Electronically Configurable Molecular-Based Logic Gates

See allHide authors and affiliations

Science  16 Jul 1999:
Vol. 285, Issue 5426, pp. 391-394
DOI: 10.1126/science.285.5426.391


Logic gates were fabricated from an array of configurable switches, each consisting of a monolayer of redox-active rotaxanes sandwiched between metal electrodes. The switches were read by monitoring current flow at reducing voltages. In the “closed” state, current flow was dominated by resonant tunneling through the electronic states of the molecules. The switches were irreversibly opened by applying an oxidizing voltage across the device. Several devices were configured together to produce AND and OR logic gates. The high and low current levels of those gates were separated by factors of 15 and 30, respectively, which is a significant enhancement over that expected for wired-logic gates.

The fields of microelectronics and computation have advanced at an amazing rate over the past few decades, and at the heart of these advances has been the technology of the complementary metal-oxide semiconductor (CMOS)–based integrated circuit. However, both the fundamental physics of CMOS devices and the increasingly prohibitive costs associated with constructing next-generation fabrication facilities indicate that this technology is rapidly maturing. Nevertheless, the physics of computation is still in its infancy (1). Many researchers are exploring alternatives to CMOS-based computational machinery, including such options as single-electron transistors (2), quantum cellular automata (3, 4), molecular computation (5), and chemically assembled electronic nanocomputers (CAENs) (6). CAENs are proposed as extensions of conventional circuitry and architecture and will be based on chemically synthesized and assembled nano- or molecular-scale components. Very small machines containing vast amounts of resources (that is, moles of devices) may be possible, yet their basic operating and programming principles will be essentially the same as for present systems. However, there are many obstacles to the construction of an operational CAEN. For example, any chemically prepared system will be ordered (that is, crystalline) and defective (due to finite chemical reaction yields), whereas any reliable computational machine requires perfect complexity. Thus, any CAEN will have to be able to tolerate an extremely large number of defects.

Some of us recently reported on the Teramac CMOS-based supercomputer and commented on its relevance to an architecture for a CAEN (7). Teramac was constructed from a very large number of partially defective resources and, as built, was not hard-wired as a computer at all. Instead, software routines were used to identify defective components and to subsequently download the logical architecture of a desired computer into a configuration memory that appropriately connected the usable hardware resources. Although there were several keys to the hardware architecture that made this possible, one of the primary lessons of Teramac for nanotechnology was that wires and configurable switches were likely to be the most plentiful and therefore important resources in a defect-tolerant CAEN. The configured logic elements of Teramac were “fine grained,” meaning that the basic unit of control was the connection of two wires, not two large busses. Thus, to be economically viable, CAENs must be assembled from extremely inexpensive wires and configurable switches with simple techniques that are amenable to mass production.

Here we describe a first experimental step toward extending the architectural ideas of the Teramac computer to a CAEN. We have developed an electronically (singly) configurable junction that consists of a molecular monolayer and a tunneling barrier sandwiched between lithographically fabricated metal wires (Fig. 1). We demonstrate that this junction can be used as a switch, and that several devices, fabricated in a linear array structure, could be used as electronically configurable wired-logic gates (an AND and an OR gate). Because these switches are only singly configurable, they cannot be used for random access memory (RAM) applications, although programmable read only memory (PROM) applications are possible. However, they do have several advantages. First, and most importantly, they should scale down to molecular dimensions without appreciable loss of performance (8). Second, when the molecular switches are closed, current flows by resonant tunneling into the molecular electronic states. The net result of this resonant tunneling is that the high and low current levels of the logic gate truth tables are widely separated. This should lead to good noise immunity in future logic circuits built with this technology. Third, molecular switches are voltage addressable rather than field addressable. This makes the device robust with respect to dimensional tolerances in manufacturing. In a CMOS-based configurable device (9), two wires (address lines) are used to configure the switch, whereas two different wires (data lines) are used to read it. For our junctions, only two wires are necessary to achieve both functions. One voltage is applied to read the device, and a voltage of opposite polarity is applied to configure it. The configuration voltage of our devices can be a factor of 2 greater than the logic levels used when they are operating. This means that it should be easy to design circuits that are safe from accidental reconfiguration under operation conditions. Finally, this architecture is highly modular. We report here on the properties of three different molecular switches incorporated into sandwich devices composed of various combinations of metallic wires.

