High-Throughput Image Analysis of Fibrillar Materials: A Case Study

Sep 27, 2017 - Measuring fiber counts, lengths, and packing density, among other properties, requires the vectorization of individual fibers, that is,...
0 downloads 11 Views 4MB Size
Subscriber access provided by Gothenburg University Library

Article

High-throughput Image Analysis of Fibrillar Materials: A Case Study on Polymer Nanofiber Packing, Alignment, and Defects in OFETs Nils Persson, Joshua Rafshoon, Kaylie Naghshpour, Tony Fast, Ping-Hsun Chu, Michael McBride, Bailey Risteen, Martha A. Grover, and Elsa Reichmanis ACS Appl. Mater. Interfaces, Just Accepted Manuscript • DOI: 10.1021/acsami.7b10510 • Publication Date (Web): 27 Sep 2017 Downloaded from http://pubs.acs.org on September 27, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

ACS Applied Materials & Interfaces is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

High-throughput Image Analysis of Fibrillar Materials: A Case Study on Polymer Nanofiber Packing, Alignment, and Defects in OFETs Nils E. Persson†*, Joshua Rafshoon†, Kaylie Naghshpour†, Tony Fastϒ, Ping-Hsun Chu†, Michael McBride†, Bailey Risteen†, Martha Grover†*, Elsa Reichmanis†‡§* †

School of Chemical & Biomolecular Engineering, Georgia Institute of Technology, Atlanta, GA 30332,

USA ϒ

George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta,

Georgia 30332, USA ‡

School of Chemistry & Biochemistry, Georgia Institute of Technology, Atlanta, Georgia 30332, USA

§

School of Materials Science & Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332,

USA *

Authors

to

whom

correspondence

should

be

addressed:

[email protected], [email protected]

ACS Paragon Plus Environment

[email protected],

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 30

Keywords OFET, conjugated polymers, image analysis, fibers, solution processing, crystallization, alignment

ACS Paragon Plus Environment

1

Page 3 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

ABSTRACT. High-throughput discovery of process-structure-property relationships in materials through an informatics-enabled empirical approach is an increasingly utilized technique in materials research due to the rapidly expanding availability of data. Here, process-structure-property relationships are extracted for the nucleation, growth and deposition of semiconducting poly(3-hexylthiophene) (P3HT) nanofibers used in organic field effect transistors, via high-throughput image analysis. This study is performed using an automated image analysis pipeline combining existing open-source software and new algorithms, enabling the rapid evaluation of structural metrics for images of fibrillar materials, including local orientational order, fiber length density, and fiber length distributions. We observe that microfluidic processing leads to fibers that pack with unusually high density, while sonication yields fibers that pack sparsely with low alignment. The is attributed to differences in their crystallization mechanisms. P3HT nanofiber packing during thin film deposition exhibits behavior suggesting that fibers are confined to packing in two-dimensional layers. We find that fiber alignment, a feature correlated with charge carrier mobility, is driven by increasing fiber length, and that shorter fibers tend to segregate to the buried dielectric interface during deposition, creating potentially performancelimiting defects in alignment. Another barrier to perfect alignment is the curvature of P3HT fibers; we propose a mechanistic simulation of fiber growth that reconciles both this curvature and the log-normal distribution of fiber lengths inherent to the fiber populations under consideration.

ACS Paragon Plus Environment

2

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 30

INTRODUCTION Conjugated polymers are driving a revolution in printable electronics. These mechanically flexible, semiconducting materials are being incorporated into devices ranging from solar cells to biosensors.1–3 Conjugated-polymer-based organic field effect transistors (OFETs) are on the brink of commercial viability for use in flexible display applications.4 Enabling these advances is a growing fundamental understanding of the process-structure-property relationships governing conjugated polymer thin films. While novel synthetic conjugated polymers continue to push the boundaries of performance, greater control of the processing of classical semi-crystalline polymers has led to a refined understanding of polymer crystallization, assembly, and long-range microstructure.5–8 Recent progress by our group in controlling the nucleation, growth and alignment of poly(3-hexylthiophene) (P3HT) nanofibers in OFET devices has produced a rich library of images of fibrillar morphologies containing valuable information on the microstructural evolution of conjugated polymer thin films.9–14 Here, we discuss the process-structure-property relationships learned from analyzing over 100 images of polymeric transistor morphology spanning a wide range of solution processing techniques. We also detail modifications to existing open-source software and models that were developed to aid with fiber extraction, structural order parameter identification, and data visualization. While P3HT has a long history in the organic electronics literature, it still finds frequent use in transistors, solar cells, and organic electronic devices with higher-level functionality.15–18 This is likely because of its widening commercial availability as well as its complex processing behavior due to its formation of striking nanofibrillar aggregates. Many techniques to induce the nucleation and growth of P3HT nanofibers have been previously reported: heating/cooling cycles, mixed solvent approaches, simple aging and shear-induced growth techniques are some notable examples.19–22 Within our group, sonication,23,24 poor solvent addition,9,25 UV irradiation,26 microfluidic processing,13 and aging14 have been demonstrated, sometimes in combination with one another, as tunable and scalable methods for the controlled fabrication of P3HT nanofibers. Controlled deposition methods such as blade coating have afforded macroscopically aligned fibrillar structures, with commensurate increases in field effect mobility in thin film transistors.12,14 Despite these rapid advances in processing, a fundamental and intuitive understanding of the structural evolution of P3HT-nanofiber-based thin films is lacking in the literature. Newbloom et al. revealed important aspects of fiber branching and network geometry through neutron scattering experiments, but these studies were performed on structures that were otherwise disordered at long range.27,28 As nanofiber-based thin films progressed toward more tightly-controlled and aligned structures, a more detailed and comprehensive model of fibrillar structure was required. UV-Vis and GIWAXS characterization provide helpful information on bulk properties such as the fraction of aggregated P3HT, its exciton bandwidth, and its out-of-plane orientation, 29–31 but they provide no local insight and limited information on in-plane orientation, which has proven useful in other conjugated polymer systems.32,33 Furthermore, none of these characterization methods approach the problem at the level of the nanofiber itself. P3HT nanofibers are readily observable through AFM, which is frequently presented alongside these techniques, but infrequently analyzed in quantitative detail.34–36

ACS Paragon Plus Environment

3

Page 5 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

The primary difficulty in the analysis of fibrillar structures from AFM images lies in the image processing techniques required to achieve the reliable extraction and measurement of fibers from images. In a previous study, we combined an anisotropic diffusion filter with a skeletonization algorithm to extract orientation distributions from images of P3HT nanofibers. This helped quantify both the radial alignment inherent to the spin-coating process and the linear aligning effect of blade coating on P3HT nanofibers.37,38 However, this approach was limited in that it never allowed analysis of the length distribution of nanofiber populations, arguably their most fundamental and important property. Recently, however, an open-source software tool called FiberApp was introduced, containing a comprehensive suite of analytical tools for images of fibers.39 However, it required semi-manual tracing, which is prohibitively tedious for datasets as large as ours, while also presenting problems with reproducibility. It seemed reasonable, then, to unite the automated skeletonization approach with the powerful fiber vectorization tools from FiberApp – a programming challenge which comprises a large portion of the Supporting Information in this manuscript. In the course of this study, it was also found that the model of orientational order provided in FiberApp, having been developed for the analysis of amyloid fibrils at liquid interfaces,40 was not wellsuited for the quantification of the specific structural features observed in our image library. As part of this study, we present modifications to this model, integrating concepts from the analysis of orientational order of block copolymers as detailed by Murphy et al.41 The automated image analysis tools developed herein are available for download at [gtfiber.github.io]. Additionally, we introduce a Python-based interactive data visualization that aids significantly in the analysis of structural order parameters across large libraries of structural imagery, accessible at [ZoomImgs.github.io]. This manuscript proceeds as follows: first, the raw data is described, comprised of images from past studies as well as new experiments designed to fill out the full range of process parameters used by our group. Then, the algorithmic details and validation of our automated fiber extraction method are covered. The curvature of P3HT nanofibers is analyzed in terms of its persistence length, deformation, and growth mechanisms. Sonication and microfluidic processing are investigated for their extreme values of fiber packing density. Fiber alignment is characterized across the entire dataset, revealing that P3HT nanofibers tend to form quasi-two-dimensional layers during thin film deposition processes. The relationship between fiber length, alignment, and charge carrier mobility is explored through a detailed characterization of the air and dielectric interfaces of an OFET device. It is shown that perfect alignment is currently limited by packing defects due to shorter fibers that preferentially segregate to the dielectric interface. We expect these results and software tools to be of interest specifically to the organic electronics community and more broadly to researchers working on fibrillar materials and composites of any kind. RESULTS AND DISCUSSION Solution Processing and Thin Film Deposition The solution processing methods and raw data considered in this study are illustrated in Figure 1. The images in Figure 1b-e are AFM tapping mode phase images taken from thin films of regioregular

