Optical calculation method and system based on on-chip diffraction optical feature extraction and chip

Abstract

The invention provides an optical calculation method and system based on on-chip diffractive optical feature extraction and a chip. The method comprises the following steps: loading data of an input end to an input optical signal; performing optical domain preprocessing on the input optical signal to obtain a preprocessed optical signal; the preprocessing mode comprises power distribution, time delay, amplitude and phase adjustment; performing feature extraction calculation on the preprocessed optical signal through an on-chip diffraction optical feature extractor to obtain optical feature information which is output in parallel; the optical characteristic information is transmitted to a receiving end. According to the method, feature extraction and preliminary calculation are carried out in the optical domain, the calculation time is greatly shortened, the high-speed requirement can be met, redundant data can be compressed, the speed, bandwidth and efficiency of the whole calculation system are improved, and the severe bottleneck caused by electronic calculation is solved.

Description

Technical Field

[0001] The present invention relates to the field of optical computing technology, and in particular to an optical computing method, system and chip based on on-chip diffraction optical feature extraction. Background Art

[0002] In the era of vigorous development of information technology, a huge amount of multimodal information is generated from industries such as industry, finance, and services. These data contain a lot of key information about the operation of a large number of industries, and the information in these data has become an important factor of production. However, the information in most of the data is implicit, and important features need to be extracted to support analysis. In high-real-time computing scenarios, ms or even sub-us computing response delays are necessary. However, in the traditional "sense, transmission, and calculation" link, the perception, transmission, and calculation of information are separated, which seriously hinders the end-to-end computing delay of the entire computing system. In addition, the data volume is much larger than the information volume, and there is a lot of redundancy. Therefore, the latency and processing speed of the entire computing system face severe bottlenecks. Summary of the invention

[0003] The present invention provides an optical computing method, system and chip based on on-chip diffraction optical feature extraction, which is used to solve the defects of the delay and processing speed of the entire computing system in the prior art. The present invention performs feature extraction and preliminary calculation in the optical domain, greatly reduces the computing time, can meet high-speed requirements, compress redundant data, improve the speed, bandwidth and efficiency of the entire computing system, and solve the severe bottleneck brought by electronic computing.

[0004] The present invention provides an optical computing method based on on-chip diffraction optical feature extraction, comprising: loading data at an input end onto an input optical signal; performing optical domain preprocessing on the input optical signal to obtain a preprocessed optical signal; the preprocessing method comprises power allocation, time delay, amplitude and phase adjustment; performing feature extraction calculation on the preprocessed optical signal by an on-chip diffraction optical feature extractor to obtain optical feature information output in parallel; and the optical feature information is used to be transmitted to a receiving end.

[0005] According to an optical computing method based on on-chip diffraction optical feature extraction provided by the present invention, the data at the input end is loaded onto the input optical signal, comprising: when the data at the input end is image data, converting the image data into a time series signal; and using an electrical signal generated by an arbitrary waveform generator to control an off-chip phase modulator to map the time series signal onto the phase of the input optical signal.

[0006] According to an optical computing method based on on-chip diffraction optical feature extraction provided by the present invention, the optical domain preprocessing of the input optical signal to obtain a preprocessed optical signal includes: the input optical signal enters a waveguide through an input coupling grating; the optical signal in the waveguide is distributed to different paths through the reconfigurable power divider to obtain an optical signal after power distribution; the power distribution ratio of the optical signal on each path can be dynamically adjusted; the optical signal after power distribution is delayed through a delay line to obtain a delayed optical signal; the optical signals on different paths have different delays after passing through the delay line; the amplitude of the delayed optical signal is adjusted through the amplitude modulator to obtain an optical signal after amplitude adjustment; the phase of the optical signal after amplitude adjustment is adjusted through the phase modulator, and the optical signal after phase adjustment is used as the preprocessed optical signal.

[0007] According to an optical computing method based on on-chip diffraction optical feature extraction provided by the present invention, after performing feature extraction calculation on the pre-processed optical signal by an on-chip diffraction optical feature extractor to obtain optical feature information, the method includes: performing incoherent power summing of multiple groups of parallel optical feature information to obtain a summed optical signal; and the summed optical signal is used to transmit to the receiving end.

