Quasi-unipolar pentacene films embedded with fullerene for non-volatile organic transistor memories Free

In the last few decades, organic field-effect transistors have attracted a tremendous amount of attention due to their wide applicability as functional elements in low-cost, flexible, and large-area electronic devices.^1–5 Significant advances have been made in organic transistors toward realizing rewritable memory devices such as non-volatile organic transistor memory (NOTM) that offers non-destructive read-out with good compatibility with complementary circuits.^6–8 The operation of such devices relies on a variation in the current signal before and after exerting a voltage pulse at the gate to change the electrical environment of the gate dielectric located next to the semiconductor channel. However, practical applications of these devices still requires for their performance to be further improved. The key device parameters that need to be addressed include generating distinct electrical signals between the programmed and erased states of memory cells; reducing the voltage needed programming/erasing the cells, increasing the switching speed; and providing a long retention time for the programmed signals. Accordingly, significant efforts have been made to develop functional gate dielectric systems that are capable of programming and erasing electric signals by using ferroelectric materials,^9,10 self-assembled monolayers,^4,11 nano-floating gate systems using nanoparticles,^12–15 or simple non-polar polymer layers.^16,17

An alternative approach involves devising a new channel architecture for NOTM. NOTM can be classified into unipolar or ambipolar types depending on their electrical characteristics. The unipolar NOTM operates through the transport of either only electrons or only holes, and their programmed/erased signal ratio is relatively large. The large signal ratio is mainly a result of unipolar transistors yielding a low current signal, particularly when the devices are in their completely off state. Although this low current level can be configured as either the programmed or erased signal level for the device, relatively high voltages need to be applied in order to program or erase the memory cells. This is because only one type of charge carriers is involved in programming and erasing unipolar devices.^6,12,16,17 Meanwhile, ambipolar NOTM is capable of simultaneously utilizing both electrons and holes for transport and programming process, and as a result, the advantages and disadvantages are reversed.^18–20 Since both types of charges are useful for both programming and erasing, the voltage required is relatively small. Although the reduction in the necessary applied voltage is an advantage, the resulting programmed/erased signal ratio is relatively poor. This is because sufficiently low current levels cannot be attained by typical ambipolar transistors: a complete charge depletion is hardly achievable as either electrons or holes are present in the channel.²¹ Therefore, a method capable of adopting the respective advantages of these two different types of NOTM can greatly improve device performance.

In this study, we propose a quasi-unipolar channel layer for NOTM in order to improve both the operating voltage and the resulting signal ratio. The quasi-unipolar channel layer was formed by discontinuously embedding an n-type fullerene layer within a p-type pentacene layer through subsequent thermal evaporation. The top n-type layer serves as the source for electrons that allows the device channel to harness both electrons and holes for signal storage processes, while its discontinuous nature allows the actual channel conduction of the device to rely solely on hole transport through the underneath p-type pentacene film. This is the key working principle of the quasi-unipolar approach. The resulting devices exhibited a good programmed/erased signal ratio (the advantage of unipolar NOTM) while utilizing the electrons in the fullerene layer to reduce the voltage necessary to program/erase the devices (the advantage of ambipolar NOTMs). The quasi-unipolar devices required program/erase input signal voltages lower than those necessary for unipolar NOTM, while still retaining the excellent programmed/erased output signal of typical unipolar NOTM. At the same time, the quasi-unipolar devices also exhibited an improved programmed/erased signal ratio relative to ambipolar NOTM under the same operational voltages. Although the results of the present study were achieved by using benchmark fullerene and pentacene, we believe that the combined strategy introduced herein can be readily expanded to recently developed advanced materials systems and can thereby make a great contribution to further improve the performances of NOTM.

