Employing suitable materials and device engineering is one of the crucial methods toward the realization of multifunctional memristive devices for constructing bioinspired neuromorphic systems. In this work, dual-functional memristors composed of eco-friendly natural silk sericin, coexistently enabling the achievement of threshold switching and memory switching triggered by adjusting the compliance current value, have been fabricated with a specific two-terminal device structure: Ag/Ag−In−Zn−S/silk sericin/W. Experimentally, the as-manufactured memristors exhibit several desirable qualities, such as low switching voltage (< 0.7 V), relatively small cycle-to-cycle and device-to-device variabilities, nonvolatile multilevel storage characteristics, and rapid switching speed (40 ns). Beyond these qualities, fundamental synaptic behaviors, such as paired-pulse facilitation and spike-timing-dependent plasticity (STDP), have been mimicked. This was made possible by a filamentary mechanism based on Ag migration. The fitted time constants corresponding to the STDP potentiation and depression are about 30 ms, and the highest changes in synaptic weight for positive and negative voltage pulses are 84.4% and 61.7%, respectively. Furthermore, the typical coincidence detection task has been executed, demonstrated by simulation based on the fitted STDP's parameters of the sericin-based device. The results from this study indicate that the sericin-based memristors, as designed, have the potential to be employed in the creation of versatile neuromorphic devices for neuromorphic computing systems.
Innovative neuromorphic computing, a potential alternative to traditional computing, has received tremendous attention from scientific researchers and industries due to its high computing efficiency and low operating energy, which is expected to effectively break through the dilemma of the conventional von Neumann architecture.1,2 With regard to the construction of a neuromorphic computing system with logical operations and data storage capabilities, artificial synapses mimicking synaptic plasticity are gaining popularity and have been deemed one of fundamental constituent elements.3,4 Such inspiration comes from faithfully simulating the human brain that enables parallel processing of a great capacity of data with extremely low power consumption.5 Recently, numerous efforts have been made in the realm of exploring various functionalized artificial synapses.6,7 A prime consideration toward that is to take advantage of advanced memristors based on potential materials and an innovative device structure with appropriate physical properties, such as two-dimensional (2D) layered materials, low dimensional quantum dots (QDs), and biomaterials.8–10
A recent trend in neuromorphic engineering is significantly tied to the utilization of biomaterial-based devices.11–14 Silk fibroin, a noteworthy biomaterial harvested from silkworm cocoons,15 has seen extensive research for its potential in the production of biocompatible memristors.11,12 However, silk sericin (SS), the secondary protein in natural silk often deemed as waste and removed through a degumming process, is garnering renewed interest as a viable biopolymer.16 This interest in SS stems from its unique properties including superior antioxidant ability, remarkable hydrophilicity, exceptional physicochemical stability, and a rich supply of polar functional groups (e.g., hydroxy and amine groups) and amino acids (e.g., tyrosine).17 Another crucial factor is that the hydroxy groups in biomaterials can create a chemical interaction with metal ions, a process known as the chelating mechanism.18,19 These characteristics potentially make SS ideal for achieving high-performance memristors by regulating the ionic process.18 However, the focused development of SS-based memristors remains nascent, with the versatile SS-based memristors capable of achieving dual-functional resistive switching (RS) characteristics yet to be fully realized.20,21 Therefore, it is imperative to delve into the comprehensive RS traits of SS biomaterial-based devices. Additionally, hydroxyl group can lower the energy barrier for the reduction of active metallic ions like silver, facilitating the formation of conductive filaments (CFs).22 Inspired by that, utilizing an active metal electrode is a pivotal consideration in investigating advanced SS-based memristors.
In this work, SS biomaterial has been spin-coated to form a switching functional layer to fabricate two-terminal memristors. Considering the material feature and interface engineering, active Ag electrode and Ag−In−Zn−S (AIZS) QDs have been leveraged as part of device construction, where Ag electrode is involved in the RS process via redox effect while the intercalated AIZS QDs may further guide the formation of CFs.23 The as-fabricated filamentary memristors manifest many advantaged performance including the reversible transition between threshold switching (TS) and memory switching (MS) behaviors, nonvolatile multilevel storage characteristics, fast switching speed, and fundamental synaptic functions. More importantly, we demonstrate the coincidence detection by employing the fitted spike-timing-dependent plasticity (STDP) parameters. Benefitting from the aforementioned excellent characteristics, the SS-based memristors can be regarded as a promising electronic platform for advanced neuromorphic applications.
