3D electrical element, machine learning system including the same, and manufacturing methods thereof

Abstract

A 3-dimensional electrical element 10 comprises four or more non-linear parts 11, each exhibiting non-linear current-voltage characteristics, and conductors 12 connecting the non-linear parts 11, and the non-linear parts 11 are arranged in a three-dimensional shape. A machine learning system 20 comprises: a 3-dimensional electrical element 10 that has four or more non-linear parts 11, each exhibiting non-linear current-voltage characteristics, and conductors 12 connecting the non-linear parts 11, the non-linear parts 11 being arranged in a three-dimensional shape; and an input electrode 13 and an output electrode 14 that are each connected to the 3-dimensional electrical element 10.

Description

Technical Field

[0001] The present invention relates to a three-dimensional electrical element that functions as a neuron element mimicking neurons (nerve cells), a machine learning system including the same, and manufacturing methods thereof.

Background Art

[0002] Various studies have been made to construct a neural network system. Examples of methods for constructing a neural network system include designing neuron elements with electronic circuits and making functional molecular elements function as neuron elements (see Patent Document 1). By using functional molecular elements, it is possible to make the neural network system more compact than an electronic circuit.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, there has been a limit to the compactification of a neural network system employing conventional functional molecular elements. Further, the compactification of the system is not a problem that appears only in neural networks, but is a problem common to the entire machine learning system. The present invention has been made in view of such circumstances, and an object thereof is to provide a three-dimensional electrical element capable of achieving compactification of a machine learning system, a machine learning system including the same, and manufacturing methods thereof.

Means for Solving the Problems

[0005] The three-dimensional electrical element according to the first invention that meets the above object includes four or more non-linear parts each showing non-linear current-voltage characteristics, and a conductor connecting the non-linear parts, and the non-linear parts are arranged in a three-dimensional shape.

[0006] The machine learning system according to the second invention that meets the above object includes a three-dimensional electrical element having four or more non-linear parts each showing non-linear current-voltage characteristics and a conductor connecting the non-linear parts, and the non-linear parts are arranged in a three-dimensional shape, and an input electrode and an output electrode each connected to the three-dimensional electrical element.

[0007] The manufacturing method of the three-dimensional electrical element according to the third invention that meets the above object includes a step of immersing a dispersion liquid into which four or more non-linear parts each showing non-linear current-voltage characteristics and a conductor connecting the non-linear parts are introduced into a porous structure to obtain a dispersion liquid-containing body, and a step of curing the dispersion liquid-containing body with a curable resin to obtain a three-dimensional electrical element in which the non-linear parts are arranged in a three-dimensional shape.

[0008] The manufacturing method of the machine learning system according to the fourth invention that meets the above object includes a step of immersing a dispersion liquid into which four or more non-linear parts each showing non-linear current-voltage characteristics and a conductor connecting the non-linear parts are introduced into a porous structure to obtain a dispersion liquid-containing body, a step of curing the dispersion liquid-containing body with a curable resin to obtain a three-dimensional electrical element in which the non-linear parts are arranged in a three-dimensional shape, and a step of connecting an input electrode and an output electrode to the three-dimensional electrical element.

Advantages of the Invention

[0009] The three-dimensional electrical element according to the first invention and the machine learning system according to the second invention each include four or more nonlinear portions that exhibit nonlinear current-voltage characteristics, and conductors that connect the nonlinear portions. Since the nonlinear portions are arranged in a three-dimensional shape (three-dimensional structure), it is possible to achieve compactification compared to a conventional machine learning system in which the nonlinear portions are arranged in a planar shape (two-dimensional structure). Further, the method for manufacturing a three-dimensional electrical element according to the third invention and the method for manufacturing a machine learning system according to the fourth invention each manufacture the three-dimensional electrical element and the machine learning system.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Modes for Carrying Out the Invention

[0011] Next, with reference to the attached drawings, embodiments embodying the present invention will be described to facilitate understanding of the present invention. As shown in FIG. 1, a three-dimensional electrical element 10 according to an embodiment of the present invention includes four or more nonlinear portions 11 each exhibiting non-linear current-voltage characteristics, and a conductor 12 electrically connecting the nonlinear portions 11. In this embodiment, all the nonlinear portions 11 are electrically connected by one or more conductors 12.