ACS Paragon Plus Environment

4

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 30

P3HT. The fibrillar regions are π-π-stacked crystallites that are dispersed in an otherwise disordered matrix; film thickness ranges from 20–40 nm. Solutions of P3HT (Mn ranging from 20 – 41 kD) in chloroform were prepared at a concentration of 5 mg/mL in all cases. Nanofibers were nucleated using either UV irradiation or sonication, resulting in the appearance of small fiber nuclei, as seen in the blade coated film shown in Figure 1b. Further growth was promoted by either allowing solutions to age or by addition of a poor co-solvent, in this case 2-methylpentane. Figure 1c shows a thin film morphology after a sonicated solution was aged for two days and blade coated. Nucleation and growth were combined in a single microfluidic flow system through a cooling and UV-assisted growth step, with the resulting morphology shown in Figure 1d after spin coating. Finally, a highly aligned morphology is shown in Figure 1e, the result of blade coating a solution processed with UV and aged for two days. Further details on the processing methods are available in the Methods section or in the papers that introduced them.9,11,13,14,23,24,26

Figure 1. Processing of P3HT nanofibers and progression of fibrillar morphologies as observed in 5 µm AFM phase images. (a) Illustrations of solution processing and thin film deposition methods considered in this study: UV irradiation and sonication induce nucleation, while aging and poor solvent addition lead to extended growth. Nucleation and growth are combined in a microfluidic-cooling-UV system. Thin film deposition methods considered include spin coating and blade coating. (b) Small fiber nuclei after sonication. (c) Longer fibers formed after aging a sonicated solution for two days. (d) Densely packed fibers formed in the microfluidic process and deposited by spin coating. (e) Highly aligned fibers formed by UV irradiation and two days of aging followed by blade coating at 3 mm/s.

The morphological space spanned by the above processing methods is highly diverse: the four images presented in Figure 1 give some idea as to the range of fiber lengths, packing densities, and orientation order present in this system, but qualitative observations from raw images are insufficient to convey and quantify the process-structure-property relationships present in this material system. Morphological imagery was available from 137 unique samples: 33 images sourced from previously published studies and the remaining 104 consisting of either previously unpublished data or samples fabricated specifically for this study to fill out the process parameter domain as well as the

ACS Paragon Plus Environment

5

Page 7 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

morphological space. Image sizes range from 2 to 10 µm, but most of the analysis in this study is limited to 5 and 7 µm images to ensure consistency in structural measurements, which brings the dataset down to an even 100. To obtain a satisfying analysis, it is necessary to extract fibers and quantify their length distributions, packing behavior and orientational order in terms of structural order parameters. We begin by outlining the fiber extraction and analysis procedure developed for this study, which integrates the approaches of previously available open source software packages. Image Processing and Analysis Starting from the phase channel of a tapping mode AFM image taken from a P3HT thin film, an image is processed through an anisotropic diffusion filter (Figure 2b),42 an adaptive thresholding step (Figure 2c), and a skeletonization step (Figure 2d). The image skeleton is limiting, however. Measuring fiber counts, lengths, and packing density, among other properties, requires the vectorization of individual fibers; that is, identifying which pixels belong to the same fiber, and converting those pixels into a string of vectors, referred to as a contour. This is the approach implemented in FiberApp,39 in which users manually select points from which contours are initialized. Here, we provide all of the initialization points for FiberApp’s contour-fitting algorithm by breaking the skeletonized image into isolated, unbranched strings of pixels (segments) as illustrated in Figure 2e. This algorithm is discussed in detail in supporting information Section S2: Segment Vectorization.

Figure 2. Extraction and vectorization of fibers from an AFM image. (a) A typical AFM phase image of P3HT nanofibers, (b) Result of anisotropic diffusion and top hat filtering, (c) Result of adaptive thresholding, (d) Result of skeletonization and

ACS Paragon Plus Environment

6

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 30

fringe removal, (e) Fitting vectorized contours to the unbranched skeletal segments using Active Contours at a 30 nm step length, (f) Line plot of the vectorized segments, (g) Illustration of segment matching and fiber reconstruction, (h) Line plot of vectorized fibers.

Overlapping fibers or fibers with adjacent ends tend to be combined into larger branched areas of the skeleton, so after breaking the skeleton into unbranched, vectorized segments, these segments must be reconstructed into linear fibers, illustrated in Figure 2g. Segments whose endpoints are close to one another are matched based on two criteria: the maximum tolerable gap between the segments and the additional curvature created by stitching them together. Two segments are matched if the gap between their endpoints is less than the maximum gap, and if stitching them does not introduce significant additional curvature to their contours. Two matches are shown with green arrows in Figure 2g. A detailed explanation of the segment reconstruction algorithm is provided in the supporting information Section S3: Fiber Reconstruction. The result of this sequence of image processing steps is represented by Figure 2h, a plot of the vectorized fibers. While reconstruction is never a perfect process, strong visual agreement between the original image in Figure 2a and the vectorized representation in Figure 2h indicates that the segmentation and vectorization were effective. To quantify the accuracy of the algorithm, manually traced images were used to train the image processing parameters, presented in supporting information Section S5: Accuracy and Sensitivity Analysis. Structural measurements can be obtained within 10% of their true values. The more detailed results presented herein, especially those concerning fiber length, were confirmed with manual tracing. Supporting Information Section S7 contains a complete discussion of accuracy, sensitivity, failure modes and limitations of the accompanying software, while Section S8 contains a guide to the user interface and parameter selection, should researchers wish to apply this analysis to their own dataset. Persistence Length The most basic property of a fiber is its persistence length, which can be estimated using the 2-D worm-like chain model, similar to polymer chains.43 This analysis was carried out using manually traced fibers in FiberApp, as shown in Figure 3a.39 Two factors complicate the estimation of their persistence length: (1) the fibers we observe are entropically constrained from being embedded in a thin film, surrounded by disordered polymer material and other fibers, and (2) the observed fibers are not typically long enough to obtain an equilibrated estimate of persistence length via mean-squared end-to-end distance measurements. As shown in Figure 3b, the estimated persistence length increases dramatically with the length scale of its calculation, from two to approximately seven microns, and appears that it would increase further if longer fibers were available for analysis. This suggests that the worm-like chain model is not well-suited to P3HT nanofibers, because they are flexible at short range but more rigid at long range.

ACS Paragon Plus Environment

7

Page 9 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Figure 3. Curvature of P3HT nanofibers. (a) Manually traced fibers (highlighted in green) in FiberApp software on a 5 µm AFM image – the longest observed fibers were selected for the analysis of persistence length. (b) Calculation of persistence length through mean-squared end-to-end distance at increasing length scales (error bars represent 95% confidence intervals). (c) Cropped section of a higher-resolution 2 µm image, showing short-range fiber curvature even in aligned structures. (d) Proposed mechanisms of deformation-induced curvature. (e) Proposed mechanism of growth-induced curvature with accompanying to-scale Monte Carlo simulation of fiber structure.

