Optical computing unit and calculation method using the same

Abstract

The present invention provides an optical computing unit that is miniaturized and capable of performing optical parallel integration. [Solution] The optical computing unit comprises an image sensor 10 having pixels 11 arranged in two dimensions, a light source unit 20 having an optical output unit corresponding to each pixel 11, and a liquid crystal panel 31 provided between the light source unit 20 and the image sensor 10, having liquid crystal cells corresponding to each pixel 11. The light source unit 20 can control the amount of light from its optical output unit, and the liquid crystal panel can control the transmittance of its liquid crystal cells. With the amount of light from the optical output unit of the light source unit 20 set according to the elements of the first matrix, and the transmittance of the liquid crystal cells of the liquid crystal panel 31 set according to the elements of the second matrix, the product of the first matrix and the second matrix is ​​calculated based on the amount of light received by the pixels 11 in the image sensor 10.

Description

Technical Field

[0004]

[0001] The present disclosure relates to an optical arithmetic unit capable of executing super-parallel optical input matrix arithmetic and the like.

Background Art

[0002] In Patent Document 1, an optical vector matrix arithmetic unit that constitutes an optical input / output type optical computer used for optical parallel information processing and the like is disclosed. In this optical vector matrix arithmetic unit, an optical intensity attenuator array group composed of a plurality of optical intensity attenuator arrays respectively corresponding to parts that contribute to the arithmetic in the matrix is used. An optical distributor distributes a two-dimensional optical input image corresponding to an input vector to the optical intensity attenuator array group, and an optical collector adds up the optical intensities of some pixels of the output of the optical intensity attenuator array group.

[0003] In Patent Document 2, a multifunctional and multipurpose liquid crystal arithmetic unit using liquid crystal cells is disclosed. This liquid crystal arithmetic unit has a structure in which a pair of liquid crystal cells for transmitting light having a specific polarization plane are stacked. The first liquid crystal cell has an electrode for applying a voltage according to the first data, and includes a liquid crystal layer that rotates the polarization plane according to the applied voltage. The second liquid crystal cell has an electrode for applying a voltage according to the second data, and includes a liquid crystal layer that rotates the polarization plane according to the applied voltage. By utilizing the additivity of the rotation angles of the polarization planes, the first data and the second data are optically added.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] In Patent Document 1, the matrix corresponding to the optical intensity attenuator array in the optical vector matrix calculator is fixed. Therefore, it can be said that it is not suitable for general-purpose calculation. In Patent Document 2, a liquid crystal calculator is shown with a configuration in which liquid crystal cells are stacked, but no special consideration is given to performing parallel multiplication, such as matrix multiplication.

[0006] In view of the foregoing, this disclosure aims to provide an optical computing unit that can be miniaturized and capable of performing optical parallel integration. [Means for solving the problem]

[0007] An optical computing unit according to one aspect of the present disclosure comprises an image sensor having pixels arranged in a two-dimensional manner; a light source unit having light output units arranged in a two-dimensional manner corresponding to each of the pixels, and capable of controlling the amount of light from the light output units; and a first liquid crystal panel provided between the light source unit and the image sensor, having liquid crystal cells arranged in a two-dimensional manner corresponding to each of the pixels, and capable of controlling the transmittance of the liquid crystal cells. The optical computing unit is configured to calculate the product of the first matrix and the second matrix based on the amount of light received by the pixels in the image sensor, with the amount of light from the light output units in the light source unit set according to the elements of a first matrix and the transmittance of the liquid crystal cells in the first liquid crystal panel set according to the elements of a second matrix. [Effects of the Invention]

[0008] This disclosure makes it possible to realize a compact optical computing unit capable of performing optical parallel integration. [Brief explanation of the drawing]

