A data processing method and system based on an artificial intelligence chip

Abstract

The present invention discloses a data processing method and system based on an artificial intelligence chip, relating to the field of artificial intelligence technology. The method includes: acquiring original data to be processed, performing preprocessing to extract feature vectors, and converting the feature vectors into sparse pulse sequences through a spatio-temporal encoder; a photon pulse stream transmitted through the shortest optical path triggers a memristive array through optoelectronic conversion, and performs multiply-accumulate operations in the memristive cells based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule to generate a final pulse sequence; performing error code detection on the final pulse sequence, and when it is detected that the error rate exceeds a set threshold, switching to a standby nanowire electronic link for parity bit transmission and retransmission request, and outputting a verification result; the present invention ensures the integrity and accuracy of data by performing error code detection on the final pulse sequence and switching to a standby nanowire electronic link for parity bit transmission and retransmission request when necessary.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a data processing method and system based on an artificial intelligence chip. Background Art

[0002] With the development of artificial intelligence technology, data processing methods have become increasingly complex and diverse. Current processing methods mostly rely on traditional computing architectures, which face many challenges when processing large-scale image and speech stream data. Traditional methods usually involve complex feature extraction processes and use CPUs or GPUs for subsequent data processing, which not only consumes a large amount of computing resources but also has problems with processing latency. In recent years, technologies such as convolutional neural networks (CNNs) and Mel-frequency cepstral coefficients (MFCCs) have been widely used in image and speech data processing, but they are often limited by hardware performance and algorithm efficiency, and there is still room for improvement in terms of real-time performance and accuracy.

[0003] Although the existing technologies perform well in certain specific application scenarios, there are still deficiencies in processing high-priority tasks. For example, in the field of autonomous driving, the requirement for real-time performance is extremely high, and any delay may lead to serious consequences. In addition, the existing technologies lack an effective error detection mechanism during data transmission, which easily leads to data loss or errors and affects the accuracy of the final result. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a data processing method based on an artificial intelligence chip to solve the balance problem between the real-time performance of high-priority tasks and the accuracy of data transmission.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a data processing method based on an artificial intelligence chip, which includes,

[0008] Obtaining the original data to be processed, performing preprocessing to extract feature vectors, converting the feature vectors into sparse pulse sequences through a spatio-temporal encoder, and attaching priority tags to the sparse pulse sequences according to the real-time requirements of the tasks;

[0009] Based on the sparse pulse sequences with attached priority tags, using a dynamic routing algorithm to allocate them to the corresponding wavelengths for transmission, and using time-division multiplexing and wavelength-division multiplexing technologies to calculate the shortest optical path;

[0010] The photon pulse stream transmitted through the shortest optical path triggers the memristor array after optoelectronic conversion, performs multiplication and accumulation operations in the memristor unit based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule to generate the final pulse sequence;

[0011] Error code detection is performed on the final pulse sequence. When the detected error rate exceeds the set threshold, the system switches to the standby nanowire electronic link for parity bit transmission and retransmission requests, and outputs the verification result.

[0012] As a preferred solution of the data processing method based on the artificial intelligence chip of the present invention, wherein: the original data to be processed includes image data and voice stream data;

[0013] For the image data, convolution operation is performed using a convolution kernel to identify edge and texture features in the image, and a high-dimensional feature vector is obtained;

[0014] For the voice stream data, it is processed through a Mel frequency cepstral coefficient filter bank to obtain an MFCC feature vector;

[0015] The high-dimensional feature vector in the image data and the MFCC feature vector in the voice stream are converted into feature vectors suitable for pulse coding processing.