[0012] In this embodiment, the nonlinear portion 11 is a particle or a void enabling a tunnel junction, and can be formed by a conductive molecule, ion, complex, polymer, metal, organic material, inorganic material, organic-inorganic hybrid material, and a mixture thereof. For example, particles of polyoxometalate can be employed as the nonlinear portion 11.

[0013] The non-linear current-voltage characteristics exhibited by the non-linear portion 11 mean characteristics in which the voltage value (current value to be energized) applied to the non-linear portion 11 and the voltage value (current value) output from the non-linear portion 11 have a non-linear relationship. Examples of non-linear current-voltage characteristics include rectifying properties due to a pn junction of a semiconductor, rectifying properties due to a Schottky junction, or electrical properties of a tunnel junction or Coulomb blockade, and resistance changes of a memristor element with respect to an applied voltage.

[0014] Also, the conductor 12 may be any conductive material that connects the non-linear portions 11. For example, an organic nanowire such as a carbon nanotube, a metallic nanowire composed of one or more elements selected from the group consisting of Ag, Au, Ni, Cu, Pd, Pt, Rh, Ir, Ru, Os, Fe, Co, Sn, and an oxide nanowire composed of one or more oxides selected from the group consisting of IrO2, In2O3, SnO2, ITO, a polymerizable polymer wire, or a composite wire in which the surface of an insulating nanowire such as DNA is covered with a conductive material can be employed as the conductor 12.

[0015] The connection of the non-linear part 11 by the conductor 12 may be any connection method that allows the electrical signal output from one non-linear part 11 to be transmitted to another non-linear part 11. The non-linear parts 11 can be physically, electrically, magnetically, optically, or chemically connected, or two or more of these can be combined for connection.

[0016] Here, taking an integer of 4 or more as N, one three-dimensional electrical element 10 has N non-linear parts 11, and these N non-linear parts 11 are three-dimensionally arranged via the conductor 12. Here, the fact that the N non-linear parts 11 are three-dimensionally arranged means that there is at least one non-linear part 11 among the N non-linear parts 11 that does not contact the virtual plane on which the geometric centers of any three non-linear parts 11 arbitrarily selected from the N non-linear parts 11 are arranged. Therefore, N is 4 or more. From the perspective of the three-dimensional electrical element 10 functioning stably as a neuron element, it is preferable that N is 30 or more, more preferably 100 or more, and even more preferably as large a number as possible of 1000 or more.

[0017] By arranging the N non-linear parts 11 three-dimensionally, the three-dimensional electrical element 10 can be made more compact compared to the case where the N non-linear parts 11 are arranged in a plane. From the perspective of making the three-dimensional electrical element 10 more compact, the ratio of the non-linear parts 11 not located on the virtual plane among the N non-linear parts 11 is preferably 95% or more, 90% or more, 80% or more, 70% or more, 60% or more, 50% or more in that order.

[0018] Also, in this embodiment, a neural network system 20, which is an example of a machine learning system, is configured to have a three-dimensional electrical element 10 and an input electrode 13 and an output electrode 14 respectively connected to the three-dimensional electrical element 10. In this embodiment, the input electrode 13 and the output electrode 14 are each in the shape of a line segment and are formed using a conductive material (for example, copper) as a raw material. The input electrode 13 is an electrode for inputting an electrical signal into the three-dimensional electrical element 10, and the output electrode 14 is an electrode from which an electrical signal is output from the three-dimensional electrical element 10. Note that the positions of the input electrode 13 and the output electrode 14 with respect to the three-dimensional electrical element 10 are arbitrary and are not limited to the positions shown in FIG. 1.

[0019] When an electrical signal (e.g., a pulse signal) is input from the input electrode 13 into the three-dimensional electrical element 10, the three-dimensional electrical element 10 functions as a neuron element in a neural network. As a result, an electrical signal is output from the three-dimensional electrical element 10 to the output electrode 14. In this embodiment, the neural network system 20 is provided with one input electrode 13 and three (i.e., a plurality of) output electrodes 14, but at least one input electrode 13 and at least one output electrode 14 are sufficient.

