Optical computing chips, systems and data processing technologies

Abstract

An optical computing chip, system and data processing technology. The optical computing chip includes a light source array, a first concave mirror and a modulator array. The light source array is located on the object plane focal plane of the first concave mirror. The modulator array is located on the image plane focal plane of the first concave mirror. The light source array is used to generate a first light signal according to first data. The first concave mirror is used to output a first reflected light signal according to the first light signal. The modulator array is used to receive the first reflected light signal, obtain first spectrum surface distribution data according to the first reflected light signal, and modulate the first spectrum surface distribution data on the modulator array.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an optical computing chip, system and data processing technology. Background Art

[0002] Due to the increase in the amount of Internet data and the rapid development of artificial intelligence (AI), deep learning (DL) has been widely used in image recognition, speech recognition, natural language processing and other fields. Deep learning is a neural network that imitates the structure of the human brain and can achieve better recognition results than traditional shallow learning methods. Due to the complexity of deep learning algorithms and the huge amount of calculations, traditional central processing units are inefficient in processing large-scale operations, and hardware research for AI acceleration has gradually become a research hotspot.

[0003] Compared with traditional microelectronic chips, the performance of optical computing in certain applications has been greatly improved. For example, when optical computing involves convolution calculations of neural networks, the calculation speed is greatly improved. Analog optical computing is a type of optical computing. Analog optical computing uses the physical properties of the optical device itself to complete the calculation of the corresponding mathematical process. At present, analog optical computing is mainly based on the classic 4F system. Relying on the Fourier transform effect of the lens, two modulators are used to process the input data in the spatial frequency domain to complete the corresponding calculation. In the existing 4F system, since the convex lens used is a three-dimensional device, it cannot be integrated on the chip. In addition, in the existing 4F system, additional computing equipment is required to first calculate the spectrum data of the data according to the Fourier transform, and then modulate the spectrum data on the modulator, and the implementation process is relatively complicated. Summary of the invention

[0004] The present application provides an optical computing chip, system and data processing technology, which can realize optical computing on the chip and improve data computing efficiency.

[0005] In a first aspect, an embodiment of the present invention provides an optical computing chip, comprising a first concave mirror, a light source array, and a modulator array. The light source array is located on the object focal plane of the first concave mirror, and the modulator array is located on the image focal plane of the first concave mirror. The light source array is used to generate a first optical signal according to first data. The first concave mirror is used to output a first reflected optical signal according to the first optical signal. The modulator array is used to receive the first reflected optical signal, obtain first spectrum surface distribution data according to the first reflected optical signal, and modulate the first spectrum surface distribution data on the modulator array.

[0006] The optical computing chip described in the embodiment of the present invention uses a concave mirror, which is a one-dimensional device and is easier to manufacture and integrate on the chip, making it possible to implement optical computing on the chip. In addition, since the modulator in the optical computing chip can generate current according to the intensity of the incident light, the first spectrum surface distribution data of the first reflected light signal can be obtained. Therefore, in the process of implementing optical computing, the spectrum surface distribution data can be directly obtained, and the obtained spectrum surface distribution data can be modulated on the modulator array. This method does not require the use of additional computing equipment to assist in the acquisition of spectrum surface data. Furthermore, the computing efficiency is improved in the process of optical computing, and the implementation is simple and efficient.

[0007] In combination with the first aspect, in a first possible implementation, the light source array is further used to generate a second light signal according to second data. The first concave mirror is further used to output a second reflected light signal according to the second light signal. The modulator array is further used to obtain a third light signal according to the second reflected light signal and the first spectrum surface distribution data.

[0008] In another possible implementation, the optical computing chip further includes a second concave mirror and a detector array. The modulator array is also located on the object focal plane of the second concave mirror, and the detector array is located on the image focal plane of the second concave mirror. The second concave mirror is used to receive the third optical signal and output a third reflected optical signal according to the third optical signal. The detector array is used to detect the third reflected optical signal, and the distribution of the third reflected optical signal on the detector array is used to indicate the convolution result of the first data and the second data.

[0009] The optical computing chip described in the embodiment of the present invention uses a modulator array that can directly obtain optical spectrum surface distribution data based on the reflected light signal and can modulate the first data on the modulator array. Therefore, in the process of implementing the convolution calculation of the first data and the second data, there is no need to use additional computing equipment to assist in completing the acquisition of the spectrum surface data of the first data, thereby improving the computing efficiency.

[0010] In yet another possible implementation, the modulator array includes a plurality of modulators, and a transmittance of each modulator to the first reflected light signal is used to indicate a value in the first frequency spectrum distribution data.

[0011] In yet another possible implementation, the modulator is implemented by at least one of the following devices: a doped silicon waveguide, an electro-absorption modulator, and a semiconductor optical amplifier SOA.

[0012] In another possible implementation, the light source array includes a plurality of light emitting elements, each of which is used to generate incoherent light. Since the light emitted by the light source array used in the embodiment of the present invention is incoherent light, the function of data modulation can be taken into account at the same time, so compared with the existing spatial optical computing system, the IO speed of the optical computing chip is greatly improved.

[0013] In another possible implementation, the light source array and the detector array are located on the same side of the chip. In this implementation, the structure of the optical computing chip can be made more compact and the chip size can be reduced.

[0014] In yet another possible implementation, the first concave mirror and the second concave mirror are parabolic concave mirrors.

[0015] In another possible implementation, the light source array includes a plurality of stacked light source sub-arrays, the modulator array includes a plurality of stacked modulator sub-arrays, and the detector array includes a plurality of stacked detector sub-arrays. In this implementation, convolution calculations on multiple rows and columns of data can be performed simultaneously.

[0016] In a second aspect, the present application provides an optical computing system, the optical computing system comprising a processor and the optical computing chip in the first aspect or any possible implementation of the first aspect. The processor is used to input first data to the optical computing chip.