[0016] As a preferred solution of the data processing method based on the artificial intelligence chip of the present invention, wherein: converting the feature vector into a sparse pulse sequence through a spatio-temporal encoder includes the following steps,

[0017] For each eigenvalue of the feature suitable for pulse coding processing, according to the position of each eigenvalue in the feature vector, the feature vector is converted into a sparse pulse sequence by applying the spatio-temporal coding formula, and the expression is:

[0018] ;

[0019] Wherein, is the sparse pulse sequence of time is the current time point, is the eigenvalue index in the feature vector, is the number of features in the feature vector, represents the step function, is the th eigenvalue in the feature vector, is the feature scaling factor, is the past time point relative to the current time is the time decay constant, is the threshold for controlling the frequency of pulse triggering. ​​

[0020] As a preferred solution of the data processing method based on an artificial intelligence chip according to the present invention, wherein: attaching a priority tag to the sparse pulse sequence according to the real-time requirement of the task includes the following steps,

[0021] Define the application scenarios of the artificial intelligence chip, and set priority levels for different application scenarios;

[0022] Based on the converted sparse pulse sequence, determine the task type corresponding to the sparse pulse sequence;

[0023] According to the priority level corresponding to the task type, convert the corresponding task into a binary tag of the corresponding priority level;

[0024] Insert the converted binary tag of the priority level into the head of the sparse pulse sequence to generate a new sparse pulse sequence.

[0025] As a preferred solution of the data processing method based on an artificial intelligence chip according to the present invention, wherein: based on the sparse pulse sequence with an attached priority tag, use a dynamic routing algorithm to allocate it to the corresponding wavelength for transmission, and use time-division multiplexing and wavelength-division multiplexing technologies to calculate the shortest optical path, including the following steps,

[0026] According to the binary priority tag at the head of the new sparse pulse sequence, determine the priority level of each pulse stream;

[0027] Divide the available time resources into multiple time periods of a fixed length. For each pulse stream to be transmitted, determine the corresponding time period according to the priority level, and combine the wavelength-division multiplexing technology to perform wavelength allocation on optical signals of different wavelengths on the same optical fiber;

[0028] During the operation of the network, continuously monitor the load conditions on each wavelength, and adjust the pulse stream allocation on each wavelength in real time according to the current network load conditions;

[0029] Construct a directed weighted graph based on the actual physical layout and logical connection relationship, and determine the starting node where the pulse stream is located and the end node corresponding to the target computing module;

[0030] Create a distance array, a predecessor node array, and a priority queue, and calculate the distances from all nodes to the starting node, and add them to the priority queue;

[0031] Take out the node with the smallest distance from the priority queue, and calculate the distance from the node with the smallest distance to its adjacent node. When the distance from the smallest node to its adjacent node is less than the smallest distance taken out from the priority queue, update the distance of the adjacent node until the path to the end point is found;

[0032] The predecessor node array traces back the path to the end point in reverse to obtain the shortest optical path from the starting point to the end point.

[0033] As a preferred embodiment of the data processing method based on an artificial intelligence chip according to the present invention, wherein: the photon pulse stream transmitted through the shortest optical path passes through optoelectronic conversion to trigger the memristor array, and performs a multiply-accumulate operation in the memristive unit based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule. Generating the final pulse sequence includes the following steps.

[0034] Each photon pulse in the pulse stream is absorbed by a photodetector during transmission along the shortest optical path, and a corresponding electrical signal is generated according to the intensity of the photon pulse.

[0035] The electrical signal activates the memristor array, and each memristive unit in the memristor array performs a multiply-accumulate operation according to its conductance value to obtain the sum of the output currents of all memristive units.

[0036] Based on the STDP rule, when a new pulse stream event occurs, the weight change amount is calculated according to the time difference between the front and rear pulses, and the conductance value of the corresponding memristive unit is updated.

[0037] A voltage threshold is defined. When the sum of the output currents of all memristive units exceeds the voltage threshold, a new pulse output is triggered to generate the final pulse sequence.

