Direct printing of functional 3D objects using polymerization-induced phase separation

3D printing has enabled materials, geometries and functional properties to be combined in unique ways otherwise unattainable via traditional manufacturing techniques, yet its adoption as a mainstream manufacturing platform for functional objects is hindered by the physical challenges in printing multiple materials. Vat polymerization offers a polymer chemistry-based approach to generating smart objects, in which phase separation is used to control the spatial positioning of materials and thus at once, achieve desirable morphological and functional properties of final 3D printed objects. This study demonstrates how the spatial distribution of different material phases can be modulated by controlling the kinetics of gelation, cross-linking density and material diffusivity through the judicious selection of photoresin components. A continuum of morphologies, ranging from functional coatings, gradients and composites are generated, enabling the fabrication of 3D piezoresistive sensors, 5G antennas and antimicrobial objects and thus illustrating a promising way forward in the integration of dissimilar materials in 3D printing of smart or functional parts.


3
Strategies to seamlessly integrate multiple materials into objects using 3D printing will enable the generation of new or improved material properties and advance 3D printing as a mainstream approach to manufacture functional and smart objects. [1][2][3][4][5][6][7][8][9][10][11][12][13][14] Using reactive precursors, vat polymerization 3D printing provides a unique opportunity to spatially control materials from the surface to deep within the object. 7,8,12,[14][15][16][17] For instance, the spatial, temporal, chromatic and intensity characteristics of light have been used in vat polymerization to pattern material properties. Elegant examples include 3D printing using two wavelengths and orthogonal chemistries to spatially control polymerization, 13 light intensity and oxygen inhibition to modulate the crosslinking density 17 and photochromic molecules in combination with two wavelengths to control polymerization resulting in bioinspired materials of soft and hard sections. 18 Moore and Barbera have recently demonstrated the demixing of precursors to yield bicontinuous phases of polymer and pre-ceramic precursor with domain sizes controlled by light intensity. 12 Light-based printing techniques have also been used to photo-reduce in situ silver precursors yielding silver nanoparticles during the printing process. [19][20][21][22][23] Here, we demonstrate that by using purposely formulated resins, material phases can be controlled within objects using vat polymerization. The method utilizes polymerization-induced phase separation (PIPS), a process previously used to generate 2D patterns in holographic polymerization. [24][25][26][27][28][29] Exploiting concomitant changes in the thermodynamics of mixing that occur during polymerization, as well as spatio-temporal variations in monomer to polymer conversion, materials can be spatially directed towards the surface of the 3D printed object. The flux of functional material towards the surface of the printed object is controlled by balancing the kinetics of gelation, crosslinking density and rates of diffusion of the resin components. This approach has the benefit of generating material domains on the nanoscale and, thus, provides a 4 means to combine macro-and micron-scale 3D designs with nanoscale material phases, features not easily achieved with nanoscale printing approaches such as two photon polymerization, 30 localized electroplating 31,32 or metal ion reduction. [33][34][35] This report is the first to explore how resin formulation influences PIPS in vat polymerization (3D PIPS) and provides the insight needed to control material placement in printed objects. The use of 3D PIPS to spatially control material phases within printed objects opens up new opportunities to create functional coatings directly from printing or to generate material gradients that are essential to reduce stresses of integrating dissimilar materials. 36 We demonstrate the utility of the approach by producing conductive metallic silver features, enabling the fabrication of a dipole antenna array, strain sensors as well as antibacterial surfaces. Using the principles described herein, this new freedom to design material complexity directly into 3D printed objects can be envisioned for generating complexes surfaces for 3D catalysis, improving the wettability of biocompatible resins with hydroxyapatite particles, or embedding anti-viral agents to minimize the transmission of pathogenic viruses in medical tools and devices and will pave the way to new technologies in structural electronics, 37 shape responsive parts for soft robotics, as well as smart objects with embedded sensors for Internet of Things and wearables. 5

Crosslinkers drive the spatial distribution of silver
