Silkworm Hemolymph Resistance Random Access Memory with High Stability and Low Power Consumption

Most current resistive memory has the problems of high and unstable threshold voltages and high device misread rates caused by low current switching ratios. To address these problems, an Al/poly(methyl methacrylate) (PMMA)/silkworm hemolymph:gold nanoparticles/PMMA/indium tin oxide memory device is fabricated by adding PMMA layers above and below the active layer. The device not only has stable bipolar switching characteristics with a high ON/OFF current ratio but also has a lower and more stable threshold voltage. Potentiation, depression, and spike‐time‐dependent plasticity at biological synapses are realized using this device. The device is successfully fabricated on a flexible substrate, and the device can still maintain a stable working state after 104 bending cycles. This research opens a new door for the future realization of artificial synapses in neural network hardware.


Introduction
The human brain, as the most powerful and efficient computing system, has ultralow power consumption while performing a series of advanced learning activities. The brain is composed of ≈10 11 neurons and 10 15 interconnected synapses, and the ultralow energy consumption of each synaptic event is only ≈1 −10 fJ. [1] The neuromorphic computing system inspired by the brain has the dual capabilities of data storage and logical operations. It can perform complex tasks and achieve functions similar to the brain. In recent years, it has been intensively studied as an alternative to digital computing systems. [2] The study of artificial synapses that can simulate the synaptic function of the human brain is a prerequisite for building a neuromorphic computing system. Therefore, there is an urgent need for a low-power, high-stability device as an artificial www.advelectronicmat.de plasticity (triple STDP) rules and can be used for pattern recognition of biological systems. [10] It has been reported that many materials have been used in the production of RRAM, such as perovskite, [11] metal oxide, [5c,12] 2D materials, [13] and organic materials. [14] However, devices based on some materials have complicated processes and high costs, and the generated electronic waste may pollute the environment. Biomaterials have unique advantages, such as degradability and biocompatibility, so they are widely used in the fields of synaptic simulation and biomedicine. [15] Some researchers have begun to use biological materials to make RRAM. RRAM made of chitosan has an ON/OFF ratio of greater than 10 and a durability of 100 times. [16] Devices made of bovine serum albumin and silver nanoclusters have a switch ratio of 10 3 and a durability of 100 times. [17] In the active layer of RRAM, the wool keratin (WK) molecule is used as the mediating molecule, and gold nanoclusters are embedded in the silk fibroin, which enhances the stability of the device. The ON/OFF current ratio reaches 10 2 , and the durability is 100 times. [18] In 2019, Lin et al. used human hair biocompatible keratin film as the medium layer of memristors. [19] The prepared Ag/keratin/FTO devices showed good electrical properties, high light transmittance, and physical transient characteristics. In 2020, Wang et al. prepared Ag/BSA-Ag/ZrO 2 / Pt devices by inserting a zirconia layer. [20] The performance of devices based on Ag-doped bovine serum albumin (BSA) was improved, such as a reduced switching voltage, uniform distribution of setting and reset voltages, better holding characteristics, fast switching speed, and low switching power. Dwipak Prasad Sahu studied the detection of bovine serum albumin (BSA) protein with TiO 2 and TiO 2 /graphene oxide (GO)-based memristors. [21] Both devices show good bipolar resistive switches, with ON/OFF ratios of 73 and 100, respectively, which proves that devices with GO can distinguish resistance states with high accuracy. Although some bio-RRAM devices have been reported, the device performance still needs to be improved. On the one hand, the ON/OFF current ratio of bio-RRAM is not high, which will increase the probability of misreading of the system. On the other hand, the performance of the device is unstable, such as the threshold voltage, resistance of each resistance state, holding time, etc., which will make it difficult for the device to truly use. Therefore, further improving the performance of bio-RRAM systems is an urgent problem to be solved.
The switch current ratio of the memristor with SH as the main material of the switch layer is superior to other protein-based memristors. While the size and uniformity of the threshold voltage are important standards to measure the power consumption and stability of devices, SH-based memristors still need to be improved. Therefore, our research group reduces the power consumption of devices and improves their stability by adding poly(methyl methacrylate) (PMMA) layers and Au NPs. The performance of Al/PMMA/SH:Au NPs/PMMA/ITO/ glass devices prepared in this work has been greatly improved compared with previous devices. In addition to the smaller V set and V reset voltages, the stability of the device threshold voltage has also been improved. Compared with other protein-based memristors, the switching current ratio of this device also has obvious advantages. In addition, neural behaviors such as enhancement, inhibition, and STDP are simulated using the device, and the conductive mechanism of the device is studied.  Figure 1c shows an SEM photo of the cross-sectional view of the Al/PMMA/SH:Au NPs/PMMA/ITO/glass device without a top electrode. The thicknesses of the upper PMMA layer, SH:Au NP layer, lower PMMA layer, and ITO layer are 64, 53, 65, and 198 nm, respectively. Figure 1d is a TEM image of Au NPs, and it can be seen that the diameter of gold nanoparticles is ≈10-20 nm. Figure 2a shows the I-V curve of Al/PMMA/SH:Au NPs/ PMMA/ITO and Al/SH:AuNPs/ITO devices, and the scanning direction is shown by the arrow in the figure. As shown in Figure 2b, the ON/OFF current ratio of the two devices was measured. The maximum ON/OFF current ratio of the Al/ PMMA/SH:Au NPs/PMMA/ITO device was ≈1.56 × 10 5 , and the maximum ON/OFF current ratio of the Al/SH:Au NPs/ ITO device was ≈1.43 × 10 5 . Figure 2c shows the I-V characteristic curve under 100 DC voltage scanning for Al/SH:Au NPs/ITO devices. Figure 2d shows the I-V characteristic curve under 200 DC voltage scanning for Al/PMMA/SH:Au NPs/ PMMA/ITO devices. It can be seen that the Al/PMMA/SH:Au NPs/PMMA/ITO devices have a better repeatability of continuous switching. As shown in Figure 2d, the resistance cumulative probability of the device at 1 V is plotted. The coefficient of variation of the resistance value of the device under a low resistance state (LRS) is 0.11, and the coefficient of variation under a high resistance state (HRS) is 1.09. Figure 2e shows the endurance of the read cycle times of the device at 1 V, and the endurance of the device is more than 200 times. As shown in Figure 2f, to analyze the stability of the threshold voltage of the device, the statistics of the device under 100 cycles of continuous scanning of a single cell are calculated. The standard deviation of the set voltage (V set ) of devices is 0.26, the average is −0.63 V. Under the setting voltage of 0.63 V (V set ), the current is 5 × 10 −7 A, and the power consumption is 3.2 × 10 −7 W. The standard deviation of the reset voltage (V reset ) is 0.30 V, and the average is 2.36 V. Table 1 shows the performance comparison of SH-based devices proposed in this work and previous work. The performance of the device has been greatly improved. After adding the PMMA layer, not only is the ON/ OFF current ratio of the device increased but also the threshold voltage of the device is decreased, and the threshold voltage stability is improved. To test the retention performance of the different resistance states of the device, as shown in Figure 2g, the resistances in the HRS and LRS of the device were tested at 1 V. The devices remained stable within 10 4 s, which reflects the good retention performance of the device. To test the stability between the devices prepared in this work, the electrical characteristics of 25 devices were tested under the same test conditions. The results are shown in Figure 2h, where each color represents a device. It is observed that the I-V characteristic curve distribution of 25 devices (Figure 2h) is similar to that of a single device for multiple cycles (Figure 2d), so it can be considered that the devices prepared in this work have good consistency. As shown in Figure 2i, the I-V curve www.advelectronicmat.de of the device under positive bias is redrawn under double log coordinates.