[0008] According to an optical computing method based on on-chip diffraction optical feature extraction provided by the present invention, before loading the information of the input end onto the input optical signal, it also includes: calibrating multiple reconfigurable power dividers, amplitude modulators and phase modulators.

[0009] According to an optical calculation method based on on-chip diffraction optical feature extraction provided by the present invention, the correction of multiple reconfigurable power dividers, amplitude modulators and phase modulators includes: using a femtosecond laser as an input light source, the receiving end includes an oscilloscope and an optical power meter connected to a photodetector; initializing multiple reconfigurable power dividers and amplitude modulators; maximizing the output power of the optical power meter so that the multiple amplitude modulators are in a fully open state; adjusting the multiple reconfigurable power dividers in sequence from front to back, minimizing the output power of the optical power meter so that all input powers are injected into the longest Delay path; scanning the voltage bias of multiple reconfigurable power dividers from back to front in sequence until the power ratio of the outputs with different delays obtained on the oscilloscope meets the preset ratio requirement, and the calibration of the reconfigurable power divider and the amplitude modulator is completed; switching the input light source to a continuous laser; scanning the loading bias of multiple phase modulators from front to back in sequence, and aligning the phases by maximizing the output power of the optical power meter; adjusting the voltage bias of multiple phase modulators again according to the phase-voltage mapping curve, so that the phase relationship meets the preset phase requirement, and the calibration of the phase modulator is completed.

[0010] According to an optical computing method based on on-chip diffraction optical feature extraction provided by the present invention, the on-chip diffraction optical feature extractor includes an input array, a diffraction zone and an output array; the on-chip diffraction optical feature extractor performs feature extraction calculation on the preprocessed optical signal to obtain optical feature information, including: the preprocessed optical signal is diffracted when entering the diffraction zone through the input array; the optical signal is diffused while propagating in the diffraction zone; and the edge features of the preprocessed signal are extracted in the output array through coherent superposition of optical signals.

[0011] According to an optical computing method based on on-chip diffraction optical feature extraction provided by the present invention, the communication band of the on-chip diffraction optical feature extractor is 1.55μm communication band, the delay range of the delay line is 80-240ps, the frequency range of the input optical signal is 4.16-12.5Gbaud / s, and the area range of the diffraction zone is 10×8μm 2 .

[0012] The present invention also provides an optical computing system based on on-chip diffraction optical feature extraction, comprising: a loading module, used to load data from an input end onto an input optical signal; a preprocessing module, used to perform optical domain preprocessing on the input optical signal to obtain a preprocessed optical signal; the preprocessing method comprises power allocation, time delay, amplitude and phase adjustment; an on-chip diffraction optical feature extractor, used to perform feature extraction calculation on the preprocessed optical signal to obtain optical feature information output in parallel; the optical feature information is used to be transmitted to a receiving end.

[0013] The present invention also provides a chip, comprising the above-mentioned optical computing system based on on-chip diffraction optical feature extraction.

[0014] The present invention provides an optical computing method, system and chip based on on-chip diffraction optical feature extraction, the method comprising: loading the data at the input end onto the input optical signal; preprocessing the input optical signal in the optical domain to obtain a preprocessed optical signal; the preprocessing method includes power allocation, delay, amplitude and phase adjustment; performing feature extraction calculation on the preprocessed optical signal through an on-chip diffraction optical feature extractor to obtain parallel output optical feature information; the optical feature information is used to transmit to the receiving end. The present invention performs feature extraction and preliminary calculation in the optical domain, greatly reducing the calculation time, being able to meet high-speed requirements, compressing redundant data, improving the speed, bandwidth and efficiency of the entire computing system, and solving the severe bottleneck brought by electronic computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 It is a flow chart of an optical calculation method based on on-chip diffraction optical feature extraction provided by the present invention.

[0017] Figure 2 It is a schematic diagram of the principle of an optical calculation method based on on-chip diffraction optical feature extraction provided by the present invention.

[0018] Figure 3 It is a schematic diagram of the principle of amplitude phase correction provided by the present invention.