Here, we showcase a range of material morphologies that can be generated using photoresins containing a silver precursor as the non-polymerizable functional component ( Figure   1). The silver precursor, silver neodecanoate (AgND), is ideal for this application as its molecular nature ensures higher diffusivity than larger functional materials such as nano-or microparticles. Furthermore, because AgND does not scatter light as particles do, resins containing high concentrations of the complex can be printed yielding highly conductive, 5 metallic silver surfaces through a simple post printing sintering step. 38  By adjusting the resin composition, it is possible to tune the morphology of the printed part from one where silver is concentrated at the surface forming a distinct coating to one in which the silver is dispersed throughout the object. A high concentration of silver at the surface of the printed object necessitates that the AgND migrates to the surface before becoming entrapped in the polymer network; this occurs when the kinetics of gelation are slow, and when the diffusion of AgND is not inhibited by the formation of a tight polymer network. A composite morphology, where the concentration of silver varies minimally throughout the 3D printed object occurs when the kinetics of gelation are fast and the AgND is impeded from migrating as a result of a tight polymer network. By dialing-in conditions with intermediate rates of gelation and crosslinking densities, gradients in silver concentration can be achieved as the AgND diffuses controllably away from the locus of polymerization and towards the surface of the object.
3D PIPS was first demonstrated by printing cylinders 1.5 mm in diameter and 2 cm in length and sintered at 210 °C to convert AgND into metallic silver. The morphologies formed by the various resins can be seen in the SEM images taken at the edge of the cross-sections of the cylinders (Figure 2a). The images reveal that silver accumulates towards the surface of the object with some resins forming a defined silver layer or coating (e.g. 25 wt % DA-170) while others produce a graded composition in silver (e.g. 99 wt % DA-700). These morphologies were 6 assessed by performing 15 m line scans at the edge of the cross-sections of the cylinders using Energy Dispersive X-ray Spectroscopy (EDS, Figure 2b). With the exception of two of the DA-170 resins (50 wt % and 99 wt % crosslinker), all cross-sections show that the concentration of silver increases from the core to the surface of the cylinder (Supplementary Figure 1). It should be noted that SEM images of the interface of un-sintered cylinders showed silver present at the surface demonstrating phase separation occurs prior to the sintering step (Supplementary Figure   2 and 3). All four series have similar behaviors; the lower the crosslinker concentration, the more the silver concentrates at the surface to form a silver coating. However, resins made with diacrylates with long PEG segments show more of a graded distribution in the silver at the surface. To more easily compare among resin systems, the surfaces of the cylinders were analyzed by EDS to give the wt % Ag within the first ~2 m of the object as shown in Figure 2c.
These results illustrate that for all resin systems, the amount of Ag that accumulates at the surface decreases with increasing concentration of crosslinker, in agreement with the analysis of the cross-sections of the cylinders. These results also show that the short diacrylates yield structures with a broader concentration range of surface silver than resins made with the long diacrylates. For instance, the Ag surface concentration varies from 88 % to 18 % for the DA-170 system, but only 86 % to 40 % for the DA-700 system when the crosslinker concentration increases from 25 to 99 wt %. These results demonstrate that the spatial distribution of silver in the printed object is dictated by the length of the diacrylate crosslinker and its concentration.
In most printed samples, the silver concentration at the surface is sufficient to form a conductive film once sintered. The electrical resistances of the cylinder as a function of wt % crosslinker of the resin (Figure 2d) shows that for all systems, the resistances increase with increasing fraction of crosslinker, in agreement with Figure 2c showing decreasing surface silver 7 with increasing wt % crosslinker. When the crosslinker concentrations are low, silver forms a coating with low resistance, owing to the high concentration of silver at the surface. As the crosslinker fraction increases, the coating progressively contains less silver, thus, increasing its electrical resistance. Above a certain fraction of crosslinker, the surface silver is below its percolation threshold resulting in no detectable electrical conductivity. The relative change in the resistance with increasing crosslinker lengths also agrees with the trend in surface silver; the resistance of cylinders made with short diacrylates increases more dramatically than with the longer diacrylates in concurrence with the more significant decrease in surface silver for the short diacrylates. These results demonstrate that 3D PIPS is a simple, single-step method to generate functional coatings on 3D objects and, thus, circumvents the disadvantages of two-step coating methodologies such as poor film adhesion and uniformity (see Supplementary Figure 4 for comparison).