[0009] [Figure 1] Example configuration of the optical computing unit according to the first embodiment [Figure 2] Image of calculations performed using an optical computing unit. [Figure 3] (a) to (c) are examples of calculations performed on paper. [Figure 4] (a) to (d) are examples of calculations using an optical computing unit. [Figure 5] Example system configuration including an optical computing unit [Figure 6] An example of processing in the system shown in Figure 5. [Figure 7] Example of a configuration in which an optical computing unit has a separation unit. [Figure 8] Example of a configuration in which an optical computing unit has a separation unit. [Figure 9] Example of a configuration in which an optical computing unit has a separation unit. [Figure 10] An example of a configuration in an optical computing unit that performs addition optically. [Figure 11] (a) and (b) are examples of configurations and operations in which addition is performed at the electrical signal level. [Figure 12] Image of an operation that alternates between positive and negative inversion. [Figure 13] An example of a configuration in which an optical computing unit is covered with a shielding structure. [Figure 14] Example configuration of the optical computing unit according to the second embodiment [Figure 15] Configuration example of an optical computing unit according to another embodiment [Modes for carrying out the invention]

[0010] (overview) An optical computing unit according to an aspect of the present disclosure comprises an image sensor having pixels arranged in a two-dimensional manner; a light source unit having light output units arranged in a two-dimensional manner corresponding to each of the pixels, and capable of controlling the amount of light from the light output units; and a first liquid crystal panel provided between the light source unit and the image sensor, having liquid crystal cells arranged in a two-dimensional manner corresponding to each of the pixels, and capable of controlling the transmittance of the liquid crystal cells. The optical computing unit is configured to calculate the product of the first matrix and the second matrix based on the amount of light received by the pixels in the image sensor, with the amount of light from the light output units in the light source unit set according to the elements of a first matrix and the transmittance of the liquid crystal cells in the first liquid crystal panel set according to the elements of a second matrix.

[0011] According to this configuration, the optical arithmetic unit includes an image sensor having pixels arranged in a two-dimensional manner, a light source unit having a light output unit corresponding to each pixel, and a first liquid crystal panel provided between the light source unit and the image sensor and having a liquid crystal cell corresponding to each pixel. The light source unit can control the amount of light output from the light output unit, and the first liquid crystal panel can control the transmittance of the liquid crystal cell. As a result, individual integration based on the amount of light output from the light output unit and the transmittance of the liquid crystal cell can be performed for each pixel of the image sensor, improving the area efficiency of the optical arithmetic unit. And according to this optical arithmetic unit, in a state where the amount of light output from the light output unit in the light source unit is set according to the elements of the first matrix and the transmittance of the liquid crystal cell in the first liquid crystal panel is set according to the elements of the second matrix, it is possible to obtain the product of the first matrix and the second matrix based on the amount of light received by the pixels in the image sensor.

[0012] In the optical arithmetic unit according to the above aspect, the light source unit may include a surface light source and a second liquid crystal panel provided between the surface light source and the first liquid crystal panel, having liquid crystal cells arranged in a two-dimensional manner corresponding to each pixel, and configured to be able to control the transmittance of the liquid crystal cells.

[0013] As a result, the light source unit can be configured using a surface light source and a second liquid crystal panel.

[0014] In the optical arithmetic unit according to the above aspect, the light source unit may include, as the light output unit, a micro light source capable of controlling the light emission amount.

[0015] As a result, the light source unit can be configured using a micro light source.

[0016] In the optical arithmetic unit according to the above aspect, for each liquid crystal cell of the first liquid crystal panel, a first separation unit for preventing the light signal input to the liquid crystal cell adjacent to the liquid crystal cell from being mixed in, and for each pixel of the image sensor, a second separation unit for preventing the light signal input to the pixel adjacent to the pixel from being mixed in may be provided.

[0017] This prevents the calculation results from mixing between adjacent pixels.

[0018] Furthermore, the first separation portion may be provided at the boundary between liquid crystal cells on the light-receiving side of the first liquid crystal panel and be a light-absorbing member, and the second separation portion may be provided at the boundary between pixels on the light-receiving side of the image sensor and be a light-absorbing member.

[0019] Furthermore, the first separation portion may be provided at the boundary between liquid crystal cells on the light-receiving side of the first liquid crystal panel and be a material with a relatively lower refractive index than the surrounding area, and the second separation portion may be provided at the boundary between pixels on the light-receiving side of the image sensor and be a material with a relatively lower refractive index than the surrounding area.