A combination of fullerene and pentacene was selected to construct the quasi-unipolar layer. The selection was made due to a favorable energy alignment that allows for charge transfer between both materials. The electrons in pentacene can be readily transferred to fullerene, generating holes in pentacene (Fig. 1(a)). Previous studies have demonstrated enhanced transport behavior for both electron conduction through a fullerene layer and hole conduction through an underlying pentacene layer.^22,23 Unlike previous reports that used bilayer-structures with continuous fullerene and pentacene layers, the quasi-unipolar layers constructed in our study were formed by depositing fullerene discontinuously onto the pentacene thin film.^24,25 Fig. 1(b) displays a schematic of the cross-section of the quasi-unipolar layer. For this device, electrons are generated at the fullerene/pentacene interface and can be harvested for charge storage or release, while the transport channel retains its unipolar characteristics due to the underlying p-type pentacene layer. The second reason for selecting these materials is that they are some of the most heavily studied p- and n-type organic materials and are both commercially available. This universalness has been important in attracting interest from diverse research communities that can contribute advancements in NOTM research.

The procedure to fabricate the devices is as follows. First, a heavily n-doped silicon wafer with an oxide thickness of 100 nm (specific capacitance, C_ox = 34 nF/cm²) was cleaned through subsequent sonication in acetone, methanol, iso-propanol, and deionized water for 10 min each. A non-polar polymer dielectric consisting of polystyrene (PS, molecular weight = 72 kg/mol) dispersed in toluene was spin-coated (60 s at 3000 rpm) onto the cleaned wafer, which served as a charge storage layer (electret) for the memory function. The detailed working principles of NOTMs based on electret layer is described thoroughly in previous reports.^16,26,27 The resulting film had a thickness of 30 nm. Pentacene was then thermally deposited through a shadow mask onto the PS layer. A calibrated quartz crystal microbalance (QCM) indicated that the pentacene layer had a thickness of ∼15 Å, which corresponds to that of a monolayer (ML)-thick film. Fig. 1(c) displays the morphology of the pristine monolayer-thick pentacene film, as obtained via atomic force microscopy. The formation of such a thin underlying layer of pentacene is critical. This is because the electrons in the discontinuous fullerene layer have to be transferred to or released from the PS layer beneath the pentacene layer to, respectively, generate the program and erase signals. Without breaking the vacuum in the evaporation chamber, fullerene is subsequently deposited onto the pentacene layer. The fullerene layer was further monitored via QCM, and the thickness was observed to vary between 5 and 30 Å. Unlike the planar growth of the underlying pentacene layer, the fullerene on pentacene exhibited island-type growth (Figs. 1(d) and 1(e)). Consequently, a continuous layer of fullerene could only be formed when the nominal thickness of the layer was above 30 Å (Fig. 1(f)). Finally, an additional thick pentacene layer was deposited onto the fullerene/pentacene film. This layer served as the passivating layer against direct exposure of fullerene to the ambient, leading to oxidation.²⁸ As a result, the quasi-unipolar pentacene layer was embedded with fullerene. Finally, Au contacts were thermally evaporated onto the as-prepared quasi-unipolar film by using a shadow mask. These contacts served as the top-contact source and drain electrodes with a channel length and width of 100 μm and 1 mm, respectively. The heavily doped wafer itself comprised the bottom gate electrodes for the NOTM. Additionally, a pair of unipolar and ambipolar NOTM was fabricated as reference devices: the unipolar NOTM contained a pristine pentacene layer as the channel, while the ambipolar devices contained a continuous bilayer of fullerene and pentacene. The electrical characteristics were then obtained in the dark under a vacuum (<10⁻⁵Torr) using a Janis vacuum probe-station connected to a Keithley 4200 semiconductor parameter analyzer.