[0020] For example, as shown in FIGS. 2(A) and (B), one input electrode 13' is inserted into the cubic three-dimensional electrical element 10' from above, and three output electrodes 14' are inserted from each of the four side surfaces (surfaces perpendicular to the surface into which the input electrode 13' is inserted) to form a neural network 20'. The input electrode 13' and the output electrode 14' have a rectangular cross section (including a square), and a widened portion 15' that contacts the surface of the three-dimensional electrical element 10' from the outside is provided between the region inserted into the three-dimensional electrical element 10' on one side in the longitudinal direction and the region protruding from the three-dimensional electrical element 10' on the other side in the longitudinal direction. Note that a plurality of input electrodes 13 can also be provided.

[0021] Next, with reference to FIG. 3, a method for manufacturing the three-dimensional electrical element 10 will be described. Here, an example in which particles of polyoxymetallate are used for the non-linear portion 11 and carbon nanotubes are used for the conductor 12 is shown. First, as shown in FIG. 3, a large number of non-linear portions 11 (particles of polyoxymetallate) and a large number of conductors 12 (carbon nanotubes) are introduced into isopropyl alcohol 30 to obtain a dispersion liquid 31 (Step 1).

[0022] Immerse the dispersion 31 obtained in Step 1 into a water-soluble porous structure 32, which is an example of a porous structure such as rock sugar, to form a dispersion-containing body 33 in which the non-linear part 11 and the conductor 12 are three-dimensionally arranged (Step 2). Then, immerse a polydimethylsiloxane 34, which is an example of a curable resin, into the dispersion-containing body 33 and heat it to cure, obtaining a cured product 35 (Step 3). Next, immerse the cured product 35 into hot water 36 at about 50 to 70 °C and remove the water-soluble porous structure 32 from the cured product 35, thereby obtaining a three-dimensional electrical element 10 in which the non-linear part 11 is three-dimensionally arranged (Step 4). Note that the cured polydimethylsiloxane 34 is elastic, and the three-dimensional electrical element 10 is sponge-like. Further, a machine learning system can be obtained by adding steps such as connecting an input electrode 13 and an output electrode 14 to the three-dimensional electrical element 10 obtained in Step 4.

[0023] A sheet in which the non-linear part 11 and the conductive part 12 are dispersed and arranged may be formed, and the three-dimensional electrical element may be obtained by stacking the sheets. Also, in Step 4, the created one can be used as an electrical element unit, and a plurality of these electrical element units can be joined to process them into a desired size and shape, so as to obtain a three-dimensional electrical element 10 having a size and shape suitable for the use environment.

[0024] When preparing the dispersion, the proportion of isopropyl alcohol 30 is preferably such that the dispersion (the combined non-linear part 11 and the conductor 12) is sufficiently suspended. If the dispersion is excessive, the dispersion will remain on the surface of the dispersion-containing body 33, and it cannot be denied that this may affect the characteristics. Moreover, if the dispersion becomes excessive, it may be disadvantageous in terms of cost. Examples of the dispersion solvent include pure water, ethanol, and dichloroethane, but the dispersion solvent only needs to be able to disperse the dispersion and is not limited to these.

[0025] Also, regarding the ratio of the non-linear part 11 to the conductor 12, it is desirable that it does not cause insulation or short-circuit linearly. If the ratio of the conductor 12 is too small, a circuit may not be formed and problems such as electrical insulation may occur. If the ratio of the conductor 12 is too large, problems such as short-circuiting may occur.

[0026] The water-soluble porous structure 32 was removed by the hot water 36, but depending on the specifications and usage environment of the three-dimensional electrical element 10, there may be cases where the water-soluble porous structure 32 does not need to be removed. A three-dimensional electrical element with a water-soluble porous structure 32 such as granulated sugar remaining, or a three-dimensional electrical element having an insoluble porous structure such as urethane sponge or ZnO can also exist as one form. Since the three-dimensional electrical element having a water-soluble or insoluble porous structure has a certain degree of hardness, an improvement in the arrangement accuracy of the input electrode and the output electrode can be expected.

