Percolative Instabilities and Sparse-Limit Fractality in 1T-TaS 2 Poulomi Maji, Md Aquib Molla, Koushik Dey, Bikash Das, Sambit Choudhury, Tanima Kundu, Pabitra Kumar Hazra, Mainak Palit, Sujan Maity, Bipul Karmakar, Kai Rossnagel, Sanjoy Kr Mahatha, Bhaskaran Muralidharan, Shamashis Sengupta, Sanchari Goswami, Subhadeep Datta Abstract The low-temperature metallic phase of 1T-TaS 2 may originate from current- and voltage-driven destabilization of the commensurate charge density wave (CDW) in a strongly correlated Mott-insulator, alongside the robust yet rarely realized influence of intrinsic electronic distortions. Electrical pulse-driven transport, combined with second harmonic response, reveals abrupt switching, negative differential resistance (NDR), and multiscale domain-wall reorganization. The free energy analysis identifies a critical order parameter threshold for the Mott–metal transition, with scaling exponents ( β ≈ 1.3 \beta\!\approx\!1.3 ) consistent with 2D percolation. The sparse limit fractal dimension D f ≈ 0.3 D_{f}\!\approx\!0.3 at 10 K, rising to ≈ 0.9 \approx\!0.9 at 300 K, reflects the hierarchical evolution of the conductive pathways throughout the temperature. These findings establish a direct connection between fractal percolation, pulse-induced instabilities, and correlated electron transport, offering a framework for controlled access to non-equilibrium phase transitions in low-dimensional quantum materials. I Introduction Electron-phonon interactions in transition metal dichalcogenides (TMDCs) are fundamental drivers of phase complexity, particularly through charge density wave (CDW) formation 57 , 96 . In metallic TMDCs, CDWs modulate the electron density ( \rho(r) ) periodically, coupling directly to lattice distortions 61 . This intricate interplay produces a rich phase landscape with transitions across various commensurate and incommensurate phases, each stabilized by electron-phonon coupling ( g ). A comprehensive theoretical model for CDW behavior in TMDCs, however, remains incomplete. CDW formation is often described by Peierls instability, where lattice distortions ( u(r) ) lower electronic energy when 2k_{F}=q , with the electron-phonon coupling represented by the Hamiltonian term H_{\text{CDW}}=g\sum_{r}(\rho(r)-\langle\rho(r)\rangle)u(r) 55 , 54 , 19 , 63 . Additional mechanisms like the Kohn anomaly and Fermi surface nesting have further refined this picture, but they do not yet encapsulate the full phenomenology observed in TMDCs 78 , 27 , 32 , 66 , 2 , 71 . On the transport side, Bardeen’s quantum tunneling model suggests that CDW conduction can proceed through tunneling of the condensate across impurity-induced gaps 4 , while classical approaches treat depinning using a sinusoidal pinning potential and a single CDW phase degree of freedom. These models predict distinct nonlinear responses in voltage- versus current-driven regimes: voltage bias leads to a discontinuous rise in differential resistance beyond threshold, whereas current bias results in negative differential resistance (NDR) 24 , 81 . Such NDR, prominent in NbSe 3 and TaS 3 , marks the threshold for abrupt collective CDW motion. However, these models lack an integrated understanding of criticality, fractal conductivity, and percolative instabilities intrinsic to domain evolution under pulses. 1T-TaS 2 , a layered TMDC, hosts intertwined CDW and Mott insulating (MI) phases 80 , 13 . Below 180 K, it stabilizes a commensurate CDW (C-CDW) with a \sqrt{13}\times\sqrt{13} “Star of David” (SOD) superlattice 47 , 94 , 86 , 74 , 84 , 65 , accompanied by a MI state associated with “AL” stacking 89 , 87 , 58 . The nearly degenerate stacking of “L” promotes a metallic phase (ML), allowing co-existence and phase percolation 39 , 89 , 64 . These interlayer arrangements modulate electronic bandwidth W and its competition with Coulomb repulsion U , determining the emergent transport properties. External stimuli – optical pulses, gating, or dimensional confinement – can tune interlayer hopping t_{\perp} to overcome U , triggering the MI–ML transformation, which has been extensively investigated via local transport techniques such as scanning tunneling spectroscopy (STS) 41 , 74 , 76 , 83 , 30 , 88 , 25 , 95 , 93 , 83 . Unlike smooth evolution in disordered films 16 , the transition in 1T-TaS 2 involves current-driven reorganization of the CDW domain, leading to abrupt, pulse-induced switching and NDR 82 , 93 , 28 . The emergence of sparse, metallic filaments follows a percolation-like dynamics governed by fractal geometry, local Joule heating, and electric-field focusing. The resistance scaling behavior is consistent with the modified Archie’s law and the fractal connectivity models 75 , 68 , 60 , revealing a new class of metastable, spatially inhomogeneous transport states. Despite the rich experimental observations, a quantitative framework describing the nonthermal fractal percolation kinetics and threshold-driven conductivity evolution in 1T-TaS 2 remains underexplored. This motivates our pulse-based approach, where we track transport transformations via R – T , I – V , and second