**3.1. Design Strategies of Inhomogeneous Networks.** Typical crosslinking generates short strands, while the use of two catalysts results in a combination of short and long strands, as illustrated in Fig. 1a. In this study, designing catalysts with different catalytic activities is crucial for creating heterogeneous networks. IR is selected as the soft matrix. The primary factor in forming crosslinked networks in rubber matrices is the sulfur ring-opening.20,25–27 Zn2+ from ZnO withdraws electrons from sulfur, while nucleophiles provide electrons for sulfur, both contributing to the sulfur ring-opening, as illustrated in **Figure S1**.28–30 In our study, ZDC and D are used as catalysts for crosslinking. Molecular structure analysis reveals that ZDC contains both nucleophiles and Zn2+, while D only contains nucleophiles (Fig. 1b). We denote IR with ZDC as IR-ZDC and IR with D as IR-D. To track the crosslinking process, we used a GOTECH M-3000 AU vulcameter to monitor the torque as a function of time.31Crosslinking enhances interactions among polymer chains, leading to the higher torque value of polymer matrixes. Torque *versus* time plots at various temperatures (**Figure S2** and **Figure S5**), ln(*MH*-*Mt*) *versus* (*t*-*t0*) plots (**Figure S3** and **Figure S6**), and fitting curves (**Figure S4** and **Figure S7**) are obtained in Supporting Information. From these plots, the activation energy (*Ea*) of IR-ZDC is 85 kJ/mol and that of IR-D is 137 kJ/mol in Fig. 1b, revealing that ZDC exhibits higher catalytic activity than D. As a result, D with low catalytic activities leads to low crosslinking density (long strands) in IR-D and ZDC with high catalytic activities induces high crosslinking density (short strands) in IR-ZDC, as shown in **Figure S8**. These results suggest that using two catalysts, ZDC and D, creates an inhomogeneous network with both short and long strands.

To examine the features of inhomogeneous networks, we conduct DQ NMR and DSC analyses after crosslinking. DQ NMR has been proved to be an effective method to explore the structure of crosslinking networks.27,32,33 *D**re*s (residual dipolar couplings) is sensitive to topological constraints and closely related to the number of Kuhn segments. The molecular weight between crosslinks, abbreviated as *M**c*, can be estimated from *D**res* (*D**res* ~ *M**c**−1*). The normalized DQ curves of IR and IR-ZDC-D are shown in **Figure S9** and **Figure S10**, respectively. The distribution curve of *D**res*, corresponding to the distribution of crosslinking density, is depicted in Fig. 1c. The *D**re*s/2π distribution in Fig. 1c becomes wider with the use of two catalysts. Additionally, the results of thermoporosimetry are similar to those from DQ NMR. The melting behavior of the cyclohexane entrapped in networks can be used to characterize the heterogeneity of networks in Fig. 1d. The broad distribution of cyclohexane melting temperature corresponds to the broad distribution of crosslinking density.34 The addition of two catalysts results in a wide melting peak distribution for IR-ZDC-D, indicating a broad distribution of crosslinking density. This broad distribution is ascribed to the combination of long strands and short strands within a single crosslinking network.

To investigate the effect of networks heterogeneities, rather than crosslinking density, on properties, this work designs all samples with the same average crosslinking density of approximately 7500 g/mol calculated by the equilibrium swelling method, as shown in Fig. 1e. In the case of the same average crosslinking density, all samples have the similar tensile strength in Fig. 1f and *T**g* in **Figure S11**. Compared to regular networks, the inhomogeneous networks exhibit longer the strain at break in Fig. 1f. The stretch limit of strands (*λ**lim*) is calculated as \(\:{\lambda\:}_{lim}=\frac{Nb}{\sqrt{N}b}=\sqrt{N}\), where *b* is the length of each Kuhn monomer and *N* is the number of Kuhn monomers per strands. Long strands in inhomogeneous networks increase *N*, thus enlarging *λ**lim* of inhomogeneous networks. The presence of only one transition temperature in each DSC heat flow curve indicates that the formation of long strands and short strands does not lead to microphase separation. Long strands are expected to have low modulus, while short strands contribute to high modulus. In Fig. 1g, IR and IR-ZDC-D have the similar modulus, demonstrating that the modulus of samples is not compromised by long strands.

