WO2020240623A1 - Optical computing element and multi-layer neural network - Google Patents

Abstract

Provided is an optical computing element facilitating bias term adjustment. The optical computing element comprises: a first photo-deformable member 30 that deforms in accordance with the intensity of incident light A; a first deformable lid 40 that is connected to the first photo-deformable member 30 and deformed by the first photo-deformable member 30; a refractive index-matching material 60 that matches a refractive index to allow external light B, which is externally introduced light, to pass therethrough; a matching material reservoir 50 that is covered with the first deformable lid 40 and filled with the refractive index-matching material 60; a passage 70 that guides the refractive index-matching material from the matching material reservoir 50 and has an open tip end; a first optical waveguide 80a that is tilted relative to the passage 70 and allows the external light B to pass therethrough; a second optical waveguide 80b that is arranged on the extension line of the first optical waveguide 80a with the passage 70 interposed therebetween and outputs the external light B that enters thereinto after passing through the passage 70; a second photo-deformable member 31 that deforms in accordance with the intensity of bias adjustment light D; and a bias-adjustment liquid reservoir 61 that is connected to the matching material reservoir 50 through a connection passage 90 and filled with the refractive index-matching material 60.

Description

Optical computing elements and multi-layer neural networks

The present invention relates to an optical arithmetic element and a multi-layer neural network that constitute an optical neural network.

The optical neural network models the nerve cell network in the human brain as a unit consisting of two neurons, an input layer neuron and an output layer neuron, and synapses connecting each neuron, and networked using optical signals. It is a thing.

An optical neural network is generally configured by connecting neuron elements that perform multiply-accumulate operations and non-linear operations and having multiple layers (for example, Non-Patent Document 1). The connection of neurons was realized, for example, by using a directional optical coupler using a photochromic material. The on / off of the output light of an optical neural network is determined by the intensity of the input light and the amount of photochromic material corresponding to the magnitude of the bias term.

Keiri Tanimoto and 3 others, "Research on Optical Neural Networks Using Photochromic Materials", 64th JSAP Spring Meeting

However, there is a problem that it is difficult to adjust the size of the bias term because it depends on the increase or decrease in the amount of the photochromic material. That is, since the bias term is determined by the amount of the photochromic material included in the optical arithmetic element, it cannot be changed after the shape of the optical arithmetic element is determined. In order to adjust the bias term, it is necessary to change the shape of the optical arithmetic element.

The present invention has been made in view of this problem, and an object of the present invention is to provide an optical arithmetic element and a multi-layer neural network in which the bias term can be easily adjusted.

The optical calculation element according to one aspect of the present invention includes a first light deforming member that deforms according to the intensity of input light, and a first deformation that is connected to the first light deforming member and deformed by the first light deforming member. The lid, the refractive index matching agent that matches the refractive index and transmits external light introduced from the outside, the matching agent reservoir that is covered with the first modified lid and filled with the refractive index matching agent, and the above. A path having an open tip for deriving the refractive index matching agent from the matching agent reservoir, a first optical waveguide that is inclined with respect to the path and propagates the external light, and the first optical waveguide that sandwiches the path. A second optical waveguide that is arranged on the extension line of the above and outputs the external light that has passed through the path, a second light deforming member that deforms according to the intensity of the bias adjusting light, and the second light deforming member. A second deformable lid that is connected and deformed by the second light deforming member, a bias adjusting liquid pool that is covered with the second deformable lid and filled with the refractive index matching agent, the matching agent pool, and the above. It is a gist to provide a connection path filled with the refractive index matching agent for connecting the bias adjusting liquid pool portion.

Further, the multi-layer neural network according to one aspect of the present invention is a multi-layer neural network in which N (N ≧ 2) optical arithmetic elements are connected in cascade, and n (n = 2,3, ..., N). The gist is that the input light of the optical arithmetic element of the first layer includes the output light of the optical arithmetic element of the n-1th layer.

According to the present invention, it is possible to provide an optical arithmetic element and a multi-layer neural network in which the bias term can be easily adjusted.