[0038] As a preferred embodiment of the data processing method based on an artificial intelligence chip according to the present invention, wherein: error code detection is performed on the final pulse sequence. When the detected error rate exceeds the set threshold, the system switches to a standby nanowire electronic link for parity bit transmission and retransmission request. Outputting the verification result includes the following steps.

[0039] Extract the timestamp of each photon pulse from the final pulse sequence and calculate the time interval between each pair of adjacent photon pulses.

[0040] Perform a bitwise exclusive OR operation on all the time intervals to obtain the final hash value.

[0041] Compare the final hash value with the reference hash value. When the two are inconsistent, it indicates that there is an error code.

[0042] Calculate the error rate based on the number of detected error bits and the total number of bits.

[0043] Set an error rate threshold. When the calculated error rate exceeds the error rate threshold, immediately switch to the standby nanowire electronic link for retransmission request and redundant parity bit transmission.

[0044] Receive the retransmitted data packet and redundant parity bit, and perform secondary error code detection to confirm whether the retransmission is successful.

[0045] In a second aspect, the present invention provides a data processing system based on an artificial intelligence chip, including:

[0046] A data processing module that acquires the raw data to be processed, performs preprocessing to extract feature vectors, converts the feature vectors into sparse pulse sequences through a spatio-temporal encoder, and attaches priority tags to the sparse pulse sequences according to the real-time requirements of the task;

[0047] A routing and allocation module that, based on the sparse pulse sequences with attached priority tags, uses a dynamic routing algorithm to allocate them to corresponding wavelengths for transmission, and calculates the shortest optical path using time-division multiplexing and wavelength-division multiplexing technologies;

[0048] An optoelectronic computing module where the photon pulse stream transmitted through the shortest optical path triggers a memristive array through optoelectronic conversion, performs multiply-accumulate operations in the memristive cells based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule to generate the final pulse sequence;

[0049] An error code verification module that performs error code detection on the final pulse sequence. When the detected error rate exceeds the set threshold, it switches to a standby nanowire electronic link for parity bit transmission and retransmission requests, and outputs the verification result.

[0050] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, it implements any step of the data processing method based on an artificial intelligence chip as described in the first aspect of the present invention.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, it implements any step of the data processing method based on an artificial intelligence chip as described in the first aspect of the present invention.

[0052] The beneficial effects of the present invention are as follows: Through efficient feature extraction and spatio-temporal encoding conversion, the raw data is quickly converted into sparse pulse sequences suitable for pulse coding processing, and priority tags are attached according to task requirements to ensure that critical tasks are processed first; By using a dynamic routing algorithm combined with time-division multiplexing and wavelength-division multiplexing technologies, the data transmission path is optimized, improving the transmission efficiency and reliability; By activating the memristive array through optoelectronic conversion and performing multiply-accumulate operations, and online adjusting the conductance value using the STDP rule to generate the final pulse sequence, the computing accuracy and speed are further improved; By performing error code detection on the final pulse sequence and switching to a standby nanowire electronic link for parity bit transmission and retransmission requests when necessary, the integrity and accuracy of the data are ensured. Description of the Drawings

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0054] Figure 1 It is a flowchart of the data processing method based on the artificial intelligence chip in Embodiment 1.

[0055] Figure 2 It is a system diagram of the data processing system based on the artificial intelligence chip in Embodiment 1. Detailed Embodiments

[0056] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the drawings in the specification.

[0057] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0058] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0059] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a data processing method based on an artificial intelligence chip, including the following steps:

[0060] S1. Obtain the original data to be processed, perform preprocessing to extract feature vectors, convert the feature vectors into sparse pulse sequences through a spatio-temporal encoder, and attach priority tags to the sparse pulse sequences according to the real-time requirements of the task.

[0061] S1.1. The original data to be processed includes image data and voice stream data.

[0062] S1.1.1 For image data, perform a convolution operation using a convolution kernel to identify edge and texture features in the image, obtaining a high-dimensional feature vector; for speech stream data, process it through a Mel Frequency Cepstral Coefficient (MFCC) filter bank to obtain an MFCC feature vector; convert the high-dimensional feature vector in the image data and the MFCC feature vector in the speech stream into feature vectors suitable for pulse coding processing.