[0027] In the above embodiment, the three-dimensional electrical element 10 was fabricated using one water-soluble porous structure 32. However, for example, a three-dimensional electrical element may be fabricated by combining a plurality of water-soluble porous structures of the same or different shapes. In this case, a three-dimensional electrical element of a desired shape and size can be fabricated, and a three-dimensional electrical element can be manufactured in consideration of the usage environment and the like.

[0028] When the three-dimensional electrical element 10 has a shape with corners, it is preferable to chamfer the corners or process the cross-section of the corners into an arc shape in order to prevent chipping of the corners. Specific shapes of the three-dimensional electrical element 10 include a cube, a rectangular parallelepiped, a columnar body, a conical body, a spherical body, an annular body, a hollow body, and the like. In connecting the input electrode 13 and the output electrode 14 to the three-dimensional electrical element 10, it is preferable that the shape of the three-dimensional electrical element 10 is a cube, a rectangular parallelepiped, or a columnar body, and the arrangement accuracy of the input electrode 13 and the output electrode 14 can be improved.

[0029] When connecting the input electrode 13 and the output electrode 14 to the three-dimensional electrical element 10, the tips of the respective electrodes are arranged so as to contact the surface of the three-dimensional electrical element 10 as shown in Fig. 4(A) (hereinafter referred to as "the first arrangement"), or the tips of the respective electrodes are arranged so as to pierce into the three-dimensional electrical element 10 as shown in Fig. 4(B) (hereinafter referred to as "the second arrangement"). The arrangement of the electrodes is determined according to the specifications of the three-dimensional electrical element and the like.

[0030] As an example of the electrode arrangement, it is also possible to use only one of the first arrangement and the second arrangement for the input electrode and the output power, or to use the second arrangement for one of the input electrode and the output power and the first arrangement for the other. Further, when using a plurality of output electrodes as in this embodiment, some of the output electrodes may be in the first arrangement and the remaining output power may be in the second arrangement.

[0031] Also, in this embodiment, after generating the dispersion liquid-containing body 33 in which the non-linear part 11 and the conductor 12 are three-dimensionally arranged in the water-soluble porous structure 32, the water-soluble porous structure 32 is removed after immersing and curing the polydimethylsiloxane 34, which is a curable resin, in the dispersion liquid-containing body 33. However, the manufacturing method of the three-dimensional electrical element is not limited to this. Hereinafter, an example of a manufacturing method of a three-dimensional electrical element different from this embodiment will be described.

[0032] In the manufacturing method of the three-dimensional electrical element 40 shown in Fig. 5, first, a tray 44 containing a liquid epoxy resin (for example, RSF816) 41 and a water-soluble porous structure 43 is accommodated in a chamber 45, and a vacuum pump 46 connected to the chamber 45 is operated. A part of the water-soluble porous structure 43 is immersed in the epoxy resin 41, and the epoxy resin 41 penetrates into the water-soluble porous structure 43 due to the decompression in the chamber 45 caused by the operation of the vacuum pump 46 (up to this point is Step1).

[0033] After the entire water-soluble porous structure 43 was impregnated with the epoxy resin 41, the reduced pressure state in the chamber 45 was released, and the water-soluble porous structure 43 impregnated with the epoxy resin 41 (hereinafter, the water-soluble porous structure 43 in this state is referred to as the resin-containing body 47) was taken out from the chamber 45. As shown in FIG. 5, the resin-containing body 47 was heat-treated to cure the epoxy resin 41 of the resin-containing body 47 (Step2).

[0034] Next, the resin-containing body 47 was immersed in a solvent 49 such as hot water to dissolve the water-soluble porous structure 43 of the resin-containing body 47, and a resin template 50 composed only of the cured epoxy resin 41 was obtained (Step3). If the water-soluble porous structure 43 is granulated sugar, the temperature of the solvent 49 is maintained at about 50 to 70 °C to dissolve the water-soluble porous structure 43.