[0019] Figure 4a It is a schematic diagram of an electron microscope of the on-chip diffraction optical feature extractor provided by the present invention.

[0020] Figure 4bThis is one of the electric field simulation schematic diagrams of the on-chip diffraction optical feature extractor provided by the present invention.

[0021] Figure 4c This is the second electric field simulation schematic diagram of the on-chip diffraction optical feature extractor provided by the present invention.

[0022] Figure 4d It is a relationship diagram between the output power and the sampling point phase increase value provided by the present invention.

[0023] Figure 5a It is an example input diagram of optical calculation based on on-chip diffraction optical feature extraction provided by the present invention.

[0024] Figure 5b It is an example output diagram of an optical calculation based on on-chip diffraction optical feature extraction provided by the present invention.

[0025] Figure 6 It is a structural schematic diagram of an optical computing system based on on-chip diffraction optical feature extraction provided by the present invention.

[0026] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] With the booming development of information technology, all walks of life, whether manufacturing, financial services or healthcare, are constantly generating and accumulating large amounts of data. These data are not only huge in volume, but also diverse in type, including text, images, sound and video, which together constitute the so-called "big data". The value contained in it is immeasurable and has become a key resource for driving economic growth and innovation. However, despite the huge potential of data, it faces major challenges in converting it into actual insights and value. The huge demand for computing resources and energy consumption of the traditional "sensing, transmission and computing" link have gradually become bottlenecks restricting its development.

[0029] Please refer to Figure 1 , Figure 1 A schematic flow chart of an optical calculation method based on on-chip diffraction optical feature extraction provided by the present invention.

[0030] The present invention provides an optical calculation method based on on-chip diffraction optical feature extraction, comprising: 101: Loading the data at the input end onto the input optical signal; 102: preprocessing the input optical signal in the optical domain to obtain a preprocessed optical signal; the preprocessing method includes power allocation, delay, amplitude and phase adjustment; 103: Perform feature extraction calculation on the pre-processed optical signal through the on-chip diffraction optical feature extractor 5 to obtain parallel output optical feature information; the optical feature information is used to transmit to the receiving end.

[0031] In order to solve the technical problems existing in the existing technology, the entire information processing system needs to add a lightweight high-speed computing network at the starting end to perform preliminary computing tasks to reduce data redundancy, share the pressure of the computing center, and reduce the amount of communication. At the starting end, light is used as a computing carrier, and the computing output results can directly enter the optical transmission part by taking advantage of its high parallelism, low power consumption, and fast signal processing speed. Compared with deploying an electronic computing network at the starting end, the computing time is greatly reduced due to the direct feature extraction and preliminary calculation in the optical domain, which can meet the high-speed requirements. Due to the high bandwidth of the entire system, the frequency of executing calculations can be increased, thereby achieving a higher computing rate. Upgrading the "transmission calculation" solution to "calculation transmission calculation" can realize redundant data compression, feature extraction and simple calculation at the initial end, which can further improve the speed, bandwidth, and efficiency of the entire system and solve the severe bottleneck brought by electronic computing.

[0032] Specifically, multiple modal information as data at the input end can be mapped according to certain rules and loaded onto the phase of the input optical signal at high speed through an off-chip phase modulator. Through the fully reconfigurable power division and delay path on the chip, the reconstruction processing of the fed high-speed optical signal is completed with low loss in the on-chip structure (the input optical signal is preprocessed), and the high-speed timing signal is prepared as a parallel high-speed sequence with accurate delay, amplitude and phase in a precisely corrected state. Through the on-chip diffraction optical feature extractor 5 with low structural complexity, the prepared parallel high-speed sequence is subjected to convolution calculation (feature extraction calculation is performed on the preprocessed optical signal) to obtain optical feature information. After the optical feature information is calculated and processed by the entire system and transmitted through the optical fiber, the calculation result is sent to the receiving end. Operations including image recognition, edge detection, segmentation, etc. can be completed. At the same time, the optical feature signals output by multiple on-chip diffraction optical feature extractors 5 in parallel on the chip will be incoherently summed again, so that the total output realizes the extensive perception of multi-source signals (processing high-speed multi-source information and generating decision actions). The entire system is deployed on a chip with a compact structure and stable functions.