[0063] Specifically, the convolution operation refers to selecting a set of convolution kernels (such as the Sobel operator or a custom-learned convolution kernel), and each convolution kernel focuses on detecting edges or other features in a specific direction.

[0064] The specific steps for processing the speech stream data through the Mel Frequency Cepstral Coefficient filter bank include:

[0065] Framing and windowing: Divide the continuous speech stream data into short-time frames of 20 - 40 ms, and apply a Hamming window to each frame to reduce spectral leakage.

[0066] Fast Fourier Transform (FFT): Perform a fast Fourier transform on each frame to obtain a spectrogram.

[0067] Mel filtering: Use a set of triangular filter banks to filter the spectrogram, simulating the human ear's perception of sounds at different frequencies.

[0068] Taking logarithm and DCT: Take the logarithm of the filtered result and apply DCT to obtain the final MFCC feature vector.

[0069] S1.2. For each eigenvalue of the features suitable for pulse coding processing, according to the position of each eigenvalue in the feature vector, apply a spatio-temporal coding formula to convert the feature vector into a sparse pulse sequence, with the expression:

[0070] ;

[0071] where, is the sparse pulse sequence at time , is the current time point, is the eigenvalue index in the feature vector, is the number of features in the feature vector, represents the step function, is the th eigenvalue in the feature vector, is the feature scaling factor, is the past time point relative to the current time , is the time decay constant, is the pulse trigger threshold that controls the frequency of pulse triggering.

[0072] Furthermore, the step function is used to determine whether to generate a pulse at a given time point, and the specific expression is:

[0073] ;

[0074] where, represents the result after applying the step function operation to the input variable , represents the input variable, that is, the difference between the integration result and the pulse trigger threshold , which determines whether to trigger a pulse at the time point . If the integration result minus the pulse trigger threshold is greater than 0, then output 1 (i.e., trigger a pulse); otherwise output 0 (do not trigger a pulse). means that if the value of is less than or equal to 0, the output of the step function is 0, which indicates that at the current time point , the accumulated eigenvalue does not reach the standard for triggering a pulse, so no pulse will be generated.

[0075] It should be noted that the features suitable for pulse coding processing usually refer to those features that can effectively represent and transmit key information through pulse coding. The following are several common feature selection criteria:

[0076] Significance: The eigenvalue should have a high degree of discrimination and be able to highlight important patterns or structures in the data. For example, in image processing, edge and texture features are usually significant.

[0077] Stability: The eigenvalue should be consistent and stable among different samples to avoid noise interference. For example, MFCC features are relatively stable in speech recognition tasks.

[0078] Interpretability: The eigenvalue should have physical meaning or be easy to interpret for subsequent analysis and understanding. For example, the high-dimensional feature vectors extracted by a convolutional neural network (CNN) can intuitively reflect the edge and texture information in an image.

[0079] S1.3. Attach priority labels to the sparse pulse sequence according to the real-time requirements of the task.

[0080] S1.3.1. Clarify the application scenarios of the AI chip, for example, and set priority levels for different application scenarios.

[0081] Specifically, the application scenarios include autonomous driving, speech recognition, and image classification, etc.

[0082] The setting of the priority level is represented by a 3-bit (i.e., 0 - 7) label system to indicate the priority:

[0083] Autopilot system: Since it has the highest real-time requirement, its priority is set to 7.

[0084] Speech recognition system: Compared with the autopilot, its real-time requirement is lower, but it still needs to respond quickly. Therefore, its priority is set to 5.

[0085] Image classification system: Compared with the previous two, its real-time requirement is the lowest. Therefore, its priority is set to 3.

