Turing-patterned Ta₂S₃ enables sub-2 nm diffusion barrier for advanced Cu interconnects

Abstract

As semiconductor technology advances towards the 15-Å node and beyond, the back-end-of-line (BEOL) metal interconnect technology encounters severe scaling challenge, because the conventional barrier/liner TaN/Ta(Co) bilayer cannot be proportionally thinned down as the pitch size shrinks. This challenge arouses significant resistance-capacitance (RC) delay that is currently the bottleneck for advanced integrated circuits. Herein, we propose a Turing-patterned Ta₂S₃ (Tp-Ta₂S₃) thin barrier to address the above interconnect scaling challenge. The Tp-Ta₂S₃, synthesized by a particularly designed energy-enhanced atomic layer deposition (ALD) process, features a labyrinthine stripe morphology with an intertwined layer structure, which can significantly enhance its Cu diffusion barrier performance by forcing the Cu atoms to migrate a significantly longer path through the 2D interlayer space. The synthesized Tp-Ta₂S₃ barrier with only 1.2 nm thickness exhibits excellent Cu wettability, remarkably strong adhesion, and much better barrier performance than the state-of-the-art TaN barrier.

To address the interconnect scaling challenge in semiconductor technology, researchers created a sub-2-nm Turing-patterned Ta₂S₃ barrier that can block copper diffusion better than the state-of-the-art TaN barrier

Introduction

As semiconductor technology advances towards the 15-Å node and beyond, copper interconnects in the back-end-of-line (BEOL) encounter critical scaling challenges. While the minimum feature size of the interconnect pitch must scale below 20 nm, the thickness of the state-of-the-art barrier/liner (TaN/Ta or Co) bilayer cannot be proportionally reduced and must remain around 2.5 nm to ensure adequate diffusion barrier performance^1,2. This scaling conflict will drive, for instance, a sixfold increase in the interconnect line resistance for the 15-Å node as compared to the 3-nm node, owing to the reduction of the line effective conductive volume from 75.6% to 61.6% (Fig. 1a). The resultant surge in resistance–capacitance (RC) delay severely limits the microprocessor operation frequency and therefore is the critical bottleneck in advanced integrated circuits (ICs)^3,4. As an estimation (Fig. 1b), if the barrier layer thickness can be reduced to 1.2 nm, the line effective conductive volume can accordingly be remarkably increased from 61.6% to 80.8%, thereby significantly reducing the interconnect line resistance by 41.3%. This estimation underscores the critical importance of developing thin barriers with superior performance to overcome the RC delay limitation in interconnects for the post-Moore era.

Fig. 1. Interconnect scaling challenge and the concept of the Turing-patterned barriers.

a Schematic illustration of the key bottleneck in advanced interconnect technology: given fixed barrier/liner thickness, the effective conductive volume decreases significantly, thereby steeply increasing the metal line resistance. Data are calculated at the M1 level from the IRDS™ Roadmap. (IRDS^TM: International Roadmap for Devices and Systems^TM; FEOL: Front-End-of-Line) (b) Schematic illustration of using a new thinner (1.2 nm) barrier to replace the current TaN/Co barrier/liner structure for copper interconnect. Critical dimensions (CD) = 16 nm and aspect ratio (AR) = 1.5 are used for the calculation as per the IRDS™ Roadmap¹ for the M1 lines for the 15-Å node. c Schematic illustration of the possible Cu ion diffusion paths for the conventional 2D barrier materials and Tp-Ta₂S₃ barrier. Blue lines denote the layers of conventional 2D materials (i.e., TaS₂ layers); gray areas (lower middle schematic) represent the interlayer spaces filled with bridging Ta atoms; red cross symbols denote the diffusion paths that are of significantly high migration barriers. The light and dark blue areas in the Turing pattern diagram (lower left schematic) respectively represent the activated and inhibited film growth areas following the Turing mechanism.

To address the above challenge, atomically thin two-dimensional (2D) layered materials have been considered as good candidates to surpass the thickness limitation of the state-of-the-art TaN-based barrier materials. Both theoretical predictions and experimental studies indicate that certain single-crystalline 2D materials such as graphene^5,6, MoS₂^7,8, h-BN⁷ and TaS₂⁹ outperform TaN in blocking copper diffusion, showing significant potential as sub-nanometer thin barriers. However, most single-crystalline 2D materials have to be synthesized at a very high temperature ( > 800 °C) by chemical vapor deposition (CVD)^5–7, which significantly exceeds the thermal budget limit of the back-end-of-line (BEOL) processing ( ≤ 400 °C). Other approaches based on transferring are not compatible to be integrated with modern interconnect technologies^5,6. Recently, Lo et al.⁹ proposed a transfer-free conversion approach to prepare the TaS₂ barrier from Ta metal at a BEOL-compatible temperature. However, such approach not only is difficult to implement for the BEOL damascene structure, but also inevitably generates high density of impurities and grain boundaries, which are the fast pathways for copper diffusion^9,10. Moreover, for the conventional 2D materials with van der Waals (vdW) interlayer interactions, the weak interlayer interactions not only can allow for fast copper ion diffusion through the interlayer space, but also lead to insufficient adhesion strength that is subject to mechanical failure in the follow-up processes such as chemical mechanical polishing (CMP).