The relatively abrupt short-range curvature of P3HT nanofibers can be observed even in closelypacked, aligned structures such as that shown in Figure 3c. The fact that high short-range curvature does not manifest as higher long-range curvature suggests that the curvature of P3HT nanofibers is due to several different mechanisms. Their long-range rigidity is likely due to their crystalline structure; bending the fibers by bending the crystal lattice is energetically unfavorable, and at high angles, geometrically infeasible. In Figure 3d, bending is illustrated by introducing an angle of 0.5º between each chain – with a π-π stacking distance of 0.38 nm,44 this degree of bending corresponds to a persistence length of 7 µm (〈cos 〉   / ). At this level of curvature, a bending angle of 30º could be achieved over approximately 1 µm of a fiber’s contour length, which matches well the observed curvature in Figure 3a. The short-range curvature is more likely due to the lateral translation or sliding of polymer chains along the (001) lattice vector, whether initiated through contact with another object, or through the underlying growth mechanism of the fiber itself, as illustrated in Figure 3e. The growth mechanism of P3HT nanofibers in solution has received relatively little attention in the literature and deserves a brief discussion.19,22,45 P3HT nanofibers follow a log-normal distribution of

ACS Paragon Plus Environment

8

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 30

lengths, as revealed by analysis presented later. The log-normal distribution is typical of a polycondensation mechanism, allowing both chain growth (the addition of one chain to a stack), and step growth (the combination of stacks).46 In P3HT nanofiber assembly, the driving force for growth is the physical π-π stacking interaction.47 Perfectly centered growth would be the naïve assumption in this case, with each additional chain or stack colliding directly at the center of the growth front. However, this need not be the case, partially because π-π stacking does not enforce a fixed lateral registry (as is the case with some fiber-forming proteins), and partially because the wide variety of chain lengths in a typical polymer molecular weight distribution provides for a growth front of varied width. If we assume that collisions are not perfectly centered, and instead follow a narrow Gaussian distribution of locations around the center of the fiber’s growth front – so narrow that the ends of the fiber are 40 standard deviations from the center – a to-scale simulation of an assembled fiber replicates the observed shortrange curvature quite accurately. These assumptions and the resulting simulation are presented in Figure 3e. Simulation details are provided in supporting information Section S9. Fiber Length Density

ACS Paragon Plus Environment

9

Page 11 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Whereas persistence length is a property of individual fibers, we are most concerned with longrange structure, and the most intuitive metric for fibrillar structures is likely fiber length density, abbreviated as ρFL. Fiber length density captures the packing density of fibers and is defined as the total length of fibers per unit area. Its calculation is illustrated in Figure S11. It is similar to the volume fraction of fibers in a material, but is not influenced by their width, which can be misrepresented by AFM, especially in large images where fibers are only a few pixels wide. As shown in Figure 4b, fiber length density varies significantly across our image library, from 2 – 14 µm-1. Some processing methods, such as UV + aging, produce films that span the entire range. However, sonication + aging never achieves a fiber length density greater than 10 µm-1, while microfluidic + UV processing (MFU) never produces films with less than 10 µm-1. Examples of these two structural classes are provided in Figure 4a and c. The stark difference between the two can be attributed to their respective crystallization mechanisms. Sonication induces fiber nucleation in solution by disentangling polymer chains, which reduces their energetic barrier to form ordered nuclei.34 The disentanglement, however, may also lead to a lack of connectivity between fibers, causing them to spread out from one another. On the other hand, crystallization under microfluidic flow places growing nanofibers in close, oriented proximity to one another, promoting the formation of inter-fiber connections.13

Figure 4. Fiber length density vs. processing method across the image library. (a) Vectorized fibers from an image with the highest measured fiber length density, processed with microfluidics. (b) Data sorted by decreasing median fiber length density for all processing methods and images. Boxes indicate 25th and 75th percentiles, with a horizontal line at the median for each process. Data markers are given horizontal jitter and transparency to show data density. MFU = microfluidics+UV, Son = sonication, PS = poor solvent. (c) An example of low fiber length density for fibers processed with sonication and aging.

Inter-fiber connectivity through tie chains and contact points is frequently incorporated into models of charge transport and mechanical properties in conjugated polymers, but as a structural feature, it remains difficult to quantitatively characterize.31,48–50 Fiber length density can, at the very least, quantify the feasibility of such structural features. Thin films of P3HT nanofibers tend to have spatially homogeneous fiber length density – that is, films are rarely observed in which tight bundles of fibers are

ACS Paragon Plus Environment

10

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 30

accompanied by large voids. When fiber length density is homogeneous, it is inversely correlated with average inter-fiber spacing. For example, in a film with ρFL = 10 µm-1, the average inter-fiber spacing would be 100 nm. For P3HT chains with Mn = 40 kD (used in Figure 4a and c, and most samples in this dataset), an average chain has a fully stretched contour length of 92 nm (assuming repeat unit MW = 166 g/mol and length = 0.38 nm), indicating that an average chain could bridge the gap between fibers in the microfluidic structure, but not necessarily in the sonicated and aged structure. Since polymers have broad molecular weight distributions, as shown in the simulated fiber in Figure 3e, this should not be interpreted as an absolute test of tie chain formation, but rather a likelihood. Orientational Order In addition to fiber length density, spatially-resolved orientational order can be extracted from populations of vectorized fibers. Jordens et al. developed a model of orientational order for the analysis of liquid crystallinity in amyloid fibrils that contains useful guidance for our solid-state fibrillar materials, especially considering that P3HT and other conjugated polymers have known liquidcrystalline properties in the solution state.51,52 Here, this model has been modified to fit the needs of the current dataset, as illustrated in Figure 5. The Jordens model is based on the calculation of the order parameter S2D at increasing length scales, or box sizes. S2D quantifies the alignment of a population of vectors, with 0 being isotropic and 1 being perfectly aligned. At very short length scales, orientations are correlated because the vectors within a box belong to the same fiber, but this correlation drops off as box 

size expands, yielding the decay curve denoted by 

in Figure 5b.



Among the images in our dataset,  tends to approach an asymptotic value before the box size reaches 5 µm, indicating that the images capture a volume element representative of longer-range fiber alignment. This was confirmed by taking multiple images from various devices, as well as larger 10 µm images. Orientational order is modeled by the following equation: 



   + 1   /



(1)

where Sfull captures the level of the asymptote, and the remainder of the data is modeled as exponential decay with a characteristic length of λC. This equation is guaranteed to decrease monotonically from 1 to 

Sfull, capturing the essential behavior of  with only two fitted parameters. Jordens et al. apply this equation after removing the contribution of random fiber packing to alignment, but here, the quantitative description of alignment is valuable regardless of its origin, so we have opted for this simpler approach.

ACS Paragon Plus Environment

11

Page 13 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Figure 5. Analysis of orientational order from populations of vectorized fibers. (a) A vectorized 5 µm image from our dataset, showing the progressively larger box sizes used to calculate S2D. Many locations are sampled and averaged. (b) Plot of S2D vs. box size for the image in (a). Raw data is fit with a two parameter model, yielding Sfull and λC. The area under S2Dimage can be decomposed into two components, with the lower half (dark purple) comprising S2Drand, the alignment due to random fiber packing. The parameter a represents the fraction of S2Dimage not contained under S2Drand.

