An optical computing chip matrix weight control precision measurement and calibration device and method

Abstract

A kind of optical computing chip matrix weight control precision measurement and calibration device and method, the device mainly includes: photodetector array module, for the photoelectric signal conversion of optical signal output by optical computing chip, parameter extraction and analysis module, for the collection of electric signal, and the calculation of optical computing chip matrix weight control precision and control rate;Environment compensation and closed-loop control module, for monitoring the ambient temperature of optical computing chip, and based on the change of temperature, the offset voltage of optical computing chip is compensated and calibrated control, realize accurate regulation and control to matrix weight.The device and method can realize the accurate measurement of matrix weight control precision and control rate, accurately perceive external environmental change parameter and change voltage parameter in weight control by compensation algorithm, effectively guarantee the high performance, reliable operation of optical computing chip.

Description

Technical Field

[0001] This invention relates to the field of optical computing technology, and in particular to a device and method for measuring and calibrating the accuracy of matrix weight control in optical computing chips. Background Technology

[0002] Optical computing chips are integrated circuits that process information based on photonics principles, performing data operations by manipulating the intensity and phase of optical signals. Their core mechanism utilizes the interference and diffraction effects of light to complete matrix operations, overcoming the speed and energy efficiency bottlenecks of traditional electronic chips. Optical computing chips typically integrate optoelectronic devices such as laser sources, waveguide networks, modulators, and detectors, achieving high-density integration of on-chip optical paths through modern micro-nano fabrication techniques.

[0003] Optical computing chips perform matrix multiplication operations through tunable optical device arrays, offering advantages such as high speed, low power consumption, and parallel processing. In optical computing systems, the weight matrix is ​​controlled by the bias voltage (current) of the optical modulator, and its accuracy directly affects the accuracy of the calculation results. In photonic neural networks, excessive weight errors can lead to model training bias or decreased inference accuracy.

[0004] Currently, the measurement methods for the matrix control accuracy of optical computing chips mainly rely on electrical measurement methods. These methods assess weight accuracy by monitoring the voltage stability of the modulator drive circuit, neglecting the nonlinear response of optoelectronic devices and failing to directly reflect the actual response of the optical domain weights. Static calibration methods use external measurement equipment such as high-precision optical power meters to measure the device's transfer function offline to establish a fixed error calibration table. However, this method cannot adapt to the effects of temperature drift, signal crosstalk, and other factors, has weak dynamic error monitoring capabilities, and is time-consuming for offline measurements, lacking dynamic tracking methods for real-time weight drift. Furthermore, the optical response of optical computing chips is easily affected by environmental factors such as temperature and wavelength drift, and the lack of a real-time calibration mechanism leads to significant measurement errors.

[0005] To address the aforementioned issues, there is an urgent need to invent a dynamic measurement and calibration method for the matrix control accuracy of optical computing chips, so as to directly measure the optical response of tunable optical devices, reflect the true weight accuracy, and improve measurement efficiency. Summary of the Invention

[0006] To address the issues of weak dynamic monitoring, low testing efficiency, sensitivity to environmental interference, and lack of closed-loop compensation in matrix control accuracy measurement of optical computing chips, this disclosure provides a device and method for measuring and calibrating the accuracy of matrix weight control in optical computing chips. By directly quantifying the output optical power parameters of tunable optical devices, a precise mapping between weight values ​​and optical response is established. A temperature sensor and adaptive algorithm are integrated to compensate for environmental drift in real time, and bias parameters are adjusted in a closed-loop manner through a driving circuit. Ultimately, this provides high-precision and high-stability weight control support for optical computing chips, ensuring their reliable application in matrix operation scenarios.

