Photoprocessing apparatus, photoprocessing system, and photoprocessing method

Abstract

An optical processing device (10) of the present invention comprises: an optical branching element (11) that branches input light; an electro-optical converter (12) that is provided in one optical path for propagating one light beam of the branched light, and that modulates the one light beam by means of an input electric signal; a delay waveguide (15) and a first phase shifter (14) that is provided in another optical path for propagating another light beam of the branched light, and that adjusts the phase of the another light beam; and a photoelectric conversion unit (16) that outputs the difference between the output of the one optical path and the output of the another optical path. The length of the delay waveguide is equal to the difference in optical path length between the one optical path and the another optical path. The wavelength of the input light is set so that an output from the photoelectric conversion unit corresponding to a prescribed input electric signal substantially matches a desired output corresponding to the prescribed electric signal.　Due to this configuration, the present invention can provide an optical processing device with which wavelength dependency of input / output characteristics can be suppressed, and processing (computing) accuracy and throughput in optical processing can be improved.

Description

Technical Field

[0001] The present invention relates to an optical processing device, an optical processing system, and an optical processing method that perform optical processing (computation) with wavelength multiplexing.

Background Art

[0002] In recent years, with the progress of silicon photonics technology, mass production of on-chip optical devices with low loss and high yield has been realized. Here, on-chip optical devices mainly include passive optical devices such as waveguides, splitters, combiners, and crossings, and active optical devices such as phase modulators and photodetectors. Along with this, as large-scale optical integrated circuits and their application technologies, optical processing devices, such as optical analog calculators such as optical neural network accelerators, have been researched and developed.

[0003] An optical analog calculator has the following characteristics. (1) Analog complex number calculations can be performed with zero energy consumption by optical interference (energy saving). (2) Calculation delay can be reduced by the propagation of light (low latency). (3) Multiple calculations can be realized with a single linear circuit by division multiplexing according to time, space, wavelength, etc. (high throughput).

[0004] In an on-chip optical analog calculator, since optical interference is used, if energy for maintaining the state of a specific optical interference system is not required, that is, if a fixed phase shifter that does not allow current to flow can be used, the energy required for linear calculations such as matrix calculations can be reduced to zero. Most of the power consumption in neural network calculations is occupied by the power consumption of matrix calculations. Therefore, by opticalizing matrix calculations, the power consumption can be significantly reduced (Non-Patent Document 1).

[0005] On the other hand, in an on-chip optical calculator, the input data size is limited. For example, in order to control one phase shifter, which is an element device of an optical calculator, one electrical analog input is always required. Therefore, if the size of the matrix is M, then M 2Since a control electrode for -M is required, it is considered that the limit is about 100 elements (~9900 electrodes) that can be electrically implemented. Thus, since the number of element devices of the optical calculator is limited, the matrix size that can be calculated optically at one time is limited.

[0006] Therefore, in an on-chip optical calculator, the application to a convolutional neural network that can be used even with a small matrix size is regarded as important. In the case of a convolutional neural network, the input data can be divided and input while fixing the matrix elements to predetermined values, and the calculation can be executed.

[0007] Here, the input data can be divided into a plurality of wavelength channels and the calculation (wavelength division multiplexing calculation) can be executed. In the wavelength division multiplexing calculation, after allocating data in different regions for each wavelength and simultaneously inputting them to the optical matrix calculation circuit, the output of the optical matrix calculation circuit is demultiplexed and detected, so that the calculation results of each region can be obtained at one time. That is, the matrix calculations required for the convolutional neural network can be simultaneously executed in a single circuit. As a result, the throughput can be increased by the number of wavelength channels.

[0008] In wavelength division multiplexing calculation, in order to increase the number of wavelength channels that can be multiplexed and operate the optical calculation circuit over a wide wavelength range, it is important to suppress the wavelength dependence of the optical calculation circuit. Therefore, (1) it is necessary to suppress the wavelength dependence of the optical elements (mainly 1x2 or 2x2 couplers) used in the circuit, and (2) in a configuration where a plurality of optical waveguides are arranged in parallel in the optical circuit, the lengths (optical path lengths) of the respective optical waveguides are made equal (hereinafter referred to as "equalization"). Here, equalization is required for all paths generated by the ON / OFF combination of the switch that switches the path of the signal light in the circuit. From the above, since the relative phase difference between the paths does not change even when the wavelength changes, the wavelength independence in the optical calculation circuit is maintained.

[0009] Since calculations within the optical analog arithmetic unit are performed using complex numbers based on optical interference, the results are also complex numbers. Homodyne detection allows the complex number calculation results to be extracted as an electrical signal. As shown in Figure 15, in homodyne detection, the signal light 2 and light from the same light source are input to the 2-input 2-output optical coupler 81 as a reference light (phase reference light) 3. The two optical outputs obtained by the interference of the signal light 2 and the reference light 3 are photoelectrically converted by the photodetectors 82 and 83, respectively, and the difference between the two photoelectrically converted outputs obtained by the subtractor 84 is output on the electrical circuit side (output 7 in the figure). From this electrical output, the amplitude and phase of the output light can be obtained.