The Jordens model accounts for the fact that even randomly packed fibers display some alignment at short length scales due to the correlated orientations of vectors along individual fibers’ !" backbones. Termed  and represented by the darker purple area in Figure 5b, this contribution is calculated by randomly translating and rotating the extracted fibers within the frame of the image, then 

recalculating S2D at all box sizes. The fraction of  not due to random fiber packing is denoted by a, referred to as the “fraction of nematic domains.” In the Jordens model, an additional weighting 

parameter was added to the fit of  to calculate a, but this approach resulted in inconsistent estimates of a for structures that were very similar. In one case, structures with nearly identical raw alignment data were estimated to have fractions of nematic domains differing by 100% (0.3 vs. 0.6), and additionally, these values did not appear to accurately quantify the alignment behavior. This is discussed in detail in the Supporting Information, Sections S4 and S5. To allay these concerns here, a is calculated

ACS Paragon Plus Environment

12

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 30



!" as the fraction of integrated area under  not contained under  (the lighter area in Figure 5b). This approach, combined with equation (1), led to a robust fit of orientational order data for all of the images under consideration. The process-structure relationships for P3HT nanofibers are presented in terms of the above order parameters in Figure 6. Each image is plotted in terms of Sfull and ρFL in Figure 6e and in terms of Sfull and a in Figure 6f, with marker colors corresponding to fiber nucleation and growth procedure, and shape corresponding to thin film deposition technique. Representative examples from each plot are shown in Figure 6a–d and g-j as false-colored Orientation Maps, in which the color of each fiber indicates its in-plane orientation as labeled on the attached color wheel in Figure 6a. Since fibers are vectorized at a 30 nm (3 pixel) step length, the orientation, and thus the color, varies along the length of each fiber. This type of visualization is relatively common in the analysis of orientational data because the color wheel has natural circular periodicity.33,41 In addition to providing a visualization of orientational order, the Orientation Map also enhances the readability of high-resolution AFM images, which are usually low-contrast and difficult to effectively analyze at small printed sizes.

ACS Paragon Plus Environment

13

Page 15 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Figure 6. Processed images of fibrillar morphologies and their coordinates in the space defined by the order parameters Sfull, ρFL, and λC. (a-d) Orientation Maps extracted from raw images at extreme order parameter values, with their location in each plot indicated by a gray pointer. Each image is 5 µm, and colors correspond to the in-plane orientation of each fiber’s backbone, as indicated by the attached color wheel in a (i.e. red is horizontal, cyan is vertical). (e) The complete structural image library plotted in terms of Sfull and ρFL, with solution processes coded by color and deposition methods coded by shape. Interactive version available at [ZoomImgs.github.io]. (f) Image library plotted in terms of fraction of nematic domains (a) and ρFL. (g-j) Four more Orientation Maps at extreme order parameter values.

The plot in Figure 6e shows that there is no intrinsic correlation between fiber length density and fiber alignment across the dataset. On the contrary, the ρFL/Sfull parameter space showcases the structural variance attainable through solution-processing of P3HT nanofibers. The trends from Figure 4 are still observable here: the structures formed through microfluidic processing (purple) fall to the high-ρFL-end

ACS Paragon Plus Environment

14

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 30

of the space. In spite of the high packing density, microfluidic processing can still yield relatively lowalignment structures, such as that shown in Figure 6h. Low alignment does not imply random alignment, though: in Figure 6f, it can be seen that a, the fraction of nematic domains, is greater than 0.6 for all of the microfluidic structures, meaning that 60% of their alignment cannot be accounted for by random fiber packing. In the calculation of a when fibers are randomly translated and rotated, they are not computationally restricted from overlapping each other. The fact that very little fiber overlap is observed in the original AFM images, then, offers a clue as to their packing mechanism. The film thickness, which varies from 20–40 nm, would allow for at least 2–4 layers of vertically stacked nanofibers, whose vertical (100) grain size is typically on the order of 10 nm.10,12 Due to the fibers’ long-range rigidity, it is unlikely that they are highly entangled or that they travel tortuous paths through many layers. It is more likely that the fibers at the air/solution interface (the surface being imaged) are practically constrained to packing only in the two dimensions of that interface during deposition. To achieve a high packing density with low alignment in two dimensions, fibers of similar orientations must clump together in tight bundles; bundles which are easily identified by their colored regions in Figure 6h and j. By contrast, the sonicated and aged structures in Figure 6g and i are nearly isotropic (Sfull ~ 0.1) and mostly randomly oriented (a ~ 0.2), apparently because the fibers have more rotational freedom. It is telling that the structures with the highest alignment also appear to have the longest fibers. The high concentration of highly aligned structures (Sfull > 0.8) were produced predominantly using UV irradiation, aging, and blade coating (blue squares), with examples shown in Figure 6b and d. These data points reflect the recent progress our group and others have made in obtaining highly aligned nanofibrillar structures with enhanced charge transport,12,14,53–57 and are a testament to the repeatability of the process: images for this technique were collected by at least four different co-authors of the current study. UV, aged, and blade coated films that fall below an Sfull of 0.8 were generally aged for less than 24 hours. In these structures, such as the one shown in Figure 6a, alignment is still relatively high, and fibers have considerable length, but packing is sparse. Increasing fiber length decreases their ability to rotate in the plane, virtually ensuring that they align in the direction of flow (vertical, in the case of Figure 6a). While this relationship is qualitatively apparent, a high-throughput, quantitative evaluation of fiber length has never been presented for P3HT nanofibers, which is the subject of the following section. Fiber Length, Alignment, and Charge Carrier Mobility Mean fiber length and fiber alignment are generally correlated with one another, as can be seen in Figure 7. Mean length, or number-averaged length, is not a perfect descriptor of fiber length – the length distribution is approximately log-normal, so a mean length of 500 nm means that many fibers are less than 500 nm, and some fibers are much longer than 500 nm. The longer fibers, containing more material, tend to visually dominate a given image frame. Nonetheless, it appears that beyond a mean length of 500 nm, high fiber alignment is much more likely, especially when using blade coating as a deposition method. Blade coating is not a guarantor of alignment and consideration must be taken for

ACS Paragon Plus Environment

15

Page 17 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

the coating rate and regime of operation,57 but it is worth noting that the low alignment outliers are mostly spin coated while the high alignment outliers are blade coated. Theoretical treatment of suspended fibers under flow is highly dependent on the specific situation. For concentrated colloidal rods at a low Peclet number (ratio of advective to diffusive transport rate), it has been observed that higher aspect ratio rods display greater alignment.58 While the diffusive properties of P3HT nanofibers are not known, this observation is in general agreement with other studies demonstrating that highaspect-ratio objects align with the flow direction.59–62

Figure 7. The relationship between mean fiber length and fiber alignment across the entire image library. The gray line is included to enhance readability.

The mechanistic origin of fiber alignment is important because fiber alignment seems to have a causal relationship with charge carrier mobility. While it is tempting to try to draw universal correlations between AFM images and mobility in OFETs, too many other material and device variables are uncontrolled across the dataset for this analysis to be meaningful. However, in the study of UV, aging and blade coating, which produced the highest alignment and highest mobility devices, a strong correlation was found between mobility and Sfull, as shown in Figure 8a.14 Raw transistor electrical data from this study can be found in Figure S16. Reports on other semi-crystalline conjugated polymers have identified similar behavior, with narrow distributions of grain orientations leading to more rapid charge transport.32,33,63 Perhaps counter-intuitively, the correlation between alignment and mobility is not because of percolation through a large network of fibers; rather, fiber alignment is indicative of underlying polymer chain alignment. The fibers simply serve as hubs for charges to find the next long polymer chain to travel along, as illustrated in Figure 8b.31 This is confirmed by the charge transport anisotropy observed in devices with aligned fibers: charge transport is faster perpendicular to aligned fibers than parallel to them.14 Greater alignment would also be expected to increase the planarization of inter-fiber tie chains, leading to faster inter-grain charge transport.31,64

ACS Paragon Plus Environment

16

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 30

Figure 8. A structure-property relationship. (a) Mobility vs. fiber alignment (Sfull) for devices produced via UV, aging, and blade coating, with the coating direction perpendicular to charge transport. (b) Schematic of charge transport at varied levels of fiber alignment.