Gelation rate, crosslinking density and diffusivity
With the aim to resolve differences in the behaviours of the various resin systems and to develop a predictive model for 3D PIPS, we examined how the crosslinker influences the diffusion of the AgND during phase separation. Diffusion of phase separating components, such as AgND, is influenced by the rate a homogeneous resin mixture is transformed into an insoluble gel. This rate determines whether the AgND becomes trapped by the network or diffuses freely towards the unreacted resin where mixing is more favourable due to entropic gains. We seconds. Moreover, for a given wt % crosslinker, the delay times decrease with increasing molecular weight (MW) of the crosslinker. This behavior likely results from the free end of long crosslinkers extending further from the polymer backbone, thus, increasing the probability of finding an unreacted acrylate group. 39 We note that experiments performed using photoresins with AgND show similar trends in td as a function of wt % of crosslinker (see Supplementary Figure 5). Figure 3c shows how the delay times correlate to wt % surface Ag; the longer the delay time, the greater the amount of surface silver. Therefore, resins that remain homogeneous mixtures of polymer, monomer and crosslinker for a longer duration afford more time of unimpeded migration for the AgND to reach the surface of the object.
Although Figure 3c highlights how the delay time affects surface morphology, the results reveal that the resin systems generate different amounts of surface Ag for a given delay time, indicating that the amount of Ag that reaches the surface is not solely dictated by gelation rates.
The effect is particularly pronounced at low delay times (i.e. high crosslinker concentrations) where resins made with long diacrylates yield objects with higher concentrations of surface silver than shorter diacrylates. We considered the role of miscibility between the AgND and the resin by comparing their calculated solubility parameters, δ (see Supplementary Table 1 and Supplementary Figure 6). However, for a given wt % crosslinker, the differences in solubility parameters for the different systems is marginal and do not explain the behavior highlighted in Figure 3c.
The diffusivity of AgND will impact the amount of Ag that accumulates at the surface and may explain the observed differences in surface silver for a given delay time. The diffusivity 9 of AgND will change during polymerization as a result of increases in viscosity and constraints imparted by the growing polymer network. 40 The extent to which the diffusivity of AgND will change when the resin is transformed into a polymer network will be highly dependent on the length of the spacer between reactive moieties in the crosslinker. To explore this idea, coarse- The inset to Figure 4a shows the diffusion coefficient of the probe molecule in the polymer networks as a function of crosslinker length. Of note, the diffusion coefficient for the probe molecule is larger in the L=9 network than the L=3 network outside of the error bars. This inversion could be heightened in the experiments due to the ratio between long and short crosslinkers being higher (see Methods). However, the simulations do clearly indicate that the diffusivity of AgND is very likely to be lower in the polymer network made with a tight crosslinked network. Reduced diffusivity in the tight network formed by short crosslinkers will impede the ability of AgND to migrate to the surface and, thus, provide a rationale for the lower surface Ag found in the system with shorter crosslinkers at low delay times. These results show how the interplay between the rate of network formation and the temporal changes in diffusivity of AgND that occur during polymerization affect the extent to which AgND migrates to the surface, which ultimately dictate the spatial distribution of silver in the object.

3D PIPS for smart objects
Using purposefully formulated resins to control the placement of AgND, we demonstrate the value in being able to tune the surface morphology of printed objects with particular material properties by considering three applications: strain sensors, antennas and antimicrobial objects.