[0010] Here, it is assumed that reference light 3 has the same wavelength as signal light 2. Although it is conceivable that reference light 3 could be input from a separate light source, in an on-chip optical computing unit, the same light source as signal light 2 is branched on-chip to input as reference light 3. In this case, by making the optical path length of the optical waveguide through which reference light 3 passes the same as the optical path length of the optical path through which signal light 2 passes, wavelength independence can be achieved in the optical computing circuit. [Prior art documents] [Patent Documents]

[0011] [Patent Document 1] Patent No. 6871206 [Non-patent literature]

[0012] [Non-Patent Document 1] Y. Chen et al., “Deep learning with coherent nanophotonic circuits,” Nature Photon. 11, 441-446 (2017). [Overview of the Initiative] [Problems that the invention aims to solve]

[0013] However, in optical processing devices such as optical computing units, even if the optical waveguides are made equal in length during the design of the optical circuit, manufacturing errors occur during the optical circuit fabrication process. As a result, the optical path length of each path cannot be fabricated as designed, causing fluctuations in optical path length (manufacturing errors), which leads to wavelength dependence in the optical circuits of the optical processing device. [Means for solving the problem]

[0014] To solve the problems described above, the optical processing apparatus according to the present invention includes an optical branching element for branching input light, an electro-optic converter that modulates one of the branched light beams with an input electrical signal in one optical path through which one of the branched light beams propagates, a first phase shifter for adjusting the phase of the other light beam in the other optical path through which the other of the branched light beams propagates, and a delay waveguide, and a photoelectric conversion unit that outputs the difference between the output of one optical path and the output of the other optical path, wherein the length of the delay waveguide is the difference in optical path lengths between one optical path and the other optical path, and the wavelength of the input light corresponds to the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal. The wavelength dependence is suppressed. It will be configured to do so. Furthermore, the optical processing apparatus according to the present invention includes an optical branching element for branching input light, an electro-optic converter that modulates one of the branched light beams with an input electrical signal in one optical path through which one of the branched light beams propagates, a first phase shifter for adjusting the phase of the other light beam and a delay waveguide in the other optical path through which the other of the branched light beams propagates, a photoelectric conversion unit that outputs the difference between the output of the one optical path and the output of the other optical path, the length of the delay waveguide being the difference in optical path lengths between the one optical path and the other optical path, the wavelength of the input light being set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches the desired output corresponding to the predetermined input electrical signal, and the R2 coefficient of determination characteristic based on the difference between the output of the photoelectric conversion unit and the desired output having multiple peaks, with the wavelength intervals of the peaks being equal. Furthermore, the optical processing apparatus according to the present invention includes an optical branching element for branching input light, an electro-optic converter that modulates one of the branched light beams with an input electrical signal in one optical path through which one of the branched light beams propagates, a first phase shifter for adjusting the phase of the other light beam and a delay waveguide in the other optical path through which the other of the branched light beams propagates, a photoelectric conversion unit that outputs the difference between the output of the one optical path and the output of the other optical path, the length of the delay waveguide being the difference in optical path lengths between the one optical path and the other optical path, the wavelength of the input light being set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches the desired output corresponding to the predetermined input electrical signal, the input light having multiple wavelengths, and the interval Δλ between the multiple wavelengths being expressed by equation (A). Δλ = 2πn eff ΔL / δφ (A) Here, n eff ΔL is the effective refractive index of the waveguide, including wavelength dependence, ΔL is the optical path length difference, and δφ is the phase error of the optical path. Furthermore, the optical processing apparatus according to the present invention includes an optical branching element for branching input light, an electro-optic converter for modulating one of the branched lights with an input electrical signal in one optical path through which one of the branched lights propagates, a first phase shifter for adjusting the phase of the other light and a delay waveguide in the other optical path through which the other of the branched lights propagates, a photoelectric conversion unit for outputting the difference between the output of the one optical path and the output of the other optical path, the length of the delay waveguide being the difference in optical path lengths between the one optical path and the other optical path, the wavelength of the input light being set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches a desired output corresponding to the predetermined input electrical signal, and an optical analog arithmetic circuit being provided between the electro-optic converter and the photoelectric conversion unit in the one optical path. Furthermore, the optical processing system according to the present invention comprises a light source, an electrical signal generation unit, light output from the light source, an optical processing device to which the electrical signal output from the electrical signal generation unit is input, a measurement unit to measure the output of the optical processing device, and a wavelength selection unit to which the output of the measurement unit is input, wherein the output of the wavelength selection unit is input to the light source, the optical processing device comprises an optical branching element for branching the input light, an electro-optic converter for modulating one of the branched light beams with an input electrical signal in one optical path through which one of the branched light beams propagates, a first phase shifter for adjusting the phase of the other light beam in the other optical path through which the other of the branched light beams propagates, and a delay waveguide, and comprises a photoelectric conversion unit that outputs the difference between the output of the one optical path and the output of the other optical path, wherein the length of the delay waveguide is the difference in optical path lengths between the one optical path and the other optical path, and the wavelength of the input light is set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches the desired output corresponding to the predetermined input electrical signal.

[0015] Furthermore, the present invention relates to an optical processing method in an optical processing system comprising a light source, an electrical signal generation unit, an optical processing device, a measurement unit, and a wavelength selection unit, wherein the optical processing device comprises one optical path through which one of the input light beams propagates and another optical path through which the other light beam propagates, and the one optical path and the other optical path have a difference in optical path length, wherein the light source sweeps the wavelength of light and outputs a first light beam, the electrical signal generation unit outputs a first electrical signal, and the optical processing device outputs a first processing result according to the first light beam and the first electrical signal. The process includes the steps of: measuring the first processing result; comparing the measured first processing result with a desired processing result corresponding to the first electrical signal and selecting a wavelength that substantially matches the measured first processing result and the desired processing result; the light source outputting a second light having the selected wavelength; the electrical signal generation unit outputting a second electrical signal; and the photoprocessing device performing processing using the second light and the second electrical signal. [Effects of the Invention]

[0016] According to the present invention, it is possible to provide an optical processing apparatus, an optical processing system, and an optical processing method that can suppress the wavelength dependence of input / output characteristics and improve processing (calculation) accuracy and throughput in optical processing. [Brief explanation of the drawing]