It is sectional drawing which shows typically the structural example of the optical arithmetic element which concerns on 1st Embodiment of this invention. It is a figure which shows typically the relationship between the light beam propagating through the first optical waveguide shown in FIG. 1 and reaching the path, and the refractive index matching agent moving in the path, and (a) the relationship between the light beam and refraction. The schematic diagram shown, (b) is a diagram showing an example of the relationship between the input light and the output light. It is a figure which shows the model of the nerve cell which the bias term can adjust. It is a figure which shows typically the example in which two optical arithmetic elements shown in FIG. 1 are connected vertically. It is a figure which shows typically the example which formed the multi-layer neural network by connecting the optical arithmetic elements shown in FIG. 1 in multi-layer. It is a figure which shows typically the example which arranged the optical component between the two optical arithmetic elements shown in FIG. It is a perspective view which shows typically the structural example of the optical arithmetic element which concerns on 2nd Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. The same reference numerals are given to the same objects in a plurality of drawings, and the description is not repeated.

[First Embodiment]
FIG. 1 is a cross-sectional view schematically showing a configuration example of an optical calculation element according to the first embodiment of the present invention. The optical arithmetic element 1 shown in FIG. 1 is an optical arithmetic element that can amplify light without using photoelectric conversion and can easily adjust the bias term.

(Structure of optical arithmetic element)
In the optical arithmetic element 1 shown in FIG. 1, for example, two rectangular parallelepiped housings 10 and 12 are connected by a connection path 90 to form one optical arithmetic element. FIG. 1 is a diagram showing a cross section in the long side direction cut near the center of the short sides of the housings 10 and 12 arranged in the vertical direction.

The housings 10 and 12 and the connection path 90 are made of, for example, an organic molecular polymer or quartz. In addition, these may be constructed using other materials (for example, metal).

The housing 10 includes a first light deforming member 30a, a first deforming lid 40a, a matching agent reservoir 50a, a refractive index matching agent 60, a path 70, a first optical waveguide 80a, and a second optical waveguide 80b. The inside of the matching agent reservoir 50a is filled with the refractive index matching agent 60.

The housing 12 includes a second light deforming member 30b, a second deforming lid 40b, and a bias adjusting liquid pool portion 50b. The inside of the bias adjusting liquid pool 50b is filled with the refractive index matching agent 60. The bias adjusting liquid pool 50b and the matching agent pool 50a are connected by a connection path 90 filled with the refractive index matching agent 60.

FIG. 1 shows an example in which two rectangular parallelepipeds of the housing 10 and 12 are combined to form one optical calculation element 1, but the optical calculation element 1 may be configured in one housing. As shown in FIG. 1, if the optical calculation element 1 is formed by combining the respective components, the shapes of the housings 10 and 12 are not limited to the rectangular parallelepiped. Further, the housings 10 and 12 may be formed of a frame. That is, it is not necessary to hold each component in a solid shape such as a rectangular parallelepiped.

An opening 20a is provided on one end side of the housing 10. The opening 20a in this example has a quadrangular plane and penetrates in the height direction of the housing 10. The input light A of the optical signal is input to the opening 20. Here, the direction is defined for the sake of explanation. The opening 20a side of the housing 10 is the rear side, and the opposite side is the front side.

The central portion of the inner wall on the front side of the opening 20a is hollowed out in a columnar shape to form a matching agent reservoir 50a. The first deformation lid 40a is fitted on the opening 20a side (rear side) of the matching agent reservoir 50a.

The first deformable lid 40a is made of a flexible material and deforms when a force is input. The first deformable lid 40a is made of, for example, rubber.

A locking portion 41a having a U-shaped flat surface is formed in the central portion of the first deformable lid 40a. A locking portion 11a having the same shape is also formed on the inner wall on the rear side of the opening 20a on which the locking portion 41a faces.

The first light deforming member 30a is hung between the locking portion 11a and the locking portion 41a, and both ends of the first light deforming member 30a are fixed to the locking portion 11a and the locking portion 41a, respectively. The first photodeformable member 30a connects the first deformable lid 40a and the inner wall (rear side) of the opening 20a while holding a predetermined tension.

The first light deforming member 30a is deformed according to the intensity of the input light A. As the first photodeformable member 30a, for example, a crosslinked polymer having diarylethene, cyclodextrin, and azobenzene can be used.