harmonic response under varying current bias, providing insights into bistability, scaling behavior, and evolution of the order parameter throughout the MI–ML transition. We report on transport in exfoliated and bulk 1T-TaS 2 flakes, showing an anomalous resistance drop near 100 K, deviating from conventional Mott insulating behavior. This transition, robust across thicknesses, suggests a stacking-induced instability that drives the system toward a metal-like (ML) state. Voltage-current ( V-I ) measurements reveal negative differential resistance (NDR) from 10–300 K, with threshold current I_{\mathrm{th}}(T) increasing up to 150 K, then decreasing, indicating current-driven percolative domain evolution. Short voltage or current pulses induce irreversible switching from the insulating towards ML state, highlighting the role of local Joule heating and field concentration near weak links. To describe the ML state’s growth, we construct a phenomenological free energy F[\phi] using a metallic order parameter \phi , representing volume fraction of conducting channels, with 0\leq\phi\leq 1 , whose spatial profile evolves under both diffusive and gradient-driven terms. The interface width between domains depends on the ratio of diffusive strength to free energy curvature, allowing smooth transition to a percolative, conductive state as current increases. As \phi\rightarrow 1 , sparse conductive backbones grow and merge, forming fractal pathways that abruptly reduce resistance beyond I_{\mathrm{th}} . The system’s bistability and nonlinearity emerge from this nonequilibrium, current-driven reorganization of electronic textures. These findings offer a unified framework for understanding current-induced insulator-to-metal transitions in correlated layered materials through spatially evolved percolation. II Result and Discussion 1T-TaS 2 undergoes a series of charge-density-wave (CDW) transitions upon cooling, evolving from a metallic phase into an incommensurate CDW below \sim 550 K and a nearly commensurate CDW (NC-CDW) below \sim 350 K. In the NC-CDW state, the lattice distortion condenses into hexagonal domains of Star-of-David (SoD) clusters (see Fig. S1(b)) 86 , 73 , 80 . Each SoD consists of 13 Ta atoms, where 12 peripheral atoms contract toward a central Ta, producing three inequivalent Ta sites. Upon further cooling below \sim 220 K, these domains order into a long-range \sqrt{13}\times\sqrt{13} superstructure, characteristic of the commensurate CDW (C-CDW) phase, in which only the central Ta atom hosts an unpaired 5d^{1} electron (see Fig. S2(d)) 47 , 70 , 86 , 74 , 84 , 65 . The NC-CDW to C-CDW transition is clearly observed in Fig. 1(a) for both devices near \sim 220 K for the heating cycle, and is further corroborated in Fig. S3(b). The pronounced thermal hysteresis between the heating and cooling cycles in Fig. S3(b) reflects the first-order nature of this transition 94 , 80 . Although XPS does not resolve a clear distinction between the inequivalent B and C-type Ta atoms, the charge disproportionation between the central (A-type) and peripheral sites is manifested by a splitting of the Ta 4f_{7/2} and 4f_{5/2} core levels below 350 K (see Fig. S1) 69 , 70 , 87 , 36 . The substantial electronic overlap among the peripheral Ta atoms is further supported by the calculated electron localization function of the supercell (Fig. S2). Below the C-CDW transition, the electronic ground state is governed by the out-of-plane stacking of the SoD clusters 64 , 89 , 58 . In the AL configuration, vertical alignment of the central Ta atoms enhances the interlayer Coulomb repulsion U and stabilizes a correlated Mott insulating state (Fig. S1(b)) 89 , 87 , 58 . Deviations from this stacking destabilize the Mott phase and introduce metastability 82 , 93 , 28 . Moreover, scanning tunneling microscopy (STM) studies further show that voltage pulses can locally convert these insulating domains into more conducting regions by modifying the stacking order 41 , 77 , 74 . First-principles calculations indicate that AL and L stackings are nearly degenerate, with the L stacking higher in energy by only \sim 1.1 meV per SoD 39 , 89 , 64 . In the L configuration, lateral displacement of the clusters enhances both in-plane and interlayer hopping, yielding a metallic state 39 , 70 , 56 , 89 . This near-degeneracy provides a natural route to the Mott–metal transition observed under optical and ionic-liquid gating 74 , 76 , 83 , 30 , 88 , 25 , 95 , 93 , 83 . In our measurements (Fig. 1), signatures of Mott–metal coexistence emerge below \sim 100 K. While one device exhibits a sharp resistance drop consistent with an L-stacking-dominated metallic phase (indicated in black), another (indicated in red) shows a resistivity dip between 50–100 K accompanied by a broad, Gaussian-like second-harmonic resistance peak. In contrast to the sharp peak associated with the NC–CDW to C–CDW transition, this broadened second-harmonic response signals the low-temperature coexistence of competing Mott insulating and metallic phases, reflecting a continuous crossover from a mixed AL–L regime toward the AL-stacked Mott