First‐Principles Study of the Electronic Properties of Egg Albumen Optoelectronic Artificial Synapses by Carbon Nanotube Insertion

The realization of artificial synapses based on biomaterials is of great significance for the development of environmentally friendly neuromorphic hardware systems and artificial intelligence. In this sense, a bioartificial synapse composited with egg albumen (EA) and multiwalled carbon nanotubes (MWCNTs) is fabricated. Based on the adjustable weight of the artificial synapse, the plasticity of electrical synapses is explored. Due to the photogenerated carriers and thermoelectric effects of carbon nanotubes, the device has optoelectronic properties, so the optoelectronic synaptic plasticity of the device is explored under light pulses. The device is well suited for biological synapses and shows great potential for applications in future high‐density storage and neuromorphic computing systems. In addition, to further study the physical mechanism of the conductive process of the device, the electrical characteristics of the contact interface between carbon nanotubes doped with Fe substitution and the upper electrode Al are mainly analyzed by first principles, and the adsorption, charge distribution, and band structure between them are theoretically studied.


Introduction
As one of the emerging nonvolatile memory technologies, the application range of memristors has been extended from multivalued storage, [1] integrated storage and computing [2] to hybrid edge computing architecture, [3] integrated circuits, [4] etc. Inspired by brain-like nerves, memristors are also used in simulating biological neurons [5] and artificial perception systems. [6]ue to the excellent advantages of memristors in these fields, they are expected to solve problems such as scalability, complementary metal-oxide-semiconductor (CMOS) compatibility, DOI: 10.1002/aelm.20230063112][13] As one of the common biological proteins, egg albumen (EA) does not require an additional extraction process compared to other proteins.Electronic devices made of it have strong durability and have been applied to one-writeread-many memory devices, [14] field effect transistors, [15] transistors, [16,17] and supercapacitors. [18]Memristors made of egg albumen have nonvolatile resistive switching properties; [19][20][21] However, under the same external stimulus, the resistive state of the device usually fluctuates in different write cycles.[24] Compared with using metal oxides as the active layer of memristors, egg albumen, as a biomaterial, is relatively environmentally friendly in its preparation and application process and has good degradability and biocompatibility.Egg albumen is a protein material with variable conductivity.This characteristic is beneficial for adjusting the weight of egg albumen artificial synapses to achieve synaptic plasticity.Incorporating multiwalled carbon nanotubes (MWCNTs) into the active layer of egg albumen memristors, creating an Al/EA:MWCNT/indium tin oxide (ITO) memristor, can improve its unstable resistance-switching characteristics.Additionally, investing ating the transition metal iron in ironcontaining proteins has been found to enhance the conductivity between the upper electrode Al and carbon nanotubes.
The memristor-based bipolar resistive switching mechanism can be explained as the formation and breaking of electron hopping paths between the upper and lower electrodes. [25]However, the physical mechanism regarding the role of EA:MWCNTs in determining the resistive switching performance of memristors remains unclear.There are problems such as low adsorption strength and poor interface conductivity between the Al metal matrix and carbon nanotubes due to poor wettability. [26]Doping carbon nanotubes can effectively increase the affinity of carbon nanotubes, reduce the Schottky barrier formed by the contact between carbon nanotubes and metals, and improve the charge transfer between carbon nanotubes and other matrix materials.[29] The presence of transition metals or defects on the surface of carbon nanotubes can improve the wettability between carbon nanotubes and Al. [30,31]ere, the active layer of the device fabricated in this paper is a composite material of carbon nanotubes and egg albumen, and the upper electrode is Al.However, in the active layer, ironcontaining proteins such as transferrin account for a high proportion of egg albumen.The presence of transition metal iron elements in ferritin improves the conductivity between the top electrode Al and carbon nanotubes.The presence of Fe improves the conductive performance of the device and makes the electrical properties of the device more stable, thereby increasing the controllability of the weight of the synaptic device, which helps simulate a more stable brain synaptic plasticity behavior.Firstprinciples simulations based on density functional theory were employed to investigate the adsorption of Al on intrinsic and Fedoped carbon nanotubes.The study examined the adsorption energy of Al atoms on carbon nanotubes before and after Fe doping, as well as the band structure and charge distribution of intrinsic and Fe-doped carbon nanotubes before and after Al adsorption.The aim was to explore the influence of Fe doping on the conductivity between the Al matrix and carbon nanotubes in Al-based carbon nanotube composites.Based on the discussions in this manuscript regarding the first-principles analysis of the Al/EA:MWCNTs/indium tin oxide (ITO) device, the following conclusions can be drawn: Due to the similar work function between Fe and carbon nanotubes, the substitutional doping of Fe as a transition metal enhances the charge transfer between carbon nanotubes and Al.Consequently, this improves the conductivity between the carbon nanotubes in the active layer and the top electrode Al.Additionally, since iron-containing proteins such as ovotransferrin are prevalent in egg albumen, the Fe enters the carbon nanotubes, creating new bonds and thereby enhancing the conductivity between the active layer and the top electrode.