[0017] In a possible implementation, the optical computing system further includes a light source array driving circuit and a modulator array driving circuit. The light source array driving circuit is respectively connected to the processor and the light source array of the optical computing chip, and is used to apply a first driving signal to the light source array according to the first data. The modulator array driving circuit is connected to the modulator array, and the modulator array driving circuit is used to sample the first spectrum surface distribution data obtained by the optical computing chip, and apply a first modulation signal to the optical computing chip according to the first spectrum surface distribution data. In this case, the light source array is specifically used to generate the first light signal according to the first driving signal. The modulator array is specifically used to modulate the first spectrum surface distribution data on the modulator array according to the first modulation signal.

[0018] In one possible implementation, the optical computing system further includes a detector array driving circuit. The detector array driving circuit is connected to the detector array of the optical computing chip. The detector array driving circuit is used to collect the third reflected light signal detected by the detector array, and perform analog-to-digital conversion on the third reflected light signal to obtain a convolution result of the first data and the second data.

[0019] In a third aspect, the present application further provides a data processing method performed by the optical computing chip in the first aspect or any one of the implementations of the first aspect. According to the method, after the light source array in the optical computing chip generates a first light signal according to the first data, the first concave mirror in the optical computing chip outputs a first reflected light signal according to the first light signal, the modulator array in the optical computing chip obtains first spectrum surface distribution data according to the first reflected light signal, and modulates the first spectrum surface distribution data on the modulator array.

[0020] In one possible implementation, the light source array may also generate a second light signal according to the second data. After the first concave mirror outputs a second reflected light signal according to the second light signal, the modulator array obtains a third light signal according to the second reflected light signal and the first spectrum surface distribution data. The second concave mirror in the optical computing chip outputs a third reflected light signal according to the third light signal. The detector array in the optical computing chip may detect the third reflected light signal, and the distribution of the third reflected light signal on the detector array is used to indicate the convolution result of the first data and the second data.

[0021] In a fourth aspect, the present application further provides a computer program product, comprising a program code, wherein the instructions included in the program code are executed by a computer to implement the data processing method described in the third aspect and any possible implementation manner of the third aspect.

[0022] In a fifth aspect, the present application also provides a computer-readable storage medium, which is used to store program code, and the instructions included in the program code are executed by a computer to implement the data processing method described in the third aspect and any possible implementation method of the third aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for describing the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention.

[0024] Figure 1 A schematic diagram of implementing convolution calculation using a 4F optical computing system provided by an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of the structure of an optical computing chip provided by an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of the structure of a light source array provided by an embodiment of the present invention;

[0027] Figure 4 A schematic diagram of the optical path of a spherical concave mirror provided in an embodiment of the present invention;

[0028] Figure 5 A flow chart of a method for implementing convolution calculation of data using an optical computing chip provided in an embodiment of the present invention;

[0029] Figure 6 A schematic diagram of the structure of another optical computing chip provided by an embodiment of the present invention;

[0030] Figure 7 A schematic diagram of the structure of another optical computing chip provided by an embodiment of the present invention;

[0031] Figure 8 A schematic diagram of the structure of an optical computing system provided by an embodiment of the present invention;

[0032] Fig. 9 A schematic diagram of the structure of a light source array driving circuit provided by an embodiment of the present invention;

[0033] Fig.10 A schematic structural diagram of a modulator array driving circuit provided by an embodiment of the present invention;

[0034] Fig.11 A schematic diagram of the structure of a detector array driving circuit provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the technical personnel in the field of the present invention better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments.

[0036] Artificial neural network (ANN), also known as neural network (NN) or quasi-neural network, is a mathematical model or computational model that imitates the structure and function of biological neural network (central nervous system of animals, especially the brain) in the field of machine learning and cognitive science, and is used to estimate or approximate functions. Artificial neural networks can include convolutional neural network (CNN), deep neural network (DNN), multilayer perceptron (MLP) and other neural networks. The algorithm of neural network system is complex and the amount of calculation is huge, which puts high demands on the calculation efficiency of data. In order to improve the calculation efficiency, optical calculation that uses the physical properties of optical devices to complete the corresponding mathematical operation process has been applied.

[0037] Below we first Figure 1 Taking the 4F optical computing system 100 shown in the figure as an example, the process of using the optical computing system to implement the convolution calculation in the neural network system is briefly introduced. For the sake of simplicity, in the embodiment of the present invention, the 4F optical computing system can also be referred to as a 4F system. The 4F system refers to a system composed of two convex lenses, two optical modulators and a detector. Figure 1 As shown, the first modulator 102 is located at the object plane focal position of the first convex lens 104. The second modulator 106 is located at the image plane focal position of the first convex lens 104, and is located at the object plane focal position of the second convex lens 108. The interval between the first convex lens 104 and the second convex lens 108 is the sum of the focal lengths of the two convex lenses (104 and 108). The detector 110 is located at the image plane focal position of the second convex lens 108, and the length of the entire system is 4 times the focal length. Figure 1 The 4F system shown in the figure performs convolution calculation in the neural network system on data A and data B as an example. Figure 1 The principle of optical computing realized by the 4F system shown is introduced.

[0038] like Figure 1As shown, in the process of performing the convolution operation of data A and data B, data A needs to be modulated on the first modulator 102 first, and the Fourier spectrum of data B needs to be modulated on the second modulator 106. After the laser emitted by the laser 101 passes through the first modulator 102, the optical signal passing through the first modulator 102 is the optical signal generated according to data A. When the optical signal is Fourier transformed by the first convex lens 104, an image in the spatial frequency domain will be presented on the second modulator 106, thereby completing the multiplication operation of data A and the convolution data B previously modulated on the second modulator 106. The signal output by the second modulator 106 is then Fourier transformed by the second convex lens 108, and finally the photodetector 110 detects the light intensity of the optical signal output by the second convex lens 108 to obtain the convolution result C of data A and data B. From Figure 1 It can be seen from the process of implementing the convolution calculation of the 4F system shown that, except for the additional time required to modulate the data B to the second modulator 106, the other calculation processes do not consume time, so the calculation speed is extremely fast. In addition, since the spatial light modulator is a two-dimensional device, the concurrency of the system IO is high, the number of components in the entire system is small, and the result is relatively simple.