[0017] [Figure 1] Figure 1 is a block diagram showing the configuration of an optical processing apparatus according to the first embodiment of the present invention. [Figure 2] Figure 2 is a block diagram showing an example of the configuration of an optical processing apparatus according to the first embodiment of the present invention. [Figure 3] Figure 3 is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 4A] Figure 4A is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 4B]Figure 4B is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 5] Figure 5 is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 6] Figure 6 is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 7] Figure 7 is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 8A] Figure 8A is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 8B] Figure 8B is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 9] Figure 9 is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 10A] Figure 10A is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 10B] Figure 10B is a diagram illustrating the operation of the optical apparatus according to the first embodiment of the present invention. [Figure 11] Figure 11 is a flowchart illustrating a photoprocessing method according to the first embodiment of the present invention. [Figure 12A] Figure 12A is a diagram illustrating the effects of the photoprocessing apparatus according to the first embodiment of the present invention. [Figure 12B] Figure 12B is a diagram illustrating the effects of the photoprocessing apparatus according to the first embodiment of the present invention. [Figure 13] Figure 13 is a block diagram showing the configuration of an optical apparatus according to a second embodiment of the present invention. [Figure 14] Figure 14 is a block diagram showing the configuration of an optical processing system according to a third embodiment of the present invention. [Figure 15] Figure 15 is a diagram illustrating conventional homodyne detection. [Modes for carrying out the invention]

[0018] <First Embodiment> A photoprocessing apparatus and photoprocessing method according to the first embodiment of the present invention will be described with reference to Figures 1 to 12B.

[0019] <Configuration of the optical processing device> As shown in Figure 1, the optical processing apparatus 10 according to this embodiment comprises an optical branching element 11, an electro-optical converter 12, an optical analog arithmetic circuit 13, a phase shifter 14 for reference light, a delayed waveguide 15, and a photoelectric conversion unit 16.

[0020] Two waveguides are connected to the optical branching element 11, splitting it into two optical paths. In one of the optical paths, an electro-optical converter 12 and an optical analog arithmetic circuit 13 are connected in order. Here, the electro-optical converter 12 and the optical analog arithmetic circuit 13 may be connected by an optical waveguide.

[0021] In the other optical path, a reference light phase shifter 14 and a delay waveguide 15 are connected in order. Here, the reference light phase shifter 14 and the delay waveguide 15 may be connected by an optical waveguide.

[0022] The output of the optical analog arithmetic circuit 13 and the output of the delay waveguide 15 are connected to the photoelectric conversion unit 16. Here, the output of the optical analog arithmetic circuit 13 and the photoelectric conversion unit 16 may be connected by an optical waveguide, and the output of the delay waveguide 15 and the photoelectric conversion unit 16 may be connected by an optical waveguide.

[0023] The optical splitting element 11 splits the light input from the light source 1. One of the split beams of light is input into one waveguide and propagates. The other beam of light, as reference light 3, is input into the other waveguide and propagates.

[0024] The electro-optic converter 12 modulates one of the light beams with an externally input electrical signal 4 to generate signal light 2.

[0025] The optical analog arithmetic circuit 13 performs an operation using the modulated optical signal by the electrical control signal (for optical analog arithmetic) 5.

[0026] The reference light phase shifter (first phase shifter) 14 adjusts the phase of the reference light 3 by the electrical signal 6 for the phase shifter. For example, the reference light phase shifter 14 adjusts so that the phase of the signal light 2 input to the photoelectric conversion unit 16 and the phase of the reference light 3 have a predetermined phase relationship (for example, 90°).

[0027] The delay waveguide 15 has an optical path length difference ΔL, and the optical path length (one optical path length) L through which the signal light 2 is generated and propagates Sig results in an optical path length difference ΔL with respect to the optical path length (the other optical path length) L through which the reference light 3 propagates. Ref

[0028] The optical path length (one optical path length) L through which the signal light 2 is generated and propagates Sig is the length from the output end of the optical branching element 11 to the input end of the photoelectric conversion unit 16 through the electro - optical converter 12 and the optical analog arithmetic circuit 13. Also, the optical path length (the other optical path length) L through which the reference light 3 propagates Ref is the length from the output end of the optical branching element 11 to the input end of the photoelectric conversion unit 16 through the reference light phase shifter 14 and the delay waveguide 15.

[0029] The photoelectric conversion unit 16 outputs the difference between the output of the optical analog arithmetic circuit 13 and the output of the delay waveguide 15 by homodyne detection.

[0030] Thus, the optical processing device 10 has a non - equal - length configuration having an optical path length difference ΔL between the optical path length (one optical path length) L through which the signal light 2 is generated and propagates Sig and the optical path length (the other optical path length) L through which the reference light 3 propagates Ref

[0031] ​​As an example of the optical processing apparatus 10 according to this embodiment, an electro-optic digital-to-analog converter (EO-DAC) may be used in the electro-optic converter 22, as shown in the optical processing apparatus 20 in Figure 2. The EO-DAC is a configuration that eliminates the need for a digital-to-analog conversion device on the electrical input side, and can reduce the delay of calculations.

[0032] In this embodiment, an example including an optical analog arithmetic circuit 13 is shown, but the operation of the optical processing device 10 according to this embodiment can also be performed in a configuration without the optical analog arithmetic circuit 13.

[0033] <Operation of the photonics processing device> The operation of the optical processing apparatus 10 according to this embodiment will be described with reference to Figures 3 to 8B.

[0034] Calculations were performed to verify the operation of the optical processing device 10.

[0035] Figure 3 shows a block diagram of the circuit model used in the calculation. The circuit configuration of the optical processing device 30 used in the calculation is the same as the optical processing device 10 according to the first embodiment, but with an EO-DAC used for the electro-optical converter 12 and the optical analog arithmetic circuit 13 omitted.

[0036] Figure 4A shows the detailed configuration of the non-isometric optical processing device 40_1 used in the calculations. For comparison, calculations were also performed on the circuit configuration of the isometric optical processing device 40_2 (Figure 4B).

[0037] Here, the EO-DAC22 is assumed to be 8-bit and is equipped with n (n=8) waveguide arms 220_1~n, each waveguide arm 220_1~n is equipped with a signal light phase shifter (second phase shifter) 222_1~n and a phase modulator 223_1~n. The n waveguide arms 220_1~n are connected on the input side by n Y-branch elements 221_1~n and on the output side by n Y-merging elements 224_1~n.

