Reversible memristive switching phenomena have been extensively studied by many groups in a variety of material systems for possible applications as nonvolatile memory, logic circuits and neuromorphic computing [1–4]. Considerable efforts have been made to explore new switching systems, understand their physical mechanisms and realize their integration with conventional complementary metal-oxide semiconductor (CMOS) processes to meet the ever-increasing demand for higher density and performance electronic circuits [5–10]. Almost all of the anion-based resistive switching materials reported so far have been insulating/semiconducting metal oxides because of the wide range of their electrical properties and their exquisite dependence on doping concentrations. However, non-oxide ionic insulators/semiconductors are a much larger material pool that may also exhibit memristive switching behavior, but with the exception of the chalcogenides [11, 12] have not been extensively explored. Obvious examples are semiconducting nitrides, which have been intensively studied for photonics applications but not much for memristive phenomena. In a few resistance switching devices incorporating nitrides reported in the literature, they were used either as an electrolyte material for an electrochemical metallization cell [13] or as an electron trapping medium [14] rather than for a mobile ionic species. In principle, ionic switching could also occur in nitrides, such as AlN, in a manner similar to the oxides [15]. The Al–N system has a very simple phase diagram, which may lead to high switching endurance as shown in the Ta–O and Hf–O oxides that possess similar phase diagrams [16–19]. The wide band gap, high electrical resistivity of the stoichiometric compound and high thermal conductivity exhibited by AlN could result in superior switching performance, such as a large ON/OFF ratio, a good thermal stability, etc. [20–23]. Furthermore, some metallic nitrides, such as TiN, WN and TaN, are commonly used in semiconductor fabrication foundries as electrode materials. This makes a semiconducting nitride a possibly more attractive switching material than oxides for a memristor application because the chemical complexity and potential thermodynamic instability of an interface between an electrode nitride and a switching oxide can be avoided. The demonstration of ionic switching behavior in nitrides could open up an entirely new area in resistance switches with significant opportunities.

Here we present our observed memristive switching behavior in Al nitride thin films with various electrode materials. Plasma enhanced atomic layer deposition (PEALD) was used to deposit AlN thin films that showed reproducible memristive switching phenomena. A variety of material characterizations were conducted to reveal the structural, chemical and electronic properties of the films and ensure that the observed switching did not arise from inadvertent oxide contamination and to shed light on the mechanism of nitride switching.

Typical quasi-DC current–voltage (I–V) switching loops are shown in Fig. 1. Devices were fabricated on thermally grown 200 nm-thick SiO2 on a Si substrate. AlN films 5∼8 nm thick were deposited by PEALD using Trimethylaluminum (TMA, Al(CH3)3) and N2:H2 (20:40 SCCM) mixed gas as a metal organic precursor and a reactant gas at 350 °C. For the device represented in Fig. 1, the bottom electrode was a blanket 20 nm TiN layer grown by PEALD using tetrachlorotitanium (TiCl4) and N2:H2 (4:40 SCCM) mixed gas as a precursor and reactant gas, respectively, at a wafer temperature of 350 °C. In order to avoid the oxidation of the TiN surface, the AlN film was directly grown on TiN in the same reactor without exposure to air. Finally, either a Pt or Al top electrode was deposited by electron-beam evaporation through a shadow mask (25 μm to 200 μm in diameter) onto the TiN/AlN stack to complete the device fabrication. The atomic concentrations and impurity levels of the AlN thin film were measured by Rutherford backscattering spectroscopy (RBS) and secondary ion mass spectrometry (SIMS). The four-terminal I–V characteristics of the devices were measured using a semiconductor parameter analyzer (HP-4156), which can extract the actual voltage drop on the device from the total applied voltage. A quasi-DC voltage sweep was applied to the top electrode with the bottom contact grounded at ambient temperature in all the electrical measurements.

Fig. 1
figure 1

Typical current–voltage (I–V) switching loops for AlN based memristors. 100 consecutive semi-log I–V loops for a TiN (20 nm)/AlN (8 nm)/Pt (30 nm) disc device with a diameter of 200 μm (inset). The black I–V curves marked with crosses show the electroforming step

Full size image

Figure 1 shows 100 consecutive switching I–V loops of the TiN/AlN/Pt device in semi-log scale. The left insert is the same data re-plotted on a linear scale. The switching is bi-polar with excellent repeatability and an ON/OFF conductance ratio over 100 for voltage excursions of −1.5 to 2.0 V. As shown schematically in the right insert to Fig. 1, the device size was fairly large (200 μm in diameter) and the ON current was at the mA level. Here the current compliance (I COMP) was set to 2 mA for both ON and OFF switching. Before displaying reversible switching, the devices fabricated in this study required a so-called ‘electroforming’ process [24, 25], for which the devices changed irreversibly from the highly insulating as-fabricated state using a voltage sweep that was about 1–2 times the magnitude of the reversible switching voltage to the switchable state indicated by the black curve with cross marks in Fig. 1. After electroforming, the devices exhibited reversible bipolar resistive switching between ON and OFF states under opposite applied voltage polarities.

