Low Contact Resistance in MoTe₂ Based Field-Effect Transistors

July 07, 2023 - Single-layer molybdenum ditelluride (MoTe₂) has attracted attention due to the smaller energy difference between the semiconducting and semimetallic phases compared to other two-dimensional transition metal dichalcogenides. Scientists from the Istituto Italiano di Tecnologia in Pisa and Genova, the University of Ulm in Germany, Lund University in Sweden, Trento University in Italy, as well as Sorbonne University in France have advanced the understanding of the direct synthesis of lateral heterostructures. The resulting 1H/1T' crystals are stabilized with a scalable encapsulation approach recently reported by the group [1], and their structural and chemical properties are investigated by Raman spectroscopy and the Cc/Cs-corrected “Sub-Ångstrom Low-Voltage Electron microscope” (SALVE) [2] at an acceleration voltage of 80 kV. The final 1H/1T' contact fronts can have a lateral size in the range of tens of micrometers, which is relevant for device fabrication such as field-effect transistors (FETs). These findings contribute significantly to the development of devices with low contact resistance, crucial for the future of spintronic devices and FETs.

Monolayer MoTe₂ is an extremely attractive candidate for the development of phase change devices [3] and low-resistance contacts [4]. Additional interest in these two phases arises as 1H-MoTe₂ has important implications for the development of spintronic [6], valleytronic, near-infrared optoelectronic, and quantum devices. While in the past most of the experiments have been carried out on mechanically exfoliated flakes [7], nowadays, it has become possible to synthesize 1T′ [8] and 1H as well as several different polymorphs of MoTe₂, i.e. 2H [8], Td [9], and also new forms such as 2D-Mo₅Te₈ [10] and 1D-Mo₆Te₆ nanowires [11]. However, while 1H-MoTe₂ is a well-studied material, 1T′-MoTe₂ experimental research lags behind due to the extreme air instability of this material, which rapidly degrades upon air exposure, with a lifetime in the minutes range [1].

To date, significant attention has been focused on understanding and achieving controllable 1H/2H to 1T′ structural transformation. Theoretically, different methods have been proposed such as, electronic excitation [12], strain [13], or metal atom substitution, and Te vacancies creation [12]. Phase transition (PT) has been achieved on few-layers or bulk MoTe₂ via ion liquid gating [15] or electric field in vertical RRAM devices [7]. In the monolayer limit, investigations on 1H/1T′ phase transition are highly complicated by the air instability of monolayer 1T′, although promising initial results have been reported by ionic liquid gating [15], THz laser irradiation [16] and Te vacancies creation [17]. Nowadays, growth of 1H/1T′ lateral heterostructures has been demonstrated in UHV conditions via Mo and Te evaporation on graphene substrates [18].

In Figure 1a,b, the optical micrograph of typical crystals of MoTe₂ grown on SiO₂/Si via liquid precursor CVD is shown. Crystals of different shapes are visible within the same sample: hexagonal (a) and elongated (b) ones, which are normally attributed to the 1H and 1T′ phase, respectively. The researchers optimized the growth parameters to maximize the density of MoTe₂ flakes in each sample, which in turn led to the observation of a significant number of hetero-contacts between hexagonal and elongated flakes (like those shown in Figures 1c,d and 2a). The MoTe₂ samples were encapsulated immediately after growth with monolayer CVD hBN (or graphene) to increase the lifetime of 1T′-MoTe₂ and allow further characterization, similar to what was reported in [1].

In Figure 2a, one can observe three elongated flakes contacting a hexagonal crystal, which interestingly presents different optical contrasts. A precise assignment of the structural phase of the material, carried out by Raman spectroscopy, indicates that only part of the hexagonal flake displays a 1H phase, with the remnants being 1T′ (see Figure 2b–d). Indeed, Figure 2b reports the typical Raman spectra measured in different regions (marked with crosses) of panel a. The blue and black spectra were recorded on the regions with dark and light contrast, respectively, while the red spectrum was recorded at the center of the hexagonal crystal. The blue spectrum shows typical features of a monolayer 1H phase, such as the in-plane E_12g mode at 236 cm^–1 and the out-of-plane A_1g mode at 168 cm^–1 [21], while few-layer peaks E_1g at 116 cm^–1 and B_12g at 288 cm^–1 are absent [22]. The portion of the hexagon with darker contrast in the left side of panel a is hence assigned to 1H-MoTe₂. The triangular feature with darker contrast placed at the center of the hexagon presents a Raman spectrum that is indicative of few-layer 2H-MoTe₂, as typically observed in seeding areas.