[0038] The Y-branching elements 221_1~n split the input light in equal proportions (branching ratio 1:1).

[0039] The signal light phase shifters 222_1~n adjust the phase of the light merged by the Y converging element so that it has a predetermined phase relationship (for example, in phase).

[0040] The Y-merging element 224_2~n merges the propagating light from two optical waveguides that are input (connected) to it in an equal ratio (merging ratio of 1:1) and outputs the result.

[0041] The calculation involved summing the interference light in each interference system of the signal light 2 and reference light 3 propagating through the circuit of the optical processing device 10 to determine the signal light 2 and reference light 3 to be input to the photoelectric conversion unit 16. The difference between the calculated signal light 2 and reference light 3 was calculated as the electrical output by homodyne detection. The photoelectric conversion unit 16 includes an optical coupler 161, photodetectors 162 and 163, and a subtractor 164.

[0042] Also, the arm length L of each signal light bit LSB ~L MSB The length was set to 8 mm, and the ΔL in the unequal length configuration was set to 2.4 mm.

[0043] Furthermore, the manufacturing error ΔL in the arm of each bit. LSB ~ΔL MSB Examples of uniformly distributed random numbers between 0 and 10 μm were selected and set to 1.54, 2.81, 4.40, 5.27, 4.57, 8.75, 5.18, and 9.44 μm, respectively. Here, LSB and MSB are the least significant bit and the most significant bit, respectively.

[0044] Furthermore, the calibration wavelength λ0 is 1550 nm. At the calibration wavelength (λ0 = 1550 nm), the signal light phase shifters 222_1~n inserted in each waveguide arm 220_1~n were adjusted to achieve the desired input / output characteristics. Here, as an example, a characteristic is used in which the output changes linearly at predetermined time intervals (steps) according to the bits input in time series.

[0045] The optical waveguide was assumed to have a structure in which a silicon nanowire core optical waveguide (width 440 nm, thickness 220 nm) is embedded in an SiO2 cladding. In this case, the effective refractive index n was determined by mode calculation using a simulator. eff We assumed a value of -1.2253λ + 4.42587, where λ is the input wavelength.

[0046] Furthermore, for both the Y-branching elements 221_1~n and the Y-merging elements 224_1~n, the branching / merging ratio was set to 1:1, and no manufacturing error was assumed.

[0047] In the phase modulators 223_1~n, it was assumed that the input electrical digital signal would provide a phase shift of -π / 2 for "0" and π / 2 for "1", and that a linear DAC response would be obtained when the input bit pattern was gradually changed from "00000000" to "11111111".

[0048] Figure 5, left and right, respectively, shows the calculation results of the EO-DAC output in isometric and non-isometric configurations as grayscale 2D maps. The horizontal axis represents the 256 input patterns using 8-bit bit combinations, and the vertical axis represents the input wavelength. The center of the vertical axis, 1.55 μm, corresponds to λ0 (dotted line).

[0049] In the isometric configuration (left diagram in the figure), the output pattern changes as the input wavelength deviates from the calibration wavelength λ0. This indicates that, due to the existence of manufacturing errors in the isometric configuration, the output becomes wavelength-dependent (no longer wavelength-independent).

[0050] On the other hand, in the non-isometric configuration (right diagram in the figure), the output changes significantly with wavelength, exhibiting a pattern resembling interference fringes. This is because the introduction of a constant value ΔL causes the relative phase of the reference light 3 with respect to the signal light 2 to change with wavelength, increasing the wavelength dependence.

[0051] Figure 6 shows a magnified view of the 1540-1541 nm wavelength range in the diagram. In this range, the isometric configuration exhibits a nearly constant output distribution (in the vertical axis direction) for the same bit pattern.

[0052] On the other hand, in a non-equalized configuration, the output distribution changes periodically (in the vertical direction) for the same bit pattern.

[0053] In the output distribution (2D map) shown in Figure 6, to quantitatively evaluate whether the optical processing device is performing linear DAC operation, R is used based on the difference between the calculated output distribution and the output distribution during desired linear operation. 2 The coefficient of determination was calculated. Figure 7 shows the calculation results. Here, when the optical processing device performs the desired linear DAC operation, R 2 This indicates approximately 1.

[0054] In the isochromatic configuration, R is present in the wavelength range of 1540-1541 nm. 2 This value changes between 0.75 and 0.85 (dotted line in the figure).

[0055] On the other hand, in an unequal configuration, R 2 It changes periodically, and periodic peaks appear, R 2 Wavelengths that are approximately 1 exist at 0.22 nm intervals (solid line in the figure). This means that even if fabrication errors occur in an unequal configuration, the input wavelength can be adjusted to R. 2 We demonstrate that by setting the value to be in the vicinity of 1, wavelength dependence can be suppressed at input wavelengths far from λ0, thereby compensating for input-output characteristics (e.g., linear characteristics).

[0056] Figures 8A and 8B show the R configuration with non-isometric elements. 2 is at its maximum or R 2 The EO-DAC input / output characteristics are plotted superimposed only at wavelengths where the coefficient of force is minimal (intermediate between maximum and maximum, not shown). Figure 8A shows the plot for the isometric configuration, and Figure 8B shows the plot for the non-isometric configuration.

[0057] In the plot for the isometric configuration (Figure 8A), the input / output characteristics on the short wavelength side are plotted in black (dark color), and the input / output characteristics on the long wavelength side are plotted in gray (light color). In the isometric configuration, the plot partially changes from dark to light color. This indicates that the input / output characteristics change depending on the wavelength. Thus, in the isometric configuration, R 2 As the decrease in [the specified value] occurs, wavelength independence decreases, resulting in output distortion.

