Understanding the complementary resistive switching in egg albumen-based single sandwich structure with non-inert Al electrode

Figure 1(c) shows an AFM image of the albumen film after thermal baking. Owing to self-filling induced by the free movement of the protein macromolecular chains [37], the surface of the albumen film is relatively smooth with a root mean square (rms) roughness measuring approximately 0.6 nm. This value is consistent with previous reports [33–35]. The smooth surface morphology of the albumen film is advantageous for the device fabrication, ensuring that uniform performance can be achieved [33].

The resistance switching behavior of the Al/Albumen/ITO device is characterized by I-V curves with the applied voltage ranging from −11 to +11 V. In previous reports [8, 9], an electrical-forming step is necessary to attain the CRS behaviors, which may lead to unintended complexity in manipulation. Figure 2(a) shows that typical CRS characteristics are obtained in the Al/Albumen/ITO device without forming process. The I-V curve rotates clockwise for the forward sweep in positive bias, and the resistance is relatively high in the initial '0' state. The current increases abruptly at ∼1 V, indicating the device is SET to low resistance state (LRS) at V_SET. A further voltage increase gives rise to the RESET transition, where the current decreases with increasing voltage. The positive high resistive state (PHRS) is thus obtained. The voltage value that makes the device switch back to HRS is V_RESET [16]. In reverse sweeping from +11 to 0 V, the current decreases monotonically, and the device remains in PHRS. The PHRS at low voltage bias is regarded as '1' state. Similar SET and RESET transitions are achieved when the applied voltage is set to negative bias. The I-V curve also rotates in a clockwise manner analogous to that in positive bias, thus allowing the achievement of negative LRS and HRS, denoted as NLRS and NHRS, respectively. To verify the device-to-device uniformity, the I-V characteristic curves of 10 different devices were measured and shown in the inset of figure 2(a). All the devices exhibit nearly consistent CRS behaviors under the same conditions. The endurance performance was examined using 100 cycles of voltage sweeps on a single CRS cell. Figure 2(b) shows that the shape of I-V curves and CRS switching properties are well maintained. The statistical variations of V_SET and V_RESET are acquired from the measured I-V curves and plotted as histograms in figure 2(c) to evaluate the cycling uniformity. A window of 1.4 V is clearly distinguished by comparing the distributions of V_SET and V_RESET. It is known that the logic states '0' and '1' in CRS device can be identified by applying a proper readout voltage between V_SET and V_RESET [11]. Figure 2(c) shows a remarkable operating window that ensures a reliable readout, which is a requirement for memory device applications.

Zoom In Zoom Out Reset image size

The memory performance was further evaluated by conducting the cycling endurance characteristic test by applying alternate positive and negative voltage sweeps on the Al/Albumen/ITO device [13, 17, 22]. Figure 2(d) shows the evolution of LRS and HRS in 140 consecutive switching cycles, exhibiting a favorable and robust memory margin of 10², significantly higher than the commercial standard (≥10) [10]. To assess the retention ability of our device, we plotted the HRS and LRS against time in figure 2(e), which indicates both resistance states persist for up to 5 × 10³ s without significant degradation. Additionally, as depicted in figure S2 in the supplementary material, the CRS behavior remained unaffected even after one month of storage in the air condition. These experimental observations highlight that memory cells with a single albumen resistive switching layer can achieve reliable CRS characteristics.

CRS devices are promising for addressing the sneak-path current issue in large passive crossbar arrays. Table 1 demonstrates that the performance of our albumen-based CRS device is comparable to that of oxide-based CRS devices. Suppose a crossbar array is constructed using CRS cells. In that case, the read margin (ΔV_READ) normalized to pull-up voltage (V_pu) determines the array density. The array size was evaluated using the worst-case scenario and the method presented in previous reports [5, 10]. As shown in figure 2(f), the read margin decreases as the cross-line number (N) increases. The maximum N value is found to be 192 at the 10% read margin of ΔV_READ/V_pu. This finding indicates that the anti-crosstalk features of the albumen-based device are advantageous in handling sneak-path current issue. Additionally, the low cost of fabrication, along with its simple device structure and straightforward fabrication process, renders our device a highly promising option for practical applications.

It is well accepted that the performance of RS devices can be modulated by adjusting the compliance current (I_CC) [2]; for example, BRS eventually transforms to CRS in an oxide-based device under a controlled compliance current [16]. As for our biomaterial-based device, I_CC of 1.0 mA was deliberately applied on the SET branches during the voltage sweeps. Figure 3(a) shows I-V curves of repeatable cycles exhibiting typical BRS behavior. During the retention characteristics test, no significant degradation for both HRS and LRS is observed, as shown in figure 3(b). These measurements demonstrate that the albumen-based device is a promising candidate for bioelectronic applications with crossbar architecture.