[0086] S1.3.2. Based on the converted sparse pulse sequence, determine the task type corresponding to the sparse pulse sequence; according to the priority level corresponding to the task type, convert the corresponding task into a binary label of the corresponding priority level; insert the converted binary label of the priority level into the head of the sparse pulse sequence to generate a new sparse pulse sequence.

[0087] For example, when the task type is autopilot, the priority 7 of the autopilot is converted into the binary label 111. When the task type is speech recognition, the priority 5 of the speech recognition is converted into the binary label 101. When the task type is image classification, the priority 3 of the image classification is converted into the binary label 011.

[0088] S2. Based on the sparse pulse sequence with an additional priority label, use the dynamic routing algorithm to allocate it to the corresponding wavelength for transmission, and use time-division multiplexing and wavelength-division multiplexing technologies to calculate the shortest optical path.

[0089] S2.1. According to the binary priority label at the head of the new sparse pulse sequence, determine the priority level of each pulse stream; divide the available time resources into multiple time periods of fixed length. For each pulse stream to be transmitted, determine the corresponding time period according to the priority level, and combine the wavelength-division multiplexing technology to allocate wavelengths for optical signals of different wavelengths on the same optical fiber.

[0090] Specifically, for high-priority data streams (such as priority ≥ 5), they are allocated to low-loss wavelengths, such as 1550 nm. This wavelength supports a bandwidth of up to 10 Tbps and is suitable for tasks that require quick response.

[0091] For low-priority data streams (such as priority < 5), they are allocated to higher-loss wavelengths, such as 1310 nm. Although the bandwidth supported by this wavelength is lower (about 5 Tbps), it is sufficient for tasks with less strict real-time requirements.

[0092] S2.2. During the operation of the network, continuously monitor the load conditions on each wavelength, and adjust the pulse stream allocation on each wavelength in real time according to the current network load conditions.

[0093] Specifically, for newly arrived high-priority data streams, even if their priority is ≥5, but if all high-bandwidth wavelengths are fully loaded, sub-optimal low-loss wavelengths can be selected and lower-priority data streams can be attempted to be rescheduled to other time slots or wavelengths.

[0094] For low-priority data streams, if the wavelength currently in use is congested, its transmission time can be delayed until an idle wavelength is available.

[0095] S2.3. Construct a directed weighted graph based on the actual physical layout and logical connection relationships, and determine the starting nodes where the pulse streams with allocated wavelengths and time periods are located and the end nodes corresponding to the target computing modules; create a distance array, a predecessor node array, and a priority queue, and calculate the distances from all nodes to the starting node and add them to the priority queue; take out the node with the minimum distance from the priority queue, and calculate the distance from the node with the minimum distance to its adjacent nodes. When the distance from the minimum node to the adjacent node is less than the minimum distance taken out from the priority queue, update the distance of the adjacent node until the path to the end point is found; the predecessor node array performs reverse tracing on the path to the end point to obtain the shortest optical path from the starting point to the end point.

[0096] S3. The photon pulse stream transmitted through the shortest optical path triggers the memristor array through optoelectronic conversion, performs multiply-accumulate operations in the memristor cells based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule to generate the final pulse sequence.

[0097] S3.1. Each photon pulse in the pulse stream is absorbed by a photodetector during transmission along the shortest optical path, and a corresponding electrical signal is generated according to the intensity of the photon pulse; the electrical signal activates the memristor array, and each memristor cell in the memristor array performs a multiply-accumulate operation according to its conductance value to obtain the sum of the output currents of all memristor cells. The expression is:

[0098] ;

[0099] where, represents the sum of the output currents of all memristor cells, represents the conductance value of the th memristor cell, represents the th input voltage of the memristor cell.

[0100] S3.2. Based on the STDP rule, when a new pulse stream event occurs, calculate the weight change amount according to the time difference between the front and back pulses, and update the conductance value of the corresponding memristor cell.

