High electrical contact resistance had stalled the promised performance of two-dimensional layered devices. Low-resistance metal–semiconductor contacts are now obtained by interfacing semiconducting MoS2 layers with the metallic phase of this material.
Transistors and lasers made of semiconductor materials power the information age by providing the building blocks for electronic switching, amplification and photonic communication. The electronic and photonic properties of the semiconductor play a primary role in determining the performance of such devices, yet comparable importance resides in the way this material interfaces with the external metallic circuits. A high electrical resistance due to a high energy barrier — the Schottky barrier — encountered by the charge carriers moving through the metal–semiconductor contacts saps the energy efficiency, and significantly degrades the performance of the device.
Semiconductor researchers are excited about the potential of transition metal dichalcogenide (TMD) layered semiconductors. For device applications, the excitement is fuelled by the dream of improving device performance by using a semiconductor channel that is potentially one monolayer thick1. The first generation of field-effect transistors made from MoS2 and related TMDs have shown promise, but trail the silicon and III–V semiconductor analogues significantly in performance. A high contact resistance is the root cause. Writing in Nature Materials, Rajesh Kappera and colleagues now report an unconventional method to address this problem head-on2, based on the local conversion of MoS2 semiconducting layers into a metal that lowers the barrier to the charges flowing in the external circuits.
The traditional method to lower the contact resistance between a metal and a semiconductor relies on an intermediate layer, usually a heavily doped semiconductor that helps charge carriers from the metal to be injected in the conduction and valence bands of the semiconductor. The metal is chosen based on its workfunction, so that it creates a low Schottky barrier height with the semiconductor contact region, whereas heavy doping reduces the thickness of the barrier, enabling a significant fraction of electrons to quantum mechanically tunnel through it. However, chemical doping has proven challenging with TMD semiconductors. In their work, Kappera et al. exploit the fact that MoS2 exists in two crystalline phases: the 2H and 1T types, which differ in their stacking geometry. The 1T phase is obtained by twisting one set of the 2H Mo–S tetrahedron by a 60° rotation. The stable 2H structural phase has a semiconducting behaviour and is desired as the channel of the transistor. The 1T phase is structurally metastable, but can be stabilized if electrons are pumped into it — it then becomes metallic because of a half-filled d-orbital band. This electronic stabilization of the metallic 1T phase can be achieved by converting the 2H phase by various routes, for example by irradiation with an electron beam3. Kappera and colleagues use a chemical technique, treating the MoS2 with an organometallic solution containing n-butyl lithium. Lithium donates electrons to the 2H MoS2 converting it into the 1T metallic phase. The chemical process achieves the phase transition selectively: covered areas are left semiconducting, and exposed areas become metallic, as confirmed by extensive chemical, structural and optical analytical characterization. When the solution is washed away, the 1T region is likely to attract immobile positive charges and remain stable. Such a semiconductor-to-metal phase-transition process can be amenable in a device fabrication environment.
The team then used this process to make batches of transistors — the test structures have 1T metallic MoS2 source–drain contact areas interfacing gold pads, and the control structures have direct gold–2H semiconductor MoS2 contacts. In the test devices, the contact resistance was found to drop from ~1–10 kΩ μm to ~0.2–0.3 kΩ μm. As a result, the performance of the 1T contact transistors was significantly superior across the board of metrics: the drive current, the sharpness of switching and the gain all improved. The researchers also find that the 1T contact transistors have a much higher reproducibility and yield compared with their 2H contact counterparts. The transistor characteristics are also less sensitive to the workfunction of the metals, suggesting that the contacts' performance now mainly depends on the 1T/2H interface. Other TMD semiconductors have similar metallic counterparts, implying the same principle potentially applies to them.
The phase-engineering approach to making low-resistance ohmic contacts to TMD semiconductor materials is thus an exciting advance that addresses a critical problem holding back potential applications. As with any new study, a list of unknowns remains to be worked out. By itself, the 1T metallic phase of MoS2 is negatively charged — meaning the identity of the neutralizing positive charges that are presumably immobile remains to be determined and controlled. The nature of the 1T–2H metal–semiconductor junction, the band alignments, and a potential way to contact the valence band for hole conduction need to be developed. The barrier between the metal and 1T phase due to weak van der Waals bonding needs to be investigated. The lateral diffusion of the organometallic lithium solution under the covered areas can convert part of the desired 2H MoS2 channel into the 1T phase, whereas the chemical conversion of the exposed regions is not perfect; alternative ways to seal the channel during the chemical treatment and achieve complete phase transformation in the contact areas need to be investigated. And as the researchers state, the stability of the 1T contacts under high-performance operation — for example when large currents are driven through the transistor — remains to be elucidated.
But how low a contact resistance can one obtain? This problem in various forms has been studied for more than a century. James Clerk Maxwell calculated4 the classical 'contact' resistance between two regions of conductivity σ separated by an insulator and connected by a conducting circular constriction of diameter D to be Rc ~ 1/(σD). In the 1950s and 1960s, Landauer5 and Sharvin6 gave the problem a quantum facelift. Their work predicted a minimum contact resistance value for Rc ~ h/(2e2M), where h is Planck's constant, e is the electron charge and M is the number of electron modes whose wavelength fit the narrow conductor. In other words, even for a perfect conducting channel with no scattering, only those electron modes that fit are allowed access into the channel, the rest are reflected. As the sheet density of electrons n2D = kF2/2π in a two-dimensional (2D) channel increases, the wavelength λ = 2π/kF of the energetic electrons riding the Fermi surface shortens, and more modes M ~kFW fit, where W is the width of the channel and kF is the Fermi wavevector. The minimum contact resistance is then RcW ~h/2e2kF ~0.026/√n2D kΩ μm, which depends strongly on the electron sheet density (in units of 1013 cm−2) in the semiconductor channel, and weakly on some aspects of the bandstructure7. This is the quantum limit of the contact resistance for crystalline semiconductors, shown by the dashed line in Fig. 1.
This limit has been experimentally verified in atomic break-junctions and in split-gate quantum point contacts, in which a quantized conductance was observed8. The highest-performance semiconductor transistors are also grazing this lower limit9, as shown in Fig. 1. The latest achievements for TMDs represent a major leap, yet there is still room for improvement. Recently, a joint academic and industry research team has reported a method to chemically dope MoS2 and have achieved low-resistance contacts and high-performance transistors10. Their technique is similar to the traditional method used for semiconductors, and the contact resistance values are similar to the 1T contacts discussed here. Both these old and new approaches mark important steps towards harnessing the innate potential of 2D crystal semiconductors. More importantly, metal–semiconductor junctions can enable a host of unanticipated physical phenomena exploiting the d-orbital pedigree of conduction electrons in TMDs.
- Nature Nanotech. 6, 147–150 (2011). , , , &
- Nature Mater. 13, 1128–1134 (2014). et al.
- Nature Nanotech. 9, 391–396 (2014). , , &
- 1904). A Treatise on Electricity and Magnetism (Clarendon,
- IBM J. Res. Dev. 1, 223–231 (1957).
- Sov. Phys. JETP 21, 655–656 (1965).
- Superlattices Microstructures 23, 771–780 (1998). , &
- Phys. Rev. Lett. 60, 848–850 (1988). et al.
- Appl. Phys. Lett. 102, 111605 (2013). , , , &
- 238–239 (2014); http:dx.doi.org/10.1109/VLSIT.2014.6894432 et al. VLSI Technol. Digest of Tech. Papers, 2014 Symposium on,