We have recently developed transmission electron microscopy (TEM) techniques for observing structural deformation in stretched and sheared rubbers37,38,39. Moreover, an advanced imaging approach based on nanoscale electron diffraction (ED) for semicrystalline polymers has been established by Panova et al.40, and has since been applied to organic semiconductors41, polyolefins42,43, and other semicrystalline polymers44,45,46. This technique relies on specimen scanning with a nanoscale electron beam and ED pattern collection at different points to reconstruct two-dimensional crystal distribution maps from pattern intensities. These maps provide a spatial resolution nearly three orders of magnitude higher than that of m-SXRD, enabling the precise visualisation of nanocrystals in semi-crystalline polymers. Herein, we integrate two cutting-edge electron microscopy techniques, namely, in situ deformation TEM and nanoscale ED mapping, to elucidate the spatial arrangement of strain-induced crystallites in filled and unfilled rubbers, providing nanoscopic insights into the interplay between fillers and SIC and establishing correlations between microscopic phenomena and macroscopic mechanical properties.

IR filled with silica nanoparticles (20 parts per hundred rubber (phr)) and unfilled IR were prepared to investigate the interplay between internal structure and SIC behaviour. The stress-strain (S-S) curves of the two specimens revealed distinct mechanical behaviours (Fig. 1a). The unfilled IR showed a sharp stress rise at a tensile strain (ε) of ≈4, whereas the silica-filled IR exhibited a gradual stress increase and markedly higher rupture strength.

The tensile strain-dependent intensity of the 200 WAXD peak of PI crystals (Fig. 1b, c), which is proportional to crystallinity, revealed that the unfilled and silica-filled IR exhibited different crystallisation kinetics. The unfilled IR displayed a steep crystallinity increase starting from ε ≈ 3.5, whereas the silica-filled IR showed a more gradual rise initiating at a lower strain (ε ≈ 2.0). These differences are clearly visible in the original 2D-WAXD patterns, as shown in Supplementary Fig. 1. These crystallinity profiles well matched the stress increases observed in the S-S curves (Fig. 1a). Notably, the stress of the unfilled IR began to rise at ε ≈ 4 following the onset of the crystallinity increase at ε ≈ 3.5, which suggested that SIC considerably contributed to stress enhancement. Although WAXD provides valuable information on spatially averaged SIC behaviour, the understanding of SIC in filled systems remains phenomenological due to the lack of insights into the underlying microscopic mechanisms.

The TEM images of the unfilled IR were acquired during tensile deformation up to ε = 5.8, a value comparable with the failure strain observed in macroscopic tensile tests, and featured dark areas in the bright rubber matrix (Fig. 1d). These areas were identified as vulcanisation promoters, specifically ZnO particles and stearic acid aggregates. With the increasing strain, the additives progressively separated along the stretching direction. The realisation of such a large deformation (ε ≈ 6) suggested that the damage induced by the electron beam was minimal.

Scanning transmission electron microscopy (STEM) and nanoscale ED mapping were conducted to analyse the unfilled IR stretched to ε = 5.8 (Fig. 2a). Figure 2b shows an annular dark-field (ADF)-STEM image of the region highlighted in Fig. 1d (yellow rectangle). This image exhibits a contrast inverse to that of the corresponding TEM image, with the ZnO particles and stearic acid aggregates appearing as bright features.

Nanoscale ED patterns were acquired in 40 × 40 nm regions within the area shown in Fig. 2b (purple rectangle) and referred to as 40 nm ED patterns (Fig. 2c). Diffraction spots (yellow triangles) were observed in several of these patterns, although the low electron dose complicated the differentiation between these spots and background (BG) noise. The 200 nm ED pattern generated by integrating the twenty five 40 nm ED patterns in Fig. 2c exhibited pronounced diffraction spots corresponding to the {200}, {201}, {120}, and {002} planes of the PI crystals (200, 201, 120, and 002 spots, respectively; Fig. 2d) and was consistent with the integrated ED pattern obtained from the entire field of view in Fig. 2b (Fig. 2e), specifically from the rubber matrix region, excluding the stearic acid aggregates (Supplementary Fig. 2). The 200/201 and 120 spots cannot be simultaneously detected in any single PI crystal (Supplementary Fig. 3). Thus, the occurrence of these spots in the single 200 nm ED pattern indicated the coexistence of multiple crystallites with diverse orientations within 200 × 200 nm areas. In these areas, the c-axes of the PI crystallites (i.e., the PI chain directions) were mostly aligned parallel to the stretching direction while maintaining rotational variance around their corresponding c-axes.