[0035] Thereafter, a solution 51 in which a polyacid (PMO 12 ) as an example of a non-linear part was dissolved in water or acetonitrile and a solution 52 in which carbon nanotubes as an example of a conductive part were dispersed in isopropanol were mixed, and the mixed solution was subjected to ultrasonic treatment for about 4 hours to obtain a dispersion liquid 53 in which the polyacid and the carbon nanotubes were dispersed (Step4). Then, the resin template 50 was immersed in the dispersion liquid 53 to immerse the dispersion liquid 52 in the resin template 50, and ultrasonic treatment was performed for about 5 minutes to obtain a three-dimensional electrical element 40 in which the polyacid and the carbon nanotubes were supported (fixed) on the resin template 50 and arranged three-dimensionally (Step5).

[0036] In addition, in the method for manufacturing the three-dimensional electrical element 60 shown in FIG. 6, a melamine sponge 63 formed of a melamine resin was immersed in a cellulose-containing liquid 62 containing cellulose fibers, and after ultrasonic treatment for about 1 minute was performed on the melamine sponge 63 immersed in the cellulose-containing liquid 62, a drying treatment was performed to generate a resin template (an example of a porous structure) 64 containing cellulose fibers and cellulose (Step1).

[0037] Here, the cellulose-containing liquid 62 may be adjusted using solutions with different concentrations of cellulose fibers (for example, 0.01% by mass, 0.1% by mass, 1% by mass). From the perspective of enhancing the learning performance of the three-dimensional electrical element 60, it is preferable that the concentration of cellulose fibers in the cellulose-containing liquid 62 is adjusted to approximately 1% by mass. Also, the drying treatment may be a drying treatment at about 50°C using an oven or the like, but natural drying is preferred because voids are distributed throughout the structure.

[0038] Then, as shown in FIG. 6, a solution 65 in which a polyacid (PMO 12 ) which is an example of a non-linear part is dissolved in water or acetonitrile, and a solution 66 in which carbon nanotubes, which are an example of a conductor, are dispersed in isopropanol by ultrasonic treatment for about 1 hour are mixed, and ultrasonic treatment is performed for about 12 hours to obtain a dispersion liquid 67 in which the polyacid and carbon nanotubes are dispersed (Step2).

[0039] Thereafter, with the resin template 64 immersed in the dispersion liquid 67, ultrasonic treatment is performed for about 2 minutes to impregnate the resin template 64 with the dispersion liquid 67, and after a drying treatment at about 50°C using an oven, a three-dimensional electrical element 60 in which the polyacid and carbon nanotubes are supported (fixed) on the resin template 64 and arranged three-dimensionally is obtained (Step3). Note that a protective film such as an epoxy resin may be provided on the surface of the three-dimensional electrical element 60 to enhance the strength.

[0040] Also, in the method for manufacturing the three-dimensional electrical element 70 shown in FIG. 7, a water-soluble granular material (it may also be a water-soluble powder) 71 such as sugar grains and a dispersion liquid 72 in which a polyacid, which is an example of a non-linear part, and carbon nanotubes, which are an example of a conductor, are dispersed in a solvent are mixed, and pressure or the like is applied to the mixture to form a porous template 73 in which the dispersion liquid 72 has penetrated into an aggregate (structure) of the water-soluble granular material 71 (Step1).

[0041] Next, as shown in FIG. 7, the porous template 73 is placed in a tray 75 together with the liquid thermosetting resin 74 to obtain a resin-impregnated body 76 in which the liquid thermosetting resin 74 has penetrated into the porous template 73 (Step2). The impregnation of the thermosetting resin 74 into the porous template 73 can be carried out using a vacuum pump.

[0042] Thereafter, by heating the resin-impregnated body 76, the thermosetting resin 74 (the thermosetting resin 74 that has penetrated into the porous template 73) of the resin-impregnated body 76 is cured to obtain a solidified template 77 (Step3). Then, by immersing the solidified template 77 in a solvent 78 such as hot water, the water-soluble granular material 71 is dissolved and removed from the solidified template 77, and a three-dimensional electrical element 70 in which the polyacid and the carbon nanotubes are supported (fixed) on a resinous porous body and arranged three-dimensionally is obtained (Step4). A protective film may also be provided on the surface of the three-dimensional electrical element 70 to enhance the strength.