To address the inherent limitations of the interlayer ion diffusion and weak adhesion for the conventional vdW 2D materials, we herein introduce a Turing-patterned layered structure of Ta₂S₃ as a thin barrier for copper diffusion. Turing-patterned structures have recently been applied to regulating mass transport across membranes^11,12 and catalyst engineering^13–17, and being inspired, we herein design a Turing-patterned Ta₂S₃ (Tp-Ta₂S₃) thin film featuring a labyrinthine stripe morphology with an intertwined layer structure (Fig. 1c). This unique microstructure can significantly enhance the copper barrier property, as the intertwined layer architecture can greatly extend the copper ion diffusion path by forcing the ions to transport laterally in between the 2D layers, therefore mitigating the long-lasting issue of fast diffusion/electromigration through the vertical grain boundaries for conventional polycrystalline 2D flakes. Moreover, the Ta₂S₃ structure can be viewed as a conventional TaS₂ 2D layer structure with extra Ta atoms bridging the neighboring TaS₂ layers, and the bridging Ta atoms can substantially increase the energy barrier for the copper ion lateral diffusion in between the TaS₂ layers and concurrently enhance the interlayer mechanical adhesion strength.

To experimentally realize Tp-Ta₂S₃, we invented an energy enhanced atomic layer deposition (ALD) process based on an amino-imino Ta precursor and H₂S plasma. By utilizing the autocatalytic property of the amino-imino Ta precursor together with the supply of the plasma energy to leverage the rapid lateral diffusion of inhibitors (sulfur clusters), we were able to realize the “local activation and lateral inhibition” Turing mechanism (see next section for details) to afford the Tp-Ta₂S₃ films with a thickness of only 1.2 nm at temperature sufficiently low ( ≤ 350 °C) to be compatible with the back-end-of-line (BEOL) processing. These Tp-Ta₂S₃ films exhibited much improved copper diffusion barrier performance that outperformed the 2.5 nm TaN standard reference; and at the same time, the interfacial adhesion strength was greatly enhanced to 10.33 J/m², which well satisfied the stringent requirement ( > 5 J/m²) for copper CMP¹⁸. The herein Ta₂S₃ barrier material, featuring the Turing pattern, can enlighten an effective path to tackle the progressively severe RC delay challenge in scaled metal interconnects for advanced semiconductor technology.

Results

Fabrication of Turing-patterned Ta₂S₃

In general, the formation of Turing pattern (i.e., Turing mechanism) relies on simultaneous presence of an activator and an inhibitor species in a reaction-diffusion system and the inhibitor diffuses significantly faster than the activator^19,20. Analogously, in a chemical-reaction-based thin-film growth system (such as CVD or ALD), some reaction intermediates or products can catalyze the reaction (also known as autocatalysis) and therefore can act as the activators^21,22; whereas some byproducts may inhibit the reaction and therefore can act as the inhibitors^23,24. If the inhibitors diffuse faster than the activators, a Turing pattern may be formed by the characteristic short-range activation and long-range inhibition mechanism. Following this general mechanism, we designed a new ALD approach for growing Turing-patterned TaS₂ films by using a tantalum precursor that can generate autocatalysts together with the energetic H₂S plasma that can generate sulfur cluster inhibitors and accelerate their surface diffusion (Fig. 2a). We chose to use (tert-butylimino)tris(diethylamino)tantalum (TBTDET, Ta(N^tBu)(NEt₂)₃) as the tantalum precursor, as this precursor bears the alkylamino and alkylimino ligands, which can react with surface sulfhydryl groups to generate autocatalytic products for the ALD growth of metal sulfides. The surface sulfhydryl groups are naturally formed on metal sulfides when exposed to H₂S, and these sulfhydryl groups are Brønsted acidic, so they can protonate the precursor alkylamino and/or alkylimino ligands to afford alkylamines when the metal precursor is dosed and chemisorbed onto the surface. The afforded alkylamines are weak Brønsted/Lewis bases but volatile, so they can possibly stick dynamically on the metal sulfide surface, with some extent of mobility, by forming Brønsted acid-base complexes with sulfhydryls²³ and/or Lewis acid-base complexes with the Lewis-acidic metal ions^25,26. In the follow-up ALD half-cycle of the H₂S exposure, these alkylamines can dynamically coordinate to the metal ions to labilize the remaining surface metal−ligand (metal−nitrogen) bonds for the H₂S attack; the attack affords the metal sulfide ALD product along with the byproducts of also the alkylamines. Therefore, overall, the alkylamines can be regarded as the autocatalysts for the film growth process, and thus they can act as the activator for a Turing process. On the other hand, sulfur clusters are naturally formed during the H₂S plasma step; some sulfur clusters (e.g., S₈) are fairly inert and therefore can act as the inhibitor for a Turing process. The energy supplied by the plasma power can further accelerate the surface diffusion of the sulfur clusters by kinetic energy transfer via collisions, and therefore the diffusion of the inhibitors can be much faster than the activators so as to satisfy the Turing conditions.

Fig. 2. Morphologies of the Turing-patterned films.

a Schematic illustration of the Turing pattern formation mechanism during ALD of TaS₂: the Ta precursor (Ta(N^tBu)(NEt₂)₃) affords autocatalytic alkylamines on surface (labeled as Ligand), which can act as the activator of the Turing process for TaS₂ growth, while the inert fast-diffusion sulfur clusters formed during the H₂S plasma step can act as the inhibitor of the Turing process. The light and dark blue areas (right schematic) represent the activated and inhibited film growth areas, respectively. Bright-field TEM (BF-TEM) images and the corresponding H(r) functions and their Fourier transforms for the Tp-TaS₂ films deposited by (b) 9 and (c) 13 ALD cycles. d Proposed film growth model. Blue horizontal lines represent the TaS₂ layers. AFM images and the corresponding pixel-wise height statistics plots (with fitted peaks) for the (e) as-deposited Tp-TaS₂ and (f) annealed Tp-Ta₂S₃ films (both by 13 ALD cycles).