[0020] Furthermore, the first separation unit may be a lens positioned on the liquid crystal cell on the light-receiving side of the first liquid crystal panel and having a refractive index relatively higher than that of the periphery, and the second separation unit may be a lens positioned on the pixel on the light-receiving side of the image sensor and having a refractive index relatively higher than that of the periphery.

[0021] In the optical computing unit according to the above embodiment, a first polarizing film may be disposed between the light source unit and the first liquid crystal panel, and a second polarizing film may be disposed between the first liquid crystal panel and the image sensor, the polarization angle of which is 90 degrees different from that of the first polarizing film.

[0022] In the optical computing unit according to the above embodiment, a lens may be provided on the light-ingress side of the image sensor, which collects the light outputs of multiple liquid crystal cells in the first liquid crystal panel into a single pixel in the image sensor.

[0023] This allows the individual integrated results of the light intensity of the light output unit and the transmittance of the liquid crystal cell to be optically added together.

[0024] In the optical computing unit according to the above embodiment, a circuit for adding electrical signals output from multiple pixels of the image sensor may be provided.

[0025] This allows the individual integrated results of the light intensity of the light output section and the transmittance of the liquid crystal cell to be electrically added together.

[0026] In the optical computing unit according to the above embodiment, the image sensor may be covered by a shielding structure so as not to allow light other than that from the light source to enter it.

[0027] This prevents disturbances caused by light other than that from the light source.

[0028] In the optical computing unit according to the above embodiment, a second liquid crystal panel may be provided between the first liquid crystal panel and the image sensor, having liquid crystal cells arranged in a two-dimensional manner corresponding to each pixel, and capable of controlling the transmittance of the liquid crystal cells.

[0029] This makes it possible to perform the multiplication of three or more matrices in an optical computing unit.

[0030] Furthermore, in a calculation method using the optical calculator according to the above embodiment, when performing multiple calculations with the second matrix fixed, the method may selectively perform the following steps: a first step of setting the light intensity of the light output unit in the light source unit according to the elements of the first matrix, setting the transmittance of the liquid crystal cells in the first liquid crystal panel according to the elements of the second matrix, and calculating the product of the first matrix and the second matrix from the output of the image sensor; and a second step of setting the light intensity of the light output unit in the light source unit according to the elements of the first matrix, setting the transmittance of the liquid crystal cells in the first liquid crystal panel according to the values ​​obtained by inverting the positive and negative signs of the elements of the second matrix, and calculating the product of the first matrix and the second matrix from the values ​​obtained by inverting the positive and negative signs of the output of the image sensor.

[0031] This makes it possible to avoid the influence of residual polarization on the calculation results.

[0032] (Embodiment) The embodiments will be described in detail below with reference to the drawings.

[0033] The embodiments described below are all comprehensive or specific examples. The numerical values, shapes, materials, components, arrangement and connection configurations of components, steps, and the order of steps shown in the following embodiments are examples only and are not intended to limit this disclosure. Furthermore, among the components in the following embodiments, those not described in the independent claim representing the highest-level concept will be described as optional components.

[0034] (First Embodiment) Figure 1 shows an example of the configuration of an optical computing unit according to the first embodiment. Figure 1 schematically shows a part of the cross-sectional configuration of the optical computing unit.

[0035] As shown in Figure 1, the optical processing unit includes an image sensor 10 having pixels 11 arranged in two dimensions. The optical processing unit includes a light guide plate 21 which serves as a surface light source, a liquid crystal panel 22, and a polarizing film 23. The liquid crystal panel 22 is arranged in two dimensions and has liquid crystal cells corresponding to each pixel 11 of the image sensor 10. The liquid crystal cells have, for example, TN liquid crystal, an alignment film sandwiching it, and transparent electrodes. The polarizing film 23 is placed between the light guide plate 21 and the liquid crystal panel 22. The light source unit 20 is composed of the light guide plate 21, the liquid crystal panel 22, and the polarizing film 23. In the light source unit 20, each liquid crystal cell of the liquid crystal panel 22 becomes a light output unit corresponding to each pixel 11 of the image sensor 10.

