitizer. While this accelerated the cationic polymerization rate, it was not selective due to an appreciable absorption of both the 365 and 405 nm LEDs (Fig. S8). Next, 3,6-dimethoxy-9H-thioxanthen-9-one (MeOTX)44 was used as a blue shifted photosensitizer to impart the desired selectivity, providing average molar absorptivity values of ~ 1920 M− 1·cm− 1 for the 365 nm LED, ~ 13× higher than THS, and 17 M− 1·cm− 1 for the 405 nm LED (Fig. 2B, Fig. S17 and Table S2). Furthermore, MeOTX increased the absorption contrast of UV vs. violet light relative to THS (contrast ≈ 113× for MeOTX vs. 12× for THS). Notably, we developed a scalable 2-step synthetic protocol to prepare MeOTX from commercial starting materials with an overall ~ 80% yield (Scheme S1 and Figs. S18 − 21), which facilitated systematic photocuring and 3D printing studies.
Initial resin optimization was accomplished by monitoring acrylate and epoxy conversion during UV and violet irradiation using RT-FTIR (Fig. 2C and Figs. S22 − S24 and Table S3). The resin providing an ideal balance of polymerization speed, photosystem component solubility, viscosity, and mechanical properties (described later) contained ECA (67.9 mol%), BAPO (0.5 mol%), THS (1 mol%), MeOTX (0.5 mol%), HEA (30 mol%), and TEGDA (0.1 mol%). Samples for RT-FTIR were placed between IR transparent plates (glass or salt) with a thickness of 50 µm to match printing conditions. The acrylate conversion was determined by tracking the disappearance of the sp2 C-H stretch at 3100 cm− 1.45–47 Upon exposure to either 365 or 405 nm LEDs (10 mW/cm2) acrylate rapidly polymerized (rate = 1.2 ± 0.1 M·s− 1), consistent with BAPOs non-selective absorption. Furthermore, a conversion of 50% (~ half max) was reached after ~ 2 seconds of irradiation. This was further supported using photorheology, which provided a time to gelation of ~ 2 seconds using either 365 or 405 nm LEDs (10 mW/cm2) (Fig. S25), indicating that these conditions were suitable for DLP 3D printing.
Epoxy polymerization was similarly tracked using the C–O–C overtone stretch at 3915 cm− 1 and confirmed with the C–O–C stretch at 909 cm− 1,48–52 which gave consistent results (Fig. 2D and Figs. S1-S8, S26 and Tables S1, S3, S4). In the presence of acrylate functionality, these epoxy signals were convoluted (Fig. S1), necessitating the use of a model acrylate-free resin to estimate the reactivity and selectivity of the present photosystem for epoxy polymerizations. Specifically, the proxy resin contained ECC (58.5 mol%), 3-ethyl-3-oxetanemethanol (OXA, 40 mol%), THS (1 mol%), and MeOTX (0.5 mol%). In analogy to HEA in the hybrid resin, OXA severed several roles, such as improving solubility of the photosystem components, lowering resin viscosity, and accelerating epoxy polymerization by providing additional hydroxyl initiation sites. Importantly, this photosystem was selective, initiating epoxy polymerizations upon exposure to the UV LED and not the violet one over the course of the experiment (~ one minute), each at 10 mW/cm2 (Fig. 2D). The low maximum epoxy conversion upon exposure to UV light was attributed to early onset vitrification/gelation, which may vary for the hybrid resin given the presence of acrylate diluents. Impressively, the combined incorporation of OXA and MeOTX resulted in a ~ 170× increase in epoxy polymerization rate upon exposure to UV light (365 nm, 10 mW/cm2). Specifically, the UV light-induced epoxy polymerization rate was 0.16 ± 0.16 M·s− 1, reaching 50% of maximum conversion in ~ 4 seconds, indicating that these conditions were suitable for DLP 3D printing.