The inside of the matching agent reservoir 50a is filled with the refractive index matching agent 60 and sealed with the first deformation lid 40a. A path 70 having a rectangular cross section is formed from the central portion of the front end surface of the matching agent reservoir 50.

As the refractive index matching agent 60, for example, silicone oil can be used. The refractive index of the refractive index matching agent 60 is, for example, 1.485 (25 ° C.), and has substantially the same refractive index as the first optical waveguide 80a and the second optical waveguide 80b.

The path 70 has, for example, a rectangular cross section in a direction orthogonal to the front-rear direction, and horizontally penetrates from the front end face of the matching agent reservoir 50a to the front end face of the housing 10 and the tip is open. Inside the path 70, the refractive index matching agent 60 is derived from the matching agent collecting portion 50a, and is filled with the refractive index matching agent 60 up to a position about half of the path 70 in the front-rear direction. The tip portion of the refractive index matching agent 60 in the path 70 moves in the front-rear direction in response to the first light deforming member 30a being deformed according to the intensity of the input light A.

The derivation direction of the route 70 is not limited to the horizontal direction. The refractive index matching agent 60 in the path 70 is almost unaffected by gravity due to its surface tension. Therefore, the derivation direction of the path 70 may be any of the vertical direction up and down, the diagonal up and down direction, and the like. The position of the tip portion of the refractive index matching agent 60 in the path 70 is mainly determined by the amount of deformation of the first deformation lid 40a due to the expansion and contraction of the first light deformation member 30a.

The first optical waveguide 80a is arranged so as to be inclined with respect to the path 70 of the portion where the tip portion of the refractive index matching agent 60 moves in the front-rear direction, and the tip portion thereof is in contact with the path 70. External light B is input from the side opposite to the path 70 of the first optical waveguide 80a. The external light B is the ambient light in the environment in which the optical calculation element 1 is arranged. The external light B is maintained at a constant intensity (illuminance). The intensity of the external light B may vary to some extent.

The second optical waveguide 80b is arranged on an extension line of the first optical waveguide 80a with the path 70 in between. The second optical waveguide 80b outputs the external light B transmitted through the path 70 to the outside (output light C).

The first optical waveguide 80a and the second optical waveguide 80b are made of a material having a higher refractive index than the other housings 10 when the housing 10 is made of an organic molecular polymer. The refractive index of the first optical waveguide 80a and the second optical waveguide 80b is, for example, 1.49.

An opening 20b is provided on one end side of the housing 12 as in the housing 10. The opening 20b in this example has a quadrangular plane and penetrates in the height direction of the housing 12. Bias adjustment light D of an optical signal is input to the opening 20b. A light-shielding plate (not shown) is provided between the openings 20a and 20b so that the bias adjusting light D does not affect the first light deformation member 30a. The optical calculation element 1 may be configured so that the openings 20a and 20b do not overlap in the vertical direction without providing a light-shielding plate.

The central portion of the inner wall on the front side of the opening 20b is hollowed out in a columnar shape to form the bias adjusting liquid pool portion 50b. A second deformation lid 40b is fitted on the opening 20b side (rear side) of the bias adjusting liquid pool 50b. The shapes and materials of the bias adjusting liquid pool portion 50b, the second deformable lid 40b, the second light deforming member 30b, the locking portion 11b, and the locking portion 41a are provided in the housing 10 corresponding to the numbers of reference numerals. Same as the component.

The inside of the bias adjusting liquid reservoir 50b is the same as the matching material reservoir 50a in that it is filled with the refractive index matching material 60. The bias adjusting liquid pool 50b and the matching material pool 50a are connected by a connection path 90 filled with the refractive index matching material 60.

The housings 10 and 12, the matching agent reservoir 50a, the bias adjusting liquid reservoir 50b, and the path 70 are formed by, for example, processing rectangular parallelepiped quartz by a well-known semiconductor process and micromachine processing technique.