Figure 2b–f shows the experimental results with the Turing pattern formation. The characteristic Turing pattern structure of the labyrinthine stripe morphology can be clearly seen by transmission electron microscopy (TEM) and atomic force microscopy (AFM). To quantitatively describe the Turing pattern, we employed a method based on height-height correlation function (HHCF) to analyze the TEM and AFM images^27,28. Details of this method are described in Supplementary Information. The HHCF analysis gives a function of H(r), which represents the correlation of the film height for a given lateral distance of r. The typical labyrinthine-stripe Turing pattern has a lateral periodicity in film height, so the corresponding H(r) function should show damped oscillations with respect to r; whereas for a normal film, the H(r) function is simply a decay function of r, where the correlation length reflects the film grain size²⁸. The oscillatory H(r) function can be further Fourier transformed to extract the characteristic frequency (f), and the inverse of frequency (1/f) reflects the average periodicity of the labyrinthine stripes. From the TEM images shown in Fig. 2b, c, the characteristic periodicities of the Turing patterns are extracted to be 19.6 nm and 25.0 nm for the films deposited by 9 and 13 ALD cycles, respectively. The periodicities were also confirmed by the analysis on the AFM images (Fig. 2e and Supplementary Fig. 1). Moreover, since the AFM images can directly give out the pixel-wise film height values, the frequency statistics of the height values are plotted in Fig. 2e and analyzed by peak fitting. The frequency statistics can be well fitted by two Gaussian peaks, with a peak separation of 0.70 nm. Notably, this number is approximately the thickness of a single layer TaS₂²⁹, and therefore the growth of TaS₂ can be viewed to follow the layer-by-layer Frank-van der Merwe growth mode. Based on the above analysis and discussion, we propose the growth model of Turing-patterned TaS₂ as illustrated in Fig. 2d.

To remove the excess sulfur to form the target Tp-Ta₂S₃ with a bridged layer structure, the as-deposited Tp-TaS₂ was annealed under H₂/Ar mixed gas flow at 350 °C for 30 min. As the AFM image shown in Fig. 2f, the afforded Ta₂S₃ retained the same Turing morphology after annealing, and the characteristic periodicity from the HHCF analysis was also found to be identical as before (1/f = 25.0 nm) (see Supplementary Fig. 1 for H(r)). The pixel-wise height frequency statistics are also analyzed and plotted in Fig. 2f. The frequency statistics can still be well fitted by two Gaussian peaks, with a slightly smaller peak separation of 0.63 nm. This smaller separation suggests that the film became more compact by forming Ta₂S₃, which has a bridged layer structure. It is worth noting that both Tp-TaS₂ and Tp-Ta₂S₃ films were quite smooth, with root-mean-square (rms) roughnesses (R_q) of 0.509 and 0.453 nm, respectively. These smooth films are fairly compatible with the advanced interconnect structures (Supplementary Fig. 2).

The elemental compositions of the films were analyzed by X-ray fluorescence (XRF). As shown in Fig. 3a, the XRF analysis indicated that the as-deposited TaS₂ and annealed Ta₂S₃ films (13 ALD cycles) had the S:Ta atomic ratios of 2.08 and 1.49, respectively, which are both in good consistence with their stoichiometric compositions. Assuming the bulk densities of 7.85 g/cm³ (TaS₂) and 8.20 g/cm³ (Ta₂S₃)^30,31, the thicknesses of the as-deposited TaS₂ and annealed Ta₂S₃ films were 1.3 and 1.2 nm, respectively. Therefore, the annealing induced TaS₂-to-Ta₂S₃ conversion was experimentally confirmed. In fact, from the atomistic structure point of view, the Ta₂S₃ structure can be viewed as TaS₂ layers with extra Ta atoms located at the interlayer space occupying the octahedral sites formed by the neighboring-layer sulfur atoms (Fig. 3b). The interlayer octahedral sites can be occupied at different percentages, forming in-plane superlattices with various Ta_xS_y stoichiometries. We calculated the formation energies by density functional theory (DFT) for a series of Ta_xS_y stoichiometries with various octahedral site occupation percentages (i.e., 0%, 25%, 33.3%, 66.7% for TaS₂, Ta₉S₁₆, Ta₇S₁₂, Ta₂S₃, respectively). As shown in Fig. 3c, under the S-rich condition, which corresponds to the experimental ALD condition, TaS₂ is the stable stoichiometry; whereas under the Ta-rich condition, which can be achieved by vacuum or H₂/Ar gas annealing, Ta₂S₃ becomes more stable. High-resolution TEM (HRTEM) images (Fig. 3d, e) further confirmed that both as-deposited TaS₂ and annealed Ta₂S₃ films had a similar Turing patterned morphology. Moreover, fast Fourier transform (FFT) analysis of the Ta₂S₃ area of the HRTEM image (boxed area in Fig. 3e) shows a pattern consisting of blurry crystalline spots (Fig. 3f, see also Supplementary Fig. 3), which are in good consistence with the simulated pattern for the crystalline Ta₂S₃ (Fig. 3g). The layer bridging feature of Ta₂S₃ was further evidenced by Raman analysis. As shown in Fig. 3h, aside from the pronounced Raman bands at approximately 300 and 400 cm⁻¹, which are respectively the characteristic E_g and A_1g vibrational modes of the TaS₂ layer³², the annealed Ta₂S₃ film showed an additional Raman band at the lower wavenumber regime of 100–160 cm⁻¹. According to our DFT calculation of the phonon dispersion diagrams for TaS₂ and Ta₂S₃ as shown in Fig. 3i, j, this low wavenumber band corresponds to the interlayer coupled vibrations induced by the bridging Ta atoms that bond the neighboring TaS₂ layers in the Ta₂S₃ structure. Cross-sectional scanning transmission electron microscopic (STEM) image was further taken on the annealed Ta₂S₃ film. As shown in Fig. 3k (see also Supplementary Fig. 4), a clear intertwined layer structure can be observed, which confirmed the successful synthesis of the Turing-patterned Ta₂S₃ film.