insulator, as shown in Fig. 1(a) inset and Fig. S4(a) 70 . Thus, the low-temperature transport reflects the competitive coexistence of metallic and Mott-insulating tendencies, with the metallic resistance smoothly connecting to the NC-CDW state, consistent with a supercooled NC-CDW–derived metal 93 , 92 , 91 . Thus the low temperature resistance ( R_{total} ) can be modeled as a mixture of metallic ( R_{ML} \sim{T^{2}} ) and Mott-like behavior ( R_{MI} \sim e^{\Delta/k_{B}T} ) 41 : R_{\text{total}}=\phi(\alpha+\beta T^{2})+(1-\phi)e^{\Delta/k_{B}T}, (1) presented in Fig. S4(b), where \phi is the ML fraction and \Delta is the Mott gap. Notably, this metallicity appeared spontaneously, without external perturbation. Previous studies reported that low-temperature metallicity in 1T-TaS 2 appears only in flakes thinner than \sim 31 nm, attributed to enhanced metastability at reduced thickness 94 , 93 , 95 . In contrast, our measurements on 30 flakes reveal that metallicity also occurs in much thicker samples, up to 60 nm. Statistically, \sim 5\% of flakes with thickness 40–60 nm and \sim 20\% with thickness 5–20 nm are metallic, while \sim 15\% of flakes in the 15–25 nm range remain Mott insulating to the lowest temperatures (see supplementary Information Sec. III, atomic force microscopy images with height of two such devices is presented in Fig. S3(a), 70 ). These results demonstrate that the low-temperature ground state is governed primarily by the out-of-plane stacking of the C-CDW Star-of-David structure rather than by thickness alone. A similar thickness dependence is reported in 2H-TaS 2 , where the IC-CDW stabilizes at higher temperatures in thin flakes due to enhanced unit-cell lattice instabilities 62 . Overall, while dimensional confinement modifies the competition between electronic and structural phases, in 1T-TaS 2 the ground state is predominantly controlled by stacking configuration. As discussed, spatially resolved studies of 1T-TaS 2 reveal heterogeneous domain textures, indicating that transport near the Mott transition proceeds through percolative metallic backbones embedded in an insulating matrix 41 , 77 , 74 . Similar behavior has been extensively studied in disordered Hubbard models, where interaction-driven metal–insulator transitions give rise to electronic phase separation, quantum percolation, and inhomogeneous superconductivity, with numerical studies revealing fractal metallic or superconducting clusters that govern transport through sparse connectivity 67 , 85 , 79 , 26 , 21 . While such Hubbard-type and disorder-based frameworks successfully capture equilibrium phase coexistence and percolative transport, they are typically restricted to equilibrium or disorder-averaged descriptions and do not incorporate controlled nonequilibrium driving, such as current or voltage pulses 79 , 26 , 21 . Consequently, the dynamical reshaping of phase stability under external bias remains outside their scope. Also, previous phenomenological descriptions of charge-density-wave (CDW) systems have largely employed a complex order parameter with well-defined amplitude and phase, successfully capturing equilibrium CDW phases, stiffness, and collective excitations 48 , 43 , 49 , 51 , 15 . In contrast, the present work does not focus on the microscopic dynamics of the CDW amplitude or phase, but instead addresses the mesoscale coexistence and competition between two macroscopically distinct electronic states. Therefore, we adopt a phenomenological free-energy framework in which the order parameter is defined as the local volume fraction, directly encoding the spatial distribution of the competing phases. We considered, F[\phi]=\phi^{2}(1-\phi)^{2}=\phi^{2}-2\phi^{3}+\phi^{4}, (2) The free energy F(\phi) represents a double-well potential with minima at \phi=0 (Mott insulating) and \phi=1 (metallic) and a maximum at \phi=0.5 [Fig. 1(b)], where the order parameter \phi=\text{metallic fraction}/(\text{metallic fraction}+\text{Mott fraction}) is bounded within 0\leq\phi\leq 1 . A positive coefficient of the highest-order term ensures thermodynamic stability and a finite equilibrium value of \phi . An applied current induces a transition from the localized Mott state to the delocalized metallic state through Joule heating without breaking global translational symmetry. Because \phi is a non-negative physical fraction, no symmetry requires invariance under \phi\rightarrow-\phi ; consequently, the free energy is not constrained to satisfy F(\phi)=F(-\phi) , and odd-order terms of \phi is allowed in the phenomenological Landau expansion 29 , 10 , 3 . Under an applied current bias, we phenomenologically extend the free-energy functional as F(\phi,I)=\phi^{2}(1-\phi)^{2}+aI\phi-bI^{2}\phi^{2}, (3) where the additional terms represent the leading-order coupling between the electronic state and the external perturbation. In this minimal model, the linear term describes the current-induced stabilization of the insulating configuration in an intermediate bias range, which may arise from enhanced carrier scattering at domain boundaries, increased CDW pinning, and the associated thermal