We systematically studied switching parameters, such as the OFF switching current (I OFF) and resistances of the ON and OFF states, with different device sizes. Switching parameters are more sensitive to I COMP than the device size (at least for micro-devices). Similar switching parameters can be obtained with different device sizes by using the same I COMP (here 1 mA used), as shown in Fig. 2(a). I OFF is closely related to and mainly determined by I COMP of ON switching, as shown in Fig. 2(b). The minimum I COMP that can be used to obtain a reliable non-volatile switching is smaller for a smaller device, which might be a result of a smaller leakage current in a smaller device. The nitride device with 25 μm in diameter can reliably switch under several tens of μA of I COMP, which seems to be smaller than most oxide based switches with a similar size.

Fig. 2
figure 2

Device size and current compliance (I COMP) effect on device resistance and OFF switching current (I OFF) level. (a) Device resistance and I OFF vs. device size when 1 mA I COMP was used for ON-switching. (b) Correlation between I COMP and the subsequent I OFF of devices with a diameter of 25 and 200 μm

Full size image

In order to be able to unambiguously attribute the observed switching behavior to a nitride rather than other impurity materials, such as oxides, unintentionally formed in the PEALD process, thorough material characterizations on the switching films and control samples were performed. A plan-view transmission electron microscope (TEM) image of a 32 nm AlN film grown on a carbon grid at 350 °C in Fig. 3(a) shows that the film is mainly amorphous, while the electron diffraction pattern in the insert reveals that it does contain nanocrystallites with the hexagonal wurtzite structure [26, 27]. The high power (400 W) of the plasma used for PEALD may induce the partial crystallization of the AlN thin film in spite of the relatively low substrate temperature. TEM study on AlN films grown directly on Pt sub-layer and on carbon grid reveals that both films consist of nano-crystalline wurzite AlN, with the crystallinity of the former slightly enhanced.

Fig. 3
figure 3

Structural, compositional, and optical characterizations of the nitride film. (a) Plan-view TEM images and electron diffraction patterns of 35 nm Al nitride films grown on Pt sub-layer (upper) and directly on carbon mesh (lower). (b) SIMS results showing the impurity level in the switching nitride film. (c) UV–VIS spectrum determining the optical band gap (∼5.5 eV) of nitride film

Full size image

Rutherford backscattering spectroscopy (RBS) indicated that the AlN film was significantly Al-deficient [Al:N=45(±1):54(±4)]. The secondary ion mass spectrometry (SIMS) depth profile revealed uniform atomic concentrations through most of the AlN film grown on SiO2 sub-layer and a low level of both carbon and oxygen impurities, as shown in Fig. 3(b). The carbon impurity level was below 1 %, consistent with the measurement results by both RBS and X-ray photoelectron spectroscopy (not shown here). The oxygen impurity concentration was even lower than that of carbon, as shown by SIMS. Impurities were expected to be incorporated from the metal-organic precursor molecules (TMA) and reactant gas mixture (N2+H2) under the 400 W remote plasma in the ALD process. However, as suggested by the XPS depth profile data, the oxygen impurity may have originated from surface oxidation when the AlN film samples were exposed to air for a few hours before each material characterization was conducted. However, there should not be air contamination issues for those devices with the switching interface fabricated in-situ by depositing both the electrode and the AlN layers in the ALD chamber without breaking the vacuum. UV–Visible spectroscopy revealed an absorption edge that started at ∼5 eV from the AlN film, corresponding to substitutional oxygen (ON) or vacancy-related (VAl or VN) defects as shown in Fig. 3(c) [28, 29]. The zinc-blend AlN crystallite phase could also result in the sub-band-gap absorption around 5.3 eV, which is unlikely since this phase does not appear in the electron diffraction pattern in the insert of Fig. 3(a) [30].

Despite the potentially large amount of dopant impurities (such as ON) and structural defects (VAl or VN), the pristine AlN film were highly insulating. As mentioned earlier, an electroforming step was needed to pre-condition the device to a switchable state. Similar to the electroforming process in some oxide memristors that have very insulating pristine oxide films [9, 25, 31], sometimes gas bubbles were seen in the electrodes of electroformed nitride memristors, suggesting possible nitrogen gas release from the AlN film during electroforming. Therefore, even though the RBS data suggested that the as-deposited AlN film might be N-rich, the active part of the film responsible for switching could be N-poor as a result of the electroforming process. Donors, such as intrinsic nitrogen vacancies \(\mathrm{V}_{\mathrm{N}}^{+}\) [32–35], and extrinsic oxygen \((\mathrm{O}_{\mathrm{N}}^{+})\) [32], and carbon \((\mathrm{C}_{\mathrm{Al}}^{+})\) substitutional impurities [35], are known to be deep levels in AlN because of its wide band gap. It is presently not clear how these donors affect the electrical conduction; they may be mobile ionic species or induce leakage paths [23]. The switching mechanism and detailed microscopic picture of the AlN switching will be addressed in further studies.

In summary, nitride memristors were fabricated using PEALD AlN thin films with a variety of electrode materials. Reversible and reproducible memristive switching characteristics were successfully demonstrated. Various material characterizations confirmed that the pristine PEALD films were mainly amorphous AlN with some nanocrystallites. Dopant-level impurities of O and C were also found in the films. The switching mechanism is under study.