Transistors, regardless of their size, rely on electrical gates to control the conductance between source and drain contacts. In atomic-scale transistors, this conductance is sensitive to single electrons hopping via individual orbitals1,2. Single-electron transport in molecular transistors has been previously studied using top-down approaches to gating, such as lithography and break junctions1,3,4,5,6,7,8,9,10,11. But atomically precise control of the gate—which is crucial to transistor action at the smallest size scales—is not possible with these approaches. Here, we used individual charged atoms, manipulated by a scanning tunnelling microscope12, to create the electrical gates for a single-molecule transistor. This degree of control allowed us to tune the molecule into the regime of sequential single-electron tunnelling, albeit with a conductance gap more than one order of magnitude larger than observed previously8,11,13,14. This unexpected behaviour arises from the existence of two different orientational conformations of the molecule, depending on its charge state. Our results show that strong coupling between these charge and conformational degrees of freedom leads to new behaviour beyond the established picture of single-electron transport in atomic-scale transistors.
We created molecular transistors using either phthalocyanine (H2Pc) or copper phthalocyanine (CuPc) molecules, depicted in Fig. 1a. These are planar π-conjugated molecules which readily adsorb on an InAs(111)A substrate15 at precisely defined locations above the In-vacancy site of the 2 × 2 reconstructed surface (Fig. 1b). The molecule is physisorbed on the surface by van der Waals interaction, with no significant charge transfer. Owing to weakness of this binding, the adsorbed molecules can be readily repositioned among different vacancy sites by using the tip of a cryogenic scanning tunnelling microscope (STM).
The InAs(111)A substrate surface is characterized by a low concentration of native defects, +1 ionized indium adatoms (Inad). The adatoms render the surface weakly metallic with the Fermi level pinned in the conduction band16. They can also be repositioned by the STM tip17. By arranging a group of charged adatoms in a carefully chosen configuration, we designed and created a variety of electrostatic surface-potential landscapes for the nearby region with sub-ångström accuracy.
This positional control over both the charged indium adatoms and the H2Pc molecules allows us, first, to design and create an atomically precise potential landscape using the In+1 adatoms and, then, to position the molecule within that landscape. The potential landscape provides the transistor gate18,19 by acting on the electron energy levels of the weakly coupled molecule. To complete the three-terminal transistor structure, the STM tip and InAs substrate serve as source and drain contacts, respectively. Electrostatic gating by repositionable charged defects has previously been used to shift the binding energy of dopants near a GaAs surface20. Here, we apply this concept to control the charge state of a single molecule (see Supplementary Information). Figure 1c, d shows topographic images of two Inad trimers with a H2Pc molecule in a remote position (Fig. 1c), and midway between the trimers (Fig. 1d). The molecule is imaged as a uniform protrusion because it performs rapid rotational jumps at sample biases Vb larger than ±0.1 V (ref. 15). In the midpoint position (Fig. 1d), the electrostatic gating is sufficient to shift the lowest-unoccupied molecular orbital (LUMO) below the Fermi level EF, charging the molecule negatively. As a consequence, the local tunnelling barrier21,22 is increased, which reduces the apparent molecular height compared to the neutral state. To quantify the charge state we measured the electrostatic potential at the tip position23,24 along the dashed line in Fig. 1c (Supplementary Fig. 1). Figure 1e shows the difference in potentials obtained in the presence and absence of the molecule. Clearly, the potential difference is essentially zero when the molecule is located in the remote position (green dots), confirming its neutral state there. At the midpoint position (red squares), on the other hand, the charge state is −1, as confirmed by comparison with the calculated potential (blue curve) of a point charge screened by the surface-accumulated electrons23,24,25 (see Supplementary Information).