Egg Albumen Artificial Synapse
Figure 1a shows the structural model of the Al/EA:MWCNT/ITO/glass device and investigates its electrical characteristics.Figure 1b utilizes UV-vis spectroscopy to perform optical analysis of the EA: MWCNT film.According to the formula It can be concluded that the film exhibits an absorption peak at  = 460 nm, with a bandgap width of ≈2.70 eV. Figure 1c displays 100 cycles of the device's I-V curves, demonstrating its bipolar resistive switching performance.It can be observed that the device exhibits good stability, with I-V scans repeating continuously for over 100 cycles on the same cell.To explore the synaptic plasticity of Al/EA:MWCNT/ITO/glass devices, an Al electrode was used as the presynaptic membrane, and an ITO electrode was used as the postsynaptic membrane to realize artificial synapses under different stimuli.Figures 1d,e are the I-V curves of the device when five consecutive positive and negative direct current (DC) voltage scans are applied to the presynaptic membrane.In Figure 1d, when five consecutive positive voltages (0-0.5 V) are applied, the current of the memristor decreases after each voltage sweep.In contrast, when a negative voltage (0∼−0.5 V) is applied, the negative current of the device gradually increases (see Figure 1e).To more clearly show the change in current with the number of consecutive DC scan biases, Figure 1f shows the trend of the current (conductance) value change of the device under the maximum positive and negative bias voltages during the DC scan.Repeated DC bias can change the conductance value of the synaptic device, corresponding to continuous stimulation of the synapse, which can change the synaptic weight.Figure 1g shows the longterm retention curve of the postsynaptic current within 100 s when a single stimulus is applied to the presynaptic membrane, which realizes excitatory postsynaptic current (EPSC) behavior, indicating the excellent nonvolatility of the device.As shown in Figure 1h, two consecutive stimulus signals are applied to the device, and the postsynaptic current of the second stimulus, A 2 , is larger than that of the previous stimulus, A 1 .This simulates the behavior of paired-pulse facilitation (PPF), enabling the device to achieve short-term plasticity.Figure 1i shows that under continuous positive (negative) excitation, the conductivity of the device shows a trend of depression (potentiation), which corresponds to the long-term depression (potentiation) of the neural behavior of the device.Figure 1j shows the synaptic potentiation and depression behaviors of the device repeated ten times in succession, which shows that the artificial synapse device has good synaptic plasticity.
Figure 1k shows the adjustment of the synaptic weight relative to the excitation rate, and the current difference (ΔI = I in -I 1 ) can more intuitively display the synaptic weight adjustment of the device under different rates of stimulation.The device exhibited synaptic potentiation behavior under continuous stimulation.As the excitation rate increases, the synaptic weight increases more obviously, and the synaptic weight reaches the maximum when the excitation interval is 10 μs, which realizes spike-rate-dependent plasticity (SRDP) behavior.Figure 1l  According to ultraviolet spectrum analysis, carbon nanotubes have a strong absorption peak at ≈400 nm, and the action of photogenerated carriers and the pyroelectric effect can make carbon nanotubes generate a photocurrent under specific light.Therefore, the photoelectric effect of the artificial synapse was explored by using a blue-violet light pulse light irradiation device with a wavelength of 405 nm and a power of 200 mW modulated by a transistor-transistor-logic (TTL), as shown in Figure 2a.In Figure 2b, under the action of continuous light pulses, the postsynaptic current tends to rise.When the light pulse is removed, the current slowly decreases until it reaches a stable state, which shows the optoelectronic properties of the optoelectronic synapse device.This is due to the photogenerated carriers and thermoelectric effect caused by MWCNTs when the device is irradiated with light pulses, causing the device to induce a photoelectric effect.When the light pulse is removed, the photogenerated carriers are gradually captured by the charge trapping center, and the device returns to the initial current state in a short period of time, thereby causing excitatory postsynaptic currents.Figure 2c shows the postsynaptic current change of the photoelectric synaptic device under different light pulse times.When the light pulse time increases, the increase in the postsynaptic current also increases, which indicates that longer irradiation will generate more carriers and that long-term irradiation converts more heat energy into electrical energy.When the light pulse is turned off, the time needed for the current to return to a steady state increases with the pulse duration, which shows the number-dependent nature of the photosynaptic device transition from short-term plasticity (STP) to long-term plasticity (LTP) on the number of light pulses.Next, we further explored the photoelectric synapse device to simulate the learning experience of the human brain.
As shown in Figure 2d, after continuous optical pulse stimulation is applied to the presynapse, the synaptic weight gradually increases, similar to the initial memory process in the human brain.After removing the pulses, the optoelectronic synapse spontaneously decays to an intermediate state within 8 s, representing the forgetting process after short-term memory in the human brain (Figure 2e).In Figure 2f, it is found that only a short light pulse stimulation of >2 s during the second stimulation is enough to reach the state before decay, which is far less than the 8 s light pulse stimulation needed for the first device, to cause the same increase in synaptic weight, which is similar to the relearning function of the human brain.As shown in Figure 2g, the decay of synaptic weights after the second stimulation is much smaller than that after the first stimulation, which is similar to the fact that the human brain can usually relearn the lost information of the previous memory faster, and the relearning process can significantly enhance the stability of memory.
There are many weak chemical bonds in the egg albumen molecule.After high-temperature treatment, the protein molecular chain unfolds, making the weak bonds easy to break and recombine into stable new chemical bonds.During this process, protein molecules connect adjacent amino acids to each other through new bonds, forming a network of interconnected, thermally cross-linked solid protein membranes.This thermal denaturation mechanism reduces the chance of oxygen scattering and increases the chance of formation and breakage of conductive filaments, thereby improving the switching performance of the device.
The conductive mechanism of the Al/EA:MWCNT/ITO device is shown in Figure 3.When there is no external bias, the particles are randomly distributed in the active layer (Figure 3a).When a negative bias voltage is applied to the top electrode Al, under the influence of the electric field, oxygen ions in the active layer move toward the bottom electrode, leaving behind oxygen vacancies.Simultaneously, some oxidized iron ions move toward the top electrode, where they are reduced to regain electrons (Figure 3b).As the applied negative voltage increases, the electric field continues to increase, and the generated oxygen vacancies and iron ions gradually increase.When the bias voltage reaches V SET , a conductive filament composed of oxygen vacancies and iron ions is formed between the top electrode and the bottom electrode, and the device changes from a high resistance state (HRS) to a low resistance state (LRS) (Figure 3c).When a positive bias is applied to the top electrode, the oxygen ions in the active layer move toward the top electrode under the action of the electric field force and continuously fill the oxygen vacancies.At this time, the iron ions lose electrons and are oxidized into oxidized iron ions.As the forward bias voltage increases, the electric field continues to increase, and the number of oxygen vacancies and iron ions continues to decrease.When the voltage reaches V RESET , the conductive filament breaks, and the device changes from LRS to HRS (Figure 3d).