[0039] However, in Figure 1 In the 4F system shown, the convex lens used is a three-dimensional device and cannot be integrated on a two-dimensional plane. Therefore, Figure 1 The 4F system shown cannot be integrated on-chip. Figure 1 The 4F system shown uses a single light source and emits coherent light. Therefore, during the calculation process, two modulators are required to modulate the two data to be calculated on the two modulators respectively. In addition, since the modulator in the existing 4F system cannot generate photocurrent according to the optical signal and does not have the function of recording spectrum surface data, two modulators are required during the calculation process, and additional computing equipment is required to first calculate the spectrum data of data B according to Fourier transform, and then modulate the spectrum data of data B on the second modulator 106, and the implementation process is relatively complicated.

[0040] Figure 2 A schematic diagram of the structure of an optical computing chip provided by an embodiment of the present invention. Figure 2 It is an on-chip optical computing system. Figure 2As shown, the optical computing chip 200 may include a light source array 202, a modulator array 204, a detector array 206, a first concave mirror 208, and a second concave mirror 210. The light source array 202 is located on the object focal plane of the first concave mirror 208. The modulator array 204 is located on the image focal plane of the first concave mirror 208, and the modulator array 204 is also located on the object focal plane of the second concave mirror 210. The detector array 206 is located on the image focal plane of the second concave mirror 210.

[0041] The light source array 202 is used for data modulation and transmission, and serves as a data input unit of the optical computing chip 200. The light source array 202 can generate multiple optical signals with different light intensities according to the input data. The first concave mirror 208 is used to implement a standard Fourier transform on the data optical signal sent by the light source array 202. The modulator array 204 has two working modes: a recording mode and a modulation mode. Among them, the recording mode is used to obtain the image of the spectrum surface of the data optical signal sent by the light source array 202 after passing through the first concave mirror 208. The modulation mode is used to modulate the spectrum surface image of the data optical signal sent by the light source array 202 on the modulator array 204. The second concave mirror 210 is used to implement a standard Fourier inverse transform on the optical signal after passing through the modulator array 204. The detector array 206 is used for light intensity signal detection, and serves as a result output unit of the optical computing chip 200. The specific implementation of each device in the optical computing chip 200 will be described in detail below.

[0042] In practical applications, the light source array 202 may include a plurality of light emitting elements 302, which may be arranged along a straight line. The intensity of light emitted by each light emitting element may be modulated. The distribution of the light intensity of the light source array along the straight line corresponds to the data that needs to be input into the optical computing chip. Figure 3 As shown, the light source array 202 may include a plurality of light emitting elements 302 arranged in a straight line. The light emitting elements 302 are used to generate incoherent light under the action of voltage. The intensity of the light emitted by the light emitting element 302 may vary with the change of voltage. The light emitted by the light emitting element 302 may be modulated. In one case, the light source array 202 may be implemented by a semiconductor light emitting diode (LED) array. In this case, the light emitting element 302 may be an LED. Each LED may emit incoherent light with a large divergence angle, and its luminous intensity may vary with the change of the injected voltage, so that modulation may be achieved. When an LED is used as a signal input source, the amplitude of the light emitted by the LED may represent the data to be calculated. For a set of data to be calculated f(x k ), set the light amplitude of the kth LED to:

[0043] E k (x) = f(x k )

[0044] Since each LED is independent of each other, the emitted light source can be regarded as the superposition of multiple light sources. For a single LED, the light field of its emitted light can be approximated as a Gaussian function. Finally, the light intensity I(x) of the light emitted by the light source array 202 on the object plane is the result of the superposition of all LED light sources, that is,

[0045]

[0046] Wherein, x is used to indicate the calculation data vector, n is the total number of LEDs on the light source array 202, and k ranges from 1 to n. k (x) is used to indicate the light amplitude of the light signal emitted by the kth LED.

[0047] In another case, the light source array 202 can also be implemented by a laser array. In this case, the light emitting element 302 can refer to a laser. By arranging a lens with a divergence angle in front of each laser, each laser can emit light with a larger divergence angle. In addition, light intensity modulation can be achieved by adding a variable transmittance material in front of the laser. It is understandable that in practical applications, using a laser as a light emitting element can be equivalent to the effect of using an LED as a light emitting element, but the implementation process is relatively complicated.

[0048] It should be noted that the light source array 202 in the embodiment of the present invention is Figure 1 Compared with a single laser that emits a single light source, since multiple light sources are used and incoherent light is emitted, and the intensity of each light source in the light source array can be changed, there is no need for an additional modulator to emit a light signal containing multiple light intensities according to different values ​​in the data, and the speed of generating light signals is faster.

[0049] The modulator array 204 has two working modes: a recording mode and a modulation mode. The recording mode is used to obtain the image of the spectrum surface of the data optical signal sent by the light source array 202 after passing through the first concave mirror 208. The modulation mode is used to modulate the spectrum surface image of the data optical signal sent by the light source array 202 on the modulator array 204. The modulator array 204 may include a plurality of modulators. For example, a plurality of modulators may be arranged in a straight line to obtain the modulator array 204. In an embodiment of the present invention, each modulator in the modulator array 204 may record and modulate the intensity of the received incident light. In order to realize the recording and modulation of the incident light intensity, the modulator may be implemented using structures based on different principles such as doped silicon waveguides, electro-absorption modulators, and semiconductor optical amplifiers (SOA).

