Scalable, Highly Pure, and Diameter‐Sorted Boron Nitride Nanotube by Aqueous Polymer Two‐Phase Extraction

Boron nitride nanotube (BNNT) has attracted recent attention owing to its exceptional material properties; yet, practical implementation in real‐life applications has been elusive, mainly due to the purity issues associated with its large‐scale synthesis. Although different purification methods have been discussed so far, there lacks a scalable solution method in the community. In this work, a simple, high‐throughput, and scalable purification of BNNT is reported via modification of an established sorting technique, aqueous polymer two‐phase extraction. A complete partition mapping of the boron nitride species is established, which enables the segregation of the highly pure BNNT with a major impurity removal efficiency of > 98%. A successful scaling up of the process is illustrated and provides solid evidence of its diameter sorting behavior. Last, towards its macroscopic assemblies, a liquid crystal of the purified BNNT is demonstrated. The effort toward large‐scale solution purification of BNNT is believed to contribute significantly to the macroscopic realization of its exceptional properties in the near future.

conventional solution processes have suffered severely from low yields of purified BNNT. [10] Specifically, being associated with poor dispersity of BNNT in common solvents, the necessity for high-speed centrifugation has resulted in an indiscriminative loss of BNNT that may otherwise be secured. [11] As the majority of reported solution methods are adopting centrifugation as a necessary pre-treatment, it is highly desirable that one develops a method that could possibly bypass it.
In our previous work, we reported high-purity BNNT via a solution process of gel column chromatography. Our key finding was that a small difference in the surface energy originating from the size effect can be utilized in differentiating between the impurities and the pure form of BNNT. [10a] Although the method was of very high resolution and of fast processing time, its scalability, need for high-speed centrifugation, and a relatively high cost of the sorting material have been notable disadvantages. To address this, here we demonstrate a scalable, high-throughput sorting of BNNT from its synthetic impurities, via a solution process of aqueous polymer two-phase extraction (ATPE). ATPE has widely been utilized in biology for isolating proteins, [12] and has successfully been translated into sorting different types of nanoparticles. [13] Notably, its scalability, simplicity, and high resolution have recently been highlighted in sorting CNTs of different metallicity and chirality, through costeffective benchtop processing. [14] Our first adaptation of ATPE in sorting BNNT features a systematic control over its purity and yield while avoiding the aforementioned, general limitation of the solution processes. We believe our finding here to be universally applicable to the established large-scale synthesis of BNNT, demonstrating a significant advancement towards future BNNT applications targeting macroscopic assemblies of various form factors, such as biological scaffolds, films, or wearable fibers/yarns.

Results and Discussion
As-synthesized BNNT in this study was prepared from a recently developed, large-scale synthesis via the inductively-coupled plasma (ICP) method. [8b,15] Compared with conventional synthesis, the ICP process features large yields of highly crystalline BNNT, albeit with less apparent purity. Following a convention of our previous work, the as-synthesized BNNT was thermally treated to give "crude" BNNT. It consists of nanotubular BN (i.e., pure BNNT) mixed with varying sizes and amounts of impurities, including (stacked) hexagonal BN (hBN) and boron oxide (B 2 O 3 ). [10a] No additional treatment was applied, to properly assess the effect of the purification process that ensues.
ATPE of the crude BNNT was carried out adopting a conventional aqueous two-phase system, in which water-soluble polymers of dextran (DX, MW ≈ 70 k) and poly(ethylene glycol) (PEG, MW ≈ 6 k) were utilized ( Figure 1A). [14d] Briefly, each Small Methods 2023, 7, 2201341 Figure 1. General description of the materials and processes utilized in this study. A) Materials and B) the ATPE processes. Stage 1 includes the mixing of the ATPE components, followed by manual shaking and equilibration. Stage 2 utilizes the mixing of a selected extract from Stage 1 with a complementary extract of the dummy phase, in which the SC concentration of the final mixture is adjusted through that of the dummy extract.