Calculation Method and Physical Model
First, a (5,5) armchair carbon nanotube unit cell is established, the number of carbon atoms is 100, and the distance between carbon atoms is between 1.41 and 1.42.To prevent the interaction between adjacent carbon nanotubes, a unit cell structure with a lattice constant of 20 × 20 × 13 A is set to ensure sufficient vacuum space.As shown in the ball-and-stick model of Figure 4a-c, there are intrinsic carbon nanotubes, Fe is doped into carbon nanotubes, and Al is placed above Fe atoms in the doped carbon nanotube lattice.To study the interaction of Al between intrinsic carbon nanotubes and Fe substitution-doped carbon nanotubes, geometry optimization of the preliminary structural model was carried out.The CASTEP calculation method is used for optimization.For setting the convergence tolerance optimization parameters, the energy of a single atom is 2.0 × 10 −5 eV, the maximum force between atoms is 0.05 eV, and the maximum stress   between crystals is 0.1 GPa, the maximum displacement between atoms is 2 × 10 −3 A, the maximum number of iterations is 100, and the GGA-PBE functional is selected to optimize the position between atoms and the unit cell parameters.At the same time, the plane wave cutoff energy is set to 240 eV, the self-consistent field cycle convergence tolerance is set to 2.0 × 10 −6 eV, and the total energy and charge density of the system are calculated based on the Brillouin zone whose K grid points are 1 × 1 × 2. To reduce the plane wave cardinal number of the system, an ultrasoft pseudopotential is used to describe the interaction between ion ions and valence electrons.is greater than that of intrinsic carbon nanotubes, which directly indicates that the addition of Fe leads to a greater charge transfer between carbon nanotubes and Al, resulting in larger interactions.Meanwhile, the atomic distances between Al atoms and intrinsic carbon nanotubes, as well as Fe-doped carbon nanotubes, before and after geometric optimization are shown in Tables 1  and 2, respectively.After optimization, the bond lengths between each atom increased, and the bond angles also changed.