[0043] Alternatively, as shown in FIG. 8, a water-soluble granular material 81 and a liquid curable resin 82 may be mixed to obtain a mixed template 83 (Step1). For example, a mixture of the water-soluble granular material 81 and the liquid curable resin 82 can be used as a material, and the mixed template 83 can be created by using a 3D printer. By using a 3D printer, mixed templates 83 of various shapes can be formed. The mixed template 83 becomes somewhat hard by the curing treatment of the curable resin 82 by heat treatment or the like.

[0044] As shown in FIG. 8, the mixed template 83 is immersed in a solvent 84 such as hot water to remove the granular material 81 from the mixed template 83, thereby obtaining a porous body 85 mainly composed of resin (Step2). By immersing the porous body 85 in a dispersion liquid 86 and performing a drying treatment, a three-dimensional electrical element 80 in which the non-linear portion and the conductor are arranged three-dimensionally is generated (Step3).

[0045] Here, instead of immersing the porous body 85 in the dispersion liquid 86, the dispersion liquid 86 may be sprayed onto the porous body 85 with a spraying machine or dropped onto the porous body 85 so that the dispersion liquid 86 permeates into the porous body 85. The three-dimensional electrical element 80 may also be strengthened by forming a protective film on the surface. Experimental example

[0046] Next, the experiments conducted to confirm the effects of the present invention will be described.

[0047] In the experiment, a three-dimensional electrical element (experimental example) in which particles (nonlinear portions) of polyoxometalate connected by carbon nanotubes (conductors) are arranged three-dimensionally and a two-dimensional electrical element (comparative example) in which particles (nonlinear portions) of polyoxometalate connected by carbon nanotubes (conductors) are arranged two-dimensionally were used.

[0048] The three-dimensional electrical element was obtained by curing polydimethylsiloxane and maintaining a state in which particles of polyoxometalate are arranged three-dimensionally. One needle-shaped input electrode and three needle-shaped output electrodes were inserted into this three-dimensional electrical element to fabricate a neural network system (hereinafter referred to as the "3D neural network system"). The two-dimensional electrical element is one in which particles of polyoxometalate and carbon nanotubes are dispersed on an insulating substrate. One plate-shaped input electrode and one plate-shaped output electrode were brought into contact with this two-dimensional electrical element to fabricate a neural network system (hereinafter referred to as the "2D neural network system").

[0049] First, in the 3D neural network system, a sine wave pulse signal was input from the input electrode to the three-dimensional electrical element, and learning was performed so that the waveform of the signal output from the output electrode became a desired waveform different from the input signal such as a triangular wave or a sawtooth wave. Thereafter, a sine wave pulse signal was input from the input electrode to the three-dimensional electrical element, the three-dimensional electrical element was made to predict the desired waveform, and the waveform of the predicted signal was output from the output electrode, and this output signal was detected. The same experiment was also conducted on the 2D neural network system.

[0050] The experimental results with the desired waveforms being triangular wave, rectangular wave, sawtooth wave, and cosine wave are shown in FIGS. 9, 10(A), and 10(B), respectively. In the experimental results of FIG. 9, the accuracy (learning) and accuracy (prediction) were numerical values quantifying how well the waveform of the signal output from the output electrode matched the set desired waveform during learning and prediction in the three-dimensional electrical element, meaning that the larger the numerical value, the better the match.

[0051] In the experimental results of FIGS. 10(A) and 10(B), the error was a numerical value quantifying how different the waveform of the signal output from the output electrode was from the set desired waveform, meaning that the larger the numerical value, the larger the error.

[0052] From the experimental results of FIG. 9, it was confirmed that in the 3D neural network system, the accuracy values were approximately 0.6 points or more both during learning and prediction. Also, from the experimental results of FIGS. 10(A) and 10(B), it was confirmed that the error of the 3D neural network system was smaller than the error of the 2D neural network system.

[0053] As described above, the embodiments of the present invention have been explained, but the present invention is not limited to the above-described forms, and all changes and the like that do not deviate from the gist are within the scope of application of the present invention. For example, the non-linear part does not necessarily have to be particulate, and non-linear parts such as acicular, film-like, and voids showing tunnel junctions can be adopted. Also, the shape of the conductor is not particularly limited, and for example, it may be particulate or film-like.