Figure 1

(A) Top view of a linear array of six devices, shown approximately to scale. The wires were a few microns in diameter, and each pad was a few hundred microns across to facilitate making an electrical connection to the device. (B) Side view cross section of a single device junction. Each device consisted of a monolayer of molecules sandwiched between two perpendicularly oriented electrodes and contained several million molecules. (C) The energy level diagram of one of the devices in (A). The Fermi levels (E f) of the Al electrodes are shown at both ends of the diagram, and discrete molecular redox energy levels [determined by solution-phase voltammetry measurements of the R(1) rotaxane] are shown in the middle. The oxidation states are noted with filled circles. The diagonally striped areas between the electrodes and the rotaxane energy levels are tunneling barriers. The thick barrier is the Al2O3 passivating layer (measured to be 1 to 1.5 nm), and the thin barrier is the Ti-rotaxane interface (estimated to be 0.5 nm).

Our devices are distinct from other electronically configurable two-terminal devices, such as ferromagnetic crosspoint memories (MagRAM) (10). In MagRAM devices, the high and low levels are typically separated by changes in the junction resistance of less than 50%. Although suitable for memory applications, such small changes are not suitable for computation.

Top and cross-sectional views of one of our devices, and its corresponding energy level diagram, are shown in Fig. 1. The switches were prepared by first lithographically patterning 6-μm-diameter Al wires onto a silica substrate. Second, a single monolayer of one of three different molecules was deposited as a Langmuir-Blodgett (LB) film over the entire substrate (11). Finally, a second set of 11-μm-diameter wires, consisting of a thin (5 nm) bottom layer of Ti followed by a thicker (0.1 μm) layer of Al, was evaporated through a contact shadow mask by using electron beam deposition. In each device batch, many linear arrays of junctions, combining to yield several hundred individual devices, were prepared. In >90% of the device fabrication runs, no defective devices were found.

The molecular compounds we used come from a class of molecules known as rotaxanes (12). Rotaxanes consist of two or more components mechanically interlocked with one another: One of the components is dumbbell-shaped, and the remaining components are rings that become trapped on the dumbbell component, encircling part of the dumbbell. In the class of rotaxanes used here, the dumbbell component contained two bipyridinium units and the encircling rings were bis-para-phenylene-34-crown-10. The molecular structure of one [R(1)] of the three systems we used is shown in Fig. 2. The other two compounds used were R(0), which was just the dumbbell component with no rings, and R(2), which contained two rings encircling the dumbbell component. Rotaxane-based switches (12) and gates (13) have been demonstrated in the solution phase, although their operation differs from what is described here.

Figure 2

A drawing of the R(1) rotaxane molecule used here.

A major concern was related to whether the monolayer of one of these molecules could survive the deposition of the top electrode. When the compounds are transferred as an LB film, the structures are oriented with the highly branched hydrophobic heads pointing away from the surface. This part of the compound apparently provides a buffer layer between the Ti-Al electrode and the centrally located bipyridinium rings, where oxidation or reduction takes place. The use of the Ti electrodes with molecular junctions does have literature precedent (14). Whether some reaction between the molecules and the Ti occurs is difficult to know. Nevertheless, the redox properties of these molecular compounds, as measured in solution, do translate suprisingly well into solid-state device properties (8). We successfully used Cr-Al or Ti-Au electrodes in place of the Ti-Al electrode and observed only minor changes in the measured device properties. However, some electrode combinations, such as an Al top electrode without Ti, did destroy the switching properties of the junctions, and we only observed nonresonant tunneling behavior when we probed those devices.