[0058] On the other hand, in the non-equalized configuration (Figure 8B), the input / output characteristics show linearity, and R 2 Linear DAC operation is successfully achieved in multiple wavelength ranges where the value is maximized (solid line in the figure).

[0059] Also, R 2 In the wavelength range where the coefficient of change is at its minimum, the input-output characteristics are inverted (dotted line in the figure). This allows for the addition of an inversion function by selecting the wavelength.

[0060] Thus, in an optical processing apparatus where the optical path length includes manufacturing errors, the non-isometric configuration provides a wider wavelength range than the isometric configuration. 2 A wavelength of ~1 can be secured. Therefore, in wavelength division multiplexing (processing) by the optical processing device, the input wavelength is set discretely, as shown in Figure 7. 2 By setting the input wavelength to the wavelength of the discrete peak in the coefficient of determination characteristics, the optical processing device can be used to obtain R over a wide wavelength range. 2 Wavelength division multiplexing (processing) can be performed so that the result is ~1, that is, so that the desired output is obtained.

[0061] In this case, if the wavelength stability of the light source or the temperature stability of the chip including the optical processing unit is low, and ΔL is too long, the output may become unstable. Therefore, it is necessary to set ΔL according to the wavelength channel used for wavelength division multiplexing (processing).

[0062] For example, by reducing ΔL, R shown in Figure 7 2This allows for a more gradual change in the coefficient of determination characteristic near the peak, expanding the acceptable range for setting the input wavelength. As an example of setting ΔL, when the wavelength fluctuation of the light source is 10 pm and the temperature is ±10°C, ΔL = approximately 1 mm.

[0063] Next, the operating principle of the optical processing apparatus 10 according to this embodiment will be explained with reference to Figures 9 to 10B.

[0064] As described above, the optical processing apparatus 10 allows input / output characteristics (for example, R) caused by fabrication errors to be corrected by selecting (tuning) the input wavelength. 2 The effect of the coefficient of determination can be compensated for. The reason (principle) for this is that the relative phase difference Δφ between the reference light 3 and each signal light 2 can be changed depending on the input wavelength.

[0065] Figure 9 is a conceptual diagram of the optical processing device 10 circuit. In the optical processing device 10 circuit, multiple local interference systems 52_1~n are arranged in a mesh pattern as the interference system 50 (the figure shows the case where n=14 for input light 51_1~8). The output of the interference system 50 and the reference light 3 are input to the photoelectric conversion unit 56.

[0066] In optical processing devices that include local interference systems, the optical waveguide has phase errors δφ1, δφ2, ...δφ due to fabrication errors. n It has. Since the manufacturing error is distributed non-uniformly, δφ1, δφ2, ...δφ n They are scattered.

[0067] Figures 10A and 10B show the phase error (δφ) in the isometric and non-isometric configurations, respectively. n , δφ n This shows the dependence of the input wavelength change Δλ on '). In Figure 10B, the solid and dotted lines in the figure represent δφ when ΔL is short and long, respectively. n This shows the dependence of ' on the input wavelength change Δλ. Also, the black and white circles in the figure represent the R values ​​when ΔL is short and long, respectively. 2 This shows the maximum value of the coefficient of determination.

[0068] As shown in Figure 10A, in the isometric configuration (ΔL=0), δφn This value is constant and does not depend on Δλ. Thus, in an isometric configuration, the input / output characteristics do not change even if the wavelength is changed.

[0069] On the other hand, as shown in Figure 10B, in the unequal length configuration, δφ n ' changes with respect to Δλ. Also, as ΔL increases, δφ with respect to Δλ changes. n The amount of change in ' increases, R 2 The wavelength interval between the maximum values ​​of the coefficient of determination decreases. Thus, increasing ΔL expands the number of maximum values ​​(peaks), i.e., the selectable wavelength range for optical processing (calculation). On the other hand, increasing ΔL increases wavelength fluctuations.

[0070] Furthermore, as mentioned above, reducing ΔL allows for a more gradual change near the maximum value (peak), expanding the acceptable range for setting the input wavelength.

[0071] Therefore, it is desirable that ΔL be between 100 μm and 10 mm.

[0072] Here, δφ n The linear change of ' is expressed by equations (1) and (2).

[0073] δφ n ' = δφ n + Δφ (1)

[0074] Δφ = 2πn eff ΔL / Δλ (2)

[0075] Here, n eff This is the effective refractive index of the waveguide, including its wavelength dependence.

[0076] In this way, by appropriately setting Δλ, the relative phase difference Δφ is changed, and δφ n '=0. That is, input / output characteristics can be compensated. For example, depending on the set phase of the interference system, δφ can be set in order from the point of high light concentration in the optical processing device circuit. n Set Δλ so that ' is approximately zero. This will result in R2 The coefficient of determination can be improved, meaning that the input / output characteristics can be made to the desired characteristics or characteristics close to the desired ones.

[0077] Δλ is given by δφ from equations (1) and (2). n It is represented by equation (3), which is derived by setting '=0.

[0078] Δλ = 2πn eff ΔL / δφ n (3)

[0079] Here, δφ is the phase error in the waveguide arm, for example, the phase errors δφ1, δφ2, ...δφ in each waveguide arm 220_1~n of the EO-DAC. n This is the average value of the values.

[0080] Furthermore, when a standalone EO-DAC is used for the electro-optic converter 22, the phase error δφ in the optical waveguide corresponding to the MSB occurs. MSB When ' becomes a zero-neighbor, R 2 The coefficient of determination is improved the most.

[0081] Here, δφ1, δφ2, ...δφ at each of the areas with high light concentration. n When the value of is distributed over a wide range, R can be calculated simultaneously by a single ΔL. 2 Unable to improve the coefficient of determination, R 2 The improvement effect on the coefficient of determination decreases. For example, in this case, R 2 When =1 is not obtained, the number of peaks showing extreme values ​​decreases, and the wavelength range in which the photoprocessing device can operate while suppressing wavelength dependence shrinks.