Fig. 3. Material properties of the TaS₂ and Ta₂S₃ films.

a XRF areal atomic densities for the as-deposited TaS₂ and annealed Ta₂S₃ films (13 ALD cycles). b Atomistic structures of TaS₂ and Ta₂S₃. S and Ta atoms are shown in gray and blue, respectively. c DFT-calculated formation energies for a series of Ta_xS_y stoichiometries. HRTEM images of the (d) as-deposited TaS₂ and (e) annealed Ta₂S₃ films. f FFT analysis of the boxed Ta₂S₃ area in panel e. Red circles denote the primary spots of the TaS₂ layer; pink circles denote the superstructure spots induced by the interlayer Ta atoms in the Ta₂S₃ structure. The pink dash lines are only for eye-guiding purpose. g Simulated diffraction pattern of Ta₂S₃. The pink lines are only for eye-guiding purpose. h Raman spectra of the TaS₂ and Ta₂S₃ films. DFT-calculated phonon dispersion diagrams for (i) TaS₂ and (j) Ta₂S₃. k Cross-sectional STEM image of the Tp-Ta₂S₃ film (upper) and the corresponding schematic illustration (lower).

Interconnect performance evaluation

Wettability, adhesion, and barrier properties are the core metrics for evaluating a potential new barrier material for advanced interconnect technologies. The wettability and adhesion determine the metal filling uniformity and interface mechanical reliability, while the barrier property is the key to block the copper metal from diffusing into the dielectrics.

To evaluate the wettability of the Tp-Ta₂S₃ film, an approximately 10 nm thick Cu film was thermally evaporated onto the Tp-Ta₂S₃ surface. As shown in Fig. 4a, b, using thermal oxide wafer (denoted as SiO₂ to highlight its surface property) as a benchmark, the Cu film grown on Tp-Ta₂S₃ was fairly smooth with no visible pinholes, demonstrating an excellent wettability; whereas the Cu film grown on SiO₂ was of rather poor wettability, showing many cracks and pinholes. The much-improved wettability of Tp-Ta₂S₃ can be ascribed to the formation of the relatively strong Cu−S bond at the Cu/Tp-Ta₂S₃ interface, and this property is highly beneficial for uniform Cu filling into small-size damascene trench/via structures (e.g., reflow process) in advanced interconnect technologies. Possibly, the Tp-Ta₂S₃ barrier can be used without the need of a liner, thereby saving more space for metal filling in small-size interconnect structures.

Fig. 4. Interconnect performances of the Tp-Ta₂S₃ barriers.

SEM images showing the wetting behaviors of Cu on (a) SiO₂ and (b) Tp-Ta₂S₃ surfaces. c Measured load−displacement curves for the Cu/SiO₂, Cu/s-TaS₂/SiO₂, and Cu/Tp-Ta₂S₃/SiO₂ stacks by the 4-PB tests. d Time-dependent dielectric breakdown (TDDB) results to evaluate the barrier properties for 1.5 nm s-TaS₂, 2.5 nm TaN, and 1.2 nm Tp-Ta₂S₃. Inset illustrates the capacitor stack for the TDDB measurement. DFT-calculated relative energies of the transition/intermediate states for the Cu atom migration in the layered structures of (e) TaS₂ and (f) Ta₂S₃ (IS: initial state; TS: transition state; MS: intermediate state; FS: final state).