HR-TEM allowed the researchers to identify the edges of the untransformed hexagonal 1H crystals to be armchair (AC) terminated, while 1T′ single crystals were found to elongate along the ZZ-direction (see Figure 3a,b). Via polarized Raman, they measured a maximum R-value in 1T′ flakes in horizontal configuration (i.e., long side/ZZ direction parallel to the light polarization), with the minimum measured for the orthogonal alignment configuration. For the flakes displayed in Figure 2a, the spatial distribution of the R-value allowed them to determine that the dominating crystallographic direction of the transformed hexagonal crystal corresponds to that of the lower contacting 1T′ flake (see Figure 3c). This finding suggests that the 1H–1T′ phase transition was originally triggered by the flake indicated in Figure 3c by the light blue arrow.

The researchers verified with a large number of flakes and found that in all instances the transformed 1T′ crystals have a direction that is either collinear or 60° rotated with respect to that of the original 1T′ contacting flake. Figure 3 reports the case of transformation with collinear orientations of initial 1H and 1T′ crystals (i.e., ZZ-directions originally parallel in both crystals), where the formation of a 60° rotated 1T′ domain and the incompleteness of the transition process can be observed. By inspecting 99 hetero-contacts, they concluded that there is no such indication: neither the occurrence of the phase transition nor the percentage of the transformed 1H flake is obviously related to the initial contact angle (see Figure 3f).

High-resolution transmission electron microscopy (HRTEM) was performed to validate the crystallinity of the flakes on the atomic scale and confirm the presence and nature of defects (Figure 4). To minimize beam damage in the monolayers, MoTe₂ was encapsulated with graphene. Monolayer 1T′ showed a rather defective crystalline structure (Figure 4d,e) and the hetero-contact region could not be imaged due to damage arising during transfer to the grid because of the high sensitivity of the 1T′ phase to oxygen [19]. Conversely, in the 1H region, two types of defects were identified: (i) single Te vacancies (see Figure 4a); (ii) inversion domains, mainly 4|4P Mirror Twin Boundary (MTB) domains (Figures 4b,c). These MTBs have already been observed in experiments with Mo excess forming “triangular” and “wagon-wheel” shaped domains [20], and in the newly reported hexagonal 1H–Mo₅Te₈ phase [10], ultimately confirming the Te deficiency already highlighted by XPEEM measurements.

The experimental data indicate that heterocontact-triggered polymorphism takes place: (i) independent from the 1T′/1H contact angle; (ii) in the presence of Te vacancies and MTB within the 1H original crystal; (iii) during CVD growth and while Mo is diffusing [21]. In the following text, the researchers adopt linear elasticity theory [22] and ab initio calculations to model and further understand the observed phenomenology.

With the help of first-principles calculations, the researchers considered the effects of various structural mechanisms on the energetic landscape of single-layer MoTe₂, specifically aiming at assessing the role of the heterocontact and identifying possible structural mechanisms entailing a reduction of the kinetic barrier between 1H and 1T′ phases, thus favoring the phase transition. In the presence of a high energy barrier, the 1H to 1T′ transition process might not take place or be too lengthy to be observed. Hence, they performed ab initio simulations of heterocontact-triggered transition paths for the collinear contact (i.e., 1H–ZZ/1T′–ZZ and 1H–AC/1T′–AC) reported in Figure 3d, considering transition paths involving Te atom displacements. First, they calculated energy barriers for pristine cases, referred to as α- and β-paths in Figure 5a. Both paths result in the same final 1T′ configuration but originate from the different lower Te atom displacement directions in the initial 1H cell. For the α-path, Te atoms displacement is parallel to the final 1T′–AC direction, while for β it is 45° tilted. They calculated 1.81 (905 meV/f.u.) and 1.74 (870 meV/f.u.) eV energy barriers for α- and β-paths, respectively, with final 1T′ energy 0.146 eV (73 meV/f.u.) higher than initial 1H phase due to the fixed (constant) cell condition. These results are in agreement with previously reported values for pristine cells [14]. Next, they performed simulations for direct 1H–1T′ contact in AC and ZZ direction considering supercells of 7 pristine unit cells (see Figure 5b,c). They referred to the same α and β notations corresponding to the lower Te atoms displacement. They considered broken periodicity in x and y directions that also results in possible two steps and additional β-path in AC direction corresponding to the transverse direction propagation similar to what was reported [13]. Both ZZ and AC routes show similar behavior of α-ZZ, two-step α-ZZ and β-AC paths with 0.94 eV (470 meV/f.u.), 0.84 eV (420 meV/f.u.), and 1.08 eV (540 meV/f.u.) energy barriers, respectively. The energy differences between the two phases are 0.036 (18 meV/f.u.) and 0.056 (28 meV/f.u.) eV for the ZZ and AC cases, respectively. They then extended the analysis to defective paths, considering Te vacancy presence at the hetero-interface. In general, paths that involve Te vacancies are the most promising: in the ZZ case, the presence of a Te vacancy is found to strongly suppress the kinetic barrier between the two phases from 0.94/0.84 eV (470/420 meV/f.u.) in the pristine case to 0.27 eV (135 meV/f.u.). However, the 1T′ phase becomes more unfavorable +0.170 eV (+85 meV/f.u.) with respect to the 1H phase. Conversely, in the AC case the energy barrier is not suppressed but the 1T′ becomes energetically favorable at −0.046 eV (−23 meV/f.u.) with respect to the reference 1H phase. These findings are consistent with the well-studied Te vacancy induced phase transition mechanism [12, 17]. Their simulations suggest that Te vacancies work differently in AC and ZZ directions and are not the sole responsible for the observed PT; however, they importantly modify the energetic landscape and can be involved in the stabilization of the 1T′ phase in combination with other mechanisms. Reported reduced energy barrier values and lowered ground state energy difference between 1H and 1T′ phases indeed can be responsible for the avalanche effect observed experimentally. However, in the ZZ case, Te vacancies, even with a reduced barrier value, lead to an ascending steps energy profile upon increase of the transformed unit cells number and should result in the backpropagation of 1T′ to 1H transformation of the triggering 1T′ crystal, which was never observed experimentally. In addition, Te vacancies (in both the AC and ZZ directions) tend to accumulate upon phase transition propagation and at some point terminate the transformation.