Calculation and Analysis of Single-Point Energy
Figure 5a shows the energy band structure of intrinsic carbon nanotubes, with a forbidden band width of 0.134 eV.The corresponding density of the states diagram is shown in Figure 5b, and there are mainly 3p orbital electrons of C atoms near the Fermi level.As shown in Figure 5a,c, the adsorption of Al on the intrinsic carbon nanotubes increases the Fermi level of the carbon nanotubes, and the bottom of the conduction band is located below the Fermi level.The density of states of Al adsorbed on intrinsic carbon nanotubes is shown in Figure 5d.The density of states of Al's 3p orbital electrons near the Fermi level is very large, so the adsorption of Al makes its own electrons enter the conduction band to form free electrons.As shown in Figure 5e, for carbon nanotubes, the addition of Fe lowers the Fermi level below the valence band top.Because the system is a semiconductor, the Fermi energy level should be between the price belt and the bottom of the guide belt, and the height of the price belt should be higher than the Fermi energy level.The entry of the Fermi level into the valence band can be considered p-type doping, with holes present in the system.And structurally, the substitutional doping of Fe 3+ replacing C 4+ generates holes, transforming the system into a p-type semiconductor.Figure 5f is a density-of-state diagram of a substitutionally doped Fe carbon nanotube, and the electron density of the 3p orbital of Fe is very large near the Fermi level, thereby forming an acceptor level.Comparing Figure 5e,g, the bottom of the conduction band is also below the Fermi level, so it can be concluded that the electrons of Al fill the holes in the carbon nanotubes doped by Fe, and more electrons from free electrons participate in conduction.Figure 5h is the density-ofstate diagram of Fe-doped carbon nanotubes after Al atom adsorption.The same orbital electrons of Fe and Al overlap at the peak near the Fermi surface, which indicates that there is chemical adsorption between Fe and Al.In summary, according to the comparison of the energy band structure and density of states of intrinsic carbon nanotubes and Fe-doped carbon nanotubes before and after Al adsorption, it can be concluded that the doping of Fe changes the energy band structure, and at the Fermi level, the formed half-full energy band makes the electrons provided by Al fill near the Fermi level, and Fe further promotes the charge transfer between carbon nanotubes and Al.

Population Analysis
Population analysis can indirectly indicate the charge transfer between particles by detecting the charge carried by particles and perform population analysis on the change in charge carried by each atom of intrinsic and Fe-doped carbon nanotubes before and after Al atom adsorption.According to Mulliken population theory, if the absolute value of the population number is small and close to 0, there is no chemical bond between atoms, and if the absolute value of the population number is large, there is a covalent bond between atoms. [32] Fe-doped carbon nanotubes formed Fe─Al chemical bonds, and C─Al did not form chemical bonds.

Charge Transfer Analysis
Charge density analysis represents the change in electron distribution induced when overall bonding is formed.It is very useful in illustrating how the chemical bonds of the whole system are formed through the delocalization of atomic charge density and is very suitable for the study of the formation of large systems from small systems.To more intuitively illustrate how the charge density changes when a chemical bond is broken or a molecule is bonded to a surface, the problem of charge transfer is analyzed by calculating the charge density method of the fragment.At the same time, it can also be seen from the plane that the blue region with a charge density of 5.000 × 10 −1 of Fe-doped carbon nanotubes is larger than that of intrinsic carbon nanotubes.