[0033] Please refer to Figure 2 , Figure 2A schematic diagram of the principle of an optical calculation method based on on-chip diffraction optical feature extraction provided by the present invention.

[0034] As a preferred embodiment, the data at the input end is loaded onto the input optical signal, including: when the data at the input end is image data, converting the image data into a time series signal; and using an electrical signal generated by an arbitrary waveform generator to control an off-chip phase modulator to map the time series signal onto the phase of the input optical signal.

[0035] In this embodiment, for the high-speed image edge feature extraction task, the image data to be transmitted is first prepared, and then the static image data is converted into a time series signal for transmission and processing through the optical system. And a header is added to the time series signal for synchronization, error detection or more complex data organization to ensure the integrity and identifiability of the data during transmission. The electrical signal generated by the arbitrary waveform generator is used to control the off-chip phase modulator to load the time series signal of the image onto the phase of the optical signal. The signal generated by the arbitrary waveform generator is synchronized with the time series signal to ensure accurate encoding of the information. This is the process of converting electrical signals or data into optical signals, encoding information by changing the phase of the light wave through phase modulation.

[0036] As a preferred embodiment, the input optical signal is preprocessed in the optical domain to obtain a preprocessed optical signal, including: the input optical signal enters the waveguide through the input coupling grating 7; the optical signal in the waveguide is distributed to different paths through a reconfigurable power divider 2 to obtain an optical signal after power distribution; the power distribution ratio of the optical signal on each path can be dynamically adjusted; the optical signal after power distribution is delayed through a delay line 1 to obtain a delayed optical signal; the optical signals on different paths have different delays after passing through the delay line 1; the delayed optical signal is amplitude-adjusted through an amplitude modulator 3 to obtain an optical signal after amplitude adjustment; the optical signal after amplitude adjustment is phase-adjusted through a phase modulator 4, and the optical signal after phase adjustment is used as the preprocessed optical signal.

[0037] In this embodiment, the entire optical computing system includes an input coupling grating 7, a reconfigurable power divider 2, a spiral delay line 1, an amplitude modulator 3, a phase modulator 4, an on-chip diffraction optical feature extractor 5 and an output coupling grating 8. The input coupling grating 7 introduces the input optical signal into the waveguide and starts to transmit in the optical computing system. The reconfigurable power divider 2 distributes the optical signal in the waveguide to different paths. It has reconfigurability, which means that the power distribution ratio of the optical signal on each path can be dynamically adjusted to meet different computing needs. The delay line 1 introduces different delays on different paths to achieve parallel processing of signals in time. The spiral design helps to achieve a longer optical path in a limited space, thereby introducing a larger delay. The amplitude modulator 3 corrects the amplitude of the delayed optical signal to maintain the relative amplitude relationship of the optical signal. The optical signal passing through the amplitude modulator 3 is phase-corrected by the phase modulator 4 to maintain the relative phase relationship of the signal, and the optical signal after the phase adjustment is used as the preprocessed optical signal. The preprocessed optical signal can be a plurality of parallel signals. The on-chip diffraction optical feature extractor 5 is the core part of the optical computing system. It extracts and calculates the features of parallel signals through the principle of diffraction optics. The on-chip diffraction optical feature extractor 5 can perform operations similar to convolution calculations, extract features and calculate the pre-processed optical signals, and finally obtain the calculation results. The output coupling grating 8 can export the optical signal in the waveguide to the outside, ensuring that the optical signal can effectively enter and leave the waveguide.

[0038] As a preferred embodiment, after the preprocessed optical signal is subjected to feature extraction calculation by the on-chip diffraction optical feature extractor 5 to obtain the optical feature information, the method further comprises: performing incoherent power summing 6 on multiple groups of parallel optical feature information to obtain the summed optical signal; the summed optical signal is used for transmission to the receiving end.

[0039] In this embodiment, the outputs of multiple groups of parallel on-chip diffraction optical feature extractors 5 are used to implement incoherent power summation 6 through Y branches, and the optical signal representing the calculation result is finally received by the receiving end (optical power meter, optical detector and oscilloscope) after being coupled into the optical fiber through the grating. At the receiving end, the optical signal is converted back to an electrical signal, and the original image data is recovered from it.