**3.2. Flaw Insensitivity of Inhomogeneous Networks.** The flaw sensitivity of networks is intricately linked to its crack propagation resistance.35,36 This work investigates the flaw sensitivity of inhomogeneous networks by uniaxially stretching IR and IR-ZDC-D samples with various precut notch width in Fig. 2a and Fig. 2b. In this part, all tested samples have overall dimensions of 100 mm in length, 15 mm in width, and 1 mm in thickness. In Fig. 2a, IR samples exhibit susceptibility to the propagation of existing flaws. As the flaw length increases, the stress at rupture decreases significantly from 11.3 MPa to 6.6 MPa, as shown in **Figure S12**. Figure 2b reveals that IR-ZDC-D samples are flaw-insensitive. For IR-ZDC-D with inhomogeneous networks, the strength of notched samples is comparable with that of unnotched samples in **Figure S13**. The flaw sensitivity of materials is strongly correlated with the fractocohesive length (*l**T*) which is defined as the ratio of *G**c* to the work of fracture.37,38 Such fractocohesive length can represent the stress transfer abilities of materials.39,40 To obtain the *l**T* of IR and IR-ZDC-D samples, this work first measures the energy density of the unnotched samples stretched up to catastrophic failure (the work of fracture) and *G**c*. *l**T* data of IR and IR-ZDC-D samples are exhibited in Fig. 2c. The *l**T* of inhomogeneous networks (1.73 mm) is 1.7 times more than that of regular networks (1 mm).

Mooney-Rivlin equation are used to perform an in-depth analysis of stress-strain curves for investigating the changes in entanglements (details in **Supporting Information**),41 \(\:\frac{\sigma\:}{\lambda\:-{\lambda\:}^{-2}}={E}_{c}+{E}_{e}f\left(\lambda\:\right)\), where *σ* is the engineering stress of materials, *E**c* represents the contribution from chemical crosslinking, and *E**e* is associated to physical topological constraints. Mooney-Rivlin curves of IR and IR-ZDC-D samples are shown in **Figure S14**-**Figure S17** and then we further calculate *E**e* (the contribution from physical topological constraints). Compared with regular networks, inhomogeneous networks have much more long strands. Long strands are often accompanied by much more entanglements.42 Thus, inhomogeneous networks exhibit higher *E**e* than regular networks in **Figure S18**. More entanglements of inhomogeneous networks are beneficial for transmitting tensions upon stretching.18,43–45

DIC techniques demonstrate the effective stress deconcentration of elastomers with long strands at the crack tip in Fig. 2d. For comparison, DIC techniques are also used to investigate the strain distribution of IR around crack tip. The apparent stress concentration phenomenon of IR appears at the crack tip. Therefore, compared with regular short strands, inhomogeneous networks are beneficial to transmitting tension. Such result is similar to the hydrogels with the unusually low amount of crosslinkers.18,19,46 When encountering cracks, long strands contribute to the stress deconcentration, as illustrated in Fig. 2e.

Additionally, we examine the impact of long strands and short strands on the crack propagation behaviors. Typically, crack propagation is driven by the release of stored elastic energy. The energy release rate (*G*) is calculated from the stress-stretch curves of unnotched samples of first cycle according to \(\:G=W\left(\lambda\:\right)·{H}_{0}\), where \(\:W\left(\lambda\:\right)\) is the stored elastic energy and *H**0* is the initial samples height. Under the *G* of approximately 800 J/m2, a camera records the crack length of IR and IR -ZDC-D under cyclic loads in Fig. 3a and Fig. 3b. For IR, cracks propagate obviously with the increase in the number of cycles in Fig. 3a. For IR-ZDC-D, we hardly observe the crack propagation after 100000th cycle in Fig. 3b. Also, SEM is used to investigate the surface of fatigue fracture samples in Fig. 3c. The fatigue fracture surface of IR is smooth, while that of IR-ZDC-D is rough. This rough fracture surface is ascribed to superior stress transfer capabilities from long strands.