Figure 1(a) illustrates the structure of the studied memristors: Ag/AIZS/sericin/W. Figure 1(b) presents a scanning electron microscopy (SEM) image of the cross-sectional area of the vertically aligned SS-based device, showing the SS layer's thickness at approximately 110 nm. Additionally, several characterizations of SS have been conducted including transmission electron microscopy (TEM), ultraviolet−visible (UV−vis) absorbance, circular dichroism (CD) spectroscopy, dynamic light scattering (DLS) measurements, and zeta potential assessments. Figure 1(c) showcases the morphology of SS disclosing a well-defined spherical structure with a relatively uniform nanoscale size. The UV-vis absorption spectra, visualized in Fig. 1(d), reveal an absorption peak at 272 nm corresponding to aromatic amino acids.24 The CD spectroscopy in Fig. 1(e) exhibits a relatively strong negative peak at around 226 nm indicating a predominantly beta-sheet structure. The size determined by DLS in Fig. 1(f) is mainly centered at 60 ± 10 nm, slightly larger than that observed by TEM. This is due to DLS measuring the hydrodynamic diameter of the SS peptide, while the sample should be prepared in a dried state before TEM characterization. This phenomenon has been observed for other peptides in our previous studies.25,26 Moreover, the zeta potential reveals the negative charge of the SS peptide, with the −19.27 ± 0.23 mV indicating the peptide's relative stability within the spherical structure.
Figure S1(a) displays the representative current–voltage (I–V) curve of the SS-based device under relatively low compliance current (CC) of 10 μA. The device is initially in high-resistance state (HRS) and then stimulated to low-resistance state (LRS) at a threshold voltage ∼0.4 V (SET voltage), which is termed as the SET process. However, the LRS cannot be effectively maintained and automatically recovers back to HRS when the voltage sweeps back, presenting a typical feature of one-way volatile TS. It is more inclined to form thin and unstable CFs that spontaneously dissipate upon removing external positive stimulation.27 The SS-based memristor converts into nonvolatile MS behavior as increasing CC to 600 μA in Fig. S1(b). Specially, the device maintains the LRS for about 3000 s in Fig. S2(a) after experiencing the SET process, indicating typical nonvolatility. Then, the device can be triggered from LRS to HRS at the reverse directions (RESET process), obviously exhibiting the bipolar RS behaviors. More significantly, the volatile TS and nonvolatile MS behaviors can be reversibly modulated by changing the value of CC as illustrated in Figs. S1(c) and S1(d).
Figure 2(a) exhibits the positive I–V curves of the SS-based device under different CC values by varying it from 10 to 1000 μA. As we analyzed that at relatively low CC, the extracted device's LRS is at a level comparable to the HRS in Fig. 2(b). When a relatively large CC is applied, it can be gradually reduced to two orders of magnitude and maintained, which is attributed to the fact that one or a few CFs become more stronger and thicker. To further investigate the performance of nonvolatile MS behavior, multiple I–V cycles have been plotted in Fig. 2(c). It can be found that the extracted HRS and LRS in Fig. 2(d) of bipolar RS process are concentrated around 5.5 × 105 Ω and 2.0 × 103 Ω, respectively. The switching window (resistance ratio, HRS/LRS) is calculated to be greater than two orders of magnitude, revealing that the device's state can be accurately distinguished during read operations. The SET voltage in Fig. 2(e) ranges from 0.3 to 0.57 V and can be fitted by Gaussian function, enabling the precise control of the device programming voltages in practical applications.28 To evaluate the device-to-device variability of the SS-based devices in memristor array, the distributions of HRS and LRS data of nine devices have been displayed in Fig. 2(f), where both states show extreme narrow distribution. Additionally, the statistical distribution plots of the SET and RESET voltages have been shown in Fig. S2(b), implying the relatively concentrated distribution in operation voltages. This may be attributed to the contribution of the smooth surface of spin-coated SS materials with spherical nanostructures and accompanying nanofibers, which can be unequivocally validated from the TEM image in Fig. 1(c).