[0050] For example, when SOA is used as a modulator, the incident light intensity can be recorded by detecting the size of the incident photocurrent of SOA. In addition, the intensity of the light signal passing through SOA can also be changed by changing the transmittance of light. In practical applications, SOA can be made of semiconductor quantum well materials. Since different voltages can cause different changes in the transmittance of SOA to light, the transmittance of SOA to light can be changed between 0-1 by voltage control. Specifically, when the voltage passing through SOA is a reverse bias voltage, the incident light will generate a photocurrent in SOA, so that the distribution of light intensity can be obtained by detecting the size of the photocurrent. In the present invention, by detecting the light intensity distribution reaching the modulator array in this way, the spectrum distribution data after Fourier transformation of the object surface signal of the first concave mirror 208 can be directly obtained, thereby realizing the recording function of the modulator. As the injection voltage changes, the transmittance of SOA to the incident light also changes accordingly. In an embodiment of the present invention, the change in the transmittance of SOA to the incident light can also be used to modulate the light spatial spectrum surface data. For example, different voltages (also called modulation signals) can be input to the corresponding modulators according to the values ​​in the obtained spatial spectrum surface data, wherein the transmittance of a modulator in the modulator array to light can reflect a value in the spectrum surface data. In this way, the spectrum surface data can be modulated on the modulator array to realize the modulation function of the modulator array.

[0051] It is understandable that, in practical applications, when SOA is under the action of forward voltage, it can also realize the function of recording the intensity of light, and when SOA is under the action of reverse bias voltage, it can also realize the function of modulating the intensity of light. In the embodiment of the present invention, there is no specific restriction on the material and operating voltage of the modulator array 204, as long as the modulator array 204 can realize the two functions of recording and modulation. In other words, in the embodiment of the present invention, the modulator array 204 needs to generate current after receiving the incident light, so as to obtain the light intensity distribution of the incident light by detecting the size of the photocurrent to realize the recording of the light intensity. In addition, the modulator array 204 can change the transmittance of light according to the applied voltage change to realize the modulation of the intensity of light.

[0052] The detector array 206 is used to detect the intensity of incident light as a result output unit of the optical computing chip 200. In practical applications, the detector array 206 can be implemented by a semiconductor photodiode (PD) array, a photoconductive detector array (eg, a photoresistor array), and the like.

[0053] As mentioned above, the first concave mirror 208 is used to implement a standard Fourier transform on the data optical signal sent by the light source array 202. The second concave mirror 210 is used to implement a standard inverse Fourier transform on the optical signal after passing through the modulator array 204. Those skilled in the art know that the traditional 4F system uses a convex lens to implement the Fourier transform. In the optical computing chip provided in the embodiment of the present invention, a reflective concave mirror is used to implement the Fourier transform and the inverse Fourier transform. The principle of the concave mirror to implement the Fourier transform and the inverse Fourier transform is briefly introduced below.

[0054] Figure 4 A schematic diagram of the light path reflected by a spherical concave mirror provided in an embodiment of the present invention. A person skilled in the art will know that for a reflective lens, the phase delay generated after the incident light is reflected is basically caused by the geometric shape of the reflector. As shown in the figure, when the concave mirror is a spherical surface, its reflection surface equation is:

[0055]

[0056] Where R is the radius of the circle. For light incident perpendicular to the optical axis, the optical path difference between it and the light propagating along the optical axis is:

[0057]

[0058] For paraxial light, assuming R>>x, y, we have

[0059]

[0060] After spherical reflection, the total phase delay at point (x, y) can be written as

[0061]

[0062] In this way, the phase reflection function after reflection by the concave mirror has the same form as the phase transmission function of the convex lens, that is,

[0063]

[0064] Therefore, the concave reflector can also realize Fourier transform of the incident light, corresponding to a focal length of R / 2.

[0065] In practical applications, due to the limited space of the chip, the size of the spherical concave mirror relative to the light source array 202 cannot meet the requirements of the paraxial light, which will cause a large error in the Fourier transform result. In the embodiment of the present invention, in order to reduce the calculation error, a parabolic concave reflector can be used to replace the spherical concave reflector. The light path diagram of the parabolic reflector can be referred to Figure 4 The optical path diagram of the spherical concave reflector shown is not repeated here.

[0066] When using a parabolic concave reflector, assume that the y-axis is the optical axis, p is a constant related to the parabola focus, and Z0 is the coordinate of the parabola vertex. Set the parabola type to:

[0067]

[0068] Then, for light incident parallel to the optical axis, through the same analysis process as the spherical mirror above, its phase change function can be obtained as

[0069]

[0070] Therefore, the parabolic concave reflector also has the same function as a convex lens, and can realize Fourier transform of incident light, and its corresponding focal length is p / 2. In practical applications, the size of the parabola of the parabolic concave reflector can be determined according to the ratio of the size of the light source and the size of the concave mirror.

[0071] The above briefly describes the principle of using a concave mirror to realize Fourier transform in the embodiment of the present invention. It should be noted that the present invention is not limited to using a parabolic concave mirror. In some cases, if the light emitted by the light source array 202 can meet the paraxial light of a spherical concave mirror, a spherical concave mirror may also be used.

[0072] In practical applications, during the production of chips, the reflective surfaces of the first concave mirror 208 and the second concave mirror 210 can be formed into air reflective surfaces through a deep etching process. In order to reduce losses, a high-reflective film can also be plated on the end surface. Another advantage of forming the reflective surface by etching is that any curved surface can be accurately defined, so the parabolic concave mirror as described above can be manufactured to achieve a more accurate Fourier transform effect.

[0073] The following will be combined Figure 5 right Figure 2 The optical computing chip shown provides a detailed introduction to the data processing process. Figure 5 A flow chart of a method for implementing convolution calculation of data using an optical computing chip provided in an embodiment of the present invention. The following describes how the computing chip implements convolution calculation by taking the implementation of convolution operation on first data and second data as an example. It can be understood that both the first data and the second data may include multiple real numbers. Figure 5 As shown, the method includes the following steps.