[0054] And a three-dimensional electrical element may be constituted by a conductive polymer provided with a junction (connection part) showing non-linear electrical voltage characteristics. In this case, in the conductive polymer, the junction corresponds to the non-linear part, and the part excluding the junction corresponds to the conductor. Furthermore, the shapes of the input electrode and the output electrode are not limited, and for example, they may be plate-like.

Industrial Applicability

[0055] According to the three-dimensional electrical element, the machine learning system including the same, and the manufacturing methods thereof according to the present invention, since the machine learning system can be miniaturized, it is possible to promote the mounting of the function learning system on small devices such as mobile terminals and wearable terminals.

Explanation of Reference Numerals

[0056] 10, 10': Three-dimensional electrical element, 11: Nonlinear part, 12: Conductor, 13, 13': Input electrode, 14, 14': Output electrode, 15': Widthening part, 20, 20': Neural network system, 30: Isopropyl alcohol, 31: Dispersion liquid, 32: Water-soluble porous structure, 33: Dispersion liquid-containing body, 34: Polydimethylsiloxane, 35: Cured product, 36: Hot water, 40: Three-dimensional electrical element, 41: Epoxy resin, 43: Water-soluble porous structure, 44: Tray, 45: Chamber, 46: Vacuum pump, 47: Resin-containing body, 49: Solvent, 50: Resin template, 51, 52: Solution, 53: Dispersion liquid, 60: Three-dimensional electrical element, 62: Cellulose-containing liquid, 63: Melamine sponge, 64: Resin template, 65, 66: Solution, 67: Dispersion liquid, 70: Three-dimensional electrical element, 71: Granular substance, 72: Dispersion liquid, 73: Porous template, 74: Thermosetting resin, 75: Tray, 76: Resin-impregnated body, 77: Solidified template, 78: Solvent, 80: Three-dimensional electrical element, 81: Granular substance, 82: Curable resin, 83: Mixed template, 84: Solvent, 85: Porous body, 86: Dispersion liquid

Claims

1. An N - element three - dimensional electrical element comprising N non - linear elements each exhibiting non - linear current - voltage characteristics and a conductor connecting the non - linear elements, wherein the N non - linear elements are arranged in a three - dimensional shape, and the ratio of the non - linear elements not located on a virtual plane on which the geometric centers of any three non - linear elements arbitrarily selected from the N non - linear elements are arranged is 50% or more. However, N is an integer of 4 or more.

2. The three - dimensional electrical element according to claim 1, wherein the three - dimensional electrical element is sponge - shaped.

3. A machine - learning system comprising a three - dimensional electrical element having N non - linear elements each exhibiting non - linear current - voltage characteristics and a conductor connecting the non - linear elements, wherein the N non - linear elements are arranged in a three - dimensional shape, and input electrodes and output electrodes respectively connected to the three - dimensional electrical element, wherein the ratio of the non - linear elements not located on a virtual plane on which the geometric centers of any three non - linear elements arbitrarily selected from the N non - linear elements are arranged is 50% or more. However, N is an integer of 4 or more.

4. The machine - learning system according to claim 3, wherein the three - dimensional electrical element is sponge - shaped.

5. In the machine - learning system according to claim 3, the three - dimensional electrical element has three or more side surfaces, and the output electrodes are respectively inserted into at least three of the side surfaces.

6. A method for manufacturing a three - dimensional electrical element, comprising the steps of immersing a dispersion liquid in which four or more non - linear elements each exhibiting non - linear current - voltage characteristics and a conductor connecting the non - linear elements are introduced into a porous structure to obtain a dispersion - liquid - containing body, and curing the dispersion - liquid - containing body with a curable resin to obtain a three - dimensional electrical element in which the non - linear elements are arranged in a three - dimensional shape.

7. A method for manufacturing a machine - learning system, comprising the steps of immersing a dispersion liquid in which four or more non - linear elements each exhibiting non - linear current - voltage characteristics and a conductor connecting the non - linear elements are introduced into a porous structure to obtain a dispersion - liquid - containing body, curing the dispersion - liquid - containing body with a curable resin to obtain a three - dimensional electrical element in which the non - linear elements are arranged in a three - dimensional shape, and connecting input electrodes and output electrodes to the three - dimensional electrical element.

Images