Unlike the ED patterns integrated over broad regions, the individual 40 nm ED patterns (Fig. 2c) displayed 200, 201, or 120 spots. The fact that not all spots appeared simultaneously suggested that the number of crystallites within each 40 nm area was insufficient to incorporate all possible orientations. The number of crystallites within the measured volume () was estimated using the relationship , where represents the crystallinity at ε = 5.8 estimated by WAXD (≈7%), denotes the volume measured by ED (40 × 40 nm area with a specimen thickness of 60 nm at ε = 5.8), and corresponds to the dimensions of individual PI crystallites (5 × 5 × 10 nm, as previously reported). This calculation yielded  ≈ 27.

The average number of crystallites detectable via 200 spots in individual 40 nm ED patterns () was calculated as , where represents the probability of observing 200 spots in an ED pattern from PI crystallites randomly oriented around their c-axis as explained (≈2.5%, as detailed in Supplementary Fig. 3). Consequently, was calculated to be approximately 0.7. This result strongly suggested that each observed diffraction spot within these 40 nm ED patterns was due to a single crystallite.

The spatial distribution of the strain-induced crystallites was analysed using 40 nm ED patterns, which provided information on individual crystallites. To distinguish diffraction spots from BG intensity containing noise, two (BG and maximum intensity) histograms were constructed using a circular mask located at a magnitude of scattering vector (s, see Methods for definition) of 1.6 nm, corresponding to the s for 200 spots (inset, Fig. 2f). (i) The BG histogram was constructed from integrated intensities within the circular mask for each ED pattern, with measurements performed at 1° intervals over the angular range φ = -90° to -60° and 60° to 90° (Fig. 2f and Supplementary Fig. 4a). The BG histogram, representing the intensity distribution at positions devoid of the 200 spots, facilitated the discrimination between these spots and BG intensity containing noise in the ED patterns. (ii) The maximum intensity histogram was constructed from the highest integrated intensities within the circular mask for each ED pattern across the angular range φ = -90° to 90° (Fig. 2f and Supplementary Fig. 4a). This histogram characterised the intensities of 200 spots.

Crystallite positions exhibiting 200 spots were discriminated from BG intensity containing noise with 99.9% statistical confidence by selecting 40 nm ED patterns where the maximum intensities exceeded the 99.9th percentile of the BG histogram. The crystallites identified via 200 spots in 40 nm ED patterns were almost uniformly distributed across the field of view (Fig. 2g), which suggested the uniformity of environmental conditions throughout the specimens.

Figure 3 presents the TEM images of silica-filled IR recorded under tensile deformation up to ε = 5.3, showing dark silica nanoparticle aggregates dispersed in the bright rubber matrix. These images reveal that the aggregates stretched and/or split in the stretching direction under large deformations and were therefore correspondingly aligned.

Figure 4a presents the ADF-STEM image corresponding to the boxed area in the TEM image of the filled IR captured at ε = 5.3 (Fig. 3), with the silica aggregates appearing white in the dark rubber matrix. This image exhibits an alternating arrangement of silica-rich and silica-poor columns parallel to the stretching direction. This is the first study to uncover such columnar arrangements under large deformation conditions (up to ε = 5.3).

Furthermore, the rubber regions between these stretch-aligned aggregates exhibit a bright contrast in the ADF-STEM image, suggesting that the rubber in these regions has either a higher density or larger thickness. This structure may arise from the SIC behaviour and inhomogeneous strain/stress distributions, as discussed below.