[0036] The liquid crystal panel 31 is provided between the image sensor 10 and the liquid crystal panel 22. The liquid crystal panel 31 is arranged in two dimensions and has liquid crystal cells corresponding to each pixel 11 of the image sensor 10. The polarizing film 32 is provided between the liquid crystal panel 22 and the liquid crystal panel 31. The polarizing film 33 is provided between the liquid crystal panel 31 and the image sensor 31. The polarization angle of the polarizing film 33 is 90 degrees different from that of the polarizing film 32. The polarization angle of the polarizing film 23 is 90 degrees different from that of the polarizing film 32.

[0037] Figure 2 shows an image of the calculation performed using the optical calculator shown in Figure 1. For example, let's consider calculating the product of a 1x3 matrix A and a 3x1 matrix B. In this case, the orientation angles corresponding to elements a1, a2, and a3 of matrix A are set for the liquid crystal cells of the liquid crystal panel 22 by applying a voltage to the transparent electrode. Similarly, the orientation angles corresponding to elements b1, b2, and b3 of matrix B are set for the liquid crystal cells of the liquid crystal panel 31 by applying a voltage to the transparent electrode. As a result, each pixel of the image sensor 10 receives light amounts corresponding to a1b1, a2b2, and a3b3, respectively. By adding these optically or electrically, as will be described later, the product of matrix A and matrix B, i.e., a1b1 + a2b2 + a3b3, can be obtained.

[0038] (Specific example) We will explain this in detail using the product operation of matrices A and B shown in the following equation as an example.

number

[0039] In practice, the calculation is performed as follows: First, with respect to the image sensor 10, the maximum amount of light that does not saturate is set as the maximum light intensity, and the minimum amount of light that can be received is set as the minimum light intensity.

[0040] (1) The amount of light emitted from each light-emitting part of the light source unit 20, in this case the transmittance of each liquid crystal cell of the liquid crystal panel 22, is set according to matrix A. Specifically, the numerical range of matrix A is normalized using the minimum and maximum light amounts. If the numerical range of matrix A is 0 to 255, the light amounts from the minimum to the maximum light amount are associated with 0 to 255. Through this association, matrix A is transformed as follows, for example. Generally, a linear association is used, but if you want to introduce nonlinearity into the calculation, a nonlinear association may be used.

number

[0041] (2) The transmittance of each liquid crystal cell in the liquid crystal panel 31 is set according to matrix B. Specifically, the range from the minimum transmittance to the maximum transmittance is associated with the numerical range of matrix B. If the numerical range of matrix B is 0 to 255, the range from the minimum transmittance of 0 to the maximum transmittance of 1 is associated with 0 to 255. Through this association, matrix B is transformed as follows, for example. Generally, a linear association is used, but if you want to introduce nonlinearity into the calculation, you may use a nonlinear association.

number

[0042] In this case, the calculation using the optical calculator is as shown in Figure 4. In Figure 4(a), the elements of matrix A are converted to light intensity, and in Figure 4(b), the elements of matrix B are converted to transmittance. When light from the liquid crystal panel 22 passes through the liquid crystal panel 31, the image sensor 10 receives light corresponding to the value obtained by the product of the elements of matrix A and the elements of matrix B. The amount of light received at this time is as shown in Figure 4(c). If necessary, as shown in Figure 4(d), the amount of light received can be multiplied by a gain based on normalization (=255×255) to convert it into numerical data. As a result, the same result as in Figure 3 is obtained.

[0043] For liquid crystal cells, the drive voltage-transmittance characteristics should be measured in advance, and the drive voltage should be set according to the assigned transmittance. Depending on the application, the drive voltage may be determined using the same characteristic graph for the liquid crystal panel, or, in the case of high-precision driving, the drive voltage may be determined using the characteristic graph for each pixel.

[0044] Figure 5 is a block diagram showing an example of a system configuration including an optical computing unit. Figure 6 is a flowchart showing an example of processing in the system shown in Figure 5.