[0074] In step 502, the light source array 202 generates a first light signal according to the first data. As described above, the light emitting elements 302 in the light source array 202 can generate light signals of different light intensities according to the change of voltage. In this step, different voltages can be input to different light emitting elements 302 in the light source array 202 according to the numerical values ​​in the first data, so that the light emitting elements 302 in the light source array 202 emit incoherent light of different light intensities according to the different numerical values, so as to obtain the first light signal. In an embodiment of the present invention, the first light signal includes a plurality of incoherent lights emitted by different light emitting elements 302 in the light source array 202.

[0075] In step 504, the first concave mirror 208 outputs a first reflected light signal according to the first light signal. As described above, in the embodiment of the present invention, the light source array 202 is located at the focal length position of the object plane of the first concave mirror 208, and the first concave mirror 208 can receive the paraxial light emitted by the light source array 202. The first light signal is reflected by the first concave mirror 208 to output a first reflected light signal.

[0076] In step 506, the modulator array 204 obtains the first spectrum surface distribution data according to the first reflected light signal. In this step, the modulator array 204 needs to work in the recording mode. As mentioned above, the preparation material of the modulator in the modulator array 204 can enable the modulator to generate current when receiving incident light. Therefore, when the modulator array 204 receives the first reflected light signal, without applying voltage to the modulator array 204, the modulator will generate photocurrent according to the received first reflected light signal, so that the first reflected light signal can be recorded in electrical form by detecting the current intensity in the modulator to obtain the spectrum surface distribution data of the first reflected light signal in the optical space. In the embodiment of the present invention, the spectrum surface distribution data of the first reflected light signal can also be referred to as the first spectrum surface distribution data.

[0077] In step 508, the modulator array 204 modulates the first spectrum surface distribution data on the modulator array. In this step, the modulator array 204 needs to work in a modulation mode. Specifically, different voltages can be applied to the modulators according to the different data recorded by the modulators in the modulator array 204, and the spectrum surface distribution data of the first reflected light signal obtained in step 506 is modulated into the transmittance of the modulators in the modulator array 204, and the first spectrum surface distribution data is represented on the modulator. It can be understood that different voltages are applied to different data, so that different voltages applied to different modulators on the modulator array 204 may be different.

[0078] In step 510, the light source array 202 generates a second light signal according to the second data. In the embodiment of the present invention, after the first data is recorded and modulated to the modulator array 204 through steps 502 to 508, the light source array 202 can generate a second light signal according to the second data. Specifically, different voltages can be input to different light emitting elements 302 in the light source array 202 according to the numerical values ​​in the second data, so that the light emitting elements 302 in the light source array 202 emit incoherent light of different light intensities according to different numerical values, so as to obtain the second light signal.

[0079] In step 512, the first concave mirror 208 outputs a second reflected light signal according to the second light signal. In step 514, the modulator array 204 obtains a third light signal based on the second reflected light signal and the first recorded spectrum surface distribution data. As mentioned above, since the first spectrum surface distribution data of the first reflected light signal obtained according to the first data has been modulated on the modulator array 204, the first spectrum surface distribution data is expressed as the transmittance of the modulator in the modulator array 204 to light. When the modulator array 204 receives the second reflected light signal, after the second reflected light signal passes through the modulator in the modulator array 204, the third light signal can be obtained together with the first spectrum surface distribution data modulated on the modulator array 204. In this way, the spectrum surface distribution data of the second reflected light signal and the spectrum surface distribution data of the first reflected light signal will complete the spectrum surface multiplication operation of the light space through the modulator array 204. In other words, the third optical signal represents an optical signal obtained after the first reflected optical signal and the second reflected optical signal pass through the modulator array 204, and the third optical signal represents the result of the multiplication operation of the first spectrum plane distribution data and the spectrum plane distribution data of the second reflected optical signal in the optical spatial frequency domain. In the embodiment of the present invention, the spectrum plane distribution data of the second reflected optical signal may also be referred to as second spectrum plane distribution data.

[0080] In step 516, the second concave mirror 210 outputs a third reflected light signal according to the third light signal. As described above, the first concave mirror 208 and the second concave mirror 210 can perform Fourier transform and inverse Fourier transform. As described above, since the first data and the second data are both Fourier transformed after passing through the first concave mirror. In this step, after receiving the third light signal output by the modulator array 204, the second concave mirror 210 will output a third reflected light signal, which is the result of inverse Fourier transform of the convolution calculation result output by the modulator array 204.

[0081] In step 518, the detector array 206 detects the third reflected light signal. The distribution of the third reflected light signal on the detector array is used to indicate the convolution result of the first data and the second data. As mentioned above, the detectors on the detector array 206 can detect the intensity of the incident light. Thus, the convolution calculation result of the first data and the second data after the Fourier inverse transform can be obtained according to the intensity of the detected third reflected light signal.

[0082] Figure 6 A schematic diagram of the structure of another optical computing chip provided by an embodiment of the present invention. Figure 2 The on-chip integrated optical computing chip provided is different in that: Figure 6 In the optical computing chip shown, the light source array 202 and the detector array 206 are arranged on the same side of the chip, making the structure of the entire computing chip more compact and reducing the chip size. Figure 6 As shown, Figure 2 Compared with the optical computing chip shown in the figure, the positions of the first concave mirror 208, the second concave mirror 210, and the modulator array 204 have not changed, and the focal length positions of the light source array 202, the modulator array 204, and the detector array 206 relative to the first concave mirror 208 and the second concave mirror 210 have not changed. Figure 6 The implementation of each device shown can refer to Figure 2 Description of the various devices in the optical computing chip shown. Figure 6 The process of implementing the convolution calculation of data by the computing chip shown can also be referred to Figure 2 and Figure 5 The description will not be repeated here.