polymer was dissolved first in water to give stock solutions (20 wt.% DX and 50 wt.% PEG, respectively), which were then mixed together to give a mixture containing 9 wt.% of each polymer. A well-dispersed aqueous solution of the crude BNNT (2 mg mL −1 ) enabled by a tip sonication, with sodium cholate (SC) as a surfactant, was then added to this mixture (further experimental details in Table S1 to Table S3, Supporting Information). A brief, manual mixing first results in a metastable mixture of the components, which then begins to phase separate spontaneously. Alongside the phase separation of the polymers, during which each polymer-rich microemulsion is generated and coalesced, the dispersed particles in the mixture partition into one with a matching surface energy window. [14d] Eventually, the system is finalized with a bottom layer consisting of a DX-rich phase and a top layer with a PEG-rich phase, with the partitioned particles inside. While the phase separation occurs spontaneously, one is generally recommended to have an additional step of low-speed centrifugation (e.g., 700 to 2000 g, 5 to 15 min). Notably, the desired selectivity is achieved by adjusting the amount of the surfactant, which affects the apparent hydrophilicity and/or hydrophobicity of each particle in the medium in situ manner (top panel, Figure 1B). [16] Another notable exploitation of ATPE is a series of partitioning, where the product of the first stage of ATPE (i.e., either the top or the bottom layer) is successively utilized in the second stage. A "dummy" phase, which serves as a blank template for the second stage, is utilized for this purpose (bottom panel, Figure 1B). [16] Importantly, ATPE does not require the utilization of high-speed centrifugation, where the partitioning is driven solely by thermodynamic spontaneity. [14d] Further, although not extensively discussed in the present study, one can adjust other ATPE parameters (e.g., temperature, polymer composition, and so on) to control the partition behavior, altogether with high resolution and fidelity. Figure 2 displays the partition behavior of the crude BNNT observed in a series of ATPE experiments, where the surfactant concentrations from 0.1 to 0.7 wt.% were applied ( Figure 2A and Figure 2B). Specifically, with an increment of 0.05 wt.%, the surfactant concentrations from 0.1 to 0.55 wt.% were tested which we refer to as "Stage 1 (Figure 2A)," while those from 0.55 to 0.7 wt.% were utilized in "Stage 2 ( Figure 2B)." Importantly, we note that Stage 2 was conducted using the top extract of Stage 1 at 0.1 wt.% SC. We use our specific notation for each of the resulting phases, as a serial combination of the component as "phase identity/surfactant concentration." With this convention, for example, the notation "B 0. 2 have the T 0.1 component ahead in common, as mentioned above.
The crude BNNT normally contains significant amounts of impurities, as shown in Figure S1, Supporting Information (note the nanotubular BN is hardly visible due to the presence of the impurities). As reported previously, the size distribution of the major impurity (i.e., hBN) goes twofold, characterized by particles of < 100 nm and > 1 µm, respectively. [10a] Therefore, it is highly expected that the size difference is directly reflected in their partition behavior, based on the size-dependent surface energy argument reported previously. [10a] Overall, the ATPE series tested here showed a notable improvement in the purity level, where systematic control over the partition behavior from either the top or the bottom series was evident. As an example, Figure 2C to Figure 2F summarize the morphological trend of the selected top series, namely, T 0.1 , T 0.2 , T 0.3 , and T 0.4 . Here, as the SC concentration goes up, the overall absence of the largesized hBN (> 1 µm) was clear, although the small-sized hBN (< 100 nm) remained. In contrast, the bottom series showed a more comprehensive partition behavior within the given surface energy window. Figure S2, Supporting Information, summarizes the partition behavior of the representative bottom series (B 0.25 , B 0.35 , B 0.45 , and B 0.55 ). We observe that the major species at low SC concentration (e.g., B 0.25 , B 0.35 ) was the largesized hBN, while it systematically switches to the mixture of nanotubular BN and the small-sized hBN at the high SC concentration (e.g., B 0.55 ) (see Figure 3 and related discussion). For an in-depth quantification of each fraction, powder X-ray diffraction (pXRD) was carried out for the selected top series (bottom panels, Figure 2C to Figure 2F). We found the pXRD results corroborate the observed morphological trend very well, where the relative peak intensity of the principal diffraction corresponding to the stacked hBN (2θ = 26.8°) reduces systematically as the SC concentration increases, compared with that of the nanotubular BN (2θ = 25.6°). [9c,10a] Note that the principal diffraction of the hBN is governed by the presence of the largesized hBN, not the small-sized one.