Intriguingly, multilevel resistance values, an essential property for high-density data storage, has been implemented by adjusting the negative RESET voltage spanned from −0.5 to −1.1 V in Fig. 2(g). The detailed programming processes are shown as follows. First, the device was SET from HRS to LRS (i.e., OFF to ON) at a higher CC, and the device's LRS can be memorized at the positive voltage region. After that, the ON state can be gradually RESET into different intermediate state (i.e., OFF1, OFF2, OFF3, OFF4) by increasing the RESET voltages. The mean values of the corresponding distinct resistance states have been listed in the inset table, which can be maintained more than 1000 s obtained at low readout voltage of 0.1 V, respectively [Fig. 2(h)], suggesting good data retention ability without any significant deviation. In addition to the detailed direct current (DC) properties, the device can be switched to LRS within approximately 40 ns by a 3 V/460 ns (pulse amplitude/width) programming voltage pulse in Fig. 2(i), showing fast switching speed under pulse mode. The SS-based devices show the relatively decent RS characteristics compared with recently reported biocompatible materials-based memristors (Tables S1 and S2), such as reversibly multiple RS behaviors, low operation voltages, fast switching speed, robust cyclic endurance, excellent capability to simulate synaptic plasticity, and significant potential in neuromorphic computing systems.
After presenting in-depth experimental analyses of the electrical performance, it is mandatory to unravel the charge transport mechanisms for better understanding of the physical operating principle. To do so, the HRS and LRS of positive nonvolatile I–V curves have been replotted with double logarithmic plot. There are three distinct fitted slopes in whole HRS in Fig. S3(a), that is, the linear fitting slope ∼1 (I ∝ V, Ohmic conduction) for lower voltage corresponding to the dominated thermally generated free carriers inside the AIZS/SS layer, as well as the slopes increases from 2.04 (I ∝ V2, Child's square law) to 3.18 (I ∝ V n, n > 2, current increase region) for the higher voltage region related to the transition from the trap-unfilled to trap-filled process, which is in good accord with the typical space charge limited current (SCLC) conduction.29 As for LRS in the SET process, the fitted slope is a constant close to 1 in Fig. S3(b), showing the Ohmic conduction related to the formation of CFs. The difference of fitting results between HRS and LRS indicates the localized behavior of CFs.30 Additionally, the dependence of HRS and LRS on the device cell area has been conducted to experimentally explore the type of RS behavior, that is, interface-type or filamentary-type switching. It is noteworthy from Fig. S4(a) that the HRS generally follows a pattern that is inversely proportional to the TE size, which is ascribed to that the homogeneous current flows through the device core area.28 While the LRS is independent of the TE area, evidently confirming the localized filamentary-type RS behavior.31
Furthermore, the electrode-cutting operation has been utilized to validate the localized conduction in Fig. S4(b). The device was first set to LRS, and its TE was cut into two parts. The resistance states of TE-2 read at 0.1 V are close to the LRS, while that of TE-1 is equal to the HRS, indicating the presence of CFs under TE-2, which further verifies the local CFs behavior.32 The characterizations of the thickness, formation, and rupture of the CFs will be carried out by using the conductive atomic force microscope measurement for a future work. It has been reported that the metal ions (e.g., Ag) can be chelated by the functional groups in SS.18,19 In other words, SS has outstanding adsorption capability of metal ions. Therefore, we speculate that the RS behaviors of the SS-based device is dominated by the metallic Ag CFs formed by the guidance of AIZS QDs and active amino groups in SS. Keeping in view of the detailed mechanism analyses mentioned above, a schematic diagram of the filamentary-type RS process has been proposed, as shown in Fig. S5. It should be pointed out that compared with the corresponding metallic forms, the metallic elements (e.g., In and Zn) in AIZS QDs are less active since they have the capability of forming covalent bonds with other metallic elements. In addition, the surface of AIZS QDs exhibits covalent cap characteristics caused by organic ligands, effectively preventing the migration of metallic elements under electrical stimulation.