bottleneck preceding collective CDW motion. This scenario is consistent with earlier reports of electric-field-induced CDW stiffening and delayed sliding 45 , 38 , 17 , 5 , and is schematically illustrated in Fig. S5(a). Where for I = 0.5, the Mott state is becoming more favorable, further increasing I to 1.2, take the system towards to metallic state ( I is a control parameter (in arbitrary units) representing the perturbation, here analogous to the applied current). The quadratic term represents the effect of Joule heating across domains ( b\propto Rt , with b>0 ), which promotes domain melting and CDW sliding toward the metallic minimum. Within this phenomenological framework, above a threshold current I_{\mathrm{th}} the metallic state becomes energetically favorable, i.e., F(\phi=1)\leq F(\phi=0) , yielding I_{\mathrm{th}}\propto a/b [inset of Fig. 1(b)]. For simplicity, a symmetric unperturbed double-well potential is assumed, although in practice the Mott-insulating state is slightly more stable than the metallic state 39 . This model provides a compact phenomenological framework that we use in the following to interpret our experimentally observed current-driven evolution between the Mott-insulating and metallic phases. Figure 2(a) displays the current-induced voltage characteristics of another Mott device. At T=10 K, the initial bias cycles ( n=1-9 ) exhibit a nearly linear response, followed by a discrete resistance jump around 2.2 mA, equivalent to free energy modification for current parameter increase up-to I=0.5 (see Fig. S5(a)), where the Mott state becomes more favorable, due to scattering at domain boundary before sliding, pinned the CDW state 45 , 38 , 17 , 5 it corresponds to a>b situation analogous with a/b=2 in Fig. 1(b) inset. From n=12 onward, a progressive decrease in resistance is observed and a negative differential resistance (NDR) region emerges beyond 2.5 mA due to increase value of b ( \propto Rt ). By n=18 , a pronounced NDR appears beyond 6 mA, culminating in a sharp upturn near 40 mA, indicative of a metallic-like regime (see inset). Due to Joule heating accumulated over cycles, the third term (joule heating) becomes more dominant than the scattering contribution, effectively corresponding to a current parameter I>0.5 analogous with a/b=0.5 (see Fig. S5(a), Fig. 1(b) inset) 31 , 40 , 8 , 44 , 90 , 46 . As a result, the metallic state becomes more favorable, leading to a sharp drop in resistance accompanied by NDR. By cycle n=18 , the increased sweep corresponds to a higher effective I , requiring a slightly larger current (6 mA) to induce NDR due to the enhanced scattering term 45 , 38 , 17 , 5 . Beyond 40 mA, the nearly linear V – I behavior indicates complete dominance of the Joule heating term and a transition of the system toward \phi=1 , corresponding to a predominantly metallic, channel-dominated state. For a fixed current value, with increase in cycle n , b will keep on increasing ultimately inducing transition towards \phi=1 state, but might require a much larger value of n . In contrast, device 1 presented in Fig. 1(a) (indicated in red), subjected to a limited current sweep ( \pm 50 \mu A), exhibits no sign of NDR, but the consistent presence of sudden increase in resistance observed within the explored range suggests I<0.5 , where scattering term completely dominates ( a>>b ), suggesting requirement of higher joule heating to induce NDR (see Fig. S6(a), supplementary section V, 70 ) 45 , 38 , 17 , 5 . However, a pristine Mott state ( \phi=0 ) which is deep in the \phi=0 valley, may not evolve into a metallic state unless the current exceeds a high threshold value I_{th}\propto 1/\phi , which risks damaging the device, observed in several devices. The behavior of the device in Fig. 2(a) is particularly noteworthy: its Mott-like response along with moderate initial resistance at 10 ~K ( R=800\ \Omega ) (see Fig. S5(b)) suggests an initial intermediate order parameter ( \phi\approx 0.5 ), enabling a feasible Mott-to-metallic evolution by applying current (Device fabrication is discussed in the supplementary section VIII 70 ). Overall, for a Joule-heating-driven Mott–metal transition to be feasible, the system must lie near \phi=0.5 in the free-energy landscape. Hence, the applied current navigates the system along three distinct trajectories: a regime with \frac{dV}{dI}>0 (yellow) corresponds to the fragmentation of domains into coexisting insulating and conductive regions; a regime with \frac{dV}{dI}<0 (pink) indicates a percolative dominance of conductive domains, leading to a reduction in overall resistance; and a regime with \frac{dV}{dI}=0 (green) marks a dynamically balanced configuration characterized by uniformly distributed resistive phases, presented in Fig. S6(b) 70 . Among the three available path which one system should follow depends on the optimal channel domain configuration which lead to least power dissipation among all admissible paths 59 , 35 , 22 , 7 . These regimes reflect current-driven non-equilibrium phase transitions mediated through real-space domain reorganization, revealing the system’s sensitivity to