[0101] STDP is a learning rule used to simulate the synaptic plasticity of biological neurons. It adjusts the synaptic weights (manifested as conductance values in memristors) based on the time difference between the firing (i.e., generating pulses) of the presynaptic neuron and the postsynaptic neuron. This mechanism enables the neural network to self-adjust according to experience, thereby achieving learning and memory functions.

[0102] The expression for the weight change is:

[0103] ;

[0104] Where, represents the change in the conductance value between the th memristive cell and the th memristive cell, represents the learning rate parameter, which controls the amplitude of each weight adjustment. A higher learning rate means a larger weight change, while a lower learning rate means a smaller weight change. represents the time difference between adjacent pulses, and represent positive and negative time constants.

[0105] S3.3. Define the voltage threshold. When the sum of the output currents of all memristive cells exceeds the voltage threshold, a new pulse output is triggered to generate the final pulse sequence.

[0106] Specifically, the final pulse sequence is different from the sparse pulse sequence. The sparse pulse sequence is a form of feature vector after being transformed by the spatio-temporal encoding formula, mainly used to represent data features, and is attached with priority tags according to task requirements. The new pulse sequence refers to the new pulse output triggered based on the calculation results of the memristive array, which is used for subsequent processing or feedback signals.

[0107] S4. Perform error detection on the final pulse sequence. When the detected error rate exceeds the set error rate threshold, switch to the standby nanowire electronic link for parity bit transmission and retransmission request, and output the verification result.

[0108] S4.1. Extract the timestamps of each photon pulse from the final pulse sequence and calculate the time interval between each pair of adjacent photon pulses; perform a bitwise exclusive OR operation on all the time intervals to obtain the final hash value.

[0109] It should be noted that the hash value provides a fast and effective error detection mechanism. By comparing the consistency of the hash values, it is judged whether there are transmission errors; the bitwise exclusive OR operation has good distribution characteristics and can effectively detect bit flip errors in a small range.

[0110] S4.2. Compare the final hash value with the reference hash value. When the two are inconsistent, it indicates the presence of error codes. Calculate the bit error rate based on the detected number of error bits and the total number of bits.

[0111] Specifically, the calculation of the bit error rate refers to the ratio of the number of error bits to the total number of bits.

[0112] S4.3. Set a bit error rate threshold. When the calculated bit error rate exceeds the threshold, immediately switch to the backup nanowire electronic link to perform a retransmission request and transmit the redundancy check bits. Receive the retransmitted data packet and the redundancy check bits, and perform a secondary error detection to confirm whether the retransmission is successful.

[0113] Further explanation: After the receiving end receives the retransmitted data packet and the redundancy check bits, it re-executes the error detection process, calculates the hash value again and compares it with the reference hash value to confirm that the data is error-free.

[0114] It should be noted that through the secondary error detection, the integrity and accuracy of the data are further verified, ensuring that the data quality after retransmission meets the requirements and enhancing the overall reliability of the system.

[0115] This embodiment also provides a data processing system based on an artificial intelligence chip, including:

[0116] A data processing module that acquires the original data to be processed, performs preprocessing to extract feature vectors, converts the feature vectors into sparse pulse sequences through a spatio-temporal encoder, and attaches priority tags to the sparse pulse sequences according to the real-time requirements of the task; a routing and allocation module that, based on the sparse pulse sequences with attached priority tags, uses a dynamic routing algorithm to allocate them to the corresponding wavelengths for transmission, and calculates the shortest optical path using time-division multiplexing and wavelength-division multiplexing technologies; an optoelectronic computing module that the photon pulse stream transmitted through the shortest optical path triggers a memristor array after optoelectronic conversion, performs multiplication and accumulation operations in the memristive units based on Ohm's law and Kirchhoff's law, and uses the STDP rule to adjust the conductance value online to generate the final pulse sequence; an error checking module that performs error detection on the final pulse sequence. When the detected bit error rate exceeds the set threshold, it switches to the backup nanowire electronic link for transmission of check bits and retransmission requests, and outputs the check result.

