Direct spectroscopic monitoring of conductance switching in polythiophene memory devices

Abstract

A large number of nonvolatile memory devices have been reported with both inorganic and organic components, and many of these involve changes in device resistance between a high conductivity “ON” state and a low conductivity “OFF” state. The mechanism of memory action in many of these devices is uncertain, and may be based on many phenomena, including redox reactions, metal filament formation, charge storage in “floating gates”, and redistribution of oxide vacancies. We report here a Raman spectroscopic probe of organic polymer memory devices which permits direct monitoring of the doping state and conductivity of polythiophene in a 3-terminal device. The polymer conductance is controlled by voltage pulses between the source and gate electrodes in FET geometry, while the conductance state is read out by a separate circuit between source and drain. The conductance was directly correlated with the Raman determination of the density of polarons in the polymer film, which was shown to control both the “electroforming” process and the conductance switching in working memory devices. The polymer conductance change requires a redox counter-reaction at the gate electrode, and atmospheric effects on performance indicate that water and oxygen reduction are involved. The observations are consistent with a redox process between the gate and source electrodes which modulates the polaron concentration and source–drain conductivity. This mechanism provides a framework for optimization of the device by changing its composition and geometry, particularly the identity of the redox counter-reaction and control of ion mobility.

Introduction

Solid-state nonvolatile memory (NVM) has enabled portable electronics in the form of cell phones, media players, and a large number of additional consumer electronic devices. The dominant technology is “flash” memory based on a floating gate field effect transistor (FET) structure using a thin layer of silicon isolated from a write/erase electrode by silicon oxide. Charge injected into the floating gate modulates the FET conductance, and the high and low conductance states are stable for typically >10 years. In terms of volume, more than 100 billion units of solid state NVM were produced as of 2009, with floating gate “flash” by far the most common [1]. While the density and utility of flash memory have increased dramatically in the past decade, it is much slower than dynamic random access memory (DRAM), requires relatively high voltage to “write” and “erase”, and has a limited cycle life (10³–10⁵) due to fatigue of the SiO₂ gate insulator. High economic value and the shortcomings of flash memory have stimulated research into a wide range of alternative NVM devices based on both “2-terminal” or “crossbar” geometry and on “3-terminal” structures resembling an FET but usually with a quite different mechanism. Reviews of inorganic [2], [3], organic [4], and polymer [5], [6] NVM approaches have appeared recently, but several organic approaches which are relevant to the current paper are noted here.

As shown in Fig. 1a, a 2-terminal cross-bar architecture consists of an organic or polymer (O) layer of different thicknesses sandwiched between two conducting electrodes (M) made of identical (MOM) [7], [8], [9], [10] or different materials (MOM′) [11], [12], [13]. When a ferroelectric polymer is used as the organic layer, bias induced switching between two stable polarization states can form the basis for memory [14], [15]. In non-ferroelectric polymers, the organic layer can also be impregnated with metal nanoparticles [16], [17], to impart or enhance the desired memory response. Conductivity switching has been reported in many cross-bar devices fabricated from a wide variety of organic molecules and polymers with and without additional stacking layers. These devices often require an initial “electroforming” step to jumpstart the device, where a high voltage is applied to initiate conductance switching [5], [6]. The device can then be subsequently operated at lower voltages. Although the mechanism of conductivity switching in these devices has often been attributed to the intrinsic properties of the stacked layer materials, clear determinations of switching mechanism are rare [4], and in many cases the data can be explained by filament formation and destruction [11], [18], [19], [20].

Three-terminal NVM devices with geometries similar to that of field effect transistors (FET) have two potential advantages over the 2-terminal devices. First, the FET geometry is already widely used in both processors and flash memory, so integration of a new 3-terminal NVM device with existing manufacturing should be possible without drastic processing changes. Second, the “write/erase” (W/E) circuit, often between the “gate” and “source” shown in Fig. 1b can be separated from the “read” circuit between the “source” and “drain”. As will be discussed later, this separation permits nondestructive readout, independent control of W/E and read events, and allows direct correlation of the conductance readout with the redox properties of the organic or polymeric semiconductor.

Existing three-terminal organic nonvolatile memory (ONVM) devices can be classified into three categories based on the mechanisms of operation, namely ferroelectric (FeFET), charge trap (CtFET), and floating gate OFET (FgFET) memory. In FeFET memory, ferroelectric polymers which have large intrinsic dipole moments are used as the gate dielectric. Reorientation of the dipole by a “write” voltage results in a persistent change in the channel (S-D) conductance. The conductive state is retained even when the power is turned off due to remanent polarization. Application of “erase” bias reverses the dipoles and returns the device to the low conducting state [6], [14], [21], [22], [23]. In CtFET memory [24], [25], [26], [27], a polymer layer called an ‘electret’ is added between the gate dielectric and the organic semiconductor (OSC) layers. For example, in CtFET with p-type OSC, the carriers accumulated in the channel upon application of −ΔV_G can tunnel and get trapped in the electret layer. The positive trapped charges screen the gate field and shifts threshold voltage (ΔV_th) to a more negative value. ΔV_th is the gate voltage at which the carrier accumulation is high enough to make the channel conducting. The trapped charges in the electrets can be detrapped by application of the reverse +ΔV_G, shifting ΔV_th to its initial value. Hence, ΔV_th of the device can be reversibly shifted between these two ΔV_th values, thereby yielding a high ON and low OFF channel currents at a particular ΔV_G, preferably at ΔV_G = 0 V. The difference in the ΔV_th, obtained on applications of ±ΔV_G, is the memory window and is proportional to the trapped charge density in the electrets. The mechanism of operation of floating gate OFET (FgFET) [28], [29], [30] memory is similar to CtFET except that the charge trapping sites in this case are made of metal or semiconducting nanoparticles or conjugated organic molecules either embedded in gate dielectric or sandwiched between two different polymer layers constituting blocking and tunnel layers. Some OFET devices exhibit a persistent “threshold voltage shift” or “bias stress effect” which is associated with device degradation [25], [31], [32], [33]. Although the voltage shift is often attributed to trapped charge, the location of that charge and the mechanism underlying the memory effect are not yet clear [31]. Our group has reported 2-terminal low-volatility memory devices based on redox reactions in small organic molecules [34], [35], TiO₂ [36], [37], [38], and conducting polymers [39], [40], [41]. We used in situ Raman and UV–Vis spectroscopy to monitor structural changes in finished devices, in order to correlate conductivity and memory effects to chemical structure. Spectroscopy of polythiophene devices showed that voltage pulses resulted in reversible oxidation of the polymer to its conducting “polaron” state, and that the process required a corresponding reduction reaction to accompany polymer oxidation. We determined that residual water in a polythiophene/oxide “stack” and a catalytically active electrode are required for long retention of the polaron state of the polymer. Investigation of a variety of device compositions containing different metal oxides and various electrode materials indicated that the catalytic reduction of H₂O to chemisorbed hydrogen and mobile hydroxide ion was the most likely counter-reaction for polymer oxidation [39]. However, the oxide layer in the 2-terminal geometry prevented direct correlation of spectroscopic results with device conductance, since the oxide itself had low conductivity. Furthermore, when the oxide is thin enough to permit electrical observation of polymer conductance changes, recombination reactions are likely which decrease retention time. In order to establish that polymer oxidation is directly responsible for observed conductance changes, we adopted the 3-terminal geometry of Fig. 1b. Not only did the approach permit determination of the conductance switching mechanism, it also revealed useful information about how to control several aspects of memory function by variation of device composition.