[0007] The optical computing chip matrix weight control accuracy measurement and calibration device disclosed herein mainly includes: The photodetector array module is used to convert the optical signal output by the optical computing chip into a photoelectric signal and feed the electrical signal into the parameter extraction and analysis module. The parameter extraction and analysis module is used to acquire electrical signals, calculate the matrix weight control accuracy and control rate of the optical computing chip, and output the corresponding results. The environmental compensation and closed-loop control module is used to monitor the ambient temperature of the optical computing chip and, based on the temperature change, to compensate and calibrate the multi-channel voltage required for the optical computing chip to construct matrix weights, thereby achieving precise control of matrix weights.

[0008] Furthermore, the parameter extraction and analysis module includes a real-time oscilloscope and a parameter extraction and analysis unit, wherein: A real-time oscilloscope is used to acquire electrical signals and extract relevant data. The parameter extractor and analyzer is used to process the data and obtain a graph showing the relationship between the normalized matrix weights and time. The graph is used to obtain the matrix weight control accuracy and modulation speed parameters.

[0009] Furthermore, the specific method by which the parameter extractor and analyzer obtains the matrix weight control accuracy and modulation speed parameters through the relationship graph includes: Based on the relationship between the normalized matrix weight values ​​and time, the total number of different data bits contained in the actual constructed matrix weight values ​​and the time interval between two data bits are obtained. Let N be the actual number of different data bits obtained. According to the relationship between the number of data bits and the precision of matrix weight control: N=2 n +1, and we get n, where n is the matrix weight control precision, in bits; The control rate of the matrix weights is the rate of one data bit, which is the reciprocal of the time T0 occupied by one data bit: 1 / T 0。

[0010] Furthermore, the environmental compensation and closed-loop control module includes: a temperature sensor and an algorithm and controller, wherein: The temperature sensor is used to monitor the ambient temperature of the optical computing chip and transmit the data to the algorithm and controller. Algorithms and controllers are used to compensate for the bias voltage of the tunable optical device array in an optical computing chip based on temperature changes, thereby achieving precise control of matrix weights.

[0011] Furthermore, the device also includes: The tunable light source and polarization control module is used to provide the optical signal required for the accuracy measurement of matrix weight control in optical computing chips, and to control the polarization mode of the optical signal. A waveguide network and a tunable optical device array module, wherein the waveguide network is used to split the optical signal and feed it into the optical computing chip, and the tunable optical device array is used to construct the matrix weights; A multi-channel programmable voltage source module is used to provide bias voltage parameters and generate voltage signals with different rates and bit numbers, which are then loaded onto the tunable optical device array module of the optical computing chip to construct matrix weights. The optical power control module is used to regulate the power of the optical signal output by the optical computing chip before inputting it into the photodetector array module.

[0012] The method for measuring and calibrating the matrix weight control accuracy of optical computing chips using the above-mentioned device mainly includes the following steps: A multi-channel programmable voltage source outputs voltage signals with different rates and bit depths through programming, which are then applied to a dimmable device. The optical computing chip generates the corresponding matrix weights; The optical signal information output by the optical computing chip is acquired by a real-time oscilloscope after power control and photoelectric conversion. The normalization process of the acquired electrical signals yields the matrix weight control accuracy and control rate of the optical computing chip. The temperature sensor on the optical computing chip is used to monitor environmental parameters in real time, establish the relationship between temperature and optical parameters, feed the error back to the drive circuit, and dynamically adjust the bias voltage of the tunable device to form a closed-loop control.

[0013] Compared with the prior art, the beneficial effects of this disclosure are: ① It can realize the accurate measurement of matrix weight control precision and control rate; ② It has environmental compensation and control functions, accurately senses the changing parameters of the external environment, and realizes the change of voltage parameters in weight control through compensation algorithm, so as to ensure that the matrix weight control precision of the optical computing chip is within a suitable range; ③ It effectively guarantees the high performance and reliable operation of the optical computing chip, which is conducive to the application of optical computing chips in high-speed data processing in the field of artificial intelligence. Attached Figure Description

[0014] The above and other objects, features and advantages of this disclosure will become more apparent from the more detailed description of exemplary embodiments of this disclosure taken in conjunction with the accompanying drawings, in which the same reference numerals generally represent the same components.