Given the importance of fiber alignment to charge transport in P3HT thin films, it would be useful to understand why alignment is currently limited at a value of Sfull = 0.9. To conclude our analysis, we demonstrate that perfect nanofiber alignment is limited by two factors. The first is the intrinsic curvature of P3HT nanofibers, as discussed in Figure 3, which is likely unavoidable. The second is packing defects, an aspect of P3HT nanofibers that has received relatively little attention.64,65 A detailed analysis of fiber packing is presented in Figure 9 for the UV, aged and blade coated processing method. For the most highly aligned structure, images of the air and SiO2 interfaces were obtained, the latter by pressing a PDMS slab onto P3HT films, submerging the entire stack in DI water for 10 minutes, then peeling off the PDMS slab with the P3HT film attached. The results tabulated in Figure 9a are the average of three images collected from each surface. The air and SiO2 interfaces of each sample are correlated, but not equivalent. It has been observed previously that thin films composed purely of P3HT nanofibers have roughly identical-looking morphologies on the top and interface, but the detailed differences have not been quantified.66

ACS Paragon Plus Environment

17

Page 19 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Figure 9. (a) Image analysis results from a sample that was UV irradiated, then aged 24h and blade coated, measured at both the air at SiO2 interface by peeling off the film (3 images at each surface). (b) Length distributions from each image, fit with log-normal probability density functions. (c-f) Raw AFM images (5 µm each) and colored Orientation Maps for each sample at each interface.

As demonstrated in Figure 9a and b, shorter fibers tend to segregate to the SiO2 interface (mean length of 359 vs. 570 nm at the air interface). This was confirmed with manual tracings as well. The difference can be rationalized by considering settling in a cereal box: the smaller items can create a more dense phase, and thus sink to the bottom. In this case, the smaller fibers are likely more mobile as well, facilitating their transport to the interface. Unfortunately, short fibers tend to cause defects in fiber alignment, as evidenced by the defects in the aligned structure in Figure 6b, the decreased alignment with shorter fibers at the interface in Figure 9f, and the general correlation between fiber length and alignment. It would seem that further improvement in mobility for pure, regioregular P3HT may be obtained by removing shorter fibers from the population. Simply aging the solution for longer times is not an option, however, because macro-scale phase separation and gelation begins to occur after 48h.28,67 Targeting crystals of a controlled size is a well-known challenge in crystallization, and will likely be the subject of further investigation.68 Interactive Exploration of Materials Image Datasets Presentation of large volumes of image data can sometimes be daunting in a print format. The image dataset presented in Figure 6 is difficult to fully convey in a static figure. As part of this study, we introduce a web-based interactive version of Figure 6e and f built with the Bokeh library for Python, illustrated in Error! Reference source not found. and available at [ZoomImgs.github.io]. When fully zoomed out, the plot looks exactly like those in Figure 6e and f. However, a user can zoom in on any data point by scrolling, similar to Google Maps, revealing the underlying image that produced each data

ACS Paragon Plus Environment

18

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 20 of 30

point. The images fade into view as the zoom level increases to avoid showing too many overlapping images at once, as shown in Figure 10a. Furthermore, when the mouse is hovered over a data marker, a box appears with all of that image’s structural measurements as well as its processing information, as shown in Figure 10b.

Figure 10. Web-based interactive data visualization for materials image data. (a) Each data point represents one image; as a user zooms in on the plot by scrolling (similar to Google Maps), the underlying image for each data point fades into view. (b) Hovering the mouse over a data marker reveals the structural measurements and processing information for that image. (c) Comparison of structural order parameters and measurements for the two images shown in b. (d) Vectorized fiber plot of left structure in b. (e) Vectorized fiber plot of right structure in b.

This tool can be helpful for understanding the structural metrics used in this study. It is also useful for exploring structural libraries: for example, it can reveal similar structures obtained by different processing methods: the two images shown in Figure 10b have nearly identical structural metrics, but one was processed by sonication, aging, and blade coating while the other was processed by UV and blade coating. The two structures can be compared by the plots of their vectorized fibers in Figure 10d and e, as well as with the table in Error! Reference source not found.Figure 10c. The analysis, simulation, and data visualization tools presented here represent an important step forward in the quantification of conjugated polymer microstructure. While P3HT is not the state of the art in terms of performance, control of orientation and alignment in any conjugated polymer thin film is recognized as one of the most important factors in determining charge transport, thermal, and mechanical properties. Since the characterization of soft materials is generally quite challenging, and

ACS Paragon Plus Environment

19

Page 21 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

imaging plays a substantial role in structural characterization, any information that can be quantitatively extracted from images is potentially useful. It is our hope that insights from the packing and defects in aligned P3HT thin films will help researchers better understand related polymer systems. Computer vision tools for materials science are also rapidly advancing as imaging technologies improve and structures become more complex. Many strategies that have been successful in biology and medicine will be applicable to materials systems as well, the present system being one example. An ideal materials imaging ecosystem would include a centralized database of structural imagery, through which one could upload their images as a “search term” and be directed to past studies that produced similar images as well as the code used to analyze those images.69 Such an approach would facilitate greater standardization in measurements from images, as well as produce a large image database that would enable the application of more advanced computer vision techniques for image segmentation, such as generative adversarial convolutional neural nets.70 This will be explored in future work. CONCLUSIONS In conclusion, we have demonstrated improvements to a powerful method for the analysis of fibrillar structures and the important process-structure-property relationships that fall out of this analysis for the P3HT nanofiber system. The results of our analysis extend the already rich knowledge base on P3HT nanofiber growth and deposition to include detailed data on meso-scale packing, alignment, and defects. The persistence length of P3HT nanofibers was analyzed, revealing that curvature can be induced through both deformation and the mechanism of fiber growth. Microfluidic crystallization was shown to produce structures with unusually high fiber length density, while sonication was shown to produce structures with lower fiber length density and alignment, due to differences in their crystallization mechanisms. Packing of P3HT nanofibers appears to occur as a constrained, quasi-twodimensional process. A strong correlation between fiber length and alignment was identified, as well as evidence that shorter fibers preferentially segregate to the buried dielectric interface, causing packing defects that potentially limit alignment and charge carrier mobility in otherwise highly aligned structures. Additionally, a new interactive data visualization was introduced for structural imagery, enabling intuitive navigation of image libraries with quantitative order parameters. These results should support the effort to bring about a more quantitative understanding of conjugated polymer self-assembly, structural ordering and charge transport, while providing foundational tools for informatics-enabled materials experimentation and the general class of fibrillar materials. METHODS Materials Regioregular poly(3-hexylthiophene) (P3HT) was obtained from Rieke Metals. Number average molecular weight (Mn) ranged from 17 kD to 40 kD, with polydispersity between 2.0 and 2.2. Regioregularity was greater than 96%. Chloroform (stabilized with amylenes) and 2-methylpentane (anhydrous) were obtained from Sigma-Aldrich.

ACS Paragon Plus Environment

20

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 22 of 30

Solution Processing Solutions were prepared at 5 mg/mL by dissolving 10 mg of polymer in 2 mL of chloroform (Sigma) in 20 mL borosilicate glass vials, tightly capped, and heated to 60 ºC for 25 minutes on a hot plate (Corning) to ensure complete dissolution. After cooling at ambient conditions for five minutes, the solutions were subjected to one or several of the following treatments: for sonication, the solution vial was dipped in a bath sonicator (Bransonic 2510, 40 kHz, 130 W) filled with tap water for two minutes. For UV irradiation, the solution vial was placed on top of a handheld UV lamp (Entela UVGL-15, 5 mW cm-2, 254 nm) which was placed on top of a magnetic stirrer (Corning), and the solution was irradiated for 8 minutes while being stirred at 300 rpm. For poor solvent addition, 2-methylpentane was slowly added to the solution of P3HT and chloroform at concentrations ranging from 0 – 40 vol%. For aging, the treated solution was capped, wrapped with parafilm, and placed in a dark drawer for the specified amount of time. Microfluidic processing involved pumping the prepared solution through a PTFE tube (300 µm i.d.) with a syringe pump at a flow rate of 1 mL/min; the tube passed through an ice bath (residence time 1s) and a zone under UV irradiation (residence time 10s) before exiting to a collection vial for deposition. OFET Device Fabrication OFET devices with bottom-gate, bottom-contact architecture were fabricated for the electrical characterization of thin films deposited from the as-prepared solutions. Highly n-doped silicon wafers with a thermally grown 300-nm SiO2 dielectric layer were used as the substrate. Source and drain electrodes were patterned by a photolithography lift-off process in a cleanroom environment, and deposited by E-beam evaporation (Denton Explorer) of 50 nm of Au with 3 nm of Cr as the adhesion layer. The device substrates were rinsed with acetone, methanol, and isopropanol, sonicated for 15 minutes in acetone, and cleaned for 30 minutes in a UV-ozone cleaner (Novascan PSD-UV) to remove residual photoresist and other organic contaminants. Thin film deposition was carried out via spin coating (WS-650MZ-23NPP, Laurell) of a 15 µL droplet at 1500 rpm for 60 s, or blade coating at a velocity of 3 mm/s. Blade coating was conducted on a motorized linear stage (A-LSQ150A- E01, Zaber) equipped with a vacuum chuck, with the blade perpendicular to the substrate at a gap height of 5 µm. A 4 µL volume of fluid was injected into the blade gap and allowed to wick the width of the blade fully before coating. Coated thin film devices were stored overnight under vacuum before any characterization was performed. Electrical Characterization Individual OFET channels were tested in a nitrogen-filled glovebox using an Agilent 4155C semiconductor parameter analyzer. The field-effect hole mobility (µ) was calculated in the saturation regime of transistor operation (VDS = -80 V) by fitting the following equation to a plot of drain current (IDS) versus gate voltage (VG): #$ 