[0040] The optical power meter is used to measure the power of the optical signal; the optical detector is used to convert the optical signal into an electrical signal; the oscilloscope is used to observe and analyze the waveform of the electrical signal output by the optical detector.

[0041] After calibration, several thermo-optical amplitude and phase modulators 4 in the optical computing system can realize parallel preparatory processing of high-speed signals and maintain the relative amplitude and phase relationship of the signals. The optical signal enters the on-chip diffractive optical feature extractor 5 for propagation, and the on-chip diffractive optical feature extractor 5 performs edge feature calculation, and the two-way output result is the edge information of the input signal, thereby realizing edge feature extraction of high-speed and high-throughput information.

[0042] As a preferred embodiment, before the information at the input end is loaded onto the input optical signal, the method further includes: calibrating a plurality of reconfigurable power dividers 2, amplitude modulators 3 and phase modulators 4.

[0043] Please refer to Figure 3 , Figure 3 A schematic diagram of the principle of amplitude and phase correction provided by the present invention.

[0044] As a preferred embodiment, multiple reconfigurable power dividers 2, amplitude modulators 3 and phase modulators 4 are calibrated, including: using a femtosecond laser as an input light source, and the receiving end includes an oscilloscope and an optical power meter connected to a photodetector; initializing multiple reconfigurable power dividers 2 and amplitude modulators 3; maximizing the output power of the optical power meter to make multiple amplitude modulators 3 in a fully open state; adjusting multiple reconfigurable power dividers 2 in sequence from front to back, and minimizing the output power of the optical power meter so that all input powers are injected into the path with the longest delay; scanning the voltage bias of multiple reconfigurable power dividers 2 in sequence from back to front until the power ratio of the outputs with different delays obtained on the oscilloscope meets the preset ratio requirement, and the calibration of the reconfigurable power divider 2 and the amplitude modulator 3 is completed; switching the input light source to a continuous laser; scanning multiple phase modulators 4 in sequence from front to back to load bias, and maximizing the output power of the optical power meter to align the phase planes; adjusting the voltage bias of multiple phase modulators 4 again according to the phase-voltage mapping curve so that the phase relationship meets the preset phase requirement, and the calibration of the phase modulator 4 is completed.

[0045] In order to achieve the expected utility of the on-chip diffraction optical feature extractor 5, it is necessary to ensure that the delay, amplitude and phase of the parallel high-speed data provided by the on-chip fully reconfigurable power division and delay path meet the expected design. Therefore, before the entire optical computing system is officially put into operation, a correction operation must be performed. In this embodiment, the power distribution correction is first performed. In this step, a femtosecond laser is used as the light source input, and the output optical signal is distributed to the power meter and the optical detector and oscilloscope to observe the signal. All power dividers and amplitude modulators 3 are randomly initialized to the non-off state, and random amplitude pulse outputs that do not overlap in time can be observed on the oscilloscope. By maximizing the output power, all amplitude modulators 3 are in a fully open state. The power divider is adjusted from front to back in sequence to minimize the output power and achieve that all light energy enters the last path. Then adjust the output of each path to meet the predetermined amplitude state. According to the expected design, the output power ratio of each path is obtained, and the voltage bias of the power divider is scanned from back to front in sequence until the output power ratio obtained on the oscilloscope meets the preset ratio requirement. After the reconfigurable power divider 2 and amplitude modulator 3 are calibrated, phase correction is performed.

[0046] Switch the light source to a continuous laser. Scan the phase modulator 4 from front to back to load the bias, take the bias position at the maximum output power, and perform multiple rounds until the maximum output power remains unchanged, thereby achieving phase plane alignment. Scan the phase modulator 4 bias again to obtain the phase power mapping relationship of each modulator, and load each phase modulator 4 to the corresponding state according to the expected designed phase relationship of each channel (preset phase requirement). At this point, the calibration process of the phase modulator 4 is completed.