[0045] The drive circuit 81 applies a drive voltage to each liquid crystal cell of the liquid crystal panel 22 of the light source unit 20, controlling the amount of light emitted by the light source unit 20. The drive circuit 82 applies a drive voltage to each liquid crystal cell of the liquid crystal panel 31, controlling its transmittance. The calculation circuit 83 gives instructions to the drive circuits 81 and 82, respectively, to perform a product operation between a matrix input by the user and a matrix stored in memory 84 (in this case, the model's weight matrix). The calculation circuit 83 calculates the matrix product from the output signal of the image sensor 10 and outputs it to the user.

[0046] In step S1, the arithmetic circuit 83 gives instructions to the drive circuit 82 according to the weight matrix of the model stored in memory 84. The drive circuit 82 drives the liquid crystal panel 31 according to the instructions from the arithmetic circuit 83. In step S2, the arithmetic circuit 83 gives instructions to the drive circuit 81 according to the matrix input by the user. The drive circuit 81 drives the light source unit 20 according to the instructions from the arithmetic circuit 83.

[0047] In step S3, the image sensor 10 receives light output from the light source unit 20 and passed through the liquid crystal panel 31, and sends a signal corresponding to the amount of light received to the calculation circuit 83. In step S4, the calculation circuit 83 performs data formatting and other processing, and when the calculation converges, it outputs the data to the user. On the other hand, if the calculation does not converge, for example, the weights in the model's weight matrix are changed, and the process is executed again from step S1.

[0048] As described above, the optical computing unit comprises an image sensor 10 having pixels 11 arranged in two dimensions, a light source unit 20 having an optical output unit corresponding to each pixel 10, and a liquid crystal panel 31 provided between the light source unit 20 and the image sensor 10, having liquid crystal cells corresponding to each pixel 11. The light source unit 20 can control the amount of light from the optical output unit, and the liquid crystal panel 31 can control the transmittance of the liquid crystal cells. This allows for individual integration of the amount of light from the optical output unit and the transmittance of the liquid crystal cells for each pixel 11 of the image sensor 10. Therefore, the optical computing unit has high area efficiency. Furthermore, with this optical computing unit, it is possible to set the amount of light from the optical output unit in the light source unit 20 according to the elements of the first matrix, and set the transmittance of the liquid crystal cells in the liquid crystal panel 31 according to the elements of the second matrix, and then calculate the product of the first matrix and the second matrix based on the amount of light received by the pixels 11 in the image sensor 10.

[0049] <Separation unit for separating light> In the optical computing unit described above, individual integration is performed for each pixel of the image sensor. However, because liquid crystal cells have angular dispersion in their output, there is a risk that light from adjacent pixels may mix with each other. When light mixes, the integration results of the pixels become mixed, making it impossible to obtain highly accurate calculation results. To solve this problem, it is preferable to provide a separation unit that separates the light between pixels.

[0050] Figure 7 shows an example in which a separation section is added to the configuration of Figure 1. In the configuration of Figure 7, a light-shielding section 61 is provided at the boundary between liquid crystal cells on the light-ingress side of the liquid crystal panel 31. Also, a light-shielding section 62 is provided at the boundary between pixels 11 on the light-ingress side of the image sensor 10. The light-shielding sections 61 and 62 can be made of any material that absorbs light, for example, a black-colored resin. The light-shielding section 61 functions as a separation section to prevent light signals input to adjacent liquid crystal cells from mixing with each liquid crystal cell of the liquid crystal panel 31. The light-shielding section 62 functions as a separation section to prevent light signals input to adjacent pixels from mixing with each pixel 11 of the image sensor 10.