The adhesion property was analyzed quantitatively by performing the four-point bending (4-PB) tests on the Cu/film/SiO₂ stacks³³. To this end, we prepared the Cu/Tp-Ta₂S₃/SiO₂ stack with the 1.2 nm Tp-Ta₂S₃ film, together with a benchmarking Cu/s-TaS₂/SiO₂ stack with a 1.5 nm s-TaS₂ film prepared by sulfidation of an e-beam evaporated Ta metal film⁹, and another benchmarking Cu/SiO₂ stack with no adhesion film for comparison. Note that the Cu film on s-TaS₂ was also of good wettability (Supplementary Fig. 5). Figure 4c shows the load-displacement curves of the 4-PB tests for these samples, where the load initially increased with displacement due to elastic deformation, and then suddenly dropped due to the crack initiated from the notch and quickly propagated, releasing the accumulated bending energy^34,35. The interfacial adhesion energies were measured to be 1.49, 3.06, and 10.33 J/m² for the Cu/SiO₂, Cu/s-TaS₂/SiO₂, and Cu/Tp-Ta₂S₃/SiO₂ samples, respectively. Apparently, the Tp-Ta₂S₃ film significantly enhanced the adhesion between Cu and SiO₂, and the enhancement was much larger than the s-TaS₂ film. This phenomenon is because the Ta₂S₃ structure contains the bridging Ta atoms that bond the S atoms of the neighboring TaS₂ layers so that the adhesive interaction between the neighboring TaS₂ layers was much stronger than the vdW interaction in the TaS₂ case. Moreover, it is particularly worth noting that the observed adhesion energy of the Cu/Tp-Ta₂S₃/SiO₂ stack was significantly larger than the requirement for the interconnect CMP process ( > 5 J/m²)¹⁸. Therefore, as one of the major concerns of using conventional 2D materials for interconnect barriers, the CMP compatibility issue can be well resolved by using the Tp-Ta₂S₃ barrier.

The barrier property was evaluated by the standard time-dependent dielectric breakdown (TDDB) measurement^36,37. As illustrated in the inset of Fig. 4d (see also Supplementary Fig. 6), the TDDB measurement was performed on a capacitor structure of Cu/barrier/SiO₂/n⁺⁺-Si³⁸, and a constant positive voltage was applied on the Cu electrode to drive the Cu ions into the SiO₂ dielectric. The Cu ions accumulated in the dielectric and eventually formed a conductive path that caused the breakdown of the dielectric³⁶. Fig. 4d displays the cumulative probability distributions of the time to breakdown (t_BD) measured at 7 MV/cm for the Cu/barrier/SiO₂/n⁺⁺-Si stacks with different barrier layers (i.e., 1.2 nm Tp-Ta₂S₃, 2.5 nm TaN, and 1.5 nm s-TaS₂) and no barrier. The cumulative probability distributions show approximately parallel shifts with a similar slope for all above stacks, which indicates that the variation of t_BD did not alter with the inserted barrier layer. This behavior can be understood as that in our case the variation of t_BD was dominated by the property of the SiO₂ dielectric and the insertion of the barrier layer uniformly delayed the diffusion of the Cu ions into the dielectric. This behavior also implies that all above barrier layers were quite uniform, because otherwise a gentler slope would have been shown in the cumulative probability distribution. To further quantify the barrier property, we took the time to 63.2% failure (TTF_63.2%) as the characteristic metric. The TTF_63.2% values were found to increase from 10.9 s for the control stack with no barrier to 41.1, 102.0, and 118.5 s for the stacks with 1.5 nm s-TaS₂, 2.5 nm TaN, and 1.2 nm Tp-Ta₂S₃ as the barriers, respectively. Remarkably, the Tp-Ta₂S₃ barrier with only 1.2 nm in thickness even outperformed the state-of-the-art TaN barrier with 2.5 nm in thickness, which demonstrates the significant advantage of the Tp-Ta₂S₃ barrier in interconnect scaling. Additionally, it is worth noting that the barrier property of 1.2 nm Tp-Ta₂S₃ was 2.88 times better than that of 1.5 nm s-TaS₂, although the latter was even thicker. This comparison well demonstrates the crucial importance of the herein synthesized Turing structure that can force the Cu atoms to transport a significantly longer distance with higher energy barrier to across the Tp-Ta₂S₃ barrier layer. We further took the cross-sectional STEM/EDS characterizations of the Cu/Tp-Ta₂S₃/SiO₂/n⁺⁺-Si and Cu/s-TaS₂/SiO₂/n⁺⁺-Si stacks after the above TDDB stress. As shown in Supplementary Figs. 7 and 8, the Tp-Ta₂S₃ barrier layer could well block Cu from diffusing through; whereas the s-TaS₂ layer failed to block Cu, showing pronounced Cu signals at the s-TaS₂/SiO₂ interface, possibly owing to the facile Cu diffusion in between the vdW 2D TaS₂ layers, as will be discussed later.

It is worth noting that the Tp-Ta₂S₃ barrier was also fairly stable. We measured the TDDB performance of the Cu/Tp-Ta₂S₃/SiO₂/n⁺⁺-Si stack after one-month storage, and as shown in Supplementary Fig. 9, no significant change was observed. In addition, we found that this Turing-patterned barrier layer strategy not only worked for Cu but also worked for other two important BEOL metals of Co and Ru. As shown in Supplementary Fig. 10, the characteristic TTF_63.2% values were also found to significantly increase from 16.1 to 186.2 s for Co and from 78.6 to 504.9 s for Ru, after the 1.2 nm Tp-Ta₂S₃ barrier layer was inserted. These results well demonstrate that the Turing-patterned barrier strategy is of good generality.