The hybrid resins comprising ECA (varied mol%), HEA (varied mol%), TEGDA (0.1 mol%), BAPO (0.5 mol%), THS (1 mol%), and MeOTX (0.5 mol%) were tested with a multicolor DLP 3D printing system equipped with 365 and 405 nm projectors (Fig. 3A). Combining and aligning the two projections through a dichroic filter provided maximum intensities of ~ 15 mW/cm2 (UV) and ~ 80 mW/cm2 (violet) at the build plane (~ 44 × 25 mm) with a pixel resolution of ~ 25 µm. Given fast acrylate conversion rates relative to epoxy under equivalent lighting conditions, the 365 nm LED was used at full intensity (15 mW/cm2), while the 405 nm LED was operated at partial power (5 or 15 mW/cm2 as noted) for all prints. For a set layer thickness of 50 µm, lightly crosslinked acrylics could be 3D printed with 4 seconds of violet light exposure and heavily crosslinked acrylate-epoxy networks with 12 seconds of UV light exposure. These conditions fulfilled the first criterion, with a z-build speed of 0.25 mm/min (12 s/50 µm layer). Although printing was possible using a 405 nm LED intensity of 5 mW/cm2, the smallest lateral features were ~ 100 µm (≈ 4 pixels), while an intensity of 15 mW/cm2 provided features < 50 µm (~ 1–2 pixels, Fig. S27). Qualitatively, stiction forces were considerably higher when using the higher violet light intensity at equal exposure times. At light intensities of 15 mW/cm2, residual monomer and photosystem components could be efficiently removed by washing with acetone. This was particularly important for the violet light (405 nm) printed objects as it provided final parts that were true elastomers (criterion two), as opposed to organogels, which precluded toxicity concerns from leaching and had the added benefit of improved stability (described later).
Upon 3D printing dogbone specimens from the hybrid resins for mechanical testing (gauge length = 20 mm, width = 4 mm, thickness = 0.5 mm) a difference in color between those prepared with UV versus violet light was apparent. UV light printed samples were yellow in color, while those prepared with violet light were colorless (Fig. 3B, inset). This fortuitously provided visual contrast that facilitated further characterization of resolution (described later). Quantifying the photochromism using UV-vis absorption spectroscopy revealed a ~ 40% increase in total photon absorption after 12 seconds of UV irradiation (Fig. S28). While the exact origin of this change in absorption was not determined, its lack of occurrence during violet light exposure suggested that it derived from the photoacid generation process (Figs. S29, S30). An added potential benefit of this photo-opacifying effect is improved vertical resolution by reducing exposure of regions past the active layer being cured.
The mechanical disparity for UV vs. violet light cured objects was subsequently assessed via uniaxial tensile testing of 3D printed dogbone specimens (Fig. 3B, Fig. S31). Optimization was accomplished by varying both the molar ratio of ECA:HEA and light exposure time per 50 µm layer. Altering the ECA:HEA ratio at a constant UV (365 nm) or violet (405 nm) light dosage per layer revealed that ratios ≥ 3:7 had comparable mechanical performance when printed with UV light, with a stark difference in elastic modulus (ΔE) of ~ 4000× between samples printed with UV or violet light (criterion three, Fig. S31). Increasing the content of HEA beyond this ratio led to a precipitous drop in E for UV light printed samples, which was attributed to a decrease in crosslink density (Fig. S32 and Table S4). Although similar for ECA:HEA ratios ≥ 3:7, the optimal difference in mechanical properties occurred at a 7:3 ratio, which was hypothesized to arise from improved dissolution of MeOTX in the resin, leading to increased epoxy conversion. At this 7:3 (ECA:HEA) ratio, samples printed with UV light showed a large increase in stiffness between 5 and 10 seconds of exposure, followed by a plateau (Fig. S33 and Table S5). As a result, a UV light exposure time of 12 seconds/50 µm layer was selected for all future prints. This provided stiff and strong plastics with an average elastic modulus (E) of 1700 ± 70 MPa, maximum strength (σm) of 76 ± 16 MPa (criterion four), and a strain at break (εf) of 7 ± 1% (Fig. 3B, Fig. S31). The stiffness of green parts produced with this UV light exposure time proved comparable to samples that received additional UV exposure in a post-curing step (E = 2108 ± 43 MPa, Figs. S32, S34), suggesting that the present dosage provided strongly interconnected networks. In stark contrast, samples printed with violet light (then acetone washed and dried) were soft and elastic, irrespective of light intensity (5 or 15 mW/cm2) and exposure time (2–15 seconds, E ≈ 0.5-1 MPa, Fig. S33, and Table S5). Specifically, an exposure time of 4 seconds/50 µm layer was selected going forward, where samples printed with 15 mW/cm2 violet light provided an E = 0.54 ± 0.05 MPa, σm = 0.6 ± 0.1 MPa, and εf = 180 ± 30% (criterion five, Fig. 3B and Fig. S31, solid line). Furthermore, it was qualitatively noticed that soft samples effectively returned to their original shape post-deformation.