The black curve in Fig. 2(a) shows a typical drain current (I_D) versus gate voltage (V_G) characteristic (the transfer characteristic) of the pristine pentacene NOTM obtained under a constant drain voltage (V_D = −8 V). The device exhibited typical p-channel behavior, such that I_D increased as V_G increased negatively. After applying a positive voltage of +70 V on the gate (V_Prog) for 1 ms, i.e., the programming process, the transfer characteristic shifted positively (the blue curve in Fig. 2(a)). We noticed that the transfer characteristic did not shift when a V_Prog smaller than +60 V was applied, perhaps because this small programming voltage was not sufficient to bring electrons into the channel to the PS electret layer. When a negative voltage of −60 V was applied to the gate (V_Era) of the programmed devices for 1 ms, i.e., the erasing process, the transfer characteristic recovered its original position (the red curve in Fig. 2(a)). Upon programming and erasing, however, the current levels at a zero gate voltage, which correspond to the resulting respective programmed and erased signals (I_Prog and I_Era, respectively), varied marginally. The resulting small I_Prog/I_Era ratio can be understood to be a result of both the blue and red transfer characteristics remaining in their off-state at V_G = 0 V. This implies that the V_Prog = +70 V or V_Era = −60 V were ineffective in generating memory signals. When a V_Prog higher than +90 V was applied, a larger positive shift in the transfer characteristics could be observed, and the resulting I_Prog/I_Era ratio became more noticeable (>10⁴) (Fig. 2(b)).

Meanwhile, the ambipolar fullerene/pentacene bilayer NOTM were made by depositing a 30 Å-thick fullerene layer onto the pentacene layer. Although the continuous fullerene layer cannot be clearly observed via atomic force microscopy (Fig. 1(f)), the observation of a typical V-shaped ambipolar transfer characteristic curve from these devices confirms that the fullerene layer was formed continuously. Since plentiful electrons and holes are available in such a system, programming and erasing could be performed at lower voltages relative to the pristine unipolar devices. For example, the blue curve in Fig. 2(c) is the transfer characteristic curve taken after the device was programmed (V_Prog = +70 V) and the red curve is that taken after the device was erased (V_Era = −60 V). In contrast to unipolar devices, for which the transfer curves shifted only marginally when this set of lower V_Prog and V_Era were applied (Fig. 2(a)), ambipolar NOTM yielded a larger shift in the curve that was comparable to that of unipolar devices at higher voltages (V_Prog = +90 V and V_Era = −70 V, Fig. 2(b)). This indicates that programming and erasing can be performed more effectively when both electrons and holes are utilized than when only holes are used. The resulting current ratio for the ambipolar-type NOTM, however, exhibited a value as low as ∼10² mainly because a true off-state with a minimal current level is hardly attained in ambipolar devices since either electrons or holes can be, at anytime, involved in charge transport for these devices. Note that the lowest current level in Fig. 2(c) is ∼10⁻¹⁰ A, which is greater than the <10⁻¹¹ A in Figs. 2(d) or 2(e). Thus, the measurements confirm the obvious advantages and disadvantages of using either ambipolar or unipolar NOTM.

Fig. 2(d) shows a series of transfer characteristics that were obtained from the pentacene NOTM embedded with a 5 Å-thick fullerene layer. The initial characteristic is in black, which after applying V_Prog = +70 V is in blue, and after applying V_Era = −60 V is in red. Despite the presence of a fullerene layer in the device, the curves revealed typical p-channel behavior such that I_D increased as V_G increased in the negative direction. This confirms that the embedded fullerene layer formed discontinuously. Upon applying V_Prog = +70 V and V_Era = −60 V, the curves shifted in a sufficiently large manner as to yield a substantially large I_Prog/I_Era value ∼10⁴. At the same V_Prog and V_Era, the pristine unipolar NOTMs yielded only a marginal shift in the curve had been obtained for pristine unipolar devices, and the resulting I_Prog/I_Era had been ∼1. When the thickness of the embedded fullerene layer increased, the transfer characteristics of the device after programming shifted more in the positive direction. In turn, an even larger ratio between I_Prog and I_Era under the same V_Prog and V_Era were observed. Fig. 2(e) displays the series of transfer characteristics that were collected from the pentacene NOTM with a 10 Å-thick embedded fullerene layer. The span further increased when the pentacene film embedded with a thicker fullerene layer was used. For example, Fig. 2(f) shows a series of transfer characteristics for a NOTM with a 20 Å-thick layer of embedded fullerene. A wider shift between the blue and red curves and an enlarged I_Prog/I_Era were clearly achieved with the use of a thicker fullerene layer.