[0015] Figure 1 This is a schematic diagram showing the module composition and connection relationship of the optical computing chip matrix weight control accuracy measurement and calibration device according to the present disclosure; Figure 2 This is a schematic diagram of the target matrix weights of an exemplary 2-bit control precision optical computing chip; Figure 3 This is a schematic diagram of the actual matrix weights of an exemplary 2-bit control precision optical computing chip; Figure 4 This is a schematic diagram of the target matrix weights of an exemplary 3-bit control precision optical computing chip; Figure 5 This is a schematic diagram of the actual matrix weights of an exemplary 3-bit control precision optical computing chip; Figure 6 This is a flowchart of a measurement and calibration process for the matrix weight control accuracy of an optical computing chip according to the present disclosure. Detailed Implementation

[0016] Preferred embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0017] This disclosure provides a device and method for measuring the accuracy of matrix weight control in optical computing chips.

[0018] In one exemplary embodiment, the composition and connection relationship of each module in the optical computing chip matrix weight control accuracy measurement and calibration device according to this disclosure are as follows: Figure 1 As shown in the attached diagram, the signal processing procedures between the modules are as follows. Figure 6 As shown. Mainly includes: The tunable light source and polarization control module 601 are used to provide the optical signal required for the precision measurement of matrix weight control in the optical computing chip and to control the polarization mode of the optical signal; wherein, the tunable light source is used to generate the optical signal, and the polarization controller realizes the control of the polarization mode of the optical signal.

[0019] The waveguide network and tunable optical device array module 602 are used to split and feed optical signals into the optical computing chip to construct matrix weights. The waveguide network distributes the beam to the branches of each tunable optical device with equal power. The tunable optical device array in the optical computing chip is an interference array based on thermo-optical or electro-optical effects, integrating heating or driving electrodes, and supporting bias voltage adjustment. It is used to construct the optical computing chip matrix, and controlling the bias voltage can be used to configure the matrix weights.

[0020] The multi-channel programmable voltage source module 603 is used to provide the voltage required for constructing the matrix weights of the tunable optical device array. By adjusting the bias voltage parameters, voltage signals with different rates and bit depths are generated and applied to the tunable optical device array module of the optical computing chip to construct the matrix weights.

[0021] The optical power control module 604 is used to regulate the power of the optical signal output by the optical computing chip. The module mainly includes an erbium-parameter fiber amplifier and an optical attenuator. The erbium-parameter fiber amplifier amplifies the power of the optical signal output by the optical computing chip; the optical attenuator appropriately attenuates the amplified optical signal to ensure a suitable optical power intensity is input to the photodetector.

[0022] The photodetector array module 605 is used to convert optical signals into electrical signals and feed the electrical signals into the parameter extraction and analysis module.

[0023] The parameter extraction and analysis module 606 is used to acquire electrical signals, calculate the matrix weight control accuracy and control rate of the optical computing chip, and output the corresponding results. It consists of a real-time oscilloscope and a parameter extractor / analyzer. The real-time oscilloscope is used for acquiring and receiving electrical signals to obtain the data corresponding to the matrix weight parameters; the parameter extractor / analyzer is used for analyzing and processing the electrical signals to calculate the accuracy of the matrix weight control and the modulation rate.

[0024] The environmental compensation and closed-loop control module 607 includes a temperature sensor, an algorithm and control module. The temperature sensor is used to monitor the ambient temperature of the optical computing chip in real time and convert the temperature information into electrical signal change information and transmit it to the algorithm and control module. The algorithm and control module outputs the calibrated voltage parameters to the programmable voltage source through the environmental compensation algorithm and the closed-loop control algorithm, so as to realize the precise control of the matrix weights and improve the control accuracy of the matrix weights of the optical computing chip.