As described above, the optical calculation element 1 according to the present embodiment is connected to the first light deforming member 30a that deforms according to the intensity of the input light and the first light deforming member 30a, and is connected by the first light deforming member 30a. The first deformable lid 40a to be deformed, the refractive index matching agent 60 that matches the refractive index and transmits external light B introduced from the outside, and the first deformable lid 40a is covered with the refractive index matching agent 60. The matching agent pool 50a, the path 70 having an open tip for deriving the refractive index matching agent 60 from the matching agent pool 50a, and the first optical waveguide 80a that is inclined with respect to the path 70 and propagates the external light B. The second optical waveguide 80b, which is arranged on the extension line of the first optical waveguide 80a with the path 70 in between and outputs the external light B transmitted through the path 70, and the bias adjusting light D are deformed according to the intensity. The second light deforming member 30b, the second deformable lid 40b connected to the second light deforming member 30b and deformed by the second light deforming member 30b, and the second deformable lid 40b are covered with the refractive index matching agent 60. It includes a filled bias adjusting liquid pool 50b and a connection path 90 filled with a refractive index matching agent 60 that connects the matching agent pool 50a and the bias adjusting liquid pool 50b.

(Action of optical arithmetic element)
FIG. 2 is a diagram schematically showing the relationship between the light beam propagating through the first optical waveguide 80a and reaching the path 70 and the refractive index matching agent 60 moving in the path 70. FIG. 2A shows a schematic diagram showing the relationship between the light beam and the refractive index matching agent 60, and FIG. 3B shows an example of the relationship between the input light A and the output light C.

The ellipse 81 shown in FIG. 3A schematically represents a light beam (hereinafter, light beam 81) that propagates through the first optical waveguide 80a and reaches the path 70. Since the first optical waveguide 80a is inclined and in contact with the path 70, the shape of the light beam 81 is elliptical.

Here, the tip of the refractive index matching agent 60 that moves in the path 70 due to the change in the intensity of the input light A is located at the front end γ of the light beam 81 when the intensity of the input light A is maximum. Suppose it is tuned. Further, it is assumed that the light beam 81 is adjusted so as to be located at the rear end α when the intensity of the input light A is the minimum.

This assumption holds when the first light deforming member 30a stretches when the intensity of the input light A is high. If the direction in which the first light deforming member 30a is deformed is opposite, the tip of the refractive index matching agent 60 is located at the rear end α of the light beam 81, for example, when the intensity of the input light A is maximum. It is adjusted so that it is located at the front end γ of the light beam 81 when the intensity of the input light A is the minimum. That is, the logic with respect to the intensity of the input light A can be reversed depending on the deformation direction of the first light deforming member 30a.

Further, the amount of the refractive index matching agent 60 existing inside the path 70 and the matching agent collecting portion 50a can be adjusted by the intensity of the bias adjusting light D. Since the volume of the matching agent reservoir 50a is fixed, the position of the tip of the refractive index matching agent 60 in the path 70 can be changed by changing the intensity of the bias adjusting light D. That is, the bias term can be adjusted.

Here, it is assumed that the position of the tip of the refractive index matching agent 60 when the intensity of the bias adjusting light D is the minimum is the end α of the above-mentioned light beam 81. Then, the second light deforming member 30b is also stretched as the intensity of the bias adjusting light D increases, similarly to the first light deforming member 30a.

Further, the position of the tip of the refractive index matching agent 60 when the intensity of the bias adjusting light D is the minimum is defined as the end α of the light beam 81. Further, the position of the tip of the refractive index matching agent 60 when the intensity of the bias adjusting light D is maximum is set to β3 in the light beam 81. Further, the position of the tip of the refractive index matching agent 60 when the intensity of the bias adjusting light D is half of the maximum and the minimum is defined as β2 in the light beam 81.

In the above assumption, the relationship between the input light A, the external light B, the output light C, and the bias adjusting light D will be described. First, the case where the intensity of the bias adjusting light D is the minimum will be described.

When the intensities of the bias adjusting light D and the input light A are the minimum, the tip portion of the refractive index matching agent 60 is located at α shown in FIG. 3A. Therefore, the light beam 81 and the refractive index matching agent 60 do not overlap. Therefore, since the refractive index of the first optical waveguide 80a and the refractive index of the path 70 do not match, most of the external light B is reflected by the path 70. Therefore, the intensity of the output light C is minimized (origin in FIG. 3B).