external bias and highlighting the intricate interplay between electronic correlations and structural inhomogeneities 81 , 37 , 45 (see supplementary section V 70 ). A similar protocol was applied to the low-temperature metal-like (ML) state of device 2 (indicated in black in Fig. 1(a)), with current swept from -60 mA to 60 mA in 1 mA steps (Fig. 2(b)). The black curve corresponding to 10 K temperature, shows a resistance drop from 176 \Omega to 58 \Omega , followed by pronounced NDR beyond 12 mA corresponding to transition towards \phi=1 state driven by dominance of joule heating term over the scattering term ( b>a ), in the very first sweep. Since the low-temperature metallic state implies a reduced population of insulating domains, scattering at CDW boundaries is significantly weaker than in the preceding regime. Rapid CDW domain reorganization (10–100 ps) lowers resistive energy, yielding a threefold drop and robust NDR 24 , 81 . Unlike the R – T behavior, where the resistance increases with T , a current bias drives a steady-state breakdown of residual Mott-insulating domains within the metallic matrix. The system thereby follows a nonequilibrium pathway characterized by \frac{dV}{dI}<0 , distinct from the equilibrium trajectory. This behavior may be understood in the spirit of the minimum power dissipation principle, which states that a system driven slightly out of equilibrium, can reorganize its internal pathways so as to reduce the overall rate of energy dissipation 59 , 35 , 22 , 7 . This problem has been addressed by Onsager and Prigogine, who showed that this holds true for a network of linear resistors with the constraints of current and voltage biasing respectively 59 , 35 , 22 , 7 . In our experiments, we encounter a more complex problem where the resistances themselves can vary with the application of a current due to phase transition physics 59 , 35 , 22 , 7 . While a concrete proof of the principle in such a case in not known, it is pertinent to check experimentally what are the limits of validity for the principle of least power dissipation in this case. Unlike classical CDW systems (NbSe 3 , TaS 3 ), NDR here spans the full temperature range and depends sensitively on the initial low- T state. Following 72 , NDR emerges when \frac{dV}{dI}=\frac{1}{\sigma_{CDW}+\sigma_{N}}-\frac{I\sigma^{\prime}_{CDW}}{(\sigma_{CDW}+\sigma_{N})^{2}}\leq 0, which requires \sigma^{\prime}_{CDW}>(\sigma_{CDW}+\sigma_{N})/I , satisfied under sudden domain fragmentation. By the principle of minimum power dissipation, this drives the system into a low-dissipation, NDR phase. At 100 K, resistance drops from 176 \Omega (10 K) to 78 \Omega , then to 28 \Omega via NDR (Fig. 2(b)). Threshold current, I_{\text{th}} , increases with T up to 150 K, then sharply decreases above 220 K (Fig. S7). NDR steps and sharpness reduce with T , while threshold power rises linearly till 100 K and drops thereafter (Fig. 2(b), inset). Beyond I_{\text{th}} , background MI domains fragment, forming percolative channels responsible for NDR. Pre-formed channels at higher T require higher current/power for fragmentation, explaining the rise in I_{\text{th}} up to 200 K. Upto 150 K the continuous decrese in voltage beyond NDR indicates, \phi corresponding to the system lies in between 0.5 and 1. Above this temperature, the system is more metallic ( \phi\to 1 ), needing less drive to disrupt remaining MI patches. As \sigma_{\text{CDW}} varies less with T , NDR weakens. At 250 K, voltage rises beyond 53 mA, deviating from the 200 K trend, marking \phi approaching towards 1; at 300 K, full fragmentation yields dI/dV>0 , indicating a mostly channel-dominated regime ( \phi\sim 1 ) (see Fig. S8(b), Fig. S8(c), all measurements details are in supplementary section VIII 70 ). A simpler resistive model is presented in the supplementary section VI to explain the observed NDR in the metallic system. For both the MI and ML states, the threshold power ( P_{\text{th}} ) remains in the microwatt range, insufficient to reach the transition temperature from C-CDW to NC-CDW, consistent with the heat balance relation P=\gamma\Omega(T_{e}^{6}-T_{ph}^{6}) , where \gamma is the electron–phonon coupling strength in \mathrm{W\,K^{-6}\,m^{-3}} , \Omega is sample volume 53 , 33 , 20 . For disordered insulators, \gamma=1.85~\mathrm{nW\,\mu m^{-3}\,K^{-6}} ; 1T-TaS 2 is known to exhibit stronger coupling, giving \gamma\sim 10^{7} – 10^{8}~\mathrm{W\,K^{-6}\,m^{-3}} . Using \Omega=40~\mu\mathrm{m}\times 10~\mathrm{nm}\times 40~\mu\mathrm{m} , the estimated T_{e} is only \sim 30 K, far below the NC-CDW transition, but sufficient to trigger thermal runaway driven by local joule heating and domain avalanches (see supplementary section VII 70 ) 37 , 14 , 1 , 8 , 11 . In the above discussion, the response of the device was governed by its position in the free-energy landscape. Figure 3(a) shows V – I characteristics of two devices with similar low- T Mott insulating states ( R>1.5 k \Omega ; inset), yet differing 50–150 K dips. The red curve ( \phi\sim 0.5 ) shows a higher dip and requires 2.5 mA for resistance surge, while the yellow