Zoom In Zoom Out Reset image size

The ON/OFF ratio shown in figure 3(b) is relatively smaller compared to previous work on the same structure [32], where an abrupt change in resistance can be observed. Conversely, the resistive states in our device gradually change under voltage scans, which is the significant feature of analog RS behavior that mimics bio-synaptic functions [39]. In the artificial synaptic device, the synaptic weight (connection strength) is deemed as the conductance of the memristor, which can be adjusted through different electrical pulses. The non-volatile change of synaptic weight, termed as synaptic plasticity, is well accepted as the mechanism of learning and memory in human brain [35]. Figure 3(c) illustrates the application of two consecutive positive voltage pulses (7 V, 100 ms) to the memory device. The results demonstrate that the first stimulated pulse increases the current response, and an enhancement of response occurs when the second pulse closely follows the first one. This feature can be described by the paired-pulse facilitation (PPF), a typical short-term plasticity (STP) [35]. When two negative voltage pulse (−15 V, 100 ms) are applied, a decrease in the response current corresponds to the paired-pulse depression (PPD), as shown in figure 3(d). These findings reveal the tunable switching behavior of the Al/Albumen/ITO cell and a promising prospect for applications in neuromorphic chips featuring crossbar arrays geometry.

To clarify the conduction mechanisms, it is a common practice to analyze the conduction behaviors at LRS and HRS by fitting the I-V characteristics [16, 19]. The I-V curve at the '0' state is replotted in double logarithmic scale and linearly fitted in the low positive voltage region, as shown in figure 4(a). The slope of the curve is 1.94, close to 2, indicating that the conduction can be described by the Mott-Gurney (MG) law I ∝ V² [40]. This suggests that the current through the Al/Albumen/ITO device is initially governed by the injected charge carriers and trap centers in the active layer [22]. When the CRS device switches to the positive LRS (PLRS), the experimental data can be well fitted with the linear relationship ln(I) ∼ V^1/2, as shown in figure 4(b). In this case, the dominated mechanism converts to the interfacial Schottky emission, given as [19]

$$\begin{eqnarray}&&I={A}^{* }{T}^{2}S\exp \left[\displaystyle \frac{-q\left({\phi }_{B}-\sqrt{qV/4\pi d\varepsilon }\right)}{kT}\right],\end{eqnarray} \tag{ 1 }$$

where A^*, S, q, ϕ_B , d, ε, k, and T represent the Richardson constant (120 A cm⁻² K⁻² for Al [41]), the area of TE, the electronic charge, the Schottky barrier height (SBH), the thickness of RS layer, the dielectric permittivity, the Boltzmann constant, and the temperature, respectively. The intercept b of ln(I) versus V^1/2 is expressed as

$$\begin{eqnarray}&&b=\,\mathrm{ln}\left({A}^{* }{T}^{2}S\right)-\displaystyle \frac{q{\phi }_{B}}{kT}\cdot \end{eqnarray} \tag{ 2 }$$

Therefore, the value of SBH ϕ_B can be estimated from the intercept b of ln(I) ∝ V^1/2 [16]. In the positive reverse sweep from 11 to 0 V, the device remains in the PHRS, and the I-V curves in the high voltage region can also be well fitted by the Schottky emission model as shown in figure 4(c). In contrast, the I-V curve in the low voltage region ('1' state), shown in figure 4(d) replotted in a double logarithm scale, shows a linear change with a slope of 1.10, which corresponds to Ohmic conduction mechanism (I ∝ V¹) [16]. This suggests that conducive channels connecting the TE and BE are formed through the albumen layer. During the negative voltage bias, similar fitting results of NLRS and NHRS are also observed, as shown in figures S3(a) and (b) in the supplementary material, proving that the interfacial Schottky emission is the dominant conduction mechanism for the switching behaviors in the Al/Albumen/ITO devices. The extracted values of b and the corresponding SBHs ϕ_B for both voltage polarities are summarized in table 2. The fitting results demonstrate that the Schottky barrier decreases as the device switches from PLRS to PHRS, whereas an enhancement of the barrier can be observed during switching from NLRS to NHRS, implying different formation mechanisms of the barriers in opposite voltage polarities.

Zoom In Zoom Out Reset image size

The dominated Schottky emission mechanism, based on curve fitting, implies the formation of a barrier at the TE/Albumen interface [19]. Considering the non-inert Al electrode has a strong tendency to be electrically oxidized by the pre-existing oxygen element in Albumen film [32], depth-profiling XPS analysis was performed to determine the chemical bonding states of Al TE layer after the I-V curves measurement. Figure 5(a) displays the normalized Al 2p XPS spectra from the Al electrodes at various Ar⁺-sputtering etching depths. The Al 2p XPS scans are deconvoluted by fitting with the Gaussian peaks. The main peak is around 72.6 eV, pertaining to the metallic Al-Al bonding [42]. Notably, a new peak deconvoluted at the small shoulder in the higher energy region emerges, corresponding to the formation of Al-O bonding of Al₂O_x [42, [43]. In addition, the relative intensity of Al-O peaks increases when the etching depth approaches the Al/Albumen interface, suggesting that the non-inert Al electrode is oxidized by absorbing oxygen elements from the albumen layer [44]. This is consistent with the detected O 1 s XPS peak centered at 531.7 eV at a depth of 30 nm, as shown in figure 5(b) [45]. Figure 5(c) depicts the depth-dependent ratios of Al and Al-O components estimated from the areas of the deconvoluted XPS spectra, which provides clear evidence of an oxide layer near the Al/Albumen interface. Therefore, the CRS behavior in the Al/Albumen/ITO device may be strongly related to the interfacial Schottky barrier.