Results and Discussion
As shown in Figure 2i, the fitting slope is approximately 0.91 when the device is in the LRS. At this time, the conduction mechanism of the device is mainly the ohmic conduction mechanism. [24] When the device is in HRS, the fitting slope of the I-V curve in the low voltage region is 1.03, and the conduction of the device is based on ohmic conduction as the main mechanism, and the fitting slope in the high voltage region is 2.26. At this time, the device conforms to the space-charge limited current (SCLC). [25] SH contains proteins, carbohydrates and various oxygen-containing functional groups (CO and OH). The switching behavior of Al/PMMA/SH:Au NPs/PMMA/ITO devices may be due to the formation and fracture of conductive filaments composed of oxygen vacancies. The conductive mechanism of the device is shown in Figure 3. First, the initial state of the device is the HRS (Figure 3a). When the negative voltage shown in Figure 3b is applied to the Al electrode of the device, the oxygen ions migrate to the ITO electrode under the action of the electric field force, leaving oxygen vacancies at the same time. Electrons enter the active layer from the Al electrode to fill traps, and some electrons will be trapped by Au NPs. When a single electron tunneled into Au NPs, it generated an additional barrier to make electron transfer more difficult. Therefore, the addition of Au NPs can effectively reduce the HRS resistance of the device and increase the switching current ratio of the device. When the bias voltage applied to the Al electrode reaches V set , the oxygen vacancy conducting filament connects the upper and lower electrodes to form a conductive channel. At this time, the device switches from the HRS to the LRS. The PMMA layer blocks the movement of electrons, so the electrons injected into the SH:Au NP layer continuously accumulate, their density gradually increases, and a local electric field is formed, reducing the randomness of the formation of conductive filaments, reducing the power consumption of the device and enhancing the stability. As shown in Figure 3c, when the forward voltage is applied to the Al electrode, the oxygen ion moves from the ITO electrode to the Al electrode and gradually fills the oxygen vacancy. When the voltage increases to V reset (Figure 3d), the conductive filament breaks, and the device switches from LRS to HRS.
Mechanical flexibility is essential for potential applications in electronic skin, biointegrated medical devices, and wearable devices. The electrical characteristics of the device prepared on the flexible PET substrate are shown in Figure 4, which shows the I-V characteristic curve of the device tested in a flat state. As shown in Figure 4b, the maximum ON/OFF current ratio of the device in the flat state is ≈1.54 × 10 6 . To test the consistency of the device, the resistance values of 20 different cells were compared at 1 V. As shown in Figure 4c, the resistance in the HRS and LRS of the 20 cells did not change much. As shown in Figure 4d,e, the I-V characteristics of the device were measured under tensile and compressive conditions. To test the mechanical durability of the device, the