curve ( \phi<0.5 ) requires only 20 \mu A but both fails to show NDR. Clearly red curve suggest more metallic fraction in the system than yellow one, as a result required larger critical current (2.5 mA) (I c ) (I c represents the current corresponding to sudden increase in resistance due to dominance of scattering term) than yellow curve, since scattering at CDW domain boundaries is weaker in the metallic state with sparse domains, hence a larger current ( I_{c} ) is required to trigger the resistance surge, where the scattering contribution becomes dominant. Another way to drive the system towards the metallic fraction limit \phi=1 is through external perturbations such as voltage pulses, which have previously shown local effects; here, however, we observe their macroscopic signature. A 2 V, 100 kHz voltage pulse applied for 60 s to a Mott device results in a sharper resistance peak in the V – I curve that shifts towards a higher critical current, indicating an increase in metallic fraction. Further in Fig. 3(b) (inset), increasing current from 3 \mu A to 30 \mu A deepens the resistance dip in the 50–150 K range, consistent with current-induced enhancement of the metallic volume fraction, i.e., \phi\to 1 . The observed changes reflect a change in the position of the system along a rugged free energy landscape where local perturbations re-normalize domain configurations. The melting of Mott insulating domains in 1T-TaS 2 is inherently spatiotemporal and inhomogeneous, occurring via nucleation, growth, and coalescence of metallic regions rather than uniform collapse 41 , 77 , 74 , 76 , 83 , 30 , 88 , 25 , 12 . Previous imaging studies have visualized such dynamic domain rearrangements, showing percolating metallic pathways within the insulating matrix 41 , 77 , 74 , 12 , while our macroscopic transport measurements reveal stochastic resistive jumps and hysteresis under current or voltage drive, consistent with nonuniform domain dynamics at larger scales. To capture this multiscale behavior, we employ a time-dependent Ginzburg–Landau (TDGL) framework and define \phi(x,t) as the local metallic fraction ( 0\leq\phi\leq 1 ), which evolves under applied current driving the system out of equilibrium 49 , 48 , 43 . Its dynamics are governed by, \frac{\partial\phi(x,t)}{\partial t}=-\Gamma\frac{\delta F}{\delta\phi}+\eta(x,t) , where F is the free-energy functional, \Gamma is a kinetic coefficient, and \eta(x,t) represents stochastic fluctuations 48 , 43 . To further account for the nonlinear and stochastic spatiotemporal evolution of domain interfaces, we incorporate a Kardar–Parisi–Zhang (KPZ)–type description 18 , 50 , 23 , 9 . Because the current-driven Mott–metal transition involves nucleation and lateral growth of metallic domains within a disordered CDW background, the resulting dynamics naturally falls into the KPZ universality class, independent of microscopic details, providing a unified description of electrically driven domain evolution and macroscopic transport. The KPZ equation gives, \partial\phi(x,t)/\partial t=\nu\nabla^{2}\phi(x,t)+2\lambda(\nabla\phi(x,t))^{2}+\eta(x,t) , captures the nonlinear growth dynamics of channels in the presence of perturbations, where \nu is a diffusion-like coefficient promoting spatial smoothing, \lambda quantifies the strength of nonlinear coupling between gradients, and \eta(x,t) represents a stochastic Gaussian white noise source. Combining the two yields the generalized evolution equation: -m\Gamma\,\delta F/\delta\phi=\nu\nabla^{2}\phi(x,t)+2\lambda(\nabla\phi(x,t)^{2}+\eta(x,t) , where m is Lagrangian multiplier. Due to application of current, and subsequently setting it zero, the system is initially driven to non equilibrium phase with order parameter having spatiotemporal dependence. Here, the interaction range p between neighboring domains is much smaller than the device dimension l (p\ll l) . Consequently, within the device-length scale probed in our measurements, the spatiotemporal evolution of the system can be described using a continuum, coarse-grained framework. Since the measured response averages over a large number of domains, the order parameter defined as the local domain fraction varies smoothly on the relevant length scales 34 , 52 . So magnitude of the wave vector of spatial variation has a range of cut off like k =2 \pi/l . In this limiting case, higher order terms of spatial derivatives of order parameters (second term of KPZ equation) is neglected and also neglected the noise term, approximating \phi near zero, higher order of \phi term has been neglected, then the combined equation for our case deduce to \nu\nabla^{2}\phi(x)=-m\Gamma\,\delta F/\delta\phi . From F = \phi^{2}(1-\phi)^{2} , \nu\nabla^{2}\phi(x)=-2m\Gamma\phi . In 1D positive value of m gives periodic solution of \phi , which is nonphysical in this stochastic dynamics. m = 0 gives trivial solution, hence considering negative value of m, and the bound nature of \phi , \phi normalized to e^{-x/d} . Where d=\sqrt{2m\Gamma/\nu} is the characteristic length scale of channel and x/d represents domain to channel ratio. As d\to 0 , the order parameter