The DFT simulations also confirm that a hetero-contact-triggered phase transition is possible for the orthogonal 1T′–AC/1H–ZZ contact, indicated by the arrow in Figure 3c. In that case, a significant lattice mismatch up to 10% at the b_1T′/2a_1H interface with approximately 5% 1H–ZZ compressive strain is expected to take place.

Finally, two possibilities are represented by other kinds of defects, namely, MTBs and Mo diffusion, since the researchers have observed MTBs and the consequences of Mo diffusion in the grown crystals (e.g., material accumulation and 1T′ post-transformation growth continuation). Thus, they performed additional DFT simulations to understand how these defects affect the transformation. They found that MTBs and Mo excess in the ZZ direction generally lead to energetically unfavorable kinetic paths in all the studied situations. More interesting is the case of excess Mo for phase transition propagation in the AC direction. They found intermediate states with double Mo chains formation, resembling the structures reported in monolayer M₄X₆ TMDs [23]. These structures are 0.7–0.77 eV (350–385 meV/f.u.) more favorable than the following 1H–1T′ cell transformation. In the end, their calculations showed a 0.54 eV diffusion barrier in the 1H phase via a substitutional mechanism [20], and 0.66 eV in the 1T′ phase along the ZZ direction via an interstitial diffusion mechanism. These barriers are significantly lower than those reported for most transition routes. Therefore, they do not exclude that transition kinetics could be limited by Mo diffusion within both the 1H and 1T′ domains and can explain why they observed uncompleted PT for different 1H–1T′ heterocontact angles. Considering few-layer MoTe₂ formation at the 1H–1T′ heterointerface and 1T′ post-transformation growth continuation, they suppose that the diffusion of excess Mo during CVD growth plays an important role and the 1H to 1T′ transformation can be stopped at high Mo concentration or in the presence of defects hindering Mo diffusion.

In conclusion, the researchers report a semiconductor to semimetal phase transition realized in monolayer MoTe₂ during CVD growth and triggered by the presence of a direct contact between 1H and 1T′ single crystals. The crystal orientation of the transformed 1T′ domains was investigated using linearly polarized Raman spectroscopy and revealed the possibility of two preferential crystal transformation routes: collinear and orthogonal to the initial 1H crystal, both resulting in domains with 60° periodicity. Chemical and structural analyses were performed via TEM on encapsulated samples to limit their degradation and indicated the presence of MTB and Te vacancies in the 1H crystals. They theoretically modeled transition pathways considering both stoichiometric and nonstoichiometric cases and identified transformation routes with a wide variety of kinetic barrier values. The pathways studied in this work can be further investigated in other TMDs, their heterojunctions, and in Janus materials [22]. A recent work has demonstrated that few-layer 2H/1T′ MoTe₂ heterocontacts can be fabricated through photolithographic techniques, an approach that has been used to recover damaged 2H-MoTe₂ [23]. Defining pathways for bottom-up synthesis of low contact resistance heterojunctions is indeed extremely appealing for the fabrication of 2D electronic and optoelectronic devices such as FETs and novel spintronic and quantum devices.

Resource: Khaustov, V. O., Convertino, D., Köster, J., Zakharov, A. A., Mohn, M. J., Gebeyehu, Z. M., Martini, L., Pace, S., Marini, G., Calandra, M., Kaiser, U., Forti, S., Coletti, C. (2023). Heterocontact-Triggered 1H to 1T′ Phase Transition in CVD-Grown Monolayer MoTe₂: Implications for Low Contact Resistance Electronic Devices. ACS Applied Nano Materials. https://doi.org/10.1021/acsanm.3c01314