To further investigate the Cu diffusion energy barriers, we performed the DFT calculations for TaS₂ and Ta₂S₃. For the conventional 2D material of TaS₂, Fig. 4e shows the lowest-energy pathway for the Cu atom to migrate from an octahedral site (IS) to a neighboring octahedral site (FS) in the interlayer space of the 2D TaS₂ layers. The migration path includes an intermediate state (MS), where Cu moves to the tetrahedral site with an energy of +0.11 eV. To arrive at this tetrahedral site, the Cu atom needs to pass the center of the shared face of the octahedron and tetrahedron (TS1 or TS2) with an energy barrier of +0.17 eV, and therefore the overall energy barrier for this pathway is +0.17 eV. Notably, this energy barrier is fairly low, and therefore Cu atoms are relatively mobile in between the vdW 2D TaS₂ layers. In contrast, for crystalline Ta₂S₃, as shown in Fig. 4f, the interlayer space is separated into isolated domains by the bridging Ta atoms, and the alien Cu atom is most stable at the center tetrahedral site of each isolated domain (IS or FS), although the octahedral site (MS1 or MS5) is only +0.02 eV higher in energy. For this Cu atom to migrate from one domain to a neighboring domain, it has to pass across the domain wall, where all the octahedral sites are occupied by the interlayer bridging Ta atoms. The lowest-energy pathway for this migration is via three neighboring tetrahedral sites around the Ta-occupied octahedral sites (MS2 → MS3 → MS4). However, the energy barrier for this pathway is significantly high, as the calculated barrier energies are +11.98 eV for the MS3 state and +22.53 eV for the corresponding transition states where Cu is at the shared edge center of two tetrahedrons (Supplementary Fig. 11). Therefore, the long-range migration of Cu atoms is rather blocked in perfect-crystalline Ta₂S₃. Even if we consider the possible presence of Ta-vacancy point defect at the domain wall, the calculated energy barrier is still quite high as +2.14 eV for the intermediate state where Cu occupies the octahedral vacancy site (Supplementary Fig. 12). Therefore, it is concluded that the bridging Ta atoms of Ta₂S₃ can significantly suppress the long-range Cu migration and therefore is pivotal for the remarkable barrier property of Ta₂S₃ as experimentally observed.

Based on the above experimental results, we further estimated the quantitative benefit of utilizing the Tp-Ta₂S₃ barrier for next-generation Cu interconnect technology. We calculated the BEOL metal line/via resistances at the M1/V1 level based on the critical dimensions (CD) of the 2 nm to 7 Å technology nodes of the IRDS™ Roadmap¹. As shown in Fig. 5a, as the interconnect CD shrinks down for smaller technology nodes (aspect ratio (AR) = 1.5), the metal line resistance (R_line) substantially increases. Because the line resistance is dominated by the volume of Cu, R_line of the 1.2 nm Tp-Ta₂S₃ case is always smaller than that of the conventional 1.5 nm TaN/1 nm Co case. Apparently, the difference in R_line becomes larger for smaller CD, because the volume fraction of the barrier progressively increases. Notably, for 1.5 nm TaN/1 nm Co, R_line exceeds the Roadmap limit for the technology nodes ≤1.5 nm; whereas for 1.2 nm Tp-Ta₂S₃, R_line is always below the Roadmap limit for the technology nodes even down to 7 Å, which suggests the high promise of the Tp-Ta₂S₃ strategy for the line barrier application. Figure 5b shows the calculated via resistances (R_via) (AR = 2). For small CD vias, R_via is dominated by the highly resistive barrier layer at the via bottom, so a bottom barrierless via structure³⁹ has been proposed to meet the Roadmap requirement. Experimentally, such bottom barrierless structure can be achieved by selective ALD of the barrier layer as shown later. As the calculation results shown in Fig. 5b, with the bottom barrierless structure, R_via for both 1.5 nm TaN/1 nm Co and 1.2 nm Tp-Ta₂S₃ cases are below the Roadmap limit. Comparatively, R_via of the Tp-Ta₂S₃ case is even better, with only half of that of the 1.5 nm TaN/1 nm Co case (21.7 versus 42.0 Ω) for the 7 Å technology node. Therefore, the Tp-Ta₂S₃ strategy is also highly promising for the via barrier application.

Fig. 5. Resistance estimation, trench coating, and selective coating of Tp-Ta₂S₃.

a Calculated R_line and (b) R_via at M1/V1 level for the advanced technology nodes of the IRDS™ Roadmap. c Cross-sectional STEM image of a high aspect ratio (AR) trench conformally coated by 1.2 nm Tp-Ta₂S₃, with the corresponding EDS (d) Ta and (e) S maps and (f) high-angle annular dark-field (HAADF) image. g SEM image and the corresponding EDS (h) Co and (i) Ta maps for a patterned Co/SiO₂ test structure with selective Tp-Ta₂S₃ coating. j EDS spectra of the boxed areas in panel g, where a.u. means arbitrary units.

To further demonstrate that the fabrication approach developed herein is also compatible with the BEOL interconnect structures, we fabricated high aspect ratio (AR) nano-trenches with 230 nm in depth and ~50 nm in opening width (AR = 4.6) and assessed the step coverage of the Tp-Ta₂S₃ barrier layer on these trenches. As shown in Fig. 5c–f, the STEM images and the corresponding EDS maps clearly show that the high-AR trench was conformally coated by the Tp-Ta₂S₃ layer with a uniform thickness of 1.2 nm throughout the entire trench wall. Notably, such high AR of the trench is considerably higher than those of the typical BEOL damascene structures⁴⁰, and therefore we speculate that the herein proposed fabrication method can be well applied to the modern BEOL interconnect technology. Moreover, to achieve the bottom barrierless metallization for the BEOL via structures, the Tp-Ta₂S₃ barrier needs to be selectively coated on the via sidewall but not on the bottom metal. To this end, we further developed a selective ALD TaS₂ method by using aniline as the inhibitor to prevent the TaS₂ growth on the metal surface. We used Co metal surface for this demonstration, as Co is the capping metal for Cu lines. Figure 5g–j shows the results of the selective Tp-Ta₂S₃ coating on a patterned test structure with exposed Co and SiO₂ surfaces. The EDS results clearly show that the Ta signal was only present on the SiO₂ surface but not on the Co surface. Therefore, selective Tp-Ta₂S₃ coating could also be achieved. Overall, our results clearly demonstrate that the herein proposed Tp-Ta₂S₃ barrier with the associated fabrication method is highly promising for the next-generation Cu interconnect technology.