saturates to \phi\to 0 , signifying a state dominated by domains i.e the Mott state, d\to\infty , the order parameter saturates to \phi\to 1 , indicating channel dominated metallic state. Fig. 4(a) shows for small value of domain to channel ratio, \phi easily achieves 1 and for larger value, \phi goes sway from 1, which well explain our observations. The KPZ framework predicts universal scaling of domain-wall fluctuations characterized by roughness and growth exponents 18 , 50 , 23 , 9 . A direct extraction of these exponents requires systematic variation of the interfacial length scale, the present work focuses on a phenomenological description of current-driven domain dynamics, while future measurements on devices with tunable channel lengths or real-space probes will enable a direct test of KPZ universality and a broader generalization of the model. Also, further experiments combining transport with in-operando structural probes will enable a more direct visualization of the domain dynamics described here. The non-linear conductivity in 1T-TaS 2 follows a power-law, \sigma\propto(I-I_{\text{th}})^{\beta} , analogous to the percolation scaling \sigma\propto(p-p_{c})^{t} , where p is the fraction of conducting sites, p_{c} the percolation threshold, and t\approx 1.3 is the universal 2D exponent 75 , 6 . For T\leq 100~\text{K} , \beta\approx 1.3 confirms a percolative transport regime. At 150 K, \beta peaks at 1.48, coinciding with maxima in I_{\text{th}} and P_{\text{th}} , where domain fragmentation dominates and sharp conductivity jumps occur. Above 150 K, \beta decreases to \approx 0.8 at 300 K, indicating a transition to a channel-driven regime with smoother I-V characteristics (Fig. 2(b)). In the sparse fractal limit, the system exhibits enhanced non-linearity due to filamentary percolation pathways that dynamically evolve under current bias. The fractal nature of the system can be realized by analyzing the area across the NDR when the system transition from higher resistive state to more metallic channel dominated lower resistive state. From statistical mechanics, the number of active conduction channels within these regions scales with characteristic linear size of these metallic patches as N\sim L^{D_{f}} , where D_{f} is the fractal dimension 42 . During a current-driven transition, for sample lengths L\gg\xi , the increase in current above the threshold, |I-I_{\mathrm{th}}| , activates additional metallic clusters, so that |I-I_{\mathrm{th}}|\sim N\sim L^{D_{f}} and thus L\sim|I-I_{\mathrm{th}}|^{1/D_{f}} . The area of NDR reflects substantial domain reconfigurations during the transition from insulating to metallic states, with the fractal dimension governing how these reconfigurations scale with the applied current, which gives, A_{\mathrm{NDR}}\sim|I-I_{\mathrm{th}}|^{2/D_{f}} . At low temperatures, D_{f}\sim 0.37 indicates that metallic regions are sparse and filamentary, while at 100 K, D_{f}\sim 0.5 suggests more connected, yet still inhomogeneous domains. As temperature increases further, D_{f} approaches \sim 0.9 at 300 K, consistent with nearly uniform metallic regions. As D_{f} increases, denser pathways form and the switching time t_{switch} decreases, highlighting the geometric control of transport kinetics. In the context of fractality, post-NDR oscillations follow conductance G\propto(I_{p}-I_{th})^{\delta} , where I p is current corresponding to each oscillation post-NDR, with \delta=0.3 for Mott state, (Fig. 2(a)), 0.4–0.5 for metallic state (Fig. 2(b)) presented in Fig. S11, reflecting a multiscale percolation hierarchy that governs non-linear conduction via evolving fractal connectivity in the sparse-limit 75 . Overall, our work provides a complementary macroscopic perspective on domain dynamics in 1T-TaS 2 . Whereas prior microscopic studies, such as STM and STS, have predominantly focused on imaging domain evolution under external perturbations at length scales of 20–500 nm, our transport measurements, combined with a phenomenological free-energy framework and KPZ dynamics, capture the collective nonequilibrium evolution of domains at the device scale 41 , 77 , 74 , 12 . In this regime, the volume fraction of domains evolves as e^{-x/d} under external perturbations, where \frac{x}{d} represents the domain-to-channel ratio. Our results are consistent with earlier observations, while further analysis demonstrates that domain reconfiguration driven by external stimuli is governed by a fractal dimension D_{f} , leading naturally to percolative transport behavior. Inspired by percolation models 75 , we construct a current-tuned site-percolation framework to explain the evolution of non-linear transport in 1T-TaS 2 . A 2D block of n\times n sites is modeled with open-site probability p , initially increasing linearly with current I . Below the percolation threshold p_{c}=0.59 , the system exhibits linear I-V behavior. Beyond a threshold current, non-linearity emerges, and the transition is captured by a modified control function (p_{c}-p)/(1-rp) , with r a coupling parameter, analogous to (I_{th}-I)/(1-rI) , provided rI<1 . Expansion yields a