[0083] The optical computing chip provided by the embodiment of the present invention uses a concave mirror, which is a one-dimensional device and is easy to manufacture on the chip, making it possible to implement optical computing on the chip. In addition, since the modulator in the optical computing chip can generate current according to the intensity of the incident light, the intensity of the received incident light can be recorded and modulated, so that in the process of implementing optical computing, data can be directly recorded and modulated on the modulator array without using additional computing equipment to assist in obtaining the spectrum surface data, thereby improving the computing efficiency and achieving simplicity and efficiency.

[0084] Furthermore, since the light emitted by the light source array used in the embodiment of the present invention is incoherent light, it can take into account the function of data modulation at the same time, so compared with the existing spatial optical computing system, the IO speed of the optical computing chip is greatly improved. In addition, since a concave mirror is used to replace the convex lens in the traditional 4F system, it is easier to manufacture and integrate on the chip. Since all devices can be integrated on the chip, compared with the existing 4F optical computing system, it has a small size, low manufacturing cost and high flexibility. Furthermore, compared with the existing optical computing system that only performs convolution operations as a multiplier, the on-chip optical computing chip can realize complex optical calculations such as Fourier transform, convolution, and autocorrelation.

[0085] Figure 2 and Figure 6 The optical computing chip described is based on a layer of on-chip structure as an example. It is understandable that due to physical implementation limitations, only a one-dimensional light source array and a one-dimensional detector array can be implemented on a layer of chip. Figure 2 and Figure 6 The optical computing chip described is a one-dimensional convolution computing system that can implement convolution computing of one-dimensional data. In another possible implementation, an embodiment of the present invention proposes an optical computing chip that can implement multi-dimensional data computing. Specifically, a multi-dimensional light source array (for example, a two-dimensional light source array) can be realized by stacking multiple single-layer light source arrays, a multi-dimensional modulator array can be realized by stacking multiple one-dimensional modulator arrays, and a multi-dimensional detector array can be realized by stacking multiple one-dimensional detector arrays. And multi-dimensional convolution computing can be realized by increasing the area of ​​the concave mirror. Figure 7 As shown, Figure 7 The optical computing chip shown may include a light source array 702, a modulator array 704, a detector array 706, a first concave mirror 708, and a second concave mirror 710. The relative positions of the above devices may be as follows: Figure 2 and Figure 6 As shown. Figure 2 and Figure 6 The difference between the optical computing chip shown is that Figure 7 In the optical computing chip shown, the light source array 702 may include multiple stacked light source sub-arrays 7022, the modulator array 704 may include multiple stacked modulator sub-arrays 7042, and the detector array 706 may include multiple stacked detector sub-arrays 7062. The structures and working principles of the light source sub-array 7021, the modulator sub-array 7042, and the detector sub-array 7062 may be respectively as follows: Figure 2 As shown in the light source array 202, modulator array 204 and detector array 206.

[0086] In practical applications, a light source subarray 7021, a modulator subarray 7042, and a detector subarray 7062 cooperate to process a row of data in the first data and a row of data in the second data to achieve the following. Figure 2 or Figure 6 For example, a light source subarray 7021 can be used to generate a first sub-light signal according to the first row of data in the first data, and generate a second sub-light signal according to the first row of data in the second data. A modulator subarray 7042 is used to record and modulate the reflected light signal according to the first sub-light signal and the second sub-light signal. A detector subarray 7062 is used to detect the calculation results of the first row of data in the first data and the first row of data in the second data. It can be understood that Figure 2 Compared with the optical computing chip shown in FIG. 1 , since the light source array 702 and the modulator array 704 are both stacked by multiple sub-arrays, the thickness of the first concave mirror 708 and the second concave mirror 710 is relatively Figure 2 The thickness of the first concave mirror 208 and the second concave mirror 210 in the embodiment of the present invention will be increased, so as to be able to reflect the optical signals emitted by the light source array 704 and the modulator array 704. It can be understood that Figure 7 The optical computing chip shown may include multiple stacked light source subarrays, multiple stacked modulator subarrays, and multiple stacked detector subarrays, and thus can implement convolution calculations of multi-dimensional data (also referred to as multi-row and multi-column data).

[0087] Those skilled in the art will know that in practical applications, the optical computing chip needs to cooperate with other auxiliary circuits to realize optical computing. Figure 8 FIG. 1 is a schematic diagram of the structure of an optical computing system provided by an embodiment of the present invention. Figure 8 As shown, the optical computing system 800 mainly includes three parts: a control plane 802, an optical computing chip 804, and a peripheral driving circuit, wherein the peripheral driving circuit may include a light source array driving circuit 806, a modulator array driving circuit 810, and a detector array driving circuit 808. In practical applications, the control plane 802 may include a device such as a processor that can realize functions such as control and processing. For example, the control plane 802 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), and other processing devices, which are not limited here. The optical computing chip 804 may be as described above Figure 2 , Figure 6 or Figure 7For example, the optical computing chip 804 may include: a light source array 8042, a modulator array 8044, a detector array 8046, a first concave mirror 8048, and a second concave mirror 8049. The specific implementation of the optical computing chip 804 may refer to the description of the above embodiment.

[0088] Since the light source array 8042, the modulator array 8044, and the detector array 8046 in the optical computing chip 804 are active devices, they need to be driven by corresponding external driving circuits. Therefore, the optical computing system 800 may also include peripheral circuits such as a light source array driving circuit 806, a modulator array driving circuit 810, and a detector array driving circuit 808. The following briefly introduces the peripheral circuits in the optical computing system by taking the aforementioned process of implementing convolution calculation of the first data and the second data as an example.