%&'( μ,- ,./  2*

ACS Paragon Plus Environment

21

Page 23 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

where W (2000 µm) and L (50 µm) are the transistor channel width and length, respectively, Vth is the threshold voltage, and COX is the capacitance per unit area of the SiO2 dielectric (1.15 × 10-8 F/cm2). Atomic Force Microscopy Thin film surface morphology was characterized with a Bruker Dimension Icon atomic force microscope operating in tapping mode with n-type silicon tips (HQ:NSC14-noAl, 5 N/m, 160 kHz, MikroMasch). Images were collected predominantly at a 5 µm scan size with 512 samples per line at 1 line/s, but some images were collected at 2, 4, 7, and 10 µm scan sizes as well. Increased drive amplitudes frequently yielded higher contrast between the fiber and amorphous phases in the phase channel. Approximately 3–5 images could be obtained per tip before loss of image quality due to tip degradation and polymer adsorption.

Acknowledgements Nils Persson gratefully acknowledges financial support from the NSF FLAMEL IGERT Traineeship program, NSF Grant No. 1258425, IGERT-CIF21: Computation-Enabled Design and Manufacturing of High Performance Materials. All researchers acknowledge support from NSF Grant No. 1264555: Morphology and Mobility Control for Functional Robust Flexible Electronics and Photovoltaics. Finally, this project would not have been possible without access to the open source code library behind the FiberApp software package. Notes The authors declare no competing financial interests.

Supporting Information • Summary of GTFiber algorithm stack • Mathematical theory of fiber vectorization • Numerical methods and algorithm for fiber reconstruction • Structural metric calculation methods • Accuracy quantification • Guide to software installation and use

ACS Paragon Plus Environment

22

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 24 of 30

References 1.

Li, G.; Zhu, R.; Yang, Y. Polymer Solar Cells. Nat. Photonics 2012, 6, 153–161.

2.

Knopfmacher, O.; Hammock, M. L.; Appleton, A. L.; Schwartz, G.; Mei, J.; Lei, T.; Pei, J.; Bao, Z. Highly Stable Organic Polymer Field-Effect Transistor Sensor for Selective Detection in the Marine Environment. Nat. Commun. 2014, 5, 2954.

3.

Zheng, H.; Zheng, Y.; Liu, N.; Ai, N.; Wang, Q.; Wu, S.; Zhou, J.; Hu, D.; Yu, S.; Han, S.; Xu, W.; Luo, C.; Meng, Y.; Jiang, Z.; Chen, Y.; Li, D.; Huang, F.; Wang, J.; Peng, J.; Cao, Y. AllSolution Processed Polymer Light-Emitting Diode Displays. Nat. Commun. 2013, 4, 1971.

4.

Sirringhaus, H. 25th Anniversary Article: Organic Field-Effect Transistors: The Path Beyond Amorphous Silicon. Adv. Mater. 2014, 26, 1319–1335.

5.

Zhang, X.; Bronstein, H.; Kronemeijer, A. J.; Smith, J.; Kim, Y.; Kline, R. J.; Richter, L. J.; Anthopoulos, T. D.; Sirringhaus, H.; Song, K.; Heeney, M.; Zhang, W.; McCulloch, I.; DeLongchamp, D. M. Molecular Origin of High Field-Effect Mobility in an Indacenodithiophene–benzothiadiazole Copolymer. Nat. Commun. 2013, 4, 2238.

6.

Takacs, C. J.; Treat, N. D.; Krämer, S.; Chen, Z.; Facchetti, A.; Chabinyc, M. L.; Heeger, A. J. Remarkable Order of a High-Performance Polymer. Nano Lett. 2013, 13, 2522–2527.

7.

Brinkmann, M.; Hartmann, L.; Biniek, L.; Tremel, K.; Kayunkid, N. Orienting Semi-Conducting π-Conjugated Polymers. Macromol. Rapid Commun. 2014, 35, 9–26.

8.

Brinkmann, M. Structure and Morphology Control in Thin Films of Regioregular poly(3Hexylthiophene). J. Polym. Sci. Part B Polym. Phys. 2011, 49, 1218–1233.

9.

Choi, D.; Chang, M.; Reichmanis, E. Controlled Assembly of Poly(3-Hexylthiophene): Managing the Disorder to Order Transition on the Nano- through Meso-Scales. Adv. Funct. Mater. 2015, 25, 920–927.

10.

Kleinhenz, N.; Persson, N.; Xue, Z.; Chu, P. H.; Wang, G.; Yuan, Z.; McBride, M. A.; Choi, D.; Grover, M. A.; Reichmanis, E. Ordering of Poly(3-Hexylthiophene) in Solutions and Films: Effects of Fiber Length and Grain Boundaries on Anisotropy and Mobility. Chem. Mater. 2016, 28, 3905–3913.

11.

Chang, M.; Lee, J.; Kleinhenz, N.; Fu, B.; Reichmanis, E. Photoinduced Anisotropic Supramolecular Assembly and Enhanced Charge Transport of Poly(3-Hexylthiophene) Thin Films. Adv. Funct. Mater. 2014, 24, 4457–4465.

12.

Chang, M.; Choi, D.; Egap, E. Macroscopic Alignment of One-Dimensional Conjugated Polymer Nanocrystallites for High-Mobility Organic Field-Effect Transistors. ACS Appl. Mater. Interfaces 2016, 8, 13484–13491.

13.

Wang, G.; Persson, N.; Chu, P.; Kleinhenz, N.; Fu, B.; Chang, M.; Deb, N.; Mao, Y.; Wang, H.; Grover, M. A.; Reichmanis, E. Microfluidic Crystal Engineering of π-Conjugated Polymers. ACS Nano 2015, 9, 8220–8230.

ACS Paragon Plus Environment

23

Page 25 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

14.

Chu, P.-H.; Kleinhenz, N.; Persson, N.; McBride, M.; Hernandez, J. L.; Fu, B.; Zhang, G.; Reichmanis, E. Toward Precision Control of Nanofiber Orientation in Conjugated Polymer Thin Films: Impact on Charge Transport. Chem. Mater. 2016, 28, 9099–9109.

15.

Lin, C.-J.; Liu, C.-L.; Chen, W.-C. Poly(3-Hexylthiophene)–graphene Composite-Based Aligned Nanofibers for High-Performance Field Effect Transistors. J. Mater. Chem. C 2015, 3, 4290– 4296.

16.

Chang, H.-C.; Liu, C.-L.; Chen, W.-C. Flexible Nonvolatile Transistor Memory Devices Based on One-Dimensional Electrospun P3HT:Au Hybrid Nanofibers. Adv. Funct. Mater. 2013, 23, 4960–4968.

17.