[0025] Application Examples According to the present disclosure, a device for measuring and calibrating the accuracy of matrix weight control in an optical computing chip is provided, as shown in the attached figure. Figure 1 As shown.

[0026] Figure 1The central location is the 4×4 optical computing chip under test; 4×4 indicates the number of signal input and output ports on the chip. The light gray boxes within the optical computing chip represent tunable optical devices, and the chip itself consists of an optical array composed of multiple tunable optical devices. Specifically: The optical signal required for the precision measurement of matrix weight control in optical computing chips is provided by a tunable laser, and a polarization controller is used to adjust the polarization of the optical signal. Based on the number of channels in the optical computing chip, a suitable beam splitter is selected to divide the power of the optical signal into multiple equal paths and feed them into the optical computing chip. A multi-channel programmable voltage source is used to provide the voltage parameters required for constructing matrix weights in tunable devices; Erbium-doped fiber amplifiers are used to amplify the power of the optical signal output from the optical computing chip; optical attenuators are used to appropriately attenuate the power of the amplified optical signal to match the photodetector at the receiving end. Photodetectors are based on the photoelectric effect generated by the interaction of light radiation and matter, which converts received light signals into electrical signals through photoelectric conversion. Real-time oscilloscopes are used to acquire electrical signals and extract relevant data; The parameter extraction and analysis device is used to process the data to obtain the relationship between the normalized weights of the matrix and time. The relationship graph is then used to further obtain the matrix weight control accuracy and modulation speed parameters.

[0027] In one exemplary embodiment, a method for accurately measuring the matrix weight control accuracy and matrix weight control rate of an optical computing chip according to the present disclosure mainly includes the following steps: (1) The multi-channel programmable voltage source outputs voltage signals with different rates and bit numbers through programming and loads them into the dimmable device; (2) The optical computing chip generates the corresponding matrix weights. (3) The output optical signal information is collected by a real-time oscilloscope after power control and photoelectric conversion for processing and analysis of matrix weight control accuracy.

[0028] The processed data result graph is as follows: Figure 2 and Figure 3 As shown. Figure 2 This is a schematic diagram of the weights of the normalized target matrix; Figure 3 This is a schematic diagram of the normalized actual matrix weights.

[0029] Based on the two diagrams, it can be concluded that due to differences in device performance and environmental factors, the actual constructed matrix weights may not be exactly the same as the target matrix weights. Observation shows that the data sequence in the diagram contains five different data bits: 0.00, 0.50, 0.75, 0.25, 1.00, 0.50, 0.75, 1.00, and 0.50, with a time interval of 0.5 milliseconds between each bit. This means that the random data sequence in the diagram is randomly selected and arranged from the five data bits: 0.00, 0.25, 0.50, 0.75, and 1.00.

[0030] like Figure 4 and Figure 5 This is a schematic diagram showing the weights of another set of normalized target matrices. Figure 4 This is a schematic diagram of the weights of the normalized target matrix; Figure 5 This is a schematic diagram illustrating the normalized actual matrix weights. The actual constructed matrix weights differ slightly from the target matrix weights. Observation shows that the data sequence in the diagram consists of 0.00, 0.125, 0.375, 0.25, 0.50, 0.00, 0.625, 0.75, 1.00, 0.875, 0.75, and 0.375, containing nine different data bits, with a time interval of 0.125 milliseconds between each data bit. This means that the random data sequence in the diagram is data randomly selected and arranged from the nine data bits: "0.00, 0.125, 0.25, 0.375, 0.5, 0.625, 0.75, 0.875, 1".

[0031] The two sets of results show that the relationship between the number of data bits and the precision of matrix weight control satisfies "N=2". n +1”, where N represents the number of data bits and n represents the precision of the n-bit matrix weight control. Figure 2 and Figure 3 The corresponding matrix weight control precision is 2 bits; Figure 4 and Figure 5 The corresponding matrix weight control precision is 3 bits.