All electrical measurements were carried out in air and at room temperature. The current-voltage (I-V) scans for an R(1)-based device are shown in Fig. 3A. R(0)- and R(2)-based devices behaved similarly, with a few subtle differences that will be discussed elsewhere. If molecular-mechanical motions were occurring in the R(1)- and R(2)-based devices, the electronic signatures of that motion were quite subtle and not obvious from our measurements. No such signatures are observed in the solution-phase electrochemistry of these particular molecules. All voltages were referenced to a grounded top electrode as defined in Fig. 1. When the device was initially probed (read) with a reducing (negative) voltage, the current increased sharply with increasing voltage, up to a few tenths of a nanoamp at –2 V. If the device was similarly probed a second time, the current fell slightly (∼20%), but on subsequent measurements the response remained stable (15). The switch was irreversibly opened by applying an oxidizing voltage of +0.7 V or more (16). Once a switch was opened, its status could again be read by applying a negative voltage. For an open switch, the measured current at −2 V was a factor of 60 to 80 less than that for a closed switch.

Figure 3

(A) Current-voltage traces that show the operation of the devices. As prepared, the molecular switches are “closed,” and the status of the devices is probed by applying a negative voltage to the bottom electrode. The switches are “opened” by oxidizing the molecules at voltages greater than +0.7 V. Finally, the open switches are again interrogated at negative bias. The current ratio at −2 V between open and closed states is between 50 and 70, depending on the specific device. (B) The same data as presented in (A), but plotted as the NDOS. When the closed switch is read, two distinct features are recorded in the NDOS. The first feature corresponds closely to the states shown in Fig. 1C. A single feature is observed in the oxidation NDOS. However, this feature is not a resolved electronic state. Rather, oxidation at around +0.7 to +0.9 V irreversibly changes the molecules, so the NDOS falls to 0 as the resonant tunneling process is quenched. When the oxidized devices are interrogated at negative voltage, the electronic states that were observed between 0 and 1 V for a “switch closed” device are now absent.

Some insight into what is happening to the actual molecular junction can be gained by re-plotting the I-V scans as a normalized density-of-states (NDOS) (17) (Fig. 3B). When the switch is shut, the NDOS at negative applied voltages is resolved into two features. The first broad peak can be understood in terms of the molecular redox states (Fig. 1B) derived from solution-phase pulsed voltammetry measurements. This feature corresponds reasonably well in energy to the lowest lying reduction features for solution-phase R(1), and is likely due to the same redox states. Thus, the onset of high current arises from resonant tunneling through these states. Tunneling mediated by molecular states has been observed previously in solution (18) and in solid-state devices (19). At voltages <−1.5 V, we believe that the device properties are not strongly correlated with the molecular electronic energy levels. We reach this conclusion because the simple linear amphiphile tunneling junctions we used as control devices exhibited similar behavior in this voltage range. After the device has been oxidized, the NDOS at negative biases between 0 and –1 V is essentially 0. The irreversible oxidation of similar molecules in solution has been previously documented (11) and is one reason that these compounds were used here. The fact that all three molecules exhibited similar behavior indicates that the dumbbell, rather than the crown ether rings, dominates the electronic properties of the junction. We successfully modeled both theI-V and the NDOS curves of Fig. 3 using the energy level diagram shown in Fig. 1C (8). This calculation provided further evidence for the correspondence between the solution-phase and solid-state redox properties of these systems.

Several “control” devices were fabricated in which LB monolayers of long-chain alkylcarboxylate salts, such as C19H39CO2 Na+, were used as the molecular layer. These linear amphiphiles acted as nonresonant tunneling barriers between the two electrodes, similar to what has been reported (20). It was possible to increase the junction resistance of the control devices by oxidizing the Ti electrode at +2-V bias. However, because this process involved the diffusion of oxygen into the electrode, an initial oxidation step had a relatively small effect on the tunneling resistance. An extended (1-hour) oxidation at +2 V eventually led to a factor of 20 increase in tunneling resistance, as measured at −2 V. No measurable change in the NDOS of the junction was detected for the control devices, indicating that the control molecules themselves were not altered by the oxidation of the Ti electrode.