dominant quadratic term \sim I^{2} , signifying local Joule heating, which activates conducting sites and triggers NDR near I_{th} (Fig. 4(b)). This model connects the emergence of NDR to a thermally driven transition from insulating ( \phi=0 ) to metallic ( \phi=1 ) states. Whereas, in the Mott phase, considering site opening is \propto current, shows absence of NDR in V-I curve presented in Fig.S10(a), suggests insufficient Joule heating prevents transition towards \phi = 1, thus suppressing NDR. The sparse fractal regime amplifies these effects due to enhanced local field concentration and filamentary conduction paths, reinforcing the observed scaling behavior and non-linear dynamics 6 . Our analysis provides a minimal yet physically consistent picture of current-induced percolative switching and NDR emergence in correlated cdw systems (see supplementary section IX, Fig. S10, 70 ). Conclusion In conclusion, our study of MI and ML states in 1T-TaS 2 under current bias combines transport experiments with a minimal free-energy model F[\phi,I] , capturing threshold-driven stabilization of metallic domains and easier transitions at high bias. Pulse-driven measurements reveal intrinsic switching, with the NDR onset exhibiting a scaling exponent consistent with 2D percolation ( \beta\approx 1.3 ). The sparse limit fractal dimension D_{f}\approx 0.3 at 10 K, rising to \approx 0.9 at 300 K, indicates temperature-driven multiscale pathway development. In the MI state, domain stability up to 190 K yields nearly constant I_{th} , while above 220 K enhanced conduction marks the emergence of a channel-dominated IC-CDW phase. In the ML state, an increase in temperature demands greater current and power to fragment domains before forming continuous pathways, elucidating the temperature-dependent interaction of percolation, domain reconfiguration, and least power dissipation. These results establish a quantitative link between current-driven phase transitions, fractal geometry, and percolative transport, offering a general framework for engineered domain architectures in low-dimensional correlated electron systems. Figure 1: Temperature-dependent resistance ( R ) and free-energy landscape F(\phi) . (a) R – T for two flakes of differing thickness, showing low- T Mott (red) and metallic (black) ground states. Inset: 2 \omega response of resistance (50–150 K) for the Mott device, where a resistance dip coincides with a Gaussian-like 2 \omega peak, indicating structural reorientation and coexistence of metallic and Mott fractions. (b) At I=0 , F(\phi) exhibits a double well with minima at \phi=0 (Mott) and \phi=1 (metallic), separated by a maximum at \phi=0.5 . Inset: I shifts the endpoint energy F(\phi=1) set by a/b ( a : scattering coefficient, b : R\times t ). Above I_{\mathrm{th}} , the metallic state is energetically favored. Figure 2: Voltage–current characteristics of MI and ML states. (a) V – I of the Mott-insulating (MI) state at 10 K for successive current loops ( n ), showing a gradual transition from \phi=0 to \phi=1 with the emergence of pronounced negative differential resistance (NDR). (b) V – I of the metal-like (ML) state from 10–300 K. At low T , strong NDR accompanies current-induced transitions to highly conductive states; with increasing T , NDR sharpness diminishes, indicating reduced resistance-switching dynamics. Inset: threshold power for NDR vs T , reflecting the temperature-dependent coupling between current-driven percolation and structural instabilities. Figure 3: Device-dependent responses and tuning of the \phi=0\rightarrow 1 transition. (a) V – I and R – T for two devices (yellow, red) with differing metallic fractions. The critical current I_{c} (where R jumps) correlates with the R – T dip (50–150 K), with larger dips yielding higher I_{c} , indicating a link between I_{c} and order parameter \phi for I\leq 0.5 . No NDR is observed in this regime. (b) Effect of voltage pulses: without pulse (blue) and after a 2 V, 100 kHz, 60 s pulse (pink). Pulse application shifts I_{c} higher for I\leq 0.5 , consistent with increased metallic fraction. Inset: R – T for the same device at I=3~\mu A and 30~\mu A; larger currents deepen the R – T dip, indicating current-induced growth of the metallic fraction. Figure 4: (a) Characteristics of order parameter ( \phi ) with channel length (d) ( \phi vs. d), for different values of domain channel ratio. Here x/d is the domain to channel ratio, for smaller value of x, \phi easily approach 1 with variation of d, as x increases sufficiently \phi can never reach 1 (metallic state). (b) Calculated V-I characteristics of the Metallic state for varying p_{c} . The Metallic state demonstrates a lower resistance and smoother transition, indicating the formation of continuous conductive channels. The inset illustrates the percolation model: red denotes closed (blocked) sites, yellow represents open (waiting) sites, and blue highlights the continuous channels formed by the connected open sites. This schematic emphasizes the role of percolation dynamics in the emergence of conductive pathways across the system, as described in the text.