When the intensity of the bias adjusting light D is the minimum and the intensity of the input light A is the maximum, the tip portion of the refractive index matching agent 60 is located at γ shown in FIG. 3A. Therefore, since the refractive index matching agent 60 occupies the entire area of the light beam 81, the refractive index of the first optical waveguide 80a and the refractive index of the path 70 match within that range. Therefore, most of the external light B passes through the path 70, and the intensity of the output light C is maximized. In this way, the intensity of the output light C can be changed (proportional in this example) in accordance with the change in the intensity of the input light A. The characteristics in this case are shown by the solid line in FIG. 3 (b).

Next, the case where the intensity of the bias adjusting light D is changed will be described.

When the intensity of the input light A is the minimum, the intensity of the vise adjustment light D is slightly increased from the minimum. Then, the second light deforming member 30b is stretched, and the refractive index matching agent 60 in the bias adjusting liquid pooling portion 50b is pushed out to the matching agent pooling portion 50a via the connection path 90. As a result, the position of the tip portion of the refractive index matching agent 60 in the path 70 moves to the front side.

It is assumed that the position of the tip portion of the refractive index matching agent 60 when the intensity of the bias adjusting light D is slightly increased from the minimum is β1 shown in FIG. 3A. When the position of the tip portion of the refractive index matching agent 60 is β1, the refractive indexes of a part of the light beam 81 are matched. As a result, the intensity of the output light C becomes the intensity corresponding to the position of β1 of the light beam 81 (β1 on the y-axis of FIG. 3B).

Therefore, the relationship between the input light A and the output light C when the intensity of the bias adjusting light D is slightly increased from the minimum is the characteristic shown by the alternate long and short dash line in FIG. 3 (b). That is, the output light C has a characteristic biased by β1.

Also, when the intensity of the input light A is the minimum, the intensity of the bias adjusting light D is reduced to half the size of the minimum and the maximum. In that case, the position of the tip portion of the refractive index matching agent 60 is assumed to be β2 shown in FIG. 3 (a). When the position of the tip portion of the refractive index matching agent 60 is β2, the external light B having a half area of the light beam 81 is transmitted to become the output light C (β2 on the y-axis in FIG. 3B).

Therefore, the relationship between the input light A and the output light C when the intensity of the bias adjusting light D is halved from the minimum and the maximum is the characteristic shown by the alternate long and short dash line in FIG. 3 (b). That is, the output light C has a characteristic biased by β2.

The relationship between the input light A and the output light C when the intensity of the bias adjusting light D is maximized is the characteristic shown by the broken line in FIG. 3 (b). The detailed description is the same as in other cases and will be omitted.

As described above, the intensity of the external light B can be changed with respect to the change in the intensity of the input light A. Further, the bias amount of the output light C can be changed depending on the intensity of the bias adjusting light D.

The intensity of external light B is constant. Therefore, by setting the intensity of the external light B to be higher than the maximum intensity of the input light A, the output light B in which the input light A is amplified can be output.

That is, according to the optical calculation element 1 according to the present embodiment, the light intensity of the input light A can be amplified without using photoelectric conversion. Further, since the optical arithmetic element 1 does not perform photoelectric conversion, it does not cause power loss and speed loss associated therewith.

Further, by applying (irradiating) a certain amount of external light B to each of the optical arithmetic elements 1 connected in multiple layers (multiple layers are longitudinally connected), a multi-layer neural network that does not cause a loss of optical signal intensity is constructed. Can be done.

FIG. 3 is a diagram showing a model of a nerve cell using the optical arithmetic element 1 according to the present embodiment. In FIG. 3, x1 to xn are inputs, w1 to wn are coupling weights, and b is a bias term. According to the optical arithmetic element 1, it is possible to realize a model of a nerve cell that can be expressed by the following equation.

The adjustment of b can be easily performed by changing the intensity of the bias adjustment light D.

(Multilayer neural network)
FIG. 4 is a diagram schematically showing a configuration in which two optical arithmetic elements 1 according to the present embodiment are connected in cascade. In FIG. 4, the description of the housing 12 and the configuration provided therein is omitted. Hereinafter, the description of the housing 12 and the configuration provided therein, which are not necessary for explanation, will be omitted.