[0047] As a preferred embodiment, the on-chip diffraction optical feature extractor 5 includes an input array, a diffraction zone and an output array; the on-chip diffraction optical feature extractor 5 performs feature extraction calculations on the preprocessed optical signal to obtain optical feature information, including: diffraction occurs when the preprocessed optical signal enters the diffraction zone through the input array; the optical signal diffuses while propagating in the diffraction zone; and edge features of the preprocessed signal are extracted in the output array through coherent superposition of the optical signals.

[0048] As a preferred embodiment, the band of the on-chip diffraction optical feature extractor 5 is the 1.55 μm communication band, the delay range of the delay line 1 is 80-240 ps, ​​the frequency range of the input optical signal is 4.16-12.5 Gbaud / s, and the area of ​​the diffraction zone is 10×8 μm 2 .

[0049] Please refer to Figure 4a , Figure 4a Schematic diagram of an electron microscope of the on-chip diffraction optical feature extractor 5 provided by the present invention.

[0050] In this embodiment, the on-chip diffraction optical feature extractor 5 can realize feature extraction of high-speed signals. The on-chip diffraction optical feature extractor 5 includes an input array, a diffraction zone and an output array. After the fully reconfigurable power division and delay path, the optical signal forms multiple groups of signals with stable delay, relatively flat phase and correct amplitude, and the signal group forms an effective sampling in the time domain. When the optical signal group passes through the end of the input array and enters the diffraction zone through the waveguide, strong diffraction occurs. While the light wave propagates in the diffraction zone, it diffuses and the energy is coupled into different output arrays. The coupling amplitude and phase relationship are related to the position and structure of the input and output waveguides. The light waves at different input array ports have different coupling amplitude phase relationships at the output, and coherent superposition is achieved here. The on-chip diffraction optical feature extractor 5 itself is equivalent to a matrix-vector multiplication calculation of the input high-speed multi-sampled signal.

[0051] When the input and output arrays are centered and the output array waveguides are symmetrical and cross-oriented, the fluctuations in the phase of the high-speed optical signal will be reflected in the deflection of the coherent superposition enhancement points after the diffraction of multiple groups of parallel signals. The upper and lower edge features of the high-speed signal can be output at different output array ports, thereby achieving effective extraction of the edge features of the high-speed signal.

[0052] The on-chip diffraction optical feature extractor 5 of the present invention operates in the near-infrared communication band (1550nm), and the adaptive fully reconfigurable power division and delay path design delay is 80ps, which can realize effective rising and falling edge feature detection for high-speed optical phase modulation signals in the frequency range of 4.16-12.5Gbaud / s.

[0053] Please refer to Figure 4b , Figure 4b This is one of the electric field simulation schematic diagrams of the on-chip diffraction optical feature extractor 5 provided by the present invention. When there is no phase difference between the input array waveguides, the output power of the output array is equal.

[0054] Please refer to Figure 4c , Figure 4c This is the second electric field simulation diagram of the on-chip diffraction optical feature extractor 5 provided by the present invention. When there is a phase difference between the input array waveguides, that is, there is edge information in the time sequence, the light spot is deflected, and the output power of the output array is unequal.

[0055] Please refer to Figure 4d , Figure 4d The relationship between the output power and the sampling point phase increase value provided by the present invention is nonlinear. The diffraction part structure occupies an area of ​​4μm×14μm, which ensures minimal calculation delay and meets real-time processing requirements.

[0056] Please refer to Figure 5a , Figure 5aAn example input diagram of optical computing based on on-chip diffraction optical feature extraction provided by the present invention.

[0057] Please refer to Figure 5b , Figure 5b An example output diagram of an optical calculation based on on-chip diffraction optical feature extraction provided by the present invention.

[0058] The optical computing system based on on-chip diffraction optical feature extraction provided by the present invention is described below. The optical computing system based on on-chip diffraction optical feature extraction described below and the optical computing method based on on-chip diffraction optical feature extraction described above can refer to each other.

[0059] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of an optical computing system based on on-chip diffraction optical feature extraction provided by the present invention.