www.advelectronicmat.de
resistance values of the device in the high-and low-resistance states under 10 000 bending times were read at 0.5. As shown in Figure 4f, with the increase in bending times, the resistance of the device is almost unchanged at 4000 bending times. After 6000 bending cycles, the performance of the device has a certain attenuation, but there is still a large switching window. The results show that the device has good mechanical strength and can be applied to wearable devices. Figure 5a, the device on a glass substrate was used to simulate neural behavior. The device was subjected to pulses with increasing amplitude (0.05 V step) from −0.2 to −0.7 V to simulate the "potentiation" behavior of biological synapses, from 1.2 to 0.3 V pulses with decreasing amplitude (0.05 V step) to simulate the "depression" behavior of biological synapses, with a time interval of 100 µs and a pulse width of 10 µs. The single test result of the device is shown in Figure 5b.

www.advelectronicmat.de
The results of applying a repeated nine times pulse train to the device are shown in Figure 5c, and the measurements demonstrate that the device successfully achieves potentiation and depression biological synaptic behavior under the applied pulses. In the construction of the neural network, the simulation of the STDP is the first. In this work, it is defined as the time difference (∆t) between prespiking and postspiking, resulting in the relative change in synaptic weight (∆W). The synapse weight is defined as ∆W = (I 2 − I 1 )/I 1 × 100%, where the current applying presynaptic pulse is I 1 , and the current applying postsynaptic pulse is I 2 . As shown in Figure 5d, the pulse applied between the top electrode and the bottom electrode of the device is defined as a presynaptic pulse (first gradually increasing from 0.5 to 0.7 V and then change to −0.7 V, the step is 0.05 V, the pulse time is 10 µs, and the pulse interval is 10 µs). As shown in Figure 5e, the pulse applied between the bottom electrode and the top electrode of the device is defined as a postsynaptic pulse (the same as the presynaptic pulse), as  shown in Figure 5f,g; ∆t = 20 µs is the net peak value imposed on the device. The test result is shown in Figure 5h. The variation in ∆W with ∆t of the device follows the relationship where τ + and τ − are the broadening of the STDP in the potentiation and depression part, and A + and A − are the maximum weights of the synapse with ∆t = 0. By fitting the experimental data, it can be obtained that A + = 124.81, τ + = 21.31 µs, A − = −114.25, and τ − = −28.75 µs, and the test results of the device are very similar to those observed in the biological synapse system, which proves that the device successfully simulates the STDP behavior. The test results show that when ∆t > 0, ∆W increases. This is because the input pulses will produce a net peak that is greater than the minimum set voltage of the device, causing the device to turn on. Conversely, when ∆t < 0, ∆W decreases, which is attributed to the input pulse generating a peak value greater than the minimum reset voltage, resulting in device turn off. As |∆t| increases, |∆W| decreases because the magnitude of the net peak imposed on the device decreases as |∆t| increases.

Conclusion
Al/PMMA/SH:Au NPs/PMMA/ITO multilayer devices were fabricated in this work, which not only have a higher ON/OFF current ratio (>10 5 ) but also have lower and more www.advelectronicmat.de stable threshold voltages (−0.63 and 2.16 V), the device still has good performance when fabricated on a flexible substrate, and the biosynaptic plasticity is successfully simulated using the device. The device is a strong candidate in both low-power memory devices and neuromorphic computing.

Experimental Section
SH (0.5 mL) and Au NP solution (7.5 mL) were mixed and sonicated for 5 min to obtain SH solution doped with Au NPs. The PMMA powder was dissolved in anisole to obtain a PMMA solution with a concentration of 70 mg mL −1 . The PMMA solution was coated on the ITO bottom electrode at 1000 rpm for 20 s, and then the substrate was dried at 100 °C for 1 h. The SH:Au NP solution was spin-coated (first at 500 rpm for 5 s and then at 4000 rpm for 40 s) on the PMMA-coated substrate and then dried at 100 °C for 10 min. Using the same process as the abovementioned PMMA layer, the PMMA insulating layer was spin-coated on the SH:Au NP solution dielectric layer and then dried. Finally, the Al electrodes were evaporated under a pressure of 2 × 10 −3 Pa to fabricate Al/PMMA/SH:Au NPs/PMMA/ITO/glass devices. Devices on PET substrates were fabricated using the same process as devices on glass substrates.