Fig. 3 provides a summary of the data collected from unipolar, quasi-unipolar, and ambipolar NOTM. The squares in Fig. 3(a) display the average voltage offsets of these devices estimated at a current level of 10⁻⁸ A between the curves taken after applying V_Prog = +70 V and V_Era = −60 V to the device. The data are provided as the average of more than three devices that were separately assembled. Under the same programming and erasing conditions, the voltage offset for the ambipolar devices was clearly larger than that for unipolar devices. The equivalent offset value for the unipolar device could be achieved when larger programming and erasing voltages were applied. For example, the red square in Fig. 3(a) was obtained from a device to which V_Prog = +90 V and V_Era = −70 V were applied. Thus, the ambipolar devices exhibit a better response than unipolar devices to programming and erasing in terms of the magnitude of the voltage that is necessary to do so. When the resulting current signal ratios were compared, however, the I_Prog/I_Era value collected for the ambipolar devices was substantially lower than that collected from unipolar devices. Fig. 3(b) displays the averaged I_Prog/I_Era values collected at V_G = 0 V after applying the programming and erasing signals. The I_Prog/I_Era value for the ambipolar device was of ∼10² (at V_Prog = +70 V and V_Era = −60 V) while that for unipolar devices was ∼10⁴ (at V_Prog = +90 V and V_Era = −70 V, which is the condition that yielded a voltage shift equivalent to that of ambipolar devices). Overall, despite the lower programming and erasing voltages that were required, the resulting signal of the ambipolar devices was inferior to that of the unipolar devices.

The quasi-unipolar NOTM allowed to selectively achieve the otherwise mutually exclusive advantages of either unipolar or ambipolar devices. The blue zone in Fig. 3(a) indicates the voltage offset for the quasi-unipolar NOTM as the thickness of the embedded fullerene layer varied. By taking advantage of the electrons that were provided from the n-type fullerene layer, an enlarged voltage offset at a current level of 10⁻⁸ A could be obtained between the curves after applied a V_Prog = +70 V and V_Era = −60 V, which is a condition that resulted in a marginal value for pristine unipolar devices. The offset increased as the thickness of the embedded layer increased, and the resulting offset values were comparable to or even larger than that of ambipolar devices. At the same time, the quasi-unipolar NOTMs yielded I_Prog/I_Era values that were substantially larger than those of ambipolar devices. These values also increased as more fullerene was embedded in the channel layer. The resulting values were, in fact, comparable to those obtained from unipolar devices exposed to larger programming and erasing voltages (at V_Prog = +90 V and V_Era = −70 V). Also, the I_Prog and I_Era data for these devices could be hold for more than 10⁵ s without significant variation, confirming the excellent retention characteristics of the quasi-unipolar memory signal. Overall, the optimal devices yielded larger I_Prog/I_Era values (the advantage of using unipolar NOTM) under reduced programming and erasing voltages (the advantage of using ambipolar NOTM), confirming that quasi-unipolar devices provide an effective approach to improve device performance.

In conclusion, we demonstrated that incorporating a quasi-unipolar channel improves the contrast between the programmed and erased signals for NOTM while also reducing the operating voltage, i.e., we selectively introduced the benefits of unipolar and ambipolar NOTM by fabricating quasi-unipolar NOTM. The quasi-unipolar layer was prepared by depositing an n-type fullerene layer discontinuously onto a thin p-type pentacene film. The resulting quasi-unipolar-type NOTM exhibited both a high I_Prog/I_Era ratio (the advantage of using unipolar NOTM) and a low programming and erasing voltages (the advantage of ambipolar NOTM). The simplicity and the combinatory nature of this approach should be applicable to a wide variety of advanced n-type and p-type organic semiconductors pairs.

This work was supported by the Soongsil University Research Fund of 2012.