To investigate the gating process systematically, we constructed Inad corrals that allowed us to fine-tune the gating through a linear potential landscape (Supplementary Fig. 2). We found that both CuPc and H2Pc can be tuned into a regime of charge bistability. STM imaging in the bistable regime revealed that the two molecular charge states have different orientational conformations in their respective ground states. The left-hand STM image in Fig. 2a shows a neutral H2Pc in one of its three equivalent conformers, A, in which the molecular lobes are oriented parallel/perpendicular to one of the three 〈110〉 in-plane directions15. On charging, the molecule performs an in-plane rotation, either clockwise or anticlockwise (centre and right STM images, respectively). There are three equivalent conformers, A, in the neutral state and six equivalent conformers, B, in the negative charge state. This suggests that the transition A → B consists of an in-plane rotation by ±15°, as well as a small lateral displacement and/or tilt26 of the molecule (Supplementary Fig. 3). Topographic images analogous to those in Fig. 2a were obtained for CuPc in its neutral and negatively charged state, respectively. The bistability observed here involves a coupled switching in charge and orientational conformation. We note that STM-driven conformational switching in adsorbed phthalocyanine-based molecules has also been reported by others27.
The electrostatic gating of the LUMO level strongly modifies the tunnelling conductance, as revealed by the current-versus-bias (I–Vb) curves in Fig. 3a, which were recorded over an extended bias range with the STM tip held fixed above a single CuPc molecule. The conductance gap is reduced as the level is brought closer to EF, consistent with the simple picture of transport via a single unoccupied level within a double-barrier tunnel junction (DBTJ; refs 28, 29). This picture is also supported by the normalized differential conductance map in Fig. 3b, which was extracted from I–Vb curves consecutively recorded at different values of the gate potential φ. At a given value, the differential conductance is sharply peaked at biases Vb for which the LUMO enters the energy window between the tip and sample Fermi levels, which leads to sequential tunnelling through the DBTJ. However, near the charge degeneracy point there is a conductance gap, ∼600 mV, which is much larger than for conventional sequential tunnelling through a gated quantum dot. This gap is substantially greater than normally found, for example, in electron transport through single molecules in the Franck–Condon blockade regime of strong electron–vibron coupling8,11,13,14.
The general features of the measured tunnelling conductance can be understood within a Coulomb blockade model30 for the DBTJ, by treating the molecule as a quantum dot attached to the STM tip and the InAs(111)A surface, which act as electron reservoirs. Motivated by the experimental observations, we assume that the molecule can exist in two conformations, A and B, and that each conformation can be either neutral or charged. Conformation A is the ground state of the neutral molecule and B the ground state of the charged molecule, as shown in Fig. 3c. We find that this assumption naturally explains the occurrence of the conductance gap in the Vb–φ plane. Indeed, both conformations give rise to a separate charge degeneracy point. The charge degeneracy point for A is shifted to higher gate voltages relative to that for B. This defines several regions in the Vb–φ plane, labelled I–IX in Fig. 3d. In regions I–III the molecule is blockaded with respect to both conformations and hence conduction is suppressed, whereas in regions IV and V the molecule is not blockaded and so conduction occurs. The remaining regions, VI–IX, are more subtle as they are conducting with respect to one conformation but blockaded with respect to the other. Franck–Condon physics strongly suppresses tunnelling-induced switching between the conformations; therefore, A is stable in regions VI and VII, whereas B is stable in VIII and IX. These correspond to the blockaded conformations and hence there is a conductance gap for all potentials φ, consistent with the experimental observations.
The bistability in charge and conformation shown in Fig. 2 occurs within the conductance gap region where sequential tunnelling is blocked. Figure 4a shows I–Vb curves recorded close to EF with the STM tip held fixed above a H2Pc molecule at the indicated φ values. Hysteresis in I–Vb is easily seen at φ = 121 mV (red), revealing bistable switching between two I–Vb curves of different slope around EF. The smaller I–Vb slope is associated with the negative charge state, concomitant with the increase in local tunnelling barrier on charging. Similar hysteretic behaviour was previously reported for the bistable charge switching of phthalocyanine-based molecules on NaCl layers grown on Cu substrates of suitable workfunction21. A slight detuning of the potential quenches the I–Vb hysteresis in Fig. 4a, as shown by the blue I–Vb curves, reflecting a dynamical crossover between the two states as detailed in the inset. Figure 4b shows the location of the hysteretic switching (red) and the crossover (blue) in the Vb–φ plane, together with the onset of sequential tunnelling via the LUMO (green). We observed similar behaviour for CuPc (Supplementary Fig. 4), for which the bistability occurs at a gating potential ∼50 mV larger than for H2Pc. Finally, the hysteretic switching dynamics is depicted in Fig. 4c, showing that the coercivity (defined as the half-width of the hysteresis) rises sharply and eventually saturates as the bias ramping speed is increased.