Synapse, a functional joint point of neurons, has been considered as a fundamental unit of the biological neural network for information transmission. Figure 3(a) schematically illustrates the diagram of a biological synapse sandwiched between presynaptic and postsynaptic neurons.2 From the biological perspective, paired-pulse facilitation (PPF), a representative synaptic behavior of short-term plasticity (STP), can be realized by a pair of identical spike pulses with predetermined time interval.33 The excitatory post-synaptic current (EPSC) triggered by the second pulse stimulation significantly increases on the basis of the first pulse stimulation, strongly revealing the potentiation of the strength of connections between adjacent neurons, as displayed in Fig. 3(b).34 Herein, the synaptic PPF behavior has been artificially implemented by our proposed SS-based memristor similarly sandwiched between the metal TE and BE. As demonstrated in the inset of Fig. 3(c), the response EPSC (A2) induced by the second voltage pulse with the same amplitude of 0.9 V is significantly increased compared to the first EPSC (A1), which may be due to the formation of stronger CFs. More importantly, the PPF index defined as the ratio of (A2 − A1)/A1 × 100% has been plotted as the function of time interval (Δt) between two consecutive voltage pulses. The short-term potentiating effect of the former pulse on the subsequent pulse is reinforced when decreasing the Δt, which is in excellent accordance with the so-called PPF phenomenon of a biological synapse.
In the coincidence detection task, the 20-input-1-output SNN is trained to trigger output spikes when receiving synchronous spikes among a large amount of asynchronous spikes as exhibited in Fig. 3(e). The top five bold and colored lines represent the synchronous spikes and have been processed with a random fluctuation within 1 ms to simulate temporal jitter. The other fifteen thin input spikes are asynchronous spikes that are randomly distributed in the time domain. It should be noted that during the unsupervised training process, the signal input in Fig. 3(e) represents one period and is repeated 10 times. The corresponding evolution of each synaptic weight is shown in Fig. 3(f), where the colored lines corresponding to the synchronous inputs have a visible increase within 2000 ms (10 periods), sharply contrasting with the unmarked lines related to the asynchronous spikes. As a consequence, the coincidence detection results can be analyzed from output spikes triggered in each period, as depicted in Fig. 3(g). In the first period, there is no output spike triggered as shown in Fig. 3(h). In the subsequent eight periods, there is only one output spike triggered. More interestingly, in the last period, there are two output spikes triggered as demonstrated in Fig. 3(i), revealing that the synchronous spikes can be more easily detected due to the enhanced weight trained by synchronous inputs.
In conclusion, we have fabricated dual-functional memristors with the structure Ag/AIZS/silk sericin/W. The key performance metrics of the device under study include repetitive RS processes, reversible transition between volatile TS and nonvolatile MS behaviors, fast switching speed (40 ns), and nonvolatile multilevel storage characteristics. Additionally, recent mechanistic discoveries pertaining to SS suggest that the Ag cations derived from TE oxidation can be directed to migrate by the AIZS QDs and subsequently chelated by the functional groups within the SS material, potentially accounting for the controllable RS characteristics. Furthermore, the prepared SS-based memristor, possessing intrinsic neuromorphic behavior, has emulated biological synaptic behaviors including the PPF and STDP characteristics. By integrating the fitted STDP's parameters into a SNN, we have demonstrated coincidence detection, revealing significant potential for hardware implementation in neuromorphic computing systems.
See the supplementary material for the reversible transition between TS and MS behaviors, data retention, distribution of operation voltage, fitting mechanism, device scaling characteristics, electrode-cutting operation, schematic of forming CFs, and performance comparison.
All the authors have given approval to the final version of the paper. This work was supported in part by 2030 Major Project of the Chinese Ministry of Science and Technology (Grant No. 2021ZD0201200), High-end foreign experts project of the Ministry of Science and Technology (Grant No. G2022178034L), Postgraduate Research and Practice Innovation Program of Jiangsu Province (Grant No. KYCX22_0928), Jiangsu Province Research Foundation (Grant No. 16KJA510003), and China Postdoctoral Science Foundation (No. 2019M651677).
Conflict of Interest
The authors have no conflicts to disclose.
Nan He: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal). Yi Tong: Formal analysis (equal); Funding acquisition (equal); Methodology (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal). Lei Zhang: Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Methodology (equal); Resources (equal); Supervision (equal). Feng Xu: Formal analysis (equal); Funding acquisition (equal); Methodology (equal); Project administration (equal); Supervision (equal). Jie Yan: Data curation (equal); Formal analysis (equal); Software (equal); Visualization (equal). Zhining Zhang: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Resources (equal). Fan Ye: Data curation (equal); Methodology (equal). Haiming Qin: Methodology (equal); Resources (equal). Er-Tao Hu: Resources (equal). Xinpeng Wang: Funding acquisition (equal); Resources (equal). Pu Chen: Methodology (equal); Resources (equal). Yang Sheng: Resources (equal).
The data that support the findings of this study are available within the article and its supplementary material.