The 3D PIPS approach provides the ability to generate strain sensors with complicated 3D geometry and holds promise in wearable electronics and motion sensing. 41 Truss structures were 3D printed to using a resin formulation that yields graded silver compositions. As these formulations yield thermosets with low stiffness, it is possible to compress the objects and modify the electrical resistances based on the applied compression. The SEM images of the surfaces and cross-sections of strain sensors in Figure 5a reveal that as the wt % crosslinker used in the resin increases, Ag particles forming the surface coating become progressively sparser 11 with more polymer inclusions in between the silver particles. The differences in morphologies lead to different electrical responses under applied compression (Figure 5c). For the truss made with 39 wt % crosslinker, the surface features a dense film of Ag nanoparticles with low electrical resistance whereas the truss made with 58 wt % crosslinker has a surface morphology with sparser particles and, correspondingly, higher electrical resistance (Supplementary Figure   7). As shown in Figure 5b, compression increases the number of particle-to-particle contacts, consequently increasing the pathway for electrical conduction; this is proportionally larger for the truss that has sparser surface Ag than the truss that has a dense film of Ag nanoparticles. 3D printing is ideally suited to fabricate millimeter wave antennas for 5G as 5G will function on small networking cells that use arrays of antennas in small geographical areas requiring a large number of integrated low lost devices. These requirements can be achieved by using 3D printing to make antennas low cost, in arrays and embedded in objects. Moreover, by suspending the antenna in air using a 3D design, signal loss can be minimized with air becoming the effective dielectric. Using the 3D PIPS approach, we fabricated an array of 3D printed dipole antennas and demonstrated transmission of 2.4 GHz waves. The dipole antenna array, shown in Figure 6a, displays the radiation pattern found in Figure 6c as measured using an anechoic chamber ( Figure 6b) and its comparison with the theoretical response for a dipole array on a ground plane. The focusing of the radiation pattern into a main lobe is the result of radiation interference between antenna elements. The half power beam width of the theoretical pattern is 12 48 ∘ compared to 45 ∘ for the measured pattern, resulting in a remarkably small difference of 3 and, thus, demonstrating the suitability of this printing process for antenna applications.
Antibacterial properties of nanoparticle silver have been used in many medical and dental applications for the prevention of infection. [42][43][44] To evaluate the antibacterial behaviors of 3D printed Ag objects, a halo inhibition zone test against E.coli as well as bacterial growth kinetics were carried out along with control objects containing no silver. A concentration of 0.5 or 1.0 wt % Ag was used in this study in order to form Ag nanoparticles rather than a film on the surface In summary, we have showcased how the temporal and spatial variation in monomer-topolymer conversion that takes place in vat 3D polymerization causes local demixing of functional materials, triggering diffusion of these materials towards the bulk resin. By harnessing the rate at which the functional materials become entrapped in the polymer network during 3D PIPS, a wide range of surface morphologies can be accessed. The insight gained in controlling 13 the material phases allows a rationalized approach to formulating resins to access a wide range of material morphologies for specific applications. Due to the universality of this approach, 3D PIPS represents a powerful method to create materials with controlled sub-phases and will accelerate the adoption of vat polymerization as a viable technique to generate functional 3D objects. Refractive index changes can be best visualized using phase-contrast imaging mode. Videos were acquired by a color camera (Luminera Infinity2) and include an initial phase without laser illumination, followed by removal of a laser shutter where resin/crosslinker exposure is initiated.

3D Printing Functional Photo-Resin Preparation: Photoresin Preparation
A small dot is observed after a few seconds which subsequently grows to an island several where td is the delay time, tc is a rate dependent parameter, and Df is a size dependent parameter.
The td, which represents the time elapsed between when the laser is turned on and the first observable sign of polymer network, serves as a measure of relative gelation time. This was repeated two more times with newly filled capillary tubes for each formulation measured.
Coarse-grain modeling of the diffusivity of a probe molecule: The simulations used standard coarse-grained (CG) polymer methodologies 46 to construct a system roughly modeled on the experimental setup. A cubic box was filled with CG polymers that represent the crosslinkers.