[0089] The light source array driving circuit 806 is used to receive the data to be calculated sent by the control plane 802, and convert the received data into a corresponding voltage to drive the light source array 8042 to emit a corresponding light signal. For example, the light source array driving circuit 806 can receive the first data and the second data sent by the control plane 802, and convert the first data and the second data into a corresponding voltage. In practical applications, Fig. 9 As shown, the light source array driving circuit 806 may include a digital-to-analog conversion circuit 8062 and an amplifier circuit 8064. Taking the reception of the first data as an example, the digital-to-analog conversion circuit 8062 in the light source array driving circuit 806 may perform digital-to-analog conversion on the received first data, and convert the first data into a corresponding voltage. Then, the amplifier circuit 8064 amplifies the voltage obtained after the conversion, and sends the corresponding amplified voltage to the light source array 8042, thereby driving the light source array 8042 to convert the received electrical signal into a corresponding optical signal, thereby converting the first data into a first optical signal.

[0090] As mentioned above, the modulator array provided in the embodiment of the present invention has two working modes: recording and modulation. Therefore, the modulator array 8044 in the optical computing chip 804 and the modulator array driving circuit 810 are a two-way data interaction process. Specifically, Fig.10As shown, the modulator array driving circuit 810 may include a first amplifier circuit 8102, an analog-to-digital conversion circuit 8104, a digital-to-analog conversion circuit 8106, and a second amplifier circuit 8108. When the modulator array 8044 works in the recording mode, the modulator array 8044 generates a corresponding current after receiving the first reflected light signal of the first concave mirror 8048. The modulator array driving circuit 810 can collect the current generated by the modulator array 8046. The photocurrent generated by the modulator array 8046 according to the received first reflected light signal is used to indicate the spectrum surface distribution data of the first reflected light signal. After the first amplifier circuit 8102 in the modulator array driving circuit 810 amplifies the received current, the analog-to-digital conversion circuit 8104 converts the amplified current into data, thereby obtaining the spectrum surface distribution data of the first reflected light signal. In other words, after the analog-to-digital conversion circuit 8104 converts the received current into a digital signal, the spectrum surface distribution data of the first reflected light signal can be recorded. After obtaining the spectrum surface distribution data of the first reflected light signal, the modulator array driving circuit 810 may send the spectrum surface distribution data of the first reflected light signal to the control plane 802, and the control plane 802 performs denoising, normalization or format conversion on the received spectrum surface distribution data of the first reflected light signal, and then sends the processed data to the modulator array driving circuit 810. The digital-to-analog conversion circuit 8106 in the modulator array driving circuit 810 may receive the processed data sent by the control plane 802, and convert the received data into a corresponding analog signal, and the second amplifier circuit 8108 inputs the corresponding voltage (also referred to as a modulation signal) into the modulator array according to the converted analog signal, thereby driving the modulator array 8044 to work in the modulation mode, and modulating the corresponding data on the modulator array 8044. In practical applications, the first amplifier circuit 8102 and the second amplifier circuit 8108 may also be implemented using one circuit, and the analog-to-digital conversion circuit 8104 and the digital-to-analog conversion circuit 8106 may also be integrated together.

[0091] like Fig.11 As shown, the detector array driving circuit 808 may also include an amplifier circuit 8082 and an analog-to-digital conversion circuit 8084. The detector array driving circuit 808 is used to receive the light intensity signal detected by the detector array 8046. It can be understood that the light intensity signal detected by the detector array 8046 is in the form of current. In practical applications, the amplifier circuit 8082 can amplify the current signal detected by the detector array 8046. Then, the analog-to-digital conversion circuit 8084 in the detector array driving circuit 808 can convert the received analog signal into a digital signal, so as to obtain the calculation results of the first data and the second data, and can send the calculation results to the control plane 802.

[0092] According to the above description of the optical computing system 800, it can be known that the data input of the entire optical computing system 800 can be realized by the control plane 802 driving the light source array 8042 through the light source array driving circuit 806, and the calculation result of the optical computing chip 804 can be collected by the detector array driving circuit 808, and the collected data is finally returned to the control plane 802.

[0093] It is to be understood that the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division, and there may be other division methods in actual implementation. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the connection between the modules discussed in the above embodiments may be electrical, mechanical or other forms. The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules. In addition, the functional modules in each embodiment of the application embodiment may exist independently or may be integrated into a processing module.

[0094] The embodiment of the present invention also provides a computer program product for data processing, including a computer-readable storage medium storing program code, wherein the program code includes instructions for executing the method flow described in any of the aforementioned method embodiments. A person of ordinary skill in the art can understand that the aforementioned storage medium includes: a U disk, a mobile hard disk, a magnetic disk, an optical disk, a random access memory (RAM), a solid state disk (SSD) or a non-volatile memory, etc., which are non-transitory machine-readable media that can store program code.

[0095] It should be noted that the embodiments provided in this application are merely illustrative. Those skilled in the art can clearly understand that, for the convenience and brevity of description, in the above embodiments, the descriptions of each embodiment have their own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. The features disclosed in the embodiments, claims, and drawings of the present invention may exist independently or in combination. The features described in the embodiment of the present invention in the form of hardware may be executed by software, and vice versa. This is not limited here.

Claims

1. An optical computing chip, characterized in that: include: The first concave mirror, A light source array is located on the object focal plane of the first concave mirror; A modulator array is located on the image focal plane of the first concave mirror; The light source array is used to generate a first light signal according to the first data; The first concave mirror is used to output a first reflected light signal according to the first light signal; The modulator array is used to: receiving the first reflected light signal; Obtaining first spectrum surface distribution data according to the first reflected light signal; The first frequency spectrum distribution data is modulated on the modulator array.

2. The optical computing chip according to claim 1, characterized in that: The light source array is further used to generate a second light signal according to the second data; The first concave mirror is further used to output a second reflected light signal according to the second light signal; The modulator array is further used to obtain a third optical signal according to the second reflected optical signal and the first frequency spectrum distribution data.