[0051] Figure 8 shows another example of the configuration in Figure 1 with the addition of a separation section. In the configuration of Figure 8, a low refractive index section 63 is provided at the boundary between liquid crystal cells on the light-receiving side of the liquid crystal panel 31. Also, a low refractive index section 64 is provided at the boundary between pixels 11 on the light-receiving side of the image sensor 10. The low refractive index sections 63 and 64 can be made of materials with a lower refractive index than the surrounding area, for example, a low refractive index resin or silicon dioxide with a refractive index lower than 1.4. The low refractive index section 63 functions as a separation section to prevent light signals input to adjacent liquid crystal cells from mixing with each liquid crystal cell of the liquid crystal panel 31. The low refractive index section 64 functions as a separation section to prevent light signals input to adjacent pixels from mixing with each pixel 11 of the image sensor 10.

[0052] Figure 9 shows another example of the configuration in Figure 1 with the addition of a separation unit. In the configuration of Figure 9, a lens 65 is provided on the light-receiving side of the liquid crystal panel 31, above the liquid crystal cells. Also, a lens 66 is provided on the light-receiving side of the image sensor 10, above the pixels 11. Lenses 65 and 66 have a relatively higher refractive index than the periphery and have the function of collecting light signals. Lens 65 functions as a separation unit for each liquid crystal cell of the liquid crystal panel 31, preventing light signals input to adjacent liquid crystal cells from mixing with each liquid crystal cell. Lens 66 functions as a separation unit for each pixel of the image sensor 10, preventing light signals input to adjacent pixels from mixing with each pixel.

[0053] By providing these separation sections, it is possible to prevent the calculation results from mixing between adjacent pixels.

[0054] <Addition> The optical computing unit described above can add up the integration results at each pixel. This addition can be performed optically or at the electrical signal level.

[0055] Figure 10 shows an example of a configuration for optically performing addition. In the configuration of Figure 10, a lens 71 is provided on the light-ingress side of the image sensor 10 to allow light from multiple liquid crystal cells to enter a common pixel 11. The lens 71 functions as an adder that collects the light outputs of multiple liquid crystal cells in the liquid crystal panel 31 into a single pixel 11 in the image sensor 10. With this configuration, the result of the sum-of-accumulate operation can be obtained from the output signal of the pixel 11.

[0056] Figure 11 shows an example of a configuration in which addition is performed at the electrical signal level, where (a) is a part of the adder circuit and (b) is a timing chart showing the operation of the circuit in (a). As shown in Figure 11, in multiplication mode, by setting signal H1 high, the signal of pixel 11 can be output to the output signal line S1, and by setting signal H2 high, the signal of pixel 21 can be output to the output signal line S1. In addition, in sum-of-products mode, by setting signals H1 and H2 high, a signal obtained by adding the signals of pixel 11 and pixel 21 can be output to the output signal line S1.

[0057] <Avoiding the effects of residual polarization> Furthermore, in the optical computing unit described above, if calculations are performed using the same matrix repeatedly, for example, the polarization of the liquid crystal may become fixed, which could prevent correct calculations from being performed after changing the matrix. To avoid this problem, for example, when performing calculations using the same matrix repeatedly, the operation should be performed alternately with positive and negative inversion operations to prevent the operation from becoming fixed. In the case of an operation with positive and negative inversion, the operation result can be reversed again in a subsequent processing step.

[0058] Figure 12 is an illustrative diagram of an operation method that alternates between positive and negative inversion operations. In Figure 12, the same matrix is ​​used to perform the product operation with the input vectors given at times t1, t2, and t3. In this case, the operation is performed directly at times t1 and t3. On the other hand, at time t2, the matrix elements are inverted, i.e., converted to two's complement (bit inversion and adding 1). Then, the result obtained from the image sensor is converted back to its original state (subtracting 1 and bit inversion). By performing the positive and negative inversion operations alternately in this way, the effects of residual polarization in the liquid crystal can be avoided.

[0059] Note that the sign reversal of a matrix does not necessarily have to be performed alternately; it can be done selectively. For example, the sign could be reversed every N (where N is an integer greater than or equal to 2) operations, or the sign reversal could be performed randomly.

[0060] <Suppression of ambient light> Furthermore, as shown in Figure 13, it is preferable that the above-mentioned optical computing unit is covered by a shielding structure 15. This prevents light other than that from the light source 20 from entering the image sensor 10. The shielding structure 15 is made of, for example, black anodized aluminum or black resin.