[0117] This embodiment also provides a computer device applicable to the case of the data processing method based on an artificial intelligence chip, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the data processing method based on an artificial intelligence chip as proposed in the above embodiment.

[0118] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, carrier networks, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads provided on the housing of the computer device. It can also be an external keyboard, touchpad, or mouse, etc.

[0119] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the data processing method based on an artificial intelligence chip as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, magnetic disks, or optical discs.

[0120] In summary, through efficient feature extraction and spatio-temporal coding conversion, the present invention quickly converts the original data into a sparse pulse sequence suitable for pulse coding processing, and attaches priority tags according to task requirements to ensure that critical tasks are processed first; adopts a dynamic routing algorithm combined with time-division multiplexing and wavelength-division multiplexing technologies to optimize the data transmission path, improve the transmission efficiency and reliability; activates the memristor array through optoelectronic conversion and performs multiply-accumulate operations, and uses the STDP rule to adjust the conductance value online to generate the final pulse sequence, further improving the calculation accuracy and speed; detects the error code of the final pulse sequence, and switches to the standby nanowire electronic link for parity bit transmission and retransmission request when necessary, ensuring the integrity and accuracy of the data.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A data processing method based on an artificial intelligence chip, characterized in that: including, obtain the original data to be processed, perform preprocessing to extract feature vectors, convert the feature vectors into sparse pulse sequences through a spatio-temporal encoder, and attach priority tags to the sparse pulse sequences according to the real-time requirements of the task; Based on the sparse pulse sequences with attached priority tags, use a dynamic routing algorithm to allocate them to the corresponding wavelengths for transmission, and use time-division multiplexing and wavelength-division multiplexing technologies to calculate the shortest optical path, including the following steps, Based on the sparse pulse sequences with attached priority tags, determine the priority levels of each pulse stream; Divide the available time resources into multiple time periods of fixed length. For each pulse stream to be transmitted, determine the corresponding time period according to the priority level, and combine wavelength-division multiplexing technology to perform wavelength allocation on optical signals of different wavelengths on the same optical fiber; During the operation of the network, continuously monitor the load conditions on each wavelength, and adjust the allocation of pulse streams on each wavelength in real time according to the current network load conditions; Construct a directed weighted graph based on the actual physical layout and logical connection relationships, and determine the starting node where the pulse stream is located and the end node corresponding to the target computing module; Create a distance array, a predecessor node array, and a priority queue, and calculate the distances from all nodes to the starting node, and add them to the priority queue; Take out the node with the minimum distance from the priority queue, and calculate the distance from the node with the minimum distance to its adjacent nodes. When the distance from the node with the minimum distance to its adjacent nodes is less than the minimum distance taken out from the priority queue, update the distance of the adjacent nodes until the path to the end point is found; The predecessor node array performs backward tracking on the path to the end point to obtain the shortest optical path from the starting point to the end point; The photon pulse stream transmitted through the shortest optical path triggers the memristive array through optoelectronic conversion, performs multiply-accumulate operations in the memristive units based on Ohm's law and Kirchhoff's law, and uses the STDP rule to adjust the conductance value online to generate the final pulse sequence; Perform error code detection on the final pulse sequence. When the detected error rate exceeds the set threshold, switch to the standby nanowire electronic link for parity bit transmission and retransmission requests, and output the verification result.

2. The data processing method based on an artificial intelligence chip according to claim 1, wherein: The original data to be processed includes image data and voice stream data; For image data, perform convolution operations using convolution kernels to identify edge and texture features in the image to obtain high-dimensional feature vectors; For voice stream data, process it through a Mel-frequency cepstral coefficient filter bank to obtain MFCC feature vectors; Convert the high-dimensional feature vectors in the image data and the MFCC feature vectors in the voice stream into feature vectors suitable for pulse coding processing.