3. The optical computing chip according to claim 2, characterized in that: Also includes: A second concave mirror and a detector array, wherein: The modulator array is also located on the object focal plane of the second concave mirror; The detector array is located on the image focal plane of the second concave mirror; The second concave mirror is used to receive the third optical signal and output a third reflected optical signal according to the third optical signal; The detector array is used to detect the third reflected light signal, and distribution of the third reflected light signal on the detector array is used to indicate a convolution result of the first data and the second data.

4. The chip according to claim 1, characterized in that: The modulator array includes a plurality of modulators, and the transmittance of each modulator to the first reflected light signal is used to indicate a value in the first frequency spectrum distribution data.

5. The chip according to any one of claims 1 to 3, characterized in that: The modulators in the modulator array are implemented by at least one of the following devices: doped silicon waveguide, electro-absorption modulator, semiconductor optical amplifier SOA.

6. The chip according to any one of claims 1 to 4, characterized in that: The light source array includes a plurality of light emitting elements, each of which is used to generate incoherent light.

7. The chip according to claim 3, characterized in that: The light source array and the detector array are located on the same side of the chip.

8. The chip according to claim 3, characterized in that: The first concave mirror and the second concave mirror are parabolic concave mirrors.

9. The chip according to claim 3, characterized in that: The light source array includes a plurality of stacked light source sub-arrays, the modulator array includes a plurality of stacked modulator sub-arrays, and the detector array includes a plurality of stacked detector sub-arrays.

10. An optical computing system, characterized in that: include: A processor, configured to input first data into the optical computing chip; The optical computing chip comprises a first concave mirror, a light source array and a modulator array, wherein the light source array is located on the object focal plane of the first concave mirror, and the modulator array is located on the image focal plane of the first concave mirror; The light source array is used to generate a first light signal according to the first data; The first concave mirror is used to output a first reflected light signal according to the first light signal; The modulator array is used to: receiving the first reflected light signal; Obtaining first spectrum distribution data according to the first reflected light signal; and The first frequency spectrum distribution data is modulated on the modulator array.

11. The optical computing system according to claim 10, characterized in that Also includes: a light source array driving circuit, connected to the processor and the light source array of the optical computing chip respectively, and used for applying a first driving signal to the light source array according to the first data; a modulator array driving circuit, connected to the modulator array, for sampling the first spectrum surface distribution data obtained by the optical computing chip according to the first reflected light signal, and applying a first modulation signal to the optical computing chip according to the first spectrum surface distribution data; The light source array is specifically configured to generate the first light signal according to the first driving signal; The modulator array is specifically used to modulate the first frequency spectrum distribution data on the modulator array according to the first modulation signal.

12. The optical computing system according to claim 11, characterized in that The processor is further configured to send second data; The light source array driving circuit is further used to generate a second driving signal according to the second data; The light source array is further configured to generate a second light signal corresponding to the second data according to the second driving signal; The first concave mirror is further used to output a second reflected light signal according to the second light signal; The modulator array is further used to obtain a third optical signal based on the second reflected optical signal and the first spectrum distribution data.

13. The optical computing system according to claim 12, characterized in that The optical computing chip further includes a second concave mirror and a detector array, the modulator array is also located on the object focal plane of the second concave mirror, and the detector array is located on the image focal plane of the second concave mirror; The second concave mirror is used to receive the third optical signal and output a third reflected optical signal according to the third optical signal; The detector array is used to detect the third reflected light signal, and distribution of the third reflected light signal on the detector array is used to indicate a convolution result of the first data and the second data.

14. The optical computing system according to claim 13, characterized in that Also includes: A detector array driving circuit is connected to the detector array of the optical computing chip and is used to: collecting the third reflected light signal detected by the detector array; Perform analog-to-digital conversion on the third reflected light signal to obtain a convolution result of the first data and the second data.

15. The optical computing system according to claim 10, characterized in that: The modulator array includes a plurality of modulators, and the transmittance of each modulator to the first reflected light signal is used to indicate a value in the first frequency spectrum distribution data.

16. The optical computing system according to claim 10, characterized in that: The modulators in the modulator array are implemented by at least one of the following devices: doped silicon waveguide, electro-absorption modulator, semiconductor optical amplifier SOA.

17. The optical computing system according to claim 10, characterized in that: The light source array includes a plurality of light emitting elements, each of which is used to generate incoherent light.

18. The optical computing system of claim 13, wherein: The light source array and the detector array are located on the same side of the optical computing chip.

19. The optical computing system of claim 13, wherein: The first concave mirror and the second concave mirror are parabolic concave mirrors.

20. The optical computing system of claim 13, wherein: The light source array includes a plurality of stacked light source sub-arrays, the modulator array includes a plurality of stacked modulator sub-arrays, and the detector array includes a plurality of stacked detector sub-arrays.

21. A data processing method, characterized in that: The method is performed by an optical computing chip, which includes a first concave mirror, a light source array, and a modulator array, wherein the light source array is located on the object focal plane of the first concave mirror, and the modulator array is located on the image focal plane of the first concave mirror; The method comprises: The light source array generates a first light signal according to the first data; The first concave mirror outputs a first reflected light signal according to the first light signal; The modulator array obtains first frequency spectrum plane distribution data according to the first reflected light signal, and modulates the first frequency spectrum plane distribution data on the modulator array.

22. The data processing method according to claim 21, characterized in that: The optical computing chip further includes a second concave mirror and a detector array, the modulator array is also located on the object focal plane of the second concave mirror, and the detector array is located on the image focal plane of the second concave mirror; The method further comprises: The light source array generates a second light signal according to the second data; The first concave mirror outputs a second reflected light signal according to the second light signal; The modulator array obtains a third optical signal according to the second reflected optical signal and the first spectrum distribution data; The second concave mirror outputs a third reflected light signal according to the third light signal; The detector array detects the third reflected light signal, and distribution of the third reflected light signal on the detector array is used to indicate a convolution result of the first data and the second data.

Images