Taking advantage of the high fidelity of the ATPE process, an additional fine-tuning of the purity (i.e., removal of the small-sized hBN) was realized ( Figure 2B). As aforementioned, the extract from the Stage 1 (T 0.1 ) was further processed with the dummy extracts, resulting in a second set of ATPE series with a new set of SC concentration window ( Figure S3, Supporting Information, 0.55-0.7 wt.%, with 0.05 wt.% interval). A champion purity was found at T 0.1 B 0.55 , where it featured essentially no presence of the stacked hBN that was dominating the crude sample. The corresponding diffraction analysis (bottom panel, Figure 2G) showed 98.5% (± 0.8%) of the hBN removal efficiency based on multiple trials (calculation details and the tabulated uncertainty with respect to the peak area and/or the multiple trials are included in the Supporting Information), while the principal diffraction from the purified BNNT showed an excellent fitting behavior to the Lorentzian (R 2 > 0.98). Note, however, that the morphological change observed at Stage 2 is not well captured by the X-ray analysis (see the spectral similarity of T 0.4 and T 0.1 B 0.55 ), as Stage 2 accompanies the removal of the small-sized hBN that does not appear strongly in the pXRD spectra. Last, Figure 2H and Figure 2I summarize the overall yield and the hBN removal efficiency based on multiple trials (hBN removal efficiency based on peak area is also summarized in Table S4 to Table S6, Supporting Information), where a general tradeoff between the two is observed.
Based on the partition behavior observed at Stage 1 and Stage 2, we built a pseudo-quantitative partition diagram as depicted in Figure 3, along with a schematic illustration of the ATPE mechanism pertaining to the sorting of BN materials. Following the general principle of ATPE, the manual mixing of the solution first results in a metastable mixture of the components, in which the formation of phase-separated microemulsions allows for the partitioning of the BN species into the ones with matching surface energy windows. [14d] Here, the emulsification is facilitated by the presence of the aggregation, forming a stable Pickering emulsion. [13b] The microemulsions of each polymer-rich phase further coalesce with each other, resulting in separated layers of the two phases ( Figure 3A). Of specific note is the role of the Pickering emulsion, which practically populates most of the aggregates onto the interface upon its coalescence. As the emulsion favors its stabilization by aggregates of either large-sized or Janus-like (i.e., a mixed aggregate of hBN and BNNT), one can argue that it replaces the conventional step of high-speed centrifugation in a more efficient way.
Regarding the partition diagram ( Figure 3B), the partition coefficient in the y-axis is defined as the ratio between the concentration of the component in the top phase to the Small Methods 2023, 7, 2201341 concentration of the same component in the bottom phase. Thus, the coefficient being unity indicates the material's trend of equal partition to both the top and the bottom phases. On the other hand, the coefficient being larger than unity means the material's partitioning is preferential to the top phase, which in our case to the PEG-rich phase (and vice versa). We believe the following observation made at Stage 1 to be strongly supportive: 1) the small-sized hBN remained along with the nanotubular BN in the top series as the SC concentration increased ( Figure 2C to Figure 2F); 2) the large-sized hBN favored its partition towards the bottom phase at the lower SC concentration regime ( Figure S2, Supporting Information); 3) the bottom series featured its dominating species in an order of the large-sized hBN, the nanotubular BN, and the smallsized hBN as the SC concentration increases. The diagram is also supported by the observation that the nanotubular BN displayed its best purity near the SC concentration of 0.55 wt.% at Stage 2 ( Figure S3). Interestingly, the pseudo-quantitative model depicted here matched nicely with the sorting behavior observed in our previous work, where hBN of varying sizes were sorted separately, with that of the nanotubular BN in between.
[10a] It strongly indicates the major sorting mechanism of the BN species follows the delicate balance between the surface energy of each particle, driven mainly by the size effect.
Another notable advantage of ATPE is its scalability, where a high level of sorting fidelity is maintained throughout its scaling up (up to 0.4 g BNNT throughput, Figure 4A, Figure  S10, and Table S4 to Table S6, Supporting Information). Figure 4B to Figure 4I summarize the corresponding morphology and the ensemble measurements of the BNNT purified through the scale-up. As is clear from the microscopy images ( Figure 4B and Figure 4C), a minimum abundance of the stacked hBN with a dense network of nanotubular BN is clearly visible. Our process features mild processing, where no significant damage to the nanotube was found ( Figure 4D and Figure 4E). Diffraction analysis of the large-scale product also showed a good fitting behavior, with no significant indication of the stacked hBN ( Figure 4F and Figure 4G, hBN removal efficiency of 98.3% ± 0.8% based on multiple trials). Additionally, the FTIR spectrum of the purified BNNT showed its characteristic vibrational modes at ≈ 1350 and ≈ 800 cm −1 for its longitudinal and radial vibration, respectively. [9c,10a,17] Note the remarkable simplification of the spectrum after the purification ( Figure 4H and Figure 4I).