[0032] The control rate of the matrix weights is the rate of one data bit, which is the reciprocal of the time T0 occupied by one data bit, 1 / T0. For example, Figure 2 and Figure 3 The time occupied by one data bit is 0.25 milliseconds (ms), and the corresponding control rate is calculated as: 1000ms / 0.5ms = 2000Hz (Hertz). Figure 4 and Figure 5 The time occupied by one data bit is 0.125 milliseconds (ms), and the corresponding control rate is calculated as: 1000ms / 0.125ms = 8000Hz (Hertz).

[0033] The above methods can be used to accurately measure the matrix weight control precision and matrix weight control rate of optical computing chips.

[0034] (4) Environmental compensation and closed-loop control process By using an on-chip high-precision temperature sensor to monitor environmental parameters in real time, an adaptive algorithm is used to establish the relationship between temperature and optical parameters, and the error is fed back to the drive circuit to dynamically adjust the bias voltage of the optical tuning device, thus forming a closed-loop control.

[0035] Specifically described as follows: A high-precision temperature sensor is placed adjacent to the optical computing chip, directly monitoring the real-time temperature of the chip area. It continuously collects temperature data to form a real-time temperature sequence, which serves as the raw input for subsequent control. The temperature data itself cannot be directly used for control; it needs to be mapped to optical parameters using an adaptive algorithm. The model can be divided into two phases: a training phase and a real-time execution phase.

[0036] During the training phase, temperature-optical parameter samples were collected at different temperatures. Within the operating temperature range, the temperature was recorded every 0.5°C. Simultaneously, optical parameters such as the corresponding center wavelength and output optical power were measured using instruments such as a spectrometer and optical power meter. Based on the sample data, the mapping pattern was learned, and a model for real-time use was generated through a neural network.

[0037] During real-time operation, the algorithm receives the real-time temperature and inputs it into the model to calculate the ideal optical parameters that the optical computing chip should achieve at that temperature. Simultaneously, it acquires the actual optical parameters and calculates the deviation to generate an error signal. This error signal is then processed through filtering and normalization to generate reliable and stable error commands.

[0038] The error command signal is fed back to the internal drive circuit of the controller. The drive circuit calculates the amount of voltage compensation to be applied based on the magnitude and direction of the error command, and converts it into a bias voltage adjustment value. The bias voltage adjustment value is input to the multi-channel programmable voltage source, and the multi-channel programmable voltage source outputs the changed voltage value to the optical computing chip's processing unit, thereby changing the optical characteristics of the processing unit.

[0039] After adjustment, the temperature sensor and photodetector will re-collect temperature and optical parameters, the adaptive algorithm will recalculate the error, and the drive circuit will readjust the bias voltage to form a closed loop until the optical parameters stabilize within the ideal range.

[0040] The above technical solutions are merely exemplary embodiments of the present invention. For those skilled in the art, based on the application methods and principles disclosed in the present invention, it is easy to make various types of improvements or modifications, and not limited to the methods described in the specific embodiments of the present invention. Therefore, the methods described above are merely preferred and not restrictive.