[0082] On the other hand, δφ1, δφ2, ...δφ at each of the areas with high light concentration n If the values ​​of are distributed within a narrow range, then R can be calculated simultaneously by a single ΔL. 2 The coefficient of determination can be improved.

[0083] As an example, as shown in the calculation results above (Figure 8B), if the fabrication error in each waveguide is within 10 μm, a good R can be achieved with a single ΔL (ΔL = 2.4 mm for a waveguide length of 8 mm). 2 The coefficient of determination characteristic can be obtained.

[0084] Thus, δφ n Depending on the distribution of light and the distribution of light concentration, R by ΔL 2 The coefficient of determination, i.e., the improvement effect on input-output characteristics, varies.

[0085] <Light treatment method> The photoprocessing method according to this embodiment will be described with reference to Figure 11. Figure 11 shows a flowchart illustrating the photoprocessing method according to this embodiment.

[0086] First, the light source sweeps the wavelength of light and outputs light (first light) (step S1).

[0087] Next, a predetermined electrical signal (the first electrical signal) is output (step S2).

[0088] Next, in response to the first light and the first electrical signal, the input / output characteristics are output as the processing (calculation) result (first processing result) (step S3).

[0089] Next, the first processing result (input / output characteristics) is measured (step S4).

[0090] Next, the first processing result (input / output characteristics) is compared with the desired processing result (input / output characteristics) corresponding to the first electrical signal, and a wavelength is selected in which the first processing result and the desired processing result substantially coincide (step S5). Here, "subjectively coincide" includes "match," and means that the difference between the measured output and the desired output is within a predetermined range, and it is sufficient if it is within a range in which wavelength division multiplexing (processing) can be performed well. Furthermore, the desired processing result (input / output characteristics) may be, for example, a characteristic in which the output changes linearly at predetermined time intervals (steps) according to bits input in time series.

[0091] Next, light having the selected wavelength (second light) is output (step S6).

[0092] Next, a second electrical signal is output (step S7).

[0093] Finally, processing (calculations) are performed according to the second light and second electrical signals, and the input / output characteristics are output as the processing result (second processing result) (step S8).

[0094] Here, in comparing the measured first processing result with the desired processing result, R 2 The coefficient of determination may also be used.

[0095] In the photoprocessing method according to this embodiment, an example is shown in which the wavelength of the input light is selected using processing results obtained by measurement (experiment), but the method is not limited to this. The wavelength may also be selected in advance based on the calculations described above. In this case, the manufacturing error is set by determining the upper limit of the manufacturing error based on the manufacturing error of a normal manufacturing process (for example, 10 μm) and using a uniform random number. Alternatively, the manufacturing error may be set based on the experimentally obtained distribution of manufacturing errors. Or, the wavelength may be selected by calculating Δλ using equation (3).

[0096] <Effects> The effects of the photoprocessing apparatus according to this embodiment will be explained with reference to the experimental results shown in Figures 12A and 12B.

[0097] Figure 12A shows the wavelength dependence of the output of the optical processing apparatus with an isometric configuration (ΔL=0). A silicon photonic EO-DAC was used as the electro-optic transducer of the optical processing apparatus. The silicon photonic EO-DAC is L LSB , L7, ...L2, L MSB Each was fabricated with a length of 8 mm. A 4-bit sawtooth wave was generated using a silicon photonics EO-DAC, and the output of the photoprocessing device was measured at wavelengths of 1550 nm, 1536 nm, and 1566 nm. Here, the calibration wavelength was 1550 nm. The symbol rate of the 4-bit sawtooth wave was 20 MS / s.

[0098] In the output of the optical processing device, the waveforms at 1536 nm and 1566 nm differ from the waveform at the calibration wavelength (1550 nm) at 0.2–0.4 μs and 0.8–1.0 μs, respectively. This is due to differences in optical path length caused by errors in the waveguide fabrication of the interferometry system that generates the bits corresponding to 0.2–0.4 μs and 0.8–1.0 μs, respectively.

[0099] Thus, in an isometric configuration, the output (characteristics) of the optical processing device depends on the wavelength.

[0100] Figure 12B shows the wavelength dependence of the output of the non-isochromatic photoprocessing apparatus. A silicon photonic EO-DAC was used as the electro-optic converter in the photoprocessing apparatus. The silicon photonic EO-DAC is L LSB , L7, ...L2, L MSB Each was fabricated with dimensions of 8 mm and ΔL = 2.4 mm. A 6-bit sawtooth wave was generated using a silicon photonics EO-DAC, and the output of the photoprocessing device was measured at wavelengths of 1550 nm, 1536 nm, and 1566 nm. The calibration wavelength was 1550 nm. The symbol rate of the 6-bit sawtooth wave was 5.3 GS / s.

[0101] In the output of the optical processing device, the waveforms at 1536nm and 1566nm nearly coincide with the waveform at the calibration wavelength (1550nm). Here, the spike-like waveform near 9ns in the measured waveform is a device-specific characteristic of the EO-DAC, and is caused by the overlap of rising and falling edges at points in the EO-DAC where there is a lot of bit switching.

[0102] Thus, in a non-isometric configuration, the output (characteristics) of the photoprocessing device are consistent at different wavelengths. In other words, wavelength dependence is suppressed.

[0103] According to the optical processing apparatus of this embodiment, the wavelength dependence of input / output characteristics can be suppressed, and the processing (calculation) accuracy and throughput in optical processing can be improved.

[0104] <Second Embodiment> A second embodiment of the present invention, an optical processing apparatus, will be described with reference to Figure 13.

[0105] <Configuration of the optical processing device> The optical processing apparatus 60 according to this embodiment includes an optical branching element 61, an electro-optical converter array 62, an M×N optical analog arithmetic circuit 63, a phase shifter array 64 for reference light, a delayed waveguide 65, and a photoelectric conversion unit 66.