Elasticity of the soft sample (15 mW/cm2, acetone washed) was quantified using cyclic tensile testing. Specifically, samples were deformed to a strain of 100% and unloaded to zero stress repeatedly over 100 cycles, using a strain rate of 20%/minute (Fig. 3C). To remove the Mullins effect (plastic deformation) commonly present in untreated elastomers53, all samples were pre-strained over 3 loading and unloading cycles prior to further testing (Fig. S35). Impressively, samples 3D printed with violet light showed an elastic recovery > 99%, along with a hysteresis loss of ~ 3–4% (criterion six). For context, natural rubber, known for its excellent elasticity, was characterized under identical conditions and provided a similar elastic recovery > 99%, but it had a hysteresis loss that was > 2× larger (~ 7–13%) relative to the soft hybrid epoxy-acrylate samples. The reduced hysteresis loss of the present soft network may result from low friction of the bulky cyclohexylepoxide sidechains.
The stability of violet light printed samples was assessed next, given the presence of unreacted epoxy functionality with the potential for deleterious mechanical instability to occur upon aging (criterion seven). Under standard aging conditions (i.e., not accelerated), acetone washed samples left on the benchtop under ambient conditions for 8 weeks showed no significant change in mechanical performance (Fig. S36 and Table S6). Next, samples were tested under accelerated aging conditions with UV light exposure post 3D printing (dosage ≈ 2.7 J/cm2). Samples rinsed only with isopropyl alcohol to remove surface resin increased in stiffness by ~ 180× (E = 362 ± 46 MPa, Fig. S37 and Table S7). In contrast, samples washed with acetone showed no significant change in stiffness (E = 0.55 ± 0.06 MPa) post-UV exposure (Fig. 3B, dashed line, Fig. S31). This was hypothesized to arise from the effective removal of photoacid generator by washing with acetone. Additionally, post-UV exposure of acetone-washed samples resulted in values for σm (= 0.89 ± 0.05 MPa) and εf (= 300 ± 60%) that were consistently superior to the corresponding green (and acetone washed) parts. This may arise from the removal of small molecules that can act as plasticizers or a very minor amount of additional crosslinking upon UV-exposure that results in a modest improvement in mechanical performance without altering stiffness. Overall, the lack of significant aging under both standard and UV-accelerated conditions underscores the significance of acetone washing as a simple post-processing procedure.
The thermal stability of 3D printed parts produced from the hybrid resin was also characterized given the known susceptibility of strained cyclic ethers (e.g., epoxides) to react upon heating (Fig. 3D). Dynamic mechanical analysis (DMA) under uniaxial tension (0.1% strain, 1 Hz frequency) for samples printed with violet light revealed a glass transition temperature (Tg) near ambient (~ 0–50°C) followed by a rubbery storage modulus of ~ 0.5 MPa, in accord with macroscopic tensile data. In contrast, the rubbery modulus of UV light printed samples, without post-treatment, was around 55 MPa (Tg ≈ 120–140°C), which indicates a ~ 55× increase in crosslinking density relative to violet light prints. For soft (violet light cured) samples, the storage modulus remained relatively constant up to ~ 190°C, after which it began to increase due presumably to thermal activation and crosslinking of unreacted epoxy functionality. The epoxy activation was corroborated using differential scanning calorimetry (DSC) of liquid monomer, which revealed an exotherm at temperatures > 150°C, notably accelerated in the presence of THS (Fig. S38). Furthermore, a thermal soak at 100°C for 10 minutes showed no significant change in storage modulus, emphasizing the thermal stability of soft parts (Fig. S39). Overall, the hybrid epoxy-acrylate objects showed good thermal stability despite the presence of unreacted epoxy groups.