Conclusion
The Al/EA: MWCNTs/ITO artificial synapse device proposed in this paper can regulate conductance to realize EPSC, synaptic potentiation and depression, PPF, SRDP, and STDP behavior.At the same time, based on the photoelectric properties of carbon nanotubes, the plasticity of photoelectric synapses, the dependence on the number of light pulses and the process of learning, forgetting, relearning, and reforgetting are realized.Second, the physical mechanism of the electrical properties of synaptic devices is further investigated by the first principles.Construct the unit cell structure and optimize the geometry to complete all calculations of the energy band structure, density of states, and charge density.Due to the close work function between Fe and carbon nanotubes, Fe as a transition metal substitution dopant enhances the charge transfer between the C and Al atoms of carbon nanotubes, thereby improving the conductivity between the carbon nanotubes and the upper electrode Al in the active layer and improving the conductivity of the device.Due to the high proportion of iron-containing proteins such as ovotransferrin in egg albumen, Fe will enter carbon nanotubes to generate new combinations, thereby improving the conductivity between the active layer and the upper electrode.

Experiments Section
First, eggs were purchased from a local supermarket, and the egg albumen was mixed with deionized water at a 1:15 ratio.The mixture underwent 15 min of ultrasonic treatment to ensure thorough mixing.Next, carboxyl-functionalized MWCNTs with a purity of over 95 wt.% and a carboxyl content of 2.00 wt.% were purchased.The MWCNTs had an inner diameter of 5-10 nanometers, an outer diameter of 10-20 nanometers, and a length of 10-30 micrometers.The MWCNTs were mixed with deionized water to form a mixture with a concentration of 1.2 wt.% and subjected to ultrasonic treatment for at least 5 h to ensure uniform dispersion of the MWCNTs throughout the solution.Then, the MWCNT dispersion was mixed with an EA solution in a 1:1 volume ratio, and 15 min of ultrasonic treatment was performed until they were thoroughly mixed.Apply the resulting mixture onto a glass substrate coated with indium tin oxide (ITO) that was cleaned with acetone, anhydrous ethanol, and deionized water using ultrasonication for 15 min.The initial spin speed was set to 500 rpm min −1 for 5 s and then increased to a final speed of 4000 rpm min −1 for 60 s.The device was dried at 105 °C for 15 min, and aluminum electrodes were deposited on the active layer.Finally, the device was annealed at 105 °C for 15 min to complete the fabrication of the Al/EA:MWCNT/ITO device structure.
The synaptic behavior of the Al/EA:MWCNTs/ITO device was investigated using a Keithley 4200 semiconductor parameter analyzer (Keithley, Solon, OH, USA).The CASTEP module of Materials Studio software based on first principles was used to study the structure of carbon nanotubes.
shows the spike-timing-dependent plasticity (STDP) neural behavior of the device.The relative variation in synaptic weight (ΔW) decreases with increasing spike time (|Δt|), and the synaptic weight increases (decreases) when the presynaptic membrane is stimulated before (after) the postsynaptic membrane.The relative change in synaptic weights was fitted with an exponential decay function as the firing interval of the pre-and postsynapse was lengthened.The scaling factors A+ and A− are 164.55 and −160.07,respectively, the time constants + and − are 104.87 and −222.95ms, respectively, and ΔW+ and ΔW-are the values of ΔW when |Δt| tends to infinity and are −3.51 and 34.55, respectively.

Figure 1 .
Figure 1.Al/EA:MWCNT/ITO/glass: a) Schematic diagram of the structure.b) UV-vis-absorption spectrum.c) Cyclic I-V curves under 100 continuous cyclic scans.d) Current modulation of the device under continuous positive voltage sweeps.e) Current modulation of the device under continuous negative voltage sweeps.f) Under the application of positive and negative continuous voltages, the change in current with the number of times.g) EPSC behavior, h) PPF behavior, i) potentiation and depression response and j) 10 consecutive cycles of this response, k) SRDP behavior, l) STDP behavior generated by the excitation device.