Each polymer consists of L beads that are linearly joined together via FENE spring bonds 47 to prevent bond crossing. Inter-and intramolecular interactions between beads were implemented using the WCA potential 48 such that there is no attraction between beads, but instead there is only short ranged repulsion that yields excluded volume. Stiffness is imparted to each polymer via a harmonic angle bond that causes a linear alignment of any three consecutive monomers to be the energetically favourable conformation. Each simulation also included a single probe molecule that represented one AgND molecule.
A length of L=3 beads was chosen to roughly correspond to the length of the DA-170 crosslinker. AgND, which is of similar length to DA-170, was also modeled as an L=3 molecule. Simulations were also performed for crosslinker molecules of length L=6 and L=9 to study how the dynamics change with crosslinker length. The L=9 molecule was the longest length that could be studied due to constraints on the simulation setup. Thus, these simulations do not replicate the crosslinker ratios studied experimentally and instead explore the dependence on crosslinker length in a more general way.
The diameter of each bead was set to σ and, thus, σ serves as the length scale for the simulation. The box length was set to 20 σ and each system was filled with enough polymers to achieve a volume fraction of ~ 49%. This density is high enough to approximate the pure polymer melt of the experiments, but is also low enough to allow for the movement of individual crosslinker molecules on simulation time scales. This middle ground then mimics the experimental setup while permitting the exploration of the dynamics of the system. The systems then consisted of: 2500 L=3 crosslinkers + 1 L=3 probe molecules, 1250 L=6 crosslinkers + 1 L=3 probe molecule, and 833 L=9 crosslinkers + 1 L=3 probe molecules. Images from the L=3 and L=9 systems are shown in the Supplementary Figure 12.
For each crosslinker length, simulations were performed for two scenarios: diffusion of the probe molecules in pure resin (no polymerization of crosslinkers) and in the final network (saturated polymerization). For both scenarios, the system was evolved in time via Langevin dynamics 46 using the HOOMD blue simulation packaged. 49,50 The setup of each system consisted of a number of preliminary steps. First, the box was filled with the specified number of polymers and one probe molecule being constructed on a grid. Second, the system was randomized with a short simulation where there were no intermolecular interactions. This allows all molecules to pass through each other and, thus, randomizing the system very quickly. Third, the excluded volume interactions between beads on different molecules was ramped up until the full potential was applied. This method yielded the initial configurations shown in the Supplementary Figure 12.
For the simulations of diffusion in pure resin, these initial states were then evolved in time and the position of the probe molecule was monitored. The mean square displacement (MSD) of the probe molecule was then calculated from these trajectories by internal averaging.
The diffusion coefficient was extracted from a linear fit of the MSD plotted against simulation time.
For the simulations of diffusion in the network, simulations were performed in two steps.
In the first step, a polymer network was formed by setting 10% of the cross-linkers to be reactive allowing polymerization to take place via diffusion. These network formation simulations were conducted for a long enough time period that the rate of adding new crosslinking points became very small. The resulting networks were then considered fully polymerized. Once the network was formed, a second simulation was conducted in which the diffusion of the L=3 probe molecule representing AgND was monitored. No further polymerization occurred during this step.
Analysis of the MSD and resulting diffusion coefficient was conducted in the same manner as the pure resin case. However, there are more sources of variability in the network case since the network is not homogenous (while the pure resin essentially is). This means that the rate of diffusion depends on the local environment and, thus, varies as the probe explores different areas of the network. Further, since the simulations are limited in the size of the network that is constructed, the results will also vary between different simulations. Even though the simulations are identical in procedure, if a different seed is used to initiate the dynamics then the network that is formed will be significantly different between simulations and, thus, the calculated diffusion coefficient may also be significantly different.
To account for these variations, simulations were performed using three different initial seeds to build three independent networks for each crosslinker length. The final diffusion coefficient given in the main manuscript is the average of the value calculated for each realization. The error bars correspond to the standard error as calculated across the ensemble of three. After predetermined times, the optical density, a measure of cell growth, was determined at 600 nm (OD600) using a microplate reader (Varioskan Flash, Thermo Scientific).