Yagi, M.; Ito, N.; Kawasaki, M.; Shimomura, T. Semiconducting Properties of P- and N-Type Organic Nanofiber/poly(methyl Methacrylate) Composite Films for Film Rectifier. Synth. Met. 2016, 213, 1–6.

18.

Li, L.; Jacobs, D. L.; Che, Y.; Huang, H.; Bunes, B. R.; Yang, X.; Zang, L. Poly(3Hexylthiophene) Nanofiber Networks for Enhancing the Morphology Stability of Polymer Solar Cells. Org. Electron. 2013, 14, 1383–1390.

19.

Sun, S.; Salim, T.; Wong, L. H.; Foo, Y. L.; Boey, F.; Lam, Y. M. A New Insight into Controlling poly(3-Hexylthiophene) Nanofiber Growth through a Mixed-Solvent Approach for Organic Photovoltaics Applications. J. Mater. Chem. 2011, 21, 377–386.

20.

Bielecka, U.; Lutsyk, P.; Janus, K.; Sworakowski, J.; Bartkowiak, W. Effect of Solution Aging on Morphology and Electrical Characteristics of Regioregular P3HT FETs Fabricated by Spin Coating and Spray Coating. Org. Electron. 2011, 12, 1768–1776.

21.

Mykhaylyk, O. O.; Chambon, P.; Impradice, C.; Fairclough, J. P. A.; Terrill, N. J.; Ryan, A. J. Control of Structural Morphology in Shear-Induced Crystallization of Polymers. Macromolecules 2010, 43, 2389–2405.

22.

Oh, J. Y.; Shin, M.; Lee, T. Il; Jang, W. S.; Min, Y.; Myoung, J.; Baik, H. K.; Jeong, U. SelfSeeded Growth of Poly(3-Hexylthiophene) (P3HT) Nanofibrils by a Cycle of Cooling and Heating in Solutions. Macromolecules 2012, 45, 7504–7513.

23.

Aiyar, A. R.; Hong, J.-I.; Izumi, J.; Choi, D.; Kleinhenz, N.; Reichmanis, E. Ultrasound-Induced Ordering in Poly(3-Hexylthiophene): Role of Molecular and Process Parameters on Morphology and Charge Transport. ACS Appl. Mater. Interfaces 2013, 5, 2368–2377.

24.

Aiyar, A. R.; Hong, J.-I.; Nambiar, R.; Collard, D. M.; Reichmanis, E. Tunable Crystallinity in Regioregular Poly(3-Hexylthiophene) Thin Films and Its Impact on Field Effect Mobility. Adv. Funct. Mater. 2011, 21, 2652–2659.

25.

Chang, M.; Choi, D.; Fu, B.; Reichmanis, E. Solvent Based Hydrogen Bonding: Impact on poly(3-Hexylthiophene) Nanoscale Morphology and Charge Transport Characteristics. ACS Nano 2013, 7, 5402–5413.

26.

Chang, M.; Lee, J.; Chu, P.-H.; Choi, D.; Park, B.; Reichmanis, E. Anisotropic Assembly of Conjugated Polymer Nanocrystallites for Enhanced Charge Transport. ACS Appl. Mater.

ACS Paragon Plus Environment

24

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 26 of 30

Interfaces 2014, 6, 21541–21549. 27.

Newbloom, G. M.; Kim, F. S.; Jenekhe, S. A.; Pozzo, D. C. Mesoscale Morphology and Charge Transport in Colloidal Networks of Poly(3-Hexylthiophene). Macromolecules 2011, 44, 3801– 3809.

28.

Newbloom, G. M.; Weigandt, K. M.; Pozzo, D. C. Electrical, Mechanical, and Structural Characterization of Self-Assembly in Poly(3-Hexylthiophene) Organogel Networks. Macromolecules 2012, 45, 3452–3462.

29.

Clark, J.; Chang, J. F.; Spano, F. C.; Friend, R. H.; Silva, C. Determining Exciton Bandwidth and Film Microstructure in Polythiophene Films Using Linear Absorption Spectroscopy. Appl. Phys. Lett. 2009, 94, 2007–2010.

30.

Rivnay, J.; Mannsfeld, S. C. B.; Miller, C. E.; Salleo, A.; Toney, M. F. Quantitative Determination of Organic Semiconductor Microstructure from the Molecular to Device Scale. Chem. Rev. 2012, 112, 5488–5519.

31.

Noriega, R.; Rivnay, J.; Vandewal, K.; Koch, F. P. V; Stingelin, N.; Smith, P.; Toney, M. F.; Salleo, A. A General Relationship between Disorder, Aggregation and Charge Transport in Conjugated Polymers. Nat. Mater. 2013, 12, 1038–1044.

32.

O’Connor, B.; Kline, R. J.; Conrad, B. R.; Richter, L. J.; Gundlach, D.; Toney, M. F.; DeLongchamp, D. M. Anisotropic Structure and Charge Transport in Highly Strain-Aligned Regioregular poly(3-Hexylthiophene). Adv. Funct. Mater. 2011, 21, 3697–3705.

33.

Zhang, X.; Hudson, S. D.; DeLongchamp, D. M.; Gundlach, D. J.; Heeney, M.; McCulloch, I. InPlane Liquid Crystalline Texture of High-Performance Thienothiophene Copolymer Thin Films. Adv. Funct. Mater. 2010, 20, 4098–4106.

34.

Zhao, K.; Khan, H. U.; Li, R.; Su, Y.; Amassian, A. Entanglement of Conjugated Polymer Chains Influences Molecular Self-Assembly and Carrier Transport. Adv. Funct. Mater. 2013, 23, 6024– 6035.

35.

Kline, R. J.; McGehee, M. D.; Kadnikova, E. N.; Liu, J.; Fréchet, J. M. J.; Toney, M. F. Dependence of Regioregular Poly(3-Hexylthiophene) Film Morphology and Field-Effect Mobility on Molecular Weight. Macromolecules 2005, 38, 3312–3319.

36.

Singh, K. a.; Young, T.; McCullough, R. D.; Kowalewski, T.; Porter, L. M. Planarization of Polymeric Field-Effect Transistors: Improvement of Nanomorphology and Enhancement of Electrical Performance. Adv. Funct. Mater. 2010, 20, 2216–2221.

37.

Persson, N. E.; McBride, M. A.; Grover, M. A.; Reichmanis, E. Automated Analysis of Orientational Order in Images of Fibrillar Materials. Chem. Mater. 2017, 29, 3–14.

38.

Persson, N. E.; Chu, P.-H.; McBride, M.; Grover, M.; Reichmanis, E. Nucleation, Growth, and Alignment of Poly(3-Hexylthiophene) Nanofibers for High-Performance OFETs. Acc. Chem. Res. 2017, 50, 932–942.

39.

Usov, I.; Mezzenga, R. FiberApp: An Open-Source Software for Tracking and Analyzing

ACS Paragon Plus Environment

25

Page 27 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Polymers, Filaments, Biomacromolecules, and Fibrous Objects. Macromolecules 2015, 48, 1269– 1280. 40.

Jordens, S.; Isa, L.; Usov, I.; Mezzenga, R. Non-Equilibrium Nature of Two-Dimensional Isotropic and Nematic Coexistence in Amyloid Fibrils at Liquid Interfaces. Nat. Commun. 2013, 4, 1917.

41.

Murphy, J. N.; Harris, K. D.; Buriak, J. M. Automated Defect and Correlation Length Analysis of Block Copolymer Thin Film Nanopatterns. PLoS One 2015, 10, 1–32.

42.

Kroon, D. J.; Slump, C. H.; Maal, T. J. J. Optimized Anisotropic Rotational Invariant Diffusion Scheme on Cone-Beam CT. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); 2010; Vol. 6363 LNCS, pp. 221–228.

43.

McCulloch, B.; Ho, V.; Hoarfrost, M.; Stanley, C.; Do, C.; Heller, W. T.; Segalman, R. a. Polymer Chain Shape of Poly(3-Alkylthiophenes) in Solution Using Small-Angle Neutron Scattering. Macromolecules 2013, 46, 1899–1907.