Claims

1. A device for measuring and calibrating the accuracy of matrix weight control in optical computing chips, characterized in that, include: The photodetector array module is used to convert the optical signal output by the optical computing chip into a photoelectric signal and feed the electrical signal into the parameter extraction and analysis module. The parameter extraction and analysis module is used to acquire electrical signals, calculate the matrix weight control accuracy and control rate of the optical computing chip, and output the corresponding results. The environmental compensation and closed-loop control module is used to monitor the ambient temperature of the optical computing chip and, based on the temperature change, to compensate and calibrate the multi-channel voltage required for the optical computing chip to construct matrix weights, thereby achieving precise control of matrix weights. The specific methods of the environmental compensation and closed-loop control module include: By using a temperature sensor located close to the optical computing chip, real-time temperature sequences are obtained and used as the raw input for regulation; A mapping model is established between an adaptive algorithm and optical parameters. The model consists of two phases: a training phase and a real-time execution phase. During the training phase, samples of "temperature-optical parameters" at different temperatures are collected. Based on the sample data, the mapping pattern is learned, and a model that can be called in real time is generated through a neural network. During the real-time operation phase, the real-time temperature is received and substituted into the model to calculate the ideal optical parameters that the optical computing chip should achieve at the current temperature. At the same time, the actual optical parameters are collected and the deviation is calculated to generate an error command signal. Based on the error command signal, the amount of voltage compensation to be applied is calculated and converted into a bias voltage adjustment value; By adjusting the bias voltage, the voltage applied to the optical computing chip's processing unit is changed, thereby altering the optical characteristics of the processing unit. After adjustment, temperature and optical parameters are re-acquired, the adaptive algorithm recalculates the error, and the drive circuit readjusts the bias voltage to form a closed loop until the optical parameters stabilize within the ideal range.

2. The apparatus according to claim 1, characterized in that, The parameter extraction and analysis module includes a real-time oscilloscope and a parameter extraction and analysis unit, wherein: A real-time oscilloscope is used to acquire electrical signals and extract relevant data. The parameter extractor and analyzer is used to process the data and obtain a graph showing the relationship between the normalized matrix weights and time. The graph is used to obtain the matrix weight control accuracy and modulation speed parameters.

3. The apparatus according to claim 2, characterized in that, The specific methods used by the parameter extraction and analyzer to obtain the matrix weight control accuracy and modulation speed parameters through the relationship diagram include: Based on the relationship between the normalized matrix weight values ​​and time, the total number of different data bits contained in the actual constructed matrix weight values ​​and the time interval between two data bits are obtained. Let N be the actual number of different data bits obtained. According to the relationship between the number of data bits and the precision of matrix weight control: N=2 n +1, and we get n, where n is the matrix weight control precision, in bits; The control rate of the matrix weights is the rate of one data bit, which is the reciprocal of the time T0 occupied by one data bit: 1 / T 0。 4. The apparatus according to claim 1, characterized in that, The environmental compensation and closed-loop control module includes: a temperature sensor and an algorithm and controller, wherein: The temperature sensor is used to monitor the ambient temperature of the optical computing chip and transmit the data to the algorithm and controller. Algorithms and controllers are used to compensate for the bias voltage of the tunable optical device array in an optical computing chip based on temperature changes, thereby achieving precise control of matrix weights.

5. The apparatus according to any one of claims 1-4, characterized in that, Also includes: The tunable light source and polarization control module is used to provide the optical signal required for the accuracy measurement of matrix weight control in optical computing chips, and to control the polarization mode of the optical signal. A waveguide network and a tunable optical device array module, wherein the waveguide network is used to split the optical signal and feed it into the optical computing chip, and the tunable optical device array is used to construct the matrix weights; A multi-channel programmable voltage source module is used to provide bias voltage parameters and generate voltage signals with different rates and bit numbers to be loaded onto the tunable optical device array module of the optical computing chip to construct matrix weights. The optical power control module is used to regulate the power of the optical signal output by the optical computing chip before inputting it into the photodetector array module.

6. A method for measuring and calibrating the matrix weight control accuracy of an optical computing chip using the device described in claim 5, characterized in that, Includes the following steps: A multi-channel programmable voltage source outputs voltage signals with different rates and bit depths through programming, which are then applied to a dimmable device. The optical computing chip generates the corresponding matrix weights; The optical signal information output by the optical computing chip is acquired by a real-time oscilloscope after power control and photoelectric conversion. The normalization process of the acquired electrical signals yields the matrix weight control accuracy and control rate of the optical computing chip. The temperature sensor on the optical computing chip is used to monitor environmental parameters in real time, establish the relationship between temperature and optical parameters, feed the error back to the drive circuit, and dynamically adjust the bias voltage of the tunable optical device to form a closed-loop control.

Images