[0060] The present invention also provides an optical computing system based on on-chip diffraction optical feature extraction, including: a loading module 610, used to load data from an input end onto an input optical signal; a preprocessing module 620, used to perform optical domain preprocessing on the input optical signal to obtain a preprocessed optical signal; the preprocessing method includes power allocation, time delay, amplitude and phase adjustment; an on-chip diffraction optical feature extractor 5, used to perform feature extraction calculation on the preprocessed optical signal to obtain optical feature information output in parallel; the optical feature information is used to be transmitted to a receiving end.

[0061] The optical computing system based on on-chip diffraction optical feature extraction of the present invention can realize the parallel processing of arbitrary amplitude and phase programmable high-speed optical signals, and a correction method for the optical computing system is proposed. After the preprocessed optical signal enters the on-chip diffraction optical feature extractor 5, the optical feature can be extracted. By designing the diffraction structure, different features can be extracted. The features can be further supplied to complex networks or directly form strategy functions. This optical computing system is compact in function, flexible in adjustment, and suitable for task environments with strong real-time requirements. The computing rate reaches 12.5 Gbaud / s, which is significantly higher than the computing rate of existing diffraction structures. At the same time, the area of ​​the diffraction part is less than 56μm 2 , the computing area is significantly reduced. The feature calculation is transferred from the back end of the system to the front end of the transmission, effectively reducing the system power consumption and latency.

[0062] The chip provided by the present invention is described below. The chip described below and the optical calculation method based on on-chip diffraction optical feature extraction described above can be referenced to each other.

[0063] The present invention also provides a chip, comprising the above-mentioned optical computing system based on on-chip diffraction optical feature extraction.

[0064] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7 As shown, the electronic device may include: a processor 710, a communications interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communications interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute an optical calculation method based on on-chip diffraction optical feature extraction, the method comprising: loading the data at the input end onto the input optical signal; preprocessing the input optical signal in the optical domain to obtain a preprocessed optical signal; the preprocessing method includes power allocation, time delay, amplitude and phase adjustment; performing feature extraction calculation on the preprocessed optical signal through the on-chip diffraction optical feature extractor 5 to obtain optical feature information output in parallel; the optical feature information is used to transmit to the receiving end.

[0065] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0066] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the optical calculation method based on on-chip diffraction optical feature extraction provided by the above-mentioned methods, and the method includes: loading the data at the input end onto the input optical signal; preprocessing the input optical signal in the optical domain to obtain a preprocessed optical signal; the preprocessing method includes power allocation, time delay, amplitude and phase adjustment; performing feature extraction calculation on the preprocessed optical signal through the on-chip diffraction optical feature extractor 5 to obtain parallel output optical feature information; the optical feature information is used to transmit to the receiving end.

[0067] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the optical calculation method based on on-chip diffraction optical feature extraction provided by the above-mentioned methods, the method comprising: loading the data at the input end onto the input optical signal; preprocessing the input optical signal in the optical domain to obtain a preprocessed optical signal; the preprocessing method comprises power allocation, time delay, amplitude and phase adjustment; performing feature extraction calculation on the preprocessed optical signal through the on-chip diffraction optical feature extractor 5 to obtain parallel output optical feature information; the optical feature information is used to transmit to the receiving end.

[0068] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0069] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optical computing method based on on-chip diffraction optical feature extraction, characterized in that: include: Loading the data at the input end onto the input optical signal; Preprocessing the input optical signal in an optical domain to obtain a preprocessed optical signal; The pre-processing methods include power allocation, time delay, amplitude and phase adjustment; The pre-processed optical signal is subjected to feature extraction calculation by an on-chip diffraction optical feature extractor to obtain optical feature information output in parallel; the optical feature information is used for transmission to a receiving end.

2. The optical computing method based on on-chip diffraction optical feature extraction according to claim 1, characterized in that: The step of loading the data at the input end onto the input optical signal comprises: In the case where the data at the input end is image data, converting the image data into a time series signal; The electrical signal generated by the arbitrary waveform generator is used to control the off-chip phase modulator to map the time series signal onto the phase of the input optical signal.