Discussion

In this work, we proposed and synthesized a Turing-patterned Ta₂S₃ (Tp-Ta₂S₃) thin layer to serve as a copper diffusion barrier to address the bottleneck BEOL interconnect scaling challenge in advanced semiconductor technology. The Turing-patterned films were synthesized based on a particularly designed energy-enhanced ALD process, which utilized the autocatalytic property of the TBTDET Ta precursor and the energetic H₂S plasma for boosting the inhibitor diffusion to achieve the “local activation and lateral inhibition” Turing pattern formation mechanism. Post annealing at 350 °C successfully converted the as-deposited Tp-TaS₂ to Tp-Ta₂S₃ while well retaining the Turing morphology with the same characteristic periodicity. The Tp-Ta₂S₃ thin film featured a labyrinthine stripe morphology with an intertwined layer structure, which significantly enhanced its barrier performance by forcing the Cu atoms to migrate a significantly longer path through the 2D interlayer space, therefore mitigating the problem of fast diffusion/electromigration via the grain boundaries as in the conventional polycrystalline 2D flakes. Moreover, the Ta₂S₃ atomistic structure contains interlayer Ta atoms, which can bridge the neighboring 2D TaS₂ layers to strengthen its mechanical adhesion and also elevate the energy barrier for the interlayer Cu atom migration, as compared to the conventional vdW 2D materials (e.g., TaS₂).

As a result, we were able to synthesize the Tp-Ta₂S₃ films with a thickness of only 1.2 nm at temperature sufficiently low ( ≤ 350 °C) to be compatible with the back-end-of-line (BEOL) processing. These Tp-Ta₂S₃ films exhibited excellent wettability with Cu and sufficiently strong adhesion with an interfacial adhesion energy of 10.33 J/m² for Cu/Tp-Ta₂S₃/SiO₂, which well surpassed the stringent requirement of >5 J/m² for the interconnect CMP process. More importantly, the TDDB barrier performance of the 1.2 nm Tp-Ta₂S₃ even outperformed the state-of-the-art TaN barrier with 2.5 nm of thickness, which well demonstrated the significant advantage of the thin Tp-Ta₂S₃ barrier for interconnect scaling. According to our estimation, by using this Tp-Ta₂S₃ barrier, the BEOL M1 line resistance can be much reduced to satisfy the IRDS^TM Roadmap requirement for the technology nodes even down to 7 Å. Conformal Tp-Ta₂S₃ coating on high-AR (4.6) small-opening ( ~ 50 nm) nano-trenches was also demonstrated, showing uniform coating thickness throughout the entire trench wall. Selective Tp-Ta₂S₃ coating on Co metal surface was also demonstrated, and therefore, the Tp-Ta₂S₃ barrier strategy could also be applied to the bottom barrierless metallization scheme for the BEOL via structures. Further estimation showed that, by using this bottom barrierless Tp-Ta₂S₃, the BEOL V1 via resistance can also satisfy the IRDS^TM Roadmap requirement for the 7-Å technology node. With the above promising results, we envision that the herein proposed sub-2 nm Ta₂S₃ barrier, featuring the Turing pattern, will enlighten an effective path to tackle the progressively severe RC delay challenge in scaled metal interconnects for advanced semiconductor technology in the post-Moore era.

Methods

Fabrication of Tp-Ta₂S₃

The ALD of Tp-TaS₂ used the Ta precursor of TBTDET and the H₂S plasma. During deposition, the TBTDET precursor was kept at 90 °C, and the ALD chamber temperature was set at 150 °C. The H₂S plasma was generated by a 60 W radio frequency (RF) power. In each ALD cycle, the TBTDET dose time was 5 s, the H₂S plasma pulse time was 20 s, and the purge time was 30 s for each half-cycle. The ALD growth saturation behavior was verified by varying the TBTDET dose time and plasma pulse time (Supplementary Fig. 13). To convert Tp-TaS₂ to Tp-Ta₂S₃, the deposited Tp-TaS₂ films were in situ annealed at 350 °C for 30 min under a mixed gas atmosphere of Ar:H₂ = 1:1. It is worth noting that Turing pattern did not show up for regular thermal ALD prepared TaS_x (Supplementary Fig. 14); also, the Turing pattern did not show in another recent ALD TaS_x report using Ta[N(CH₃)₂]₃[NC(CH₃)₃] precursor and H₂S plasma with different deposition conditions⁴¹, which suggests the importance of the ALD conditions to form the Turing pattern. High AR nano-trenches (depth 230 nm, opening width ~50 nm) were fabricated on thermal oxide wafers (Si wafers with 300 nm thermally oxidized SiO₂ on top) by e-beam lithographic patterning and inductively coupled plasma (ICP) etching. On these trench structures were conformally coated the Tp-Ta₂S₃ layer using the same abovementioned ALD TaS₂ and annealing process. To achieve selective Tp-Ta₂S₃ coating on a patterned Co/SiO₂ test structure, the sample was immersed in aniline for 30 min prior to ALD. Presumably, aniline could be selectively adsorbed on Co, thereby inhibiting the ALD of TaS₂ on Co.