Figure 2 .
Figure 2. a) Device structure under violet light irradiation.b) Excitatory postsynapic currents of the device under continuous light pulses.c) Schematic diagram of the dependence of the device on the number of light pulses.The process of d) learning, e) forgetting, f) relearning, and g) refarting generated by the light pulse excitation device.

Figure 4 .
Figure 4. a) Intrinsic carbon nanotubes under a lattice constant of 20 × 20 × 13 A unit cell, b) Fe doped into carbon nanotubes, and c) Al placed in doped carbon nanotubes above the Fe atoms in the tube lattice.d) Al placed in the intrinsic CNT lattice before and after structure optimization.e) Al placed in the Fe-doped carbon nanotube lattice before and after structure optimization.

Figure 5 .
Figure 5. a) Band structure and b) density of states diagram of intrinsic carbon nanotubes.c) Band structure and d) density of states diagram of intrinsic carbon nanotubes after Al atom adsorption.e) Band structure and f) density of states diagram of Fe-doped carbon nanotubes.g) Band structure and h) density of states diagram of Fe-doped carbon nanotubes after Al atom adsorption.
Figure 4d before and after the structural optimization of Al placed in the intrinsic carbon nanotube lattice, and Figure 4e before and after the structural optimization of Al placed in the Fe-doped carbon nanotube lattice.Carbon nanotubes form a locally outward convex deformation near the vicinity of Al.The deformation of Fe-doped carbon nanotubes Figure 6a,b is the top view of the charge transfer distribution of intrinsic and Fe-doped carbon nanotubes after Al atom adsorption.The color changes from red to blue, indicating that the charge density increases from −2.000 × 10 −1 to 5.000 × 10 −1 .It can be seen from the figure that there are a large number of blue and yellow areas distributed between Al atoms and carbon nanotubes, and the charge density is between 2.000 × 10 −1 and 5.000 × 10 −1 .The blue region where Fe-doped CNTs intersect with intrinsic CNTs is larger, corresponding to a wider distribution of charge transfer.When the effective overlap of the charge density reaches >1.000 × 10 −1 , it indicates the formation of chemical bonds.Therefore, we believe that Fe-doped carbon nanotubes are conducive to charge transfer between Al and carbon nanotubes, which is conducive to the formation of chemical adsorption and chemical bonds.All charge transfer distributions within the intrinsic and Fe-doped CNT unit cells after Al atom adsorption are shown from the plane in Figure 6c-d.The chemical bonds between atoms are surrounded by the blue area, the charge density is ≈5.000*10 −1 , the yellow area indicates the surrounding atoms, and the charge density is ≈2.000 × 10 −1 .

Figure 6 .
Figure 6.a) Charge transfer distribution of intrinsic carbon nanotubes after Al atom adsorption.b) Charge transfer distribution of Fe-doped carbon nanotubes after Al atom adsorption.c) Distribution of all charge transfers inside the intrinsic carbon nanotube cell after Al atom adsorption.d) All charge transfer distributions in the unit cell of Fe-doped CNTs after Al atom adsorption.

Table 1 .
The atomic spacing of Al atoms and intrinsic carbon nanotubes before and after geometry optimization.

Table 2 .
Al atoms and atomic distances before and after geometric optimization of Fe-doped carbon nanotubes.

Table 3 .
Population analysis of intrinsic carbon nanotubes with Al adsorption.

Table 3
shows the population number analysis of intrinsic carbon nanotubes adsorbed on Al.The average bond length between C-Al is 3.118 A, and the population number is ≈0.11.The average C─C bond length is 0.75 A, and the population number is 0.75.It can be judged that there is almost no chemical bond formed between C─Al, and C─C is a covalent bond of the carbon nanotube structure itself.Table4presents the analysis of the occupancy of Al adsorbed on Fe-doped carbon nanotubes.The average bond length between C and Al is ≈3.023Å, and the population number is ≈0.03.This indicates that the addition of Fe does not lead to the formation of chemical bonds between C and Al in the carbon nanotubes.The Fe-Al bond length is 2.164 A, the average C─Fe bond length is 1.899 A, and the population number is ≈0.3 or more, so it can be judged that a chemical bond is formed.In summary, the intrinsic carbon nanotubes adsorbed Al without the formation of chemical bonds,

Table 4 .
Population analysis of Fe-doped carbon nanotubes with Al adsorption.