3. The data processing method based on an artificial intelligence chip according to claim 2, wherein: Converting the feature vectors into sparse pulse sequences through a spatio-temporal encoder includes the following steps, For each eigenvalue of the features suitable for pulse coding processing, according to the position of each eigenvalue in the feature vector, apply the spatio-temporal coding formula to convert the feature vector into a sparse pulse sequence, and the expression is: ； Among them, is the time of the sparse pulse sequence, is the current time point, is the eigenvalue index in the eigenvector, is the number of features in the eigenvector, represents the step function, is the th eigenvalue in the eigenvector, is the feature scaling factor, represents the past time point relative to the current time , is the time decay constant, is the threshold for controlling the frequency of pulse triggering.

4. The data processing method based on an artificial intelligence chip according to claim 3, characterized in that: Attach priority tags to the sparse pulse sequences according to the real-time requirements of the task including the following steps, clarify the application scenarios of the artificial intelligence chip, and set priority levels for different application scenarios; Based on the converted sparse pulse sequences, determine the task types corresponding to the sparse pulse sequences; Convert the corresponding task into a binary tag of the corresponding priority level according to the priority level corresponding to the task type; Insert the binary tag of the converted priority level into the head of the sparse pulse sequence to generate a sparse pulse sequence with an additional priority tag.

5. The data processing method based on an artificial intelligence chip according to claim 4, wherein: The photon pulse stream transmitted through the shortest optical path undergoes optoelectronic conversion to trigger the memristor array, performs a multiply-accumulate operation in the memristor cells based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule to generate the final pulse sequence including the following steps, Each photon pulse in the pulse stream is absorbed by a photodetector during the shortest optical path transmission, and a corresponding electrical signal is generated according to the intensity of the photon pulse; The electrical signal activates the memristor array, and each memristor cell in the memristor array performs a multiply-accumulate operation according to its conductance value to obtain the sum of the output currents of all memristor cells; Based on the STDP rule, when a new pulse stream event occurs, calculate the weight change amount according to the time difference between the front and rear pulses, and update the conductance value of the corresponding memristor cell; Define a voltage threshold. When the sum of the output currents of all memristor cells exceeds the voltage threshold, trigger a new pulse output to generate the final pulse sequence.

6. The data processing method based on an artificial intelligence chip according to claim 5, wherein: Perform error code detection on the final pulse sequence. When the detected error rate exceeds the set threshold, switch to the standby nanowire electronic link for parity bit transmission and retransmission request, and output the verification result including the following steps, Extract the timestamp of each photon pulse from the final pulse sequence and calculate the time interval between each pair of adjacent photon pulses; Perform a bitwise exclusive OR operation on all the time intervals to obtain the final hash value; Compare the final hash value with the reference hash value. When the two are inconsistent, it indicates that there is an error code; Calculate the error rate based on the number of detected error bits and the total number of bits; Set an error rate threshold. When the calculated error rate exceeds the error rate threshold, immediately switch to the standby nanowire electronic link for retransmission request and redundant parity bit transmission; Receive the retransmitted data packet and redundant parity bit, and perform secondary error code detection to confirm whether the retransmission is successful.

7. A data processing system based on an artificial intelligence chip, based on the data processing method based on an artificial intelligence chip according to any one of claims 1 to 6, characterized in that: including, A data processing module that obtains the original data to be processed, performs preprocessing to extract feature vectors, converts the feature vectors into sparse pulse sequences through a spatio-temporal encoder, and attaches a priority tag to the sparse pulse sequence according to the real-time requirements of the task; A routing allocation module that, based on the sparse pulse sequence with an additional priority tag, uses a dynamic routing algorithm to allocate to the corresponding wavelength for transmission, and uses time-division multiplexing and wavelength-division multiplexing technologies to calculate the shortest optical path; An optoelectronic computing module that the photon pulse stream