Figure 4b, c shows the 40 nm ED patterns acquired in regions 1 and 2 of Fig. 4a, respectively. The ED patterns in region 1 exhibit the 200, 201, and 120 spots of PI crystallites, whereas those in region 2 show few diffraction spots. This contrast indicates the coexistence of crystallite-rich and crystallite-poor domains within the stretched specimen. The integrated 200 nm ED patterns acquired in regions 1 and 2 (Fig. 4d, e, respectively) corroborate this heterogeneous crystallite distribution. Such a heterogeneous distribution is unrecognisable in the average ED pattern acquired across the entire rubber region (Fig. 4f).

Figure 4g illustrates the spatial distribution of crystallites determined from the intensities of the 200 spots in 40 nm ED patterns (Supplementary Fig. 4). Locations with 200 spot intensities exceeding the 99.9th percentile of BG intensities are highlighted in the ADF-STEM images. Strain-induced crystallites preferentially formed along the silica-rich columns, whereas silica-poor columns exhibited fewer crystallites. A similar distribution pattern was obtained using 120 spot intensities (Supplementary Fig. 6), corroborating the results of the 200-spot-based analysis (Fig. 4g). The presence of both the 200 and 120 spots in the silica-rich columns suggested that crystallites with various c-axis orientations coexisted in these regions.

Local strain distributions were analysed by tracking characteristic points in the TEM images acquired at ε = 0.8-5.3 (Fig. 3), as reported previously. The maps of local maximum and minimum principal strains (ε and ε, respectively), as shown in Fig. 5b and Supplementary Fig. 7b, indicate a non-uniform nanoscale deformation behaviour, which is attributed to the heterogeneous distribution of silica aggregates. Crystallite locations were overlaid on these maps, as shown in Fig. 5e and Supplementary Fig. 7c. Furthermore, the histogram in Supplementary Fig. 7a illustrates the quantitative correlation between the distributions of local ε and crystallites. This quantitative analysis revealed that the local ε at locations where strain-induced crystallites exist tends to be smaller than the average strain, suggesting that regions with dense strain-induced crystallites become less deformable than those with sparse crystallites due to an increased local elastic modulus.

Subsequently, to estimate local stress distribution in the stretched silica-filled IR, we performed finite element method (FEM) simulations of tensile deformation based on the spatial distribution of silica aggregates in the TEM image at ε = 5.3. The map of the local normal stress in the stretching direction (Fig. 5c) indicated that substantial stress propagated along the silica aggregates aligned in this direction. This stress distribution was well correlated with the observed crystallite distribution (Fig. 5f). Thus, the aligned silica aggregates were concluded to enhance local stress in interstitial rubber regions, promoting site-specific crystallite formation. These crystallites, in turn, reinforced the columnar arrangements of the silica aggregates, serving as principal stress propagation pathways.

In the unfilled IR with a relatively uniform environment, SIC begins uniformly throughout the specimen at a threshold of ε ≈ 3.5, and the crystallinity steeply increases beyond this point (Fig. 1b), which results in a sharp stress increase (Fig. 1a).

In the silica-filled IR, silica aggregates align to form silica-rich columns upon stretching, which serve as stress propagation pathways by bearing substantial stress. Site-specific SIC occurs along the aligned silica columns, enhancing the elastic moduli of the stress propagation pathways, as illustrated in Fig. 6. This localised reinforcement increases the macroscopic elastic modulus and rupture strength (Fig. 1a). The earlier onset of SIC (ε ≈ 2.0) compared with that in the unfilled IR (ε ≈ 3.5) and gradual increase in crystallinity (Fig. 1b) are attributed to the substantial heterogeneity of local strain.

Our findings reveal fundamental differences in the reinforcement mechanisms of the unfilled and filled systems. The unfilled systems exhibit uniform self-reinforcement via SIC, whereas filled systems demonstrate preferential SIC reinforcement along filler-mediated stress propagation pathways (Fig. 6). This study highlights the crucial role of filler arrangement in controlling the mechanical properties of crystallisable elastomers, thereby providing essential insights for material design.