Material characterizations

The Turing morphology was examined by TEM (Jeol, JEM-3200FS, 200 kV) and AFM (Bruker, Dimension Icon). The elemental composition was analyzed XRF (Shimadzu, EDX-LE). Electron diffraction pattern was simulated by ReciPro using the kinematical and excitation error model. Raman analysis was performed using the 532 nm laser excitation (Thermo Fisher, DXR). Cross-sectional microstructures were examined by STEM (FEI, Titan Themis 200, 200 kV) equipped with EDS (Super-X EDS). Wettability and selectivity were examined by SEM (Zeiss, Supra55) equipped with EDS (Oxford Instruments, X-max^N 20). The adhesion energy was measured by 4-PB test (Lloyd Instruments, LRX Plus).

TDDB measurement

TDDB capacitor structures were fabricated on thermal oxide wafers (50 nm SiO₂ on n⁺⁺-Si wafers). Top Cu electrodes (thickness 100 nm, diameter 300 μm) were thermally evaporated through a shadow mask, and the Cu electrodes were further capped by 20 nm Al to prevent Cu oxidation. Another 20 nm Al was thermally evaporated on the backside of the n⁺⁺-Si wafer to ensure good electrical bottom contact. For the benchmarking barrier layers, the s-TaS₂ layer was synthesized by a two-step process, where a Ta pure metal layer was first e-beam evaporated on SiO₂ and then sulfidized under the H₂S plasma at 400 °C for 20 min (H₂S flow 10 sccm, RF power 70 W)⁹. The benchmarking TaN layer was synthesized by thermal ALD from TBTDET and NH₃ (deposition temperature 250 °C)⁴². The TDDB current traces were monitored by a probe station using an Agilent B1500A semiconductor parameter analyzer.

Interconnect resistance estimation

The BEOL metal line/via resistances were calculated at the M1/V1 level using the CD values from the IRDS™ Roadmap¹. The size dependences of the Cu and Co resistivities were taken into account^43,44. The resistivity of 0.9 mΩ cm was used for TaN⁴², and the resistivity of 5.0 mΩ cm was used for Tp-Ta₂S₃, as measured by the four-point probe method (Napson, RT-3000/RG-2000).

DFT calculations

The DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP 6.3.2)⁴⁵ with the projector-augmented wave (PAW)⁴⁶ method and Perdew-Burke-Ernzerhof (PBE)⁴⁷ generalized gradient approximation (GGA) exchange-correlation functional. 5p⁶5d³6s² (Ta), 3s²3p⁴ (S), and 3d¹⁰4s¹ (Cu) were treated as valence electrons. To account for van der Waals interactions, the DFT-D3 dispersion correction was applied⁴⁸. A plane-wave energy cutoff of 500 eV and a Monkhorst-Pack k-point mesh with a k-spacing of 0.03 Å⁻¹ were adopted, which could ensure the total energy convergence within 10⁻⁶ eV/atom and forces below 0.01 eV/Å for structural optimization. Gaussian smearing with 0.02 eV smearing width was used. Spin-polarized calculations were conducted to model the magnetic ordering with the initial atomic coordinates derived from the experimental crystallographic data. A vacuum spacing of 20 Å was added to the slab model to minimize the interaction between adjacent slabs. The formation energies (E_f) of Ta_xS_y (can be rewritten as Ta_m+nS_2m) were calculated by E_f = E_tot(Ta_m+nS_2m) − E_tot(Ta_mS_2m) − nμ_Ta, where E_tot(Ta_m+nS_2m) and E_tot(Ta_mS_2m) are the total energies of Ta_m+nS_2m and Ta_mS_2m, respectively, n is the number of Ta atoms intercalated into the TaS₂ bilayer structure, and μ_Ta is the chemical potential of Ta atom, which varies upon experimental conditions⁴⁹. Phonon modes were calculated using density functional perturbation theory (DFPT) with GGA. For this purpose, the structures were reoptimized using more strict convergence criteria of 10⁻⁷ eV for total energy and 10⁻⁶ eV/Å for forces, and the VASP-reported dynamical (Hessian) matrices were used to obtain the phonon modes by Phonopy^50,51. The Cu migration barriers were calculated at 0 K temperature based on Ta₁₈S₃₆ (TaS₂) and Ta₂₄S₃₆ (Ta₂S₃) structures (Supplementary Table 1) using the climbing-image nudged elastic band (CI-NEB) method with force convergence < 0.02 eV/Å, and three intermediate images were inserted along each path between neighboring local minima.

Author contributions

X.W. and G.C. supervised the project. X.T. and X.W. conceived the ideas and designed the experiments. X.T. carried out the experiments with the help of S.J., J.L., Y.D., S.Z., J.K., Y.Q. and Y.Z. S.Y. and Y.T. conducted the DFT calculations. R.L. performed the Raman measurements. X.T. and X.L. conducted the HHCF analysis. X.T. and S.Y. conducted the TEM measurements and analysis. X.T. wrote the original draft. X.W., G.C., R.W. and M.L. revised the manuscript with the input from all other authors.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data in this study are available in the manuscript and in the Supplementary information section. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71545-2.