Resolution of multi-material prints was assessed both optically, using the difference in color between UV and violet light exposed regions, and mechanically using tensile testing and nanoindentation. Samples containing lines of alternating violet and UV exposure of equal area were 3D printed with spacings ranging from 5 mm to 0.1 mm (Fig. 4A). Light microscopy visually showed sharp features down to 0.25 mm, with apparent blurring of edges occurring for 0.1 mm samples. Applying uniaxial tension to samples with lines parallel (∥) or perpendicular (⊥) to the axis of strain was then used to determine macroscopic moduli and compare the values to an ideal composite spring model (Fig. 4B). With a constant 1:1 hard-to-soft ratio, tensile modulus should be independent of feature size, providing theoretical values of 850 (E∥) and 0.9 MPa (E⊥) for lines in parallel and series, respectively. Experimentally, prints with lines ≥ 1 mm were in good agreement with the idealized spring model (Fig. 4A equation and Fig. 4B dashed lines). For example, the parallel and series moduli of 1 mm lines were 956 ± 2 MPa and 1.6 ± 0.3 MPa, respectively. However, as spacings decreased in size from 1 to 0.1 mm, the series moduli values progressively increased, deviating from the model until 0.1 mm where no apparent difference in modulus between parallel and series was observed. This was hypothesized to arise from slight overcure of the UV-light irradiated regions, which becomes more significant as the soft-hard interfacial area increases.
To test this conjecture, stiffness was mapped across soft-hard interfaces using nanoindentation. For comparison, samples were printed with both lateral (within a layer; x,y) and vertical (between layers; z) interfaces. In all cases, the contact moduli increased by ~ 3 orders of magnitude across the interface, in accord with the ΔE determined using tensile testing. Within a layer (x,y), this transition (or gradient) spanned ~ 200 µm across the soft-hard interface (Fig. 4C, left), providing a slope of ~ 8,500 MPa/mm, comparable to recent reports of mechanical patterning in 2D.31,54,55 The transition region was postulated to arise from a combination of acid diffusion, light scattering, and/or exothermic reactions causing epoxy curing in regions adjacent to illuminated areas. A comparable ~ 200 µm gradient was observed for vertical interfaces where the hard (UV light cured) layers were printed first, followed by the soft (violet light cured) layers (Fig. 4C, center). In contrast, printing hard layers onto soft ones resulted in a broader ~ 250 µm gradient and a ~ 100 µm overcure of the hard material into the soft layers. This overcure was attributed to transmission of UV light through the layer being actively printed, causing unintentional exposure to previous layers cured with violet light, which in-turn induced epoxy crosslinking (Fig. 4C, right).
To emulate structures found in nature requires the ability to program the interfacial gradient (criterion eight) from sharp (~µm scale, as seen in knee entheses56) to shallow (~ cm scale, as seen in squid beaks57,58). To create shallower soft-hard gradients we developed a method to independently overlay greyscale UV and violet light projections (Fig. 4D and Fig. S40). In this manner, the dosage of UV and violet light was spatially controlled within each layer of a 3D printed bar (l×w×h dimensions = 30×20×5 mm3). In the center of the bar over a distance of ~ 23 mm, the violet light intensity was varied from 2.5 to 0 mW/cm2 (4 s/50 µm layer), and the UV light intensity from 0 to 7.5 mW/cm2 (12 s/50 µm layer) in the same direction. The light intensity was set to half that of prior values to avoid high stiction forces from the large area projection within each layer. Furthermore, the large area projection was anticipated to accelerate curing due to the exotherm of polymerization59 and provide an extent of cure necessary for the desired mechanical gradient. Nanoindentation across the surface of this sample provided contact moduli that again spanned ~ 3 orders of magnitude (~ 1 to 2000 MPa), however this time over ~ 20 mm. This provided a gradient of ~ 100 MPa/mm, nearly 2 orders of magnitude shallower than the prior sharp gradient. Thus, multicolor greyscale DLP 3D printing represents a nascent method to program mechanical gradients between disparate domains within multimaterial objects.