On the basis of the model description outlined in Fig. 3c, d, we set up a master equation for the coupled electronic and conformational dynamics (see Supplementary Information for details). As shown in Fig. 4d, the numerical solution indeed predicts a pronounced conductance gap for all φ, in agreement with the measurements in Figs 3b and 4b. This model also explains the experimentally observed hysteretic behaviour in Fig. 4a and Supplementary Fig. 4. The central region in Fig. 4d consists of two sections, separated by the orange line, with different molecular ground states. Experimentally, these regions have different co-tunnelling conductances30, which explains why we observe bistable I–Vb characteristics (Fig. 4a, b). Moreover, as conformational changes are slow, the current can be hysteretic when varying parameters across this line. Experimentally, this is most pronounced at small biases (Fig. 4b), where electronic tunnelling (and hence the attempt rate for conformational switching) is weak. The observed coercivity as a function of ramping speed of the bias voltage (Fig. 4c) is consistent with the theoretical results from the master equation, as shown in Fig. 4e.
Although our model qualitatively explains many features of the observations, some aspects may require a more refined description. For instance, the model predicts that the lines of current onset extrapolate to cross at the pertinent charge degeneracy point in the Vb–φ plane, and thus at zero bias voltage. This is not actually the case in the experimental data in Figs 3b and 4b. This effect might be due to finite temperature or to interactions beyond our constant-capacitance model (see Supplementary Information for further discussion).
We have shown that a single-molecule transistor can be electrostatically gated by arranging individual charged atoms with sub-ångström precision using a scanning tunnelling microscope. The resulting transistor action reveals a conductance gap far larger than previously observed for Franck–Condon-blockaded transport8,11,13,14, arising from strong coupling between charge and conformational degrees of freedom. Understanding and controlling this type of coupling—and the new kinds of behaviour to which it can lead—will be crucial for integrating atomic-scale transistors and other devices with existing semiconductor technologies.
The measurements were carried out with an ultrahigh-vacuum (UHV) STM operated at a base temperature of 5 K. Electrochemically etched tungsten tips were cleaned in UHV by Ne ion sputtering and electron beam heating. Undoped InAs layers of thickness 20 nm were grown by molecular beam epitaxy (MBE) on an InAs(111)A wafer (purchased from Wafer Technology) to prepare the In-terminated InAs(111)A-(2 × 2) surface. After the MBE growth, the surface was capped with an amorphous layer of arsenic and transferred under ambient conditions to the UHV system of the STM apparatus. The As capping layer was removed by annealing at 630 K and the sample loaded into the microscope cooled down to 5 K. InAs(111)A samples prepared in this way showed the same surface features as MBE-grown and in situ investigated samples. H2Pc and CuPc specimens (purchased from Aldrich) were purified in UHV by repeated cycles of degassing. Low coverages (∼1 × 1012 molecules cm−2) were deposited directly into the microscope at sample temperatures <20 K by sublimation from a tantalum crucible held at 628 K. STM images were recorded in constant-current mode; bias voltages refer to the sample with respect to the tip. Spectroscopy measurements of the differential tunnelling conductance were performed by lock-in technique (10 mV peak-to-peak modulation at a frequency of 675 Hz) with a disabled feedback loop.
The theoretical model included four states of the H2Pc molecule, corresponding to the two conformations A and B and the two charge states for each conformation. The state of the molecule was described in terms of probabilities for each of these four states. In the weak-tunnelling limit (tunnelling rate much smaller than the temperature and spacing between the energies of the four molecular states) these probabilities can be determined from the solution of a master equation that involves transition rates for thermally induced transitions between different conformations of the molecule at constant charge, and transition rates between different charge states at the same conformation, due to tunnelling to/from the scanning tip and the substrate. The stationary solution of the master equation was used to determine the onset of current flow in Fig. 4d; a time-dependent solution with a bias voltage that varies linearly in time was used to calculate the coercivity in Fig. 4e.
This work was supported by the German Research Foundation (Collaborative Research Network SFB 658) and the Office of Naval Research through the Naval Research Laboratory’s Basic Research Program. Some computations were performed at the DoD Major Shared Resource Center at AFRL.
About this article
Nature Nanotechnology (2016)