Zoom In Zoom Out Reset image size

Note that egg albumen is composed of various proteins [33]. Conduction characteristics could be related to oxygen ions originating from the oxygen-containing functional groups in hydroxyl and carboxyl proteins [31, 32]. Oxygen ions are randomly distributed in the albumen layer prior to applying a voltage bias to the Al/Albumen/ITO device. When the bottom electrode is grounded, and the top electrode is subjected to a positive voltage, oxygen ions are driven toward the Al electrode, leaving oxygen vacancies, as shown in figure 6(a). As the applied positive voltage increases, enough generated oxygen vacancies accumulate into conductive filaments connecting the TE and BE, resulting in PLRS. Meanwhile, the Al electrode absorbs oxygen ions, forming a Al₂O_x layer at the Al/Albumen interface. The Schottky emission begins to dominate because electrons must surpass the barrier between filaments and Al oxide layer. When the positive bias exceeds V_RESET, more oxygen ions infiltrate into the TE to react with Al. The Al₂O_x barrier reduces the current intensity and the device switches to PHRS, as shown in figure 6(b). After a lower negative voltage bias is applied, the recombination of partial oxygen ions in the barrier with oxygen vacancies in the albumen layer diminishes the barrier, decreasing the overall resistance (figure 6(c)). As a result, recovery of the metallic Al electrode brings the device to NLRS. With increasing the negative voltage bias, insufficient oxygen ions in the barrier have been consumed to bond with oxygen vacancies, causing the filaments to break, preferably near the TE [19]. As shown in figure 6(d), a new barrier is formed between the Al electrode and the albumen layer, and NHRS is thus obtained. When the compliance current is applied to the device, the Al layer absorbs a limited number of oxygen ions, hindering the formation of Schottky barrier. As a result, BRS behavior is detected in figure 3(a).

Zoom In Zoom Out Reset image size

According to the proposed RS model, the origins of the Schottky barriers in positive and negative voltage sides are correlated with different behaviors of oxygen ions and oxygen vacancies. It has been revealed that the increase of oxygen vacancies would reduce the SBH of the Schottky interface [46]. Consequently, the generation of oxygen vacancies in positive voltage bias leads to the decrease of SBH as the device switches from PLRS to PHRS, as illustrated in table 2, and the consumption of oxygen vacancies under the action of negative voltage enhances the SBH when switching from NLRS to NHRS.

The above model explains the CRS behavior regarding the interfacial barrier formation, which is directly related to the non-inert electrode. This model can be partially supported by examining the conduction behaviors with voltage cycling in positive polarity. As shown in figure 7(a), the amplitude of the current decreases monotonously with the voltage sweep. No obvious CRS characteristics have been detected for the last several sweeps. After that, the chemical bonding state of the Al electrode was analyzed by XPS. Figure 7(b) shows the Al 2p spectrum and the corresponding Gaussian deconvolution fittings, taken at a sputter-etching depth of 10 nm. According to the area of the deconvoluted Al 2p XPS peaks, the ratio of the Al-O component is estimated to be 67.5%. This value is significantly higher than that (∼13.0%) shown in figure 5(a), indicating that more oxygen ions are captured to form Al₂O_x component due to continuous application of positive voltage bias.

Zoom In Zoom Out Reset image size

To further explore the role of electrode oxidation in the emergence of CRS, the Al layer was replaced with inert Pt material. Figure 8(a) shows I-V curves of repetitive voltage sweeping cycles for the Pt/Albumen/ITO device. Compared to that in devices with Al electrodes, significant deterioration in CRS features is illustrated due to the poor affinity of Pt to oxygen ions [16]. Meanwhile, measurements of endurance in the Pt/Albumen/ITO device exhibit a significant decay as cycling increases, as shown in figure 8(b). Previous studies have indicated that oxygen at the interface can diffuse into the atmosphere through Pt grain boundaries, resulting in degradation after successive voltage cycling [47]. The influence of electrode material on BRS properties has been previously established through early researches [43, 48]. Our theoretical analysis and experimental evidence suggest that the non-inert Al electrode plays a critical role in controlling CRS behavior in bio-memory cells with a single resistive switching layer.

Zoom In Zoom Out Reset image size