The optical signal output from the second optical waveguide 80b of the first optical arithmetic element 11 is input to the opening 202 of the second optical arithmetic element 12. External light B having a constant light intensity is input to the first optical waveguide 80b1 and the first optical waveguide 80b2 of the optical calculation element 11 and the optical calculation element 12, respectively.

A multi-layer neural network can be constructed by connecting two or more optical arithmetic elements 1 in a sequential manner.

FIG. 5 is a diagram schematically showing a configuration example of a multi-layer neural network according to the present embodiment. The multilayer neural network 100 shown in FIG. 5 is formed by connecting two or more layers of the above optical arithmetic elements 1 in a longitudinal manner.

The input light of the optical arithmetic element 11 of the first layer is the sum of the outputs of the two multipliers 101 and 1021 by the adder 1031. Here, the bias adjustment light D is represented by D input to the adder 1031. The same applies to the optical arithmetic elements 12 and 13 of the second layer and the third layer. The bias term can be adjusted for each layer.

The output light Z1 of the first layer optical calculation element 11 is input to the multiplier 1012 that generates the input light of the second layer optical calculation element 12. The multiplier 1012 multiplies the output light Z1 by the weight w3 and outputs it to one input of the adder 1032.

The adder 1032 adds the output of the multiplier 1012 and the output of the multiplier 1022 to generate the input light of the optical arithmetic element 12. The output of the multiplier 1022 is obtained by multiplying the output light Z2 of an optical arithmetic element (not shown) by the weight w4.

The optical arithmetic element 12 of the second layer generates the output light Z3 obtained by converting the external light B taken in from the outside by the product-sum signal output by the adder 1032 corresponding to the input light A. The output light Z5 generated by the optical calculation element 13 of the third and subsequent layers also has the same configuration for generating the output light Z5 as the optical calculation element 12 of the second layer. The reference code numbers in the figures are updated and described, and the description thereof will be omitted.

As described above, the multi-layer neural network 100 according to the present embodiment is a multi-layer neural network in which N (N ≧ 2) optical arithmetic elements 1 are connected in cascade, and n (n = 2,3, ..., N). The input light An of the optical arithmetic element of the layer) includes the output light Zn-1 of the optical arithmetic element of the n-1th layer.

According to this configuration, external light B having a constant intensity is input to each of the optical arithmetic elements 1n of each layer, and the external light B is converted by the output light Zn-1 of the optical arithmetic element 1n-1 of the front layer n-1. The output light Zn is generated. Therefore, the intensity of the output light Zn of the rear optical arithmetic element 1n connected in multiple layers is not attenuated. As a result, photoelectric conversion is not required, and the multi-layer neural network can be depowered.

Other optical components may be arranged between the optical arithmetic elements 1n of each layer. As the optical component, for example, an optical filter and a coupler can be considered.

FIG. 6 is a diagram schematically showing an example in which, for example, an optical filter 95 is arranged between the optical arithmetic element 11 and the optical arithmetic element 12 shown in FIG. The optical filter 95 may be replaced with a coupler that branches an optical signal.

In this way, a coupler or modulation that branches the output light of the nth optical arithmetic element between the n (n = 2,3, ..., N) layer optical arithmetic element and the n + 1th layer optical arithmetic element. A multi-layer neural network may be configured so that the optical fluter to be generated is arranged. According to this, the degree of freedom in designing the multi-layer neural network can be improved.

[Second Embodiment]
FIG. 7 is a perspective view schematically showing a configuration example of an optical calculation element according to a second embodiment of the present invention. FIG. 7 is a diagram corresponding to FIG.

The optical calculation element 2 shown in FIG. 7 is different from the optical calculation element 1 (FIG. 1) in that the third optical waveguide 80c is provided. The third optical waveguide 80c is for deriving the reflected light reflected by the path 70 to the outside.

It is considered that there are two types of reflected light reflected by the path 70: a reflected light that propagates through the first optical waveguide 80a and returns to the input side, and a reflected wave that repeats reflection until it disappears near the path 70. The latter reflected wave may deteriorate the signal-to-noise ratio of the calculation of the optical signal.