[0106] M+N waveguides are connected to the optical branching element 61, and the optical signal is branched into an optical path connected to the M waveguides (one optical path) and an optical path connected to the N waveguides (the other optical path).

[0107] One optical path is connected in order to the electro-optical converter array 62 and the M×N optical analog arithmetic circuit 63.

[0108] A phase shifter array 64 for reference light and a delayed waveguide 65 are connected in order to the other optical path, and the reference light 3 propagates through it.

[0109] The output of the M×N optical analog arithmetic circuit 63 and the output of the delayed waveguide 65 are connected to the photoelectric conversion unit 66.

[0110] The electro-optic transducer array 62 is composed of M electro-optic transducers, and generates signal light 2 by modulating the light input from each of the M waveguides with an externally input electrical signal.

[0111] The M×N optical analog calculation circuit 63 performs calculations on M modulated light signals using electrical control and outputs N light signals.

[0112] The reference light phase shifter array 64 is composed of N phase shifters to match the N outputs of the M×N optical analog arithmetic circuit, and adjusts the phase of the reference light 3 input from each of the N waveguides.

[0113] The delayed waveguide 65 is composed of N optical waveguides, each of which has an optical path length difference ΔL, and the optical path length L over which the signal light 2 is generated and propagates is such that Sig In contrast, the optical path length L of the reference light 3 Ref This results in an optical path length difference ΔL.

[0114] The photoelectric conversion unit 66 outputs the difference between the outputs of N optical analog arithmetic circuits and the outputs of N delayed waveguides by homodyne detection.

[0115] Thus, the optical processing device 60 generates and propagates signal light 2 through an optical path length L Sig The optical path length L through which the reference light 3 propagates. Ref It has an unequal configuration with an optical path length difference ΔL.

[0116] In the optical processing apparatus according to this embodiment, optical processing (calculation) is performed based on the optical processing method according to the first embodiment.

[0117] According to the optical processing apparatus of this embodiment, the wavelength dependence of input / output characteristics can be suppressed, and the processing (calculation) accuracy and throughput in optical processing can be improved. Furthermore, it is possible to perform N times more optical processing (calculations) compared to the first embodiment.

[0118] In this embodiment, an example is shown using an optical waveguide through which N reference beams propagate, a phase shifter array for reference beams, and a delayed waveguide consisting of N optical waveguides, but the invention is not limited to this. If it is not necessary to acquire N outputs simultaneously in the optical processing device, a 1:N optical switch can be placed after a configuration consisting of an optical waveguide through which one reference beam propagates, one phase shifter for reference beams, and a delayed waveguide consisting of one optical waveguide, to output a reference beam, and by interfering each of the N signal beams with one reference beam, the outputs of N homodyne detections can be acquired serially.

[0119] <Third Embodiment> A third embodiment of the present invention, an optical processing system, will be described with reference to Figure 14.

[0120] <Configuration of the photoprocessing system> The optical processing system 70 according to this embodiment comprises a light source 71, an electrical signal generation unit 72, an optical processing device 73, a measurement unit 74, a wavelength selection unit 75, and a storage unit 76. Here, the optical processing device 73 is the optical processing device according to the first embodiment.

[0121] The light source 71 is a multi-wavelength light source that, before performing calculations, sweeps the wavelength and outputs light (first light). Furthermore, during the execution of calculations, it outputs light (second light) of the wavelength selected by the wavelength selection unit 75.

[0122] The electrical signal generation unit 72 outputs a predetermined electrical signal (the first electrical signal) before executing the calculation. It also outputs an electrical signal (the second electrical signal) when the calculation is being executed.

[0123] Before performing the calculation, the optical processing device 73 receives a first light and a first electrical signal, and outputs the processing (calculation) result (first processing result) as input / output characteristics. Furthermore, when the calculation is performed, a second light is received from the light source 71 and a second electrical signal is received from the electrical signal generation unit 72, and the processing (calculation) result (second processing result) is output as input / output characteristics.

[0124] The measurement unit 74 measures the first processing result (input / output characteristics) from the optical processing device 73 before performing calculations.

[0125] The memory unit 76 stores the desired input / output characteristics (processing results) corresponding to the first electrical signal.

[0126] Before performing the calculation, the wavelength selection unit 75 reads the desired input / output characteristics (processing results) corresponding to the first electrical signal from the storage unit 76, compares the first processing result with the desired processing result, and selects a wavelength in which the first processing result and the desired processing result substantially match. The selected wavelength is output (feedback) to the light source 71.

[0127] In the optical processing system according to this embodiment, optical processing (calculation) is performed based on the optical processing method according to the first embodiment.

[0128] According to the optical processing system of this embodiment, the wavelength dependence of input / output characteristics can be suppressed, and the processing (calculation) accuracy and throughput in optical processing can be improved.

[0129] The optical processing apparatus according to the embodiment of the present invention may be fabricated on the same substrate (for example, a Si substrate or an SOI substrate) to constitute an on-chip optical processing apparatus.

[0130] In the embodiments of the present invention, an example is shown in which the desired input / output characteristic is one in which the output changes linearly at predetermined time intervals (steps) according to the bits input in a time series, but the invention is not limited to this. Other input / output characteristics, such as those in which the output changes quadratically, may also be used.