**3.3. Tough and Fatigue-Resistant Inhomogeneous Networks.** This part demonstrates that inhomogeneous networks hinder the crack propagation under monotonic loads or cyclic loads in **Fig. 4**. These inhomogeneous networks exhibit near-perfect elasticity in **Fig. 4b**. Stress-strain curves have negligible hysteresis, indicating the absence of sacrificial bonds in inhomogeneous networks. The near-perfect elasticity is ascribed to short strands in the matrix. For fatigue experiments on notched samples, a pre-crack is cut using a razor blade. During fatigue experiments, we obtain crack length (*Δc*) versus number of cycles (*NC*) in steady state and further get the crack propagation rate (d*c*/d*NC*) of samples by calculating the slope of plots of *Δc* versus *NC*, as shown in **Figure S19**-**Figure S22**. The *G* is obtained from stress-strain curves of unnotched samples when the cyclic loading reaches the steady state at the corresponding strain. Subsequently, d*c*/d*NC* versus *G* curves are plotted in **Fig. 4a**, allowing us to approximately determine *Γ**0* below which the fatigue crack does not propagate under infinite cycles of loads. In **Fig. 4c**, the *Γ**0* of IR (regular networks) is around 67 J/m2, which is comparable to other elastomers such as chloroprene rubber, butadiene rubber, and polydimethylsiloxane.47–49 In terms of the classical Lake-Thomas model,5 the energy required to fracture a single layer of strands per unit area is directly proportional to the monomer number of strands in polymer physics. Regular networks with short strands have low *Γ**0*. With the formation of long strands in inhomogeneous networks, the stored elastic energy in strands increases and the energy required to fracture strands also increases, improving the crack propagation resistance under monotonic or cyclic loads.

Consequently, our proposed inhomogeneous networks enhance the crack propagation resistance by reducing stress concentration and enhancing stored elastic energy, which is fundamentally different from energy dissipation toughening networks. When a crack propagates, a single layer of strands based on the Lake-Thomas model should be ruptured. Long strands in inhomogeneous networks can effectively deconcentrate the stress around the crack tip and provide more elastic energy for the matrix, contributing to the crack propagation resistance. The ability to resist the crack propagation under monotonic loads is characterized by *G**c* and the ability to resist the crack propagation under cyclic loads is characterized by *Γ**0*.50 In **Fig. 4c** and **Fig. 4d**, both *G**c* and *Γ**0* increase with the formation of inhomogeneous networks. It is worth noting that IR-ZDC-D has achieved a high threshold of 1200 J/m2, one order of magnitude higher than existing elastomers in **Fig. 4e**. The concept of inhomogeneous networks is generic to soft matter systems. In fact, the present applied crosslinked elastomers typically feature a single crosslinking network, such as natural rubber and synthetic elastomers. Through the regulation of long strands and short strands, single crosslinking networks become crack propagation-resistant structure under external loads, which is widely applicable to network designs for various applications.

**Figure 4.** Tough and fatigue-resistant inhomogeneous networks. (a) d*c*/d*NC* versus *G* curves of IR and IR-ZDC-D samples. (b) Cyclic stress-stretch curves for IR-ZDC-D with no precut cracks. (c) *Γ**0* of IR and IR-ZDC-D samples. (d) *G**c* of IR and IR-ZDC-D samples. (e) *Γ**0* of various elastomers. Elastomers with inhomogeneous networks in the current work are compared with other elastomers, such as regular elastomers by single network design,5,17,18,51–53 double elastomers,54,55 and reinforced elastomers.56–58