Three proof-of-concept demonstrations were accomplished to highlight the utility of multimaterial 3D printing in manufacturing bioinspired mechanical metamaterials (i.e., synthetic structures with unique bulk mechanical behavior) (Fig. 5). Specifically, brick-and-mortar architectures to tune tensile toughness (like nacre in shells), hard springs within soft cylinders to tune compressive damping (like spines in vertebrates), and hard “bones” connected by soft “ligaments” to provide smooth joints (like knees in humans). In the first example, a 3D printed structure comprising hard inclusions (“bricks”) from UV light exposure surrounded by a soft and stretchable matrix from violet light exposure resulted in a material with a tunable strain energy density (a.k.a., modulus of toughness) (Fig. 5A and Movie S1). Upon extension, defects in the matrix, such as cracks, were arrested upon encountering a “brick”. This resulted in stark strain-hardening followed by the formation of several new cracks within the soft material, which dissipated energy prior to macroscopic failure. This resulted in a composite structure that behaved like the soft component in its ability to deform elastically, yet like the hard component in its resistance to rupture, providing a strain energy density ~ 7× higher than the pure soft 3D printed analog. In a related effort to tune toughness, a macroscopic double network architecture was printed, where a rigid skeleton was embedded in a soft matrix (like rebar in concrete). In this example, the rigid skeleton contained thin regions that acted as sacrificial “bonds” to dissipate energy upon rupture without leading to macroscopic failure (Fig. 5A and Fig. S41). In both of these examples strain energy density could be tuned by the number, location, and geometry of hard inclusions within a given area of a soft matrix.
To showcase how compressive behavior could be tailored without altering overall geometry or creating voids, we designed a multimaterial structure that contained a hard concentric twisted coil (“spring”) of varying pitch (Fig. 5B and Movies S2-S6) and diameter (Fig. S42 − S44 and Movies S7-S9) within a soft cylinder. Additionally, hard disks were placed at the top and bottom of the soft cylinder to qualitatively showcase interfacial strength given a lack of delamination at high stress and strain. As controls, a hard-only spring and soft-only cylinder (with hard top and bottom disks) were also prepared. Compressing samples of varying spring pitch from 4 to 3 to 2 mm at a rate of 0.01 mm/s (= 0.1%/s) resulted in distinct mechanical responses, with stiffness values increasing by ~ 4×. Moreover, at an ultimate force of 50 N the global compressive strains were 28%, 18%, and 8% for pitches of 4, 3, and 2 mm, respectively. In contrast, the soft cylinder control (no spring) reached 44% strain at a force of 50 N, while a hard spring alone having a 3 mm pitch compressed to 65% strain upon applying only ~ 1–2 N of force. Thus, the multimaterial structures displayed a non-linear combination of properties, which showcases the rich landscape for materials design and testing that this platform offers. Reinforcing this point were an additional set of cylindrical structures having a hole in the center and a layered sandwich paneling architecture with alternating hard and soft disks (Fig. S45), reminiscent of the forewing in beetles.42 Altering the width of the washer-shaped soft rings led to a distinct change in the apparent stiffness by > 10× without needing to change the outer dimensions of the cylinder. These examples highlight how multimaterial fabrication can precisely tune bulk mechanical properties without altering surface geometry.
As a final demonstration, a detailed knee joint was 3D printed (Fig. 5C). Using a scaled down model of a human knee, a rigid femur, patella, and tibia were produced using UV light, while soft and stretchable connective tissue (e.g., tendons and ligaments) were simultaneously printed using violet light. The entire structure was ~ 46.5 mm tall and ~ 17.5 mm wide at the knee, with the smallest portions (ligaments/tendons) being ~ 0.6 mm in diameter. Despite the small soft features and their proximity to larger hard domains during the print, each feature remained intact and free at every point between their intended top and bottom junctions. The excellent print fidelity enabled ease of unidirectional bending upon applying a weak force by hand, followed by an elastic retraction upon removing the force (Fig. 5C, Figs. S46, S47, and Movie S10). Therefore, multimaterial 3D printing (Figs. S48-S50) using the present hybrid resins enabled the production of lifelike joints with smooth motion.