[0131] In the embodiments of the present invention, examples of the structure, dimensions, materials, etc. of each component in the optical processing apparatus, optical processing system configuration, optical processing method, etc., are shown, but the invention is not limited to these examples. Any configuration that allows the optical processing apparatus to perform its function and produce the desired effect is acceptable. [Industrial applicability]

[0132] The present invention relates to an optical processing apparatus, an optical processing system, and an optical processing method that perform optical processing (calculations) using wavelength division multiplexing, and can be applied to optical networks, optical computers, and the like. [Explanation of Symbols]

[0133] 10. Photoprocessing device 11. Optical branching element 12 Electro-optical converter 14. First phase shifter (phase shifter for reference light) 15 Delayed waveguide 16 Photoelectric conversion unit

Claims

1. An optical branching element that splits the input light, One of the branched light beams propagates through one of the optical paths, and an electro-optic converter is provided in which the light beam is modulated by an input electrical signal. The other optical path through which the other of the branched light propagates is provided with a first phase shifter for adjusting the phase of the other light and a delayed waveguide. The system includes a photoelectric conversion unit that outputs the difference between the output of one optical path and the output of the other optical path, The length of the delay waveguide is the difference in optical path lengths between the one optical path and the other optical path. A photoprocessing apparatus in which the wavelength of the input light is set such that the wavelength dependence of the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal is suppressed.

2. An optical branching element for branching input light, One of the branched light beams propagates through one of the optical paths, and an electro-optic converter is provided in which the light beam is modulated by an input electrical signal. The other optical path through which the other of the branched light propagates is provided with a first phase shifter for adjusting the phase of the other light and a delayed waveguide. The system includes a photoelectric conversion unit that outputs the difference between the output of one optical path and the output of the other optical path, The length of the delay waveguide is the difference in optical path lengths between the one optical path and the other optical path. The wavelength of the input light is set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches a desired output corresponding to the predetermined input electrical signal. A photoprocessing apparatus in which the R2 coefficient of determination characteristic based on the difference between the output of the photoelectric conversion unit and the desired output has multiple peaks, and the wavelength intervals of the peaks are equal.

3. An optical branching element for branching input light, One of the branched light beams propagates through one of the optical paths, and an electro-optic converter is provided in which the light beam is modulated by an input electrical signal. The other optical path through which the other of the branched light propagates is provided with a first phase shifter for adjusting the phase of the other light and a delayed waveguide. The system includes a photoelectric conversion unit that outputs the difference between the output of one optical path and the output of the other optical path, The length of the delay waveguide is the difference in optical path lengths between the one optical path and the other optical path. The wavelength of the input light is set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches a desired output corresponding to the predetermined input electrical signal. A photoprocessing apparatus in which the input light has multiple wavelengths, and the interval Δλ between the multiple wavelengths is represented by equation (A). Dl = 2πn ｅｆｆ ΔL / dφ (A) Here, n ｅｆｆ ΔL is the effective refractive index of the waveguide, including wavelength dependence, ΔL is the optical path length difference, and δφ is the phase error of the optical path.

4. An optical branching element for branching input light, One of the branched light beams propagates through one of the optical paths, and an electro-optic converter is provided in which the light beam is modulated by an input electrical signal. The other optical path through which the other of the branched light propagates is provided with a first phase shifter for adjusting the phase of the other light and a delayed waveguide. The system includes a photoelectric conversion unit that outputs the difference between the output of one optical path and the output of the other optical path, The length of the delay waveguide is the difference in optical path lengths between the one optical path and the other optical path. The wavelength of the input light is set such that the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal substantially matches a desired output corresponding to the predetermined input electrical signal. An optical processing apparatus comprising an optical analog arithmetic circuit between the electro-optic converter and the photoelectric conversion unit in one of the aforementioned optical paths.

5. The aforementioned optical path has M waveguides, The optical analog arithmetic circuit is an M×N optical analog arithmetic circuit, The other optical path has N waveguides, The N outputs of the M×N optical analog arithmetic circuit are connected to the photoelectric conversion unit. The aforementioned delay waveguide consists of N waveguides. The photopharmaceutical apparatus according to claim 4.

6. The aforementioned electro-optic converter n Y-branch elements connected in cascades, n Y-junction elements connected in cascades, n waveguide arms connecting one output of each of the n Y-branch elements to one input of each of the n Y-junction elements Equipped with, The waveguide arm, starting from the Y-branch element side, The second phase shifter, Phase modulator and The photocosmetic apparatus according to claim 1 or claim 2, comprising:

7. Light source and Electrical signal generation unit, A photoprocessing device to which light output from the light source and an electrical signal output from the electrical signal generation unit are input, A measuring unit for measuring the output of the aforementioned photoprocessing device, The system comprises a wavelength selection unit to which the output of the measurement unit is input, The output of the wavelength selection unit is input to the light source. The aforementioned photoprocessing apparatus is An optical branching element that splits the input light, One of the branched light beams propagates through one of the optical paths, and an electro-optic converter is provided in which the light beam is modulated by an input electrical signal. The other optical path through which the other of the branched light propagates is provided with a first phase shifter for adjusting the phase of the other light and a delayed waveguide. The system includes a photoelectric conversion unit that outputs the difference between the output of one optical path and the output of the other optical path, The length of the delay waveguide is the difference in optical path lengths between the one optical path and the other optical path. An optical processing system in which the wavelength of the input light is set to substantially coincide with the output of the photoelectric conversion unit corresponding to a predetermined input electrical signal and a desired output corresponding to the predetermined input electrical signal.

8. An optical processing method in an optical processing system comprising a light source, an electrical signal generation unit, an optical processing device, a measurement unit, and a wavelength selection unit, wherein the optical processing device comprises one optical path through which one of the input light beams propagates and another optical path through which the other light beam propagates, and the one optical path and the other optical path have a difference in optical path length, The light source provides a first light by sweeping the wavelength of light, The electrical signal generation unit outputs a first electrical signal, The optical processing apparatus outputs a first processing result in accordance with the first light and the first electrical signal, The measurement unit performs the step of measuring the first processing result, The wavelength selection unit compares the measured first processing result with a desired processing result corresponding to the first electrical signal and selects a wavelength in which the measured first processing result and the desired processing result substantially coincide. The light source outputs a second light having the selected wavelength, The steps include: the electrical signal generation unit outputs a second electrical signal; The optical processing apparatus performs processing using the second light and the second electrical signal. A light processing method comprising the following features.

Images