[0061] (Second Embodiment) Figure 14 shows an example of the configuration of an optical computing unit according to the second embodiment. Figure 14 schematically shows a part of the cross-sectional configuration of the optical computing unit. In Figure 14, components common to Figure 1 are denoted by the same reference numerals as in Figure 1, and their detailed explanation may be omitted here.

[0062] In the optical computing unit shown in Figure 14, the light source unit 40 includes micro-light sources 41 arranged in a two-dimensional manner. Each micro-light source 41 corresponds to a pixel 11 of the image sensor 10. The micro-light sources 41 are composed of, for example, LEDs or lasers, and the amount of light emitted can be controlled. In the light source unit 40, the micro-light sources 41 become the light output units corresponding to the pixels of the image sensor 10. On the other hand, the light source unit 20, namely the light guide plate 21, liquid crystal panel 22, and polarizing film 23, which are present in the optical computing unit of Figure 1, are omitted in the configuration of Figure 14.

[0063] The optical computing unit according to this embodiment provides the same effects and advantages as the first embodiment described above. Furthermore, by using a micro light source 41 as the optical output unit in the light source unit 40, it becomes possible to drive the light source unit 40 at high speed (for example, 100 Hz or higher). This makes it possible to switch input vectors and matrices used in calculations at high speed.

[0064] For the micro light source 41, the drive current-light intensity characteristics should be measured in advance, and the drive current should be set according to the amount of light to be allocated. Depending on the application, the drive voltage may be determined using the same characteristic graph for the entire light source unit 40, or, in the case of high-precision driving, the drive voltage may be determined using the characteristic graph for each micro light source 41.

[0065] In addition, as in the first embodiment, a separation unit for separating light may be provided, or an optical or electrical signal addition function may be provided.

[0066] (Other embodiments) In the above-described embodiment, the optical computing unit is configured to include a light source unit capable of controlling the light intensity of the optical output unit and a liquid crystal panel capable of controlling the transmittance of the liquid crystal cell. However, the optical computing unit according to this disclosure may further include a liquid crystal panel capable of controlling the transmittance of the liquid crystal cell. This allows the optical computing unit to perform multiplication operations on three or more matrices.

[0067] Figure 15 shows an example of the configuration of an optical computing unit according to another embodiment. In the configuration of Figure 15, in addition to the configuration of the first embodiment, a liquid crystal panel 51 and a polarizing film 52 are provided. The liquid crystal panel 51 and the polarizing film 52 are arranged between the polarizing film 33 and the image sensor 10. The liquid crystal panel 51, like the liquid crystal panel 31, has liquid crystal cells arranged in two dimensions corresponding to each pixel 11 of the image sensor 10, and the transmittance of the liquid crystal cells can be controlled. In the configuration of Figure 15, for example, by assigning the elements of matrix A to each liquid crystal cell of the liquid crystal panel 22, the elements of matrix B to each liquid crystal cell of the liquid crystal panel 31, and the elements of matrix C to each liquid crystal cell of the liquid crystal panel 51, the product of matrices A, B, and C can be calculated from the output of the image sensor 10. [Industrial applicability]

[0068] This disclosure enables the realization of a compact optical computing unit capable of performing optical parallel integration, and is therefore useful, for example, in reducing power consumption in AI (Artificial Intelligence) learning calculations. [Explanation of Symbols]

[0069] 10 Image sensor 11 pixels 15 Shielding structure 20 Light source section 21 Light guide plate 22 LCD panels 31 LCD panel 32 Polarizing film 33 Polarizing film 40 Light source section 41 Microlight 61, 62 Light-shielding part 63,64 Low refractive index region 65, 66 lenses 71 Lens

Claims

1. An image sensor having pixels arranged in a two-dimensional manner, A light source unit having light output units arranged in a two-dimensional manner corresponding to each of the aforementioned pixels, and capable of controlling the light intensity of the light output units, The system includes a first liquid crystal panel provided between the light source and the image sensor, having liquid crystal cells arranged in a two-dimensional manner corresponding to each pixel, and capable of controlling the transmittance of the liquid crystal cells, The configuration allows for the calculation of the product of the first matrix and the second matrix based on the amount of light received by the pixels in the image sensor, with the light intensity of the light output section in the light source section set according to the elements of the first matrix, and the transmittance of the liquid crystal cells in the first liquid crystal panel set according to the elements of the second matrix. Optical calculator.