3. The optical computing method based on on-chip diffraction optical feature extraction according to claim 1, characterized in that: The preprocessing of the input optical signal in the optical domain to obtain a preprocessed optical signal includes: The input optical signal enters the waveguide through the input coupling grating; The optical signal in the waveguide is distributed to different paths through a reconfigurable power divider to obtain an optical signal after power distribution; the power distribution ratio of the optical signal on each path can be dynamically adjusted; The optical signal after power distribution is delayed by a delay line to obtain a delayed optical signal; optical signals on different paths have different delays after passing through the delay line; The delayed optical signal is subjected to amplitude adjustment by an amplitude modulator to obtain an optical signal with adjusted amplitude; The optical signal after the amplitude adjustment is passed through a phase modulator to adjust the phase of the optical signal, and the optical signal after the phase adjustment is used as the pre-processed optical signal.

4. The optical computing method based on on-chip diffraction optical feature extraction according to claim 1, characterized in that: After performing feature extraction calculation on the pre-processed optical signal by the on-chip diffraction optical feature extractor to obtain optical feature information, the method includes: Incoherent power summing is performed on multiple groups of parallel optical characteristic information to obtain a summed optical signal; the summed optical signal is used to transmit to the receiving end.

5. The optical computing method based on on-chip diffraction optical feature extraction according to claim 3, characterized in that: Before the information of the input end is loaded onto the input optical signal, the method further comprises: Calibrate multiple reconfigurable power dividers, amplitude modulators, and phase modulators.

6. The optical computing method based on on-chip diffraction optical feature extraction according to claim 5, characterized in that: The correction of multiple reconfigurable power dividers, amplitude modulators and phase modulators comprises: A femtosecond laser is used as an input light source, and the receiving end includes an oscilloscope and an optical power meter connected to a photodetector; Initializing a plurality of the reconfigurable power dividers and the amplitude modulators; By maximizing the output power of the optical power meter, the plurality of amplitude modulators are in a fully-on state; The plurality of reconfigurable power dividers are sequentially adjusted from front to back, so that all input powers are injected into the path with the longest delay by minimizing the output power of the optical power meter; Scanning the voltage biases of the plurality of reconfigurable power dividers sequentially from back to front until the power ratio of the outputs with different delays obtained on the oscilloscope meets the preset ratio requirement, thereby completing the calibration of the reconfigurable power divider and the amplitude modulator; Switching the input light source to a continuous laser; Scanning the bias loading of the plurality of phase modulators sequentially from front to back to align the phases by maximizing the output power of the optical power meter; The voltage bias of the plurality of phase modulators is adjusted again according to the phase-voltage mapping curve so that the phase relationship meets the preset phase requirement, thereby completing the calibration of the phase modulator.

7. The optical computing method based on on-chip diffraction optical feature extraction according to claim 3, characterized in that: The on-chip diffractive optical feature extractor includes an input array, a diffractive region, and an output array; The on-chip diffraction optical feature extractor performs feature extraction calculation on the pre-processed optical signal to obtain optical feature information, including: The pre-processed optical signal is diffracted when entering the diffraction region through the input array; The light signal diffuses while propagating in the diffraction region; By coherent superposition of optical signals, edge features of the preprocessed signal are extracted in the output array.

8. The optical computing method based on on-chip diffraction optical feature extraction according to claim 7, characterized in that: The wavelength band of the on-chip diffraction optical feature extractor is 1.55 μm communication wavelength band, the delay range of the delay line is 80-240 ps, ​​the frequency range of the input optical signal is 4.16-12.5 Gbaud / s, and the area of ​​the diffraction zone is 10×8 μm 2 .

9. An optical computing system based on on-chip diffraction optical feature extraction, characterized in that: include: A loading module, used for loading the data at the input end onto the input optical signal; A preprocessing module, used for performing optical domain preprocessing on the input optical signal to obtain a preprocessed optical signal; The pre-processing methods include power allocation, time delay, amplitude and phase adjustment; The on-chip diffraction optical feature extractor is used to perform feature extraction calculation on the preprocessed optical signal to obtain optical feature information output in parallel; the optical feature information is used to be transmitted to the receiving end.

10. A chip, characterized in that: The optical computing system based on on-chip diffraction optical feature extraction comprises the optical computing system according to claim 9.

Images