Once we had determined the conditions for addressing and configuring the individual devices, we were able to configure linear arrays of devices into AND and OR wired-logic gates. An experimentally derived truth table for a two-input AND gate is shown in Fig. 4A. On the scale of the yaxis, experimental noise of the output current for a given input address state is not discernible. The truth table of any AND gate is such that a high response is only recorded when both inputs are high. Although the wired-AND gate presented here is simply a network of dissipative devices, its response is much better than would be expected from standard resistors. Even if one had complete flexibility in choosing the values for the resistors in a standard network, the maximum difference in current flow that one could hope for between the high and low states would be a factor of 2. However, we observed something closer to a factor of 15 because of the highly nonlinearI-V characteristics of our resonant-tunneling junctions. This nonlinearity is extremely useful because it should allow combination of wired-AND and wired-OR gates with each other to create more complex and intrinsically nonlinear logic circuits.

Figure 4

Experimentally measured truth tables for logic gates configured from linear arrays of molecular switch junctions. For all logic gates, a low input is held at ground, and a high input is held at +2 V. Arbitrary high and low output current levels are assigned on each plot. The inputs are labeled alphabetically, and one device, labeled L, was configured as a load impedance on the gate. (A) The current output of a two-terminal AND gate as a function of input address, with an accompanying schematic of how the device was configured. (B) The current output (plotted on a logarithmic scale) for a three-input OR gate (solid trace), which was subsequently reconfigured into a two-input OR gate (dotted line) by oxidizing input C. For the two-input gate, the same truth table was measured, but input C was a dummy input. The [001] address state does raise the output current level, but not nearly enough to make the output “high.”

In Fig. 4B we present truth tables for an OR gate that was initially configured as a three-input OR gate, but then reconfigured as a two-input OR gate. The truth table of any OR gate is such that if any input combination is held high, then the output is high. For this experiment, the truth table of the three-input device was measured, and then the device corresponding to input C was oxidized. The same truth table was measured again, but with input C as a dummy variable. The point of this exercise was not only to demonstrate the reconfigurability of our devices, but also to investigate the issue of crosstalk between an operating gate and an interfering device somewhere else in the same circuit (21). In this case, the interfering device was the oxidized dummy input (labeled C in Fig. 4B) to the two-terminal gate. As seen from Fig. 4B, both versions of the OR gate yield quite high discrimination between low and high output states. The difference between high and low is at least a factor of 30, even for the reconfigured two-input gate.

These logic gates are similar in principle to diode-based wired-logic gates, with the exceptions that they are electronically configurable and that their behavior remains highly nonlinear over a broad voltage range. This means that the truth tables shown in Fig. 4 remain current-scaled versions of Fig. 4 at all operating voltages from −0.5 and −2.5 V, although they are substantially noisier at the lowest voltages. Furthermore, the simplicity of the fabrication process makes them inexpensive to produce. The simplicity of this device fabrication process makes it straightforward to quickly screen molecules for useful electronic properties. A clear target to screen for is a reconfigurable molecular switch, which would allow for the demonstration of random access memory as well as all logic functions [through the use of stored look-up tables (7)]. A second challenge will be to scale down the size of the device and integrate molecular switches with molecular-scale wires, such as metallic carbon nanotubes. The fact that the electronic properties of our compounds are very similar in both the solid-state devices and in a dilute solution suggests that these devices should scale down in size without significant change in operation. The devices presented here provide a compelling argument that molecular switch devices may play an important role in future computational technologies.

  • * These authors contributed equally to this work.

  • To whom correspondence should be addressed. E-mail: heath{at}


View Abstract

Navigate This Article