Therefore, the optical arithmetic element 2 according to the present embodiment has a reflection angle of the same angle as the incident angle of the first optical waveguide 80a with respect to the path 70, and the tip portion is brought into contact with the tip portion of the first optical waveguide 80a. It is equipped with 80c. As a result, the external light B reflected by the path 70 is led out to the outside by the third optical waveguide 80c. Therefore, since the reflected wave that repeats reflection near the path 70 is reduced, the SN ratio of the calculation of the optical signal can be improved.

It is also possible to use the reflected light derived from the third optical waveguide 80c for the calculation. By using the reflected light output from the third optical waveguide 80c for the calculation, the degree of freedom in designing the multi-layer neural network can be improved.

As described above, according to the optical arithmetic elements 1 and 2 of the present embodiment, it is possible to construct a multi-layered optical neural network without performing photoelectric conversion. Further, according to the multi-layer neural network 100 of the present embodiment, since the external light B is introduced from the outside into each of the layers, it is not necessary to alternately perform the calculation using the optical signal and the electric signal. As a result, the calculation can be reduced in power. Moreover, it is possible to adjust the bias term for each layer.

In addition, since it can be composed of a non-metal material, it can be used in situations where it is difficult to apply an IoT device using a conventional semiconductor chip, including the fact that it is powerless. In addition, it also has the effect of reducing the cost and the risk of failure by reducing the number of parts by not performing photoelectric conversion.

The first and second photodeformable members 30a and 30b of the present invention have been described with the example of the so-called string-shaped form, but the present invention is not limited to this example. These light deforming members may be, for example, strip-shaped or knitted strings. In short, the photodeformable member may be any one as long as it deforms such as expansion, bending, and extension. As described above, the present invention is not limited to the above-described embodiment, and can be modified within the scope of the gist thereof.

Although the processing method of the optical arithmetic elements 1 and 2 has not been described by showing a specific example, all the existing semiconductor process and micromachine processing technology can be used for the processing.

1, 11-13, 2: Optical arithmetic elements 10, 12: Housing 20a, 20b: Opening 30a: First light deforming member 30b: Second light deforming member 40a: First deforming lid 40b: Second deforming lid 50a : Matching agent pool 50b: Bias adjusting liquid pool 60: Refractive index matching agent 70: Path 80a: First optical waveguide 80b: Second optical waveguide 80c: Third optical waveguide 90: Connection path 95: Optical filter or coupler 100 : Multilayer neural network A: Input light B: External light C: Output light D: Bias adjustment light

Claims (5)

The first light deforming member that deforms according to the intensity of the input light,
A first deformable lid that is connected to the first light deforming member and deformed by the first light deforming member.
A refractive index matching agent that matches the refractive index and transmits external light introduced from the outside,
The matching agent reservoir, which is covered with the first deformation lid and filled with the refractive index matching agent,
A path with an open tip for deriving the refractive index matching agent from the matching agent reservoir,
A first optical waveguide that is inclined with respect to the path and propagates the external light,
A second optical waveguide that is arranged on an extension of the first optical waveguide with the path in between and outputs the external light that has passed through the path.
A second light deforming member that deforms according to the intensity of the bias adjustment light,
A second deformable lid that is connected to the second light deforming member and deformed by the second light deforming member.
A bias adjusting liquid pool that is covered with the second deformable lid and filled with the refractive index matching agent.
An optical calculation element including a connection path filled with the refractive index matching agent that connects the matching agent pooling portion and the bias adjusting liquid pooling portion. The first aspect of claim 1, wherein the third optical waveguide has a reflection angle of the same angle as the incident angle of the first optical waveguide with respect to the path, and the tip portion is brought into contact with the tip portion of the first optical waveguide. Optical computing element. The optical calculation element according to claim 1 or 2, wherein the external light is shielded when the intensity of the input light is the minimum. A multi-layer neural network in which N (N ≧ 2) optical arithmetic elements according to any one of claims 1 to 3 are connected in cascade.
A multi-layer neural network characterized in that the input light of the optical arithmetic element of the n (n = 2,3, ..., N) layer includes the output light of the optical arithmetic element of the n-1th layer. A coupler that branches or modulates the output light of the nth optical arithmetic element between the n (n = 2,3, ..., N) layer optical arithmetic element and the n + 1th layer optical arithmetic element. The multi-layer neural network according to claim 4, wherein is arranged.

Images