44.

Cheung, D. L.; McMahon, D. P.; Troisi, A. Computational Study of the Structure and ChargeTransfer Parameters in Low-Molecular-Mass P3HT. J. Phys. Chem. B 2009, 113, 9393–9401.

45.

Malik, S.; Nandi, A. K. Crystallization Mechanism of Regioregular poly(3-Alkyl Thiophene)s. J. Polym. Sci. Part B Polym. Phys. 2002, 40, 2073–2085.

46.

Arosio, P.; Beeg, M.; Nicoud, L.; Morbidelli, M. Time Evolution of Amyloid Fibril Length Distribution Described by a Population Balance Model. Chem. Eng. Sci. 2012, 78, 21–32.

47.

Korevaar, P. a.; De Greef, T. F. a; Meijer, E. W. Pathway Complexity in π-Conjugated Materials. Chem. Mater. 2014, 26, 576–586.

48.

Lan, Y.-K.; Huang, C.-I. Charge Mobility and Transport Behavior in the Ordered and Disordered States of the Regioregular poly(3-Hexylthiophene). J. Phys. Chem. B 2009, 113, 14555–14564.

49.

Zhao, B.; Zikry, M. A. The Effects of Structural Disorders and Microstructural Mechanisms on Semi-Crystalline P3HT Behavior. Polymer (Guildf). 2015, 57, 1–11.

50.

Zhao, Y.; Zhao, X.; Zang, Y.; Di, C.; Diao, Y.; Mei, J. Conjugation-Break Spacers in Semiconducting Polymers: Impact on Polymer Processability and Charge Transport Properties. Macromolecules 2015, 48, 2048–2053.

51.

McCulloch, I.; Heeney, M.; Bailey, C.; Genevicius, K.; Macdonald, I.; Shkunov, M.; Sparrowe, D.; Tierney, S.; Wagner, R.; Zhang, W.; Chabinyc, M. L.; Kline, R. J.; McGehee, M. D.; Toney, M. F. Liquid-Crystalline Semiconducting Polymers with High Charge-Carrier Mobility. Nat. Mater. 2006, 5, 328–333.

52.

Kleinhenz, N.; Rosu, C.; Chatterjee, S.; Chang, M.; Nayani, K.; Xue, Z.; Kim, E.; Middlebrooks, J.; Russo, P. S.; Park, J. O.; Srinivasarao, M.; Reichmanis, E. Liquid Crystalline Poly(3Hexylthiophene) Solutions Revisited: Role of Time-Dependent Self-Assembly. Chem. Mater. 2015, 27, 2687–2694.

ACS Paragon Plus Environment

26

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 28 of 30

53.

Chang, M.; Su, Z.; Egap, E. Alignment and Charge Transport of One-Dimensional Conjugated Polymer Nanowires in Insulating Polymer Blends. Macromolecules 2016, 49, 9449–9456.

54.

Li, J.-H.; Xi, Y.; Pozzo, L. D.; Xu, J.-T.; Luscombe, C. K. Macroscopically Aligned Nanowire Arrays of π-Conjugated Polymers via Shear-Enhanced Crystallization. J. Mater. Chem. C 2017, 5, 5128–5134.

55.

Skrypnychuk, V.; Wetzelaer, G.-J. A. H.; Gordiichuk, P. I.; Mannsfeld, S. C. B.; Herrmann, A.; Toney, M. F.; Barbero, D. R. Ultrahigh Mobility in an Organic Semiconductor by Vertical Chain Alignment. Adv. Mater. 2016, 28, 2359–2366.

56.

Wang, G.; Chu, P.-H.; Fu, B.; He, Z.; Kleinhenz, N.; Yuan, Z.; Mao, Y.; Wang, H.; Reichmanis, E. Conjugated Polymer Alignment: Synergisms Derived from Microfluidic Shear Design and UV Irradiation. ACS Appl. Mater. Interfaces 2016, 8, 24761–24772.

57.

Shaw, L.; Hayoz, P.; Diao, Y.; Reinspach, J. A.; To, J. W. F.; Toney, M. F.; Weitz, R. T.; Bao, Z. Direct Uniaxial Alignment of a Donor–Acceptor Semiconducting Polymer Using Single-Step Solution Shearing. ACS Appl. Mater. Interfaces 2016, 8, 9285–9296.

58.

Mukhija, D.; Solomon, M. J. Nematic Order in Suspensions of Colloidal Rods by Application of a Centrifugal Field. Soft Matter 2011, 7, 540–545.

59.

Papathanasiou, T. D.; Guell, D. C. Flow-Induced Alignment in Composite Materials; 1st ed.; Elsevier, 1997.

60.

Natarajan, B.; Lachman, N.; Lam, T.; Jacobs, D.; Long, C.; Zhao, M.; Wardle, B. L.; Sharma, R.; Liddle, J. A. The Evolution of Carbon Nanotube Network Structure in Unidirectional Nanocomposites Resolved by Quantitative Electron Tomography. ACS Nano 2015, 9, 6050–6058.

61.

Katz, E.; Yarin, A. L.; Salalha, W.; Zussman, E. Alignment and Self-Assembly of Elongated Micronsize Rods in Several Flow Fields. J. Appl. Phys. 2006, 100, 34313.

62.

Lee, P.; Lin, R.; Moon, J.; Lee, L. P. Microfluidic Alignment of Collagen Fibers for in Vitro Cell Culture. Biomed. Microdevices 2006, 8, 35–41.

63.

Himmelberger, S.; Dacuña, J.; Rivnay, J.; Jimison, L. H.; McCarthy-Ward, T.; Heeney, M.; McCulloch, I.; Toney, M. F.; Salleo, A. Effects of Confinement on Microstructure and Charge Transport in High Performance Semicrystalline Polymer Semiconductors. Adv. Funct. Mater. 2013, 23, 2091–2098.

64.

Jimison, L. H.; Toney, M. F.; McCulloch, I.; Heeney, M.; Salleo, A. Charge-Transport Anisotropy Due to Grain Boundaries in Directionally Crystallized Thin Films of Regioregular Poly(3-Hexylthiophene). Adv. Mater. 2009, 21, 1568–1572.

65.

Kwiatkowski, J. J.; Jimison, L. H.; Salleo, A.; Spakowitz, A. J. A Boltzmann-Weighted Hopping Model of Charge Transport in Organic Semicrystalline Films. J. Appl. Phys. 2011, 109, 1–6.

66.

Choi, D.; Kim, H.; Persson, N.; Chu, P. H.; Chang, M.; Kang, J. H.; Graham, S.; Reichmanis, E. Elastomer-Polymer Semiconductor Blends for High-Performance Stretchable Charge Transport Networks. Chem. Mater. 2016, 28, 1196–1204.

ACS Paragon Plus Environment

27

Page 29 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

67.

Kao, K. Y.; Lo, S. C.; Chen, H. L.; Chen, J. H.; Chen, S. A. Gelation of a Solution of poly(3Hexylthiophene) Greatly Retards Its Crystallization Rate in the Subsequently Cast Film. J. Phys. Chem. B 2014, 118, 14510–14518.

68.

Griffin, D. J.; Grover, M. A.; Kawajiri, Y.; Rousseau, R. W. Data-Driven Modeling and Dynamic Programming Applied to Batch Cooling Crystallization. Ind. Eng. Chem. Res. 2016, 55, 1361– 1372.

69.

DeCost, B. L.; Holm, E. A. A Computer Vision Approach for Automated Analysis and Classification of Microstructural Image Data. Comput. Mater. Sci. 2015, 110, 126–133.

70.

Rawat, W.; Wang, Z. Deep Convolutional Neural Networks for Image Classification: A Comprehensive Review. Neural Comput. 2017, 1–98.

ACS Paragon Plus Environment

28

ACS Applied Materials & Interfaces

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 30 of 30

TOC Graphic

ACS Paragon Plus Environment

29