2. In the optical computing unit according to claim 1, The aforementioned light source unit is A surface light source and The system comprises a second liquid crystal panel provided between the surface light source and the first liquid crystal panel, having liquid crystal cells arranged in a two-dimensional manner corresponding to each of the pixels, and configured to allow control of the transmittance of the liquid crystal cells. Optical calculator.

3. In the optical computing unit according to claim 1, The aforementioned light source unit is The aforementioned light output unit includes a micro light source with controllable light emission amount. Optical calculator.

4. In the optical computing unit according to claim 1, A first separation unit is provided for each liquid crystal cell of the first liquid crystal panel to prevent light signals input to adjacent liquid crystal cells from mixing with the liquid crystal cell in question. The image sensor includes a second separation unit that prevents light signals input to adjacent pixels from mixing with each pixel of the image sensor. Optical calculator.

5. In the optical computing unit according to claim 4, The first separation portion is provided at the boundary between liquid crystal cells on the light-receiving side of the first liquid crystal panel, and is a light-absorbing member. The second separation portion is provided at the boundary between pixels on the light-receiving side of the image sensor and is a light-absorbing material. Optical calculator.

6. In the optical computing unit according to claim 4, The first separation portion is provided at the boundary between liquid crystal cells on the light-receiving side of the first liquid crystal panel, and is a material with a relatively lower refractive index than the surrounding area. The second separation portion is provided at the boundary between pixels on the light-receiving side of the image sensor and is made of a material with a relatively lower refractive index than the surrounding area. Optical calculator.

7. In the optical computing unit according to claim 4, The first separation unit is a lens that is positioned on the liquid crystal cell on the light-receiving side of the first liquid crystal panel and has a relatively higher refractive index than the surrounding area. The second separation unit is a lens positioned on the pixel on the light-receiving side of the image sensor, and has a relatively higher refractive index than the surrounding area. Optical calculator.

8. In the optical computing unit according to claim 1, A first polarizing film is disposed between the light source and the first liquid crystal panel, The system comprises a second polarizing film disposed between the first liquid crystal panel and the image sensor, the second polarizing film having a polarization angle that differs by 90 degrees from that of the first polarizing film. Optical calculator.

9. In the optical computing unit according to claim 1, The image sensor is provided on the light-receiving side and includes a lens that focuses the light output of multiple liquid crystal cells in the first liquid crystal panel onto a single pixel in the image sensor. Optical calculator.

10. In the optical computing unit according to claim 1, The image sensor includes a circuit that adds up the electrical signals output from multiple pixels. Optical calculator.

11. In the optical computing unit according to claim 1, The image sensor is covered by a shielding structure to prevent light other than that from the light source from entering it. Optical calculator.

12. In the optical computing unit according to claim 1, The system further includes a second liquid crystal panel provided between the first liquid crystal panel and the image sensor, having liquid crystal cells arranged in a two-dimensional manner corresponding to each pixel, and capable of controlling the transmittance of the liquid crystal cells. Optical calculator.

13. A calculation method using the optical computing unit described in claim 1, When performing multiple operations with the second matrix fixed, The first step involves setting the light intensity of the light output section in the light source section according to the elements of the first matrix, setting the transmittance of the liquid crystal cells in the first liquid crystal panel according to the elements of the second matrix, and calculating the product of the first matrix and the second matrix from the output of the image sensor. A second step is selectively performed which involves setting the light intensity of the light output section in the light source section according to the elements of the first matrix, setting the transmittance of the liquid crystal cells in the first liquid crystal panel according to the values ​​obtained by inverting the sign of the elements of the second matrix, and calculating the product of the first matrix and the second matrix from the output of the image sensor obtained by inverting the sign of the second matrix. Calculation method.

Images