Overall, although we took advantage of the size range of hBN that the ICP method features, we would like to reiterate that the determinant of the optimal condition for the successful isolation of the nanotubular BN was the diameter (and its associated curvature effect). Therefore, for example, BNNT prepared by the high temperature and pressure (HTP) method that is generally known to result in a similar average diameter (≈ 5 nm) compared with that of the ICP method, [9a] is expected to be equally applicable to the described partition procedure. However, following the same analogy, further optimization of the surfactant concentration is expected to be required for BNNT having a vastly different average diameter. Following this argument, one can also conclude that the ATPE process should not be susceptible to the batch-to-batch variability that is common to large-scale synthesis. Figure S11, Supporting Information, summarizes the variability of the BNNT batches utilized in this study (see the nanotubular BN nearly invisible in the second batch), where the optimal condition for the process practically remains the same. Another feature that is worth noting regarding the solution process of ATPE is its near complete Small Methods 2023, 7, 2201341 removal of the boron allotropes at all the fractions under observation. Specifically, these impurities were found to be readily soluble in the aqueous solution, especially during the tip sonication step. As we mainly utilized vacuum filtration to collect the partitioned fractions, the whereabouts of the boron allotropes were unclear.
Apart from the scalability, early reports on the ATPE of CNT have focused on its diameter sorting characteristics. [14b,16] We note that, although a similar precision is not required for BNNT owing to its diameter-invariant band-gap, certain properties (e.g., mechanical) of its macroscopic assembly should exhibit a significant diameter dependence. Based on this idea, we checked the diameter sorting behavior of our ATPE series ( Figure 5). The crude BNNT from the ICP process generally featured its diameter distribution as illustrated in Figure 5C, in which the nanotubes have diameters of 4.9 (± 0.25 nm) and 14.5 nm (± 0.50 nm) being the most abundant species. The ATPE series showed a clear indication of the diameter sorting, where different fractions of the crude BNNT partitioned (e.g. B 0.4 and B 0.5 ), exhibited a marked difference in their average diameters. Figure 5D and Figure 5G display the corresponding statistical analysis, where B 0.4 featured the narrowest diameter distribution centered at 4.02 nm (± 0.04 nm), while that of B 0.5 showed the BNNT with an average diameter of 4.55 nm (± 0.09 nm) (note the distribution of B 0.5 significantly narrower than that of the crude BNNT). Note the diameter series showed different morphological behavior consistent with our original hypothesis. Specifically, the early diameter series (e.g., B 0.4 ) showed a notably "wavy" nature compared with the later series (e.g., B 0.5 ) which showed more "stiff" nanotubular characteristics. The anisotropic, rod-like nature of BNNT forms a lyotropic liquid crystal phase when dispersed in an appropriate medium. [18] We prepared an SC dispersion (1 wt.%) of the purified BNNT at a BNNT concentration of 1 mg mL −1 , using a bath sonicator. To reach a sufficient concentration for the liquid crystal formation, we evaporated the water to give a nominal BNNT concentration of ≈ 7 wt.%. A sample cell with a thickness of 20 µm was prepared, which was then observed through a polarized optical microscope with a rotating stage. Figure 6A displays the corresponding polarized light images, which clearly show a Schlieren texture that is characteristic of a nematic phase. Specifically, the dark brushes appear when the directors of a liquid crystal are either parallel or perpendicular to the plane of the polarization, being related to the singularities of the nematic liquid crystal. As the sample rotates, we observed the dark brushes rotating along the director as well (Movie S1, Supporting Information). Note, however, that the nominal concentration reported here is a significant overestimation of the actual concentration at which the solution starts to form the liquid crystal. Specifically, we found increasing amounts of aggregations as the BNNT gets thicker, due mainly to the inherent solubility limit of BNNT in common solvents. Even so, the successful formation of the nematic phase strongly indicates the capability of our ATPE process in generating BNNT of as high as a liquid crystal grade.

Conclusions
While different solution purification protocols for BNNT have been reported so far, there lacked a method that is scalable to maintain the high purity of BNNT. In this work, we reported a high-throughput, simple, and scalable purification of BNNT via ATPE, that effectively differentiates between the pure form of BNNT and the persisting impurities based on their surface energy difference. We illustrated a systematic control over the composition and showed a complete partition diagram governing the solution purification of the BN species. We successfully demonstrated a scaling-up of the protocol, and maintained the high sorting fidelity. We believe our advancement towards a large-scale solution purification signifies its importance in achieving the desired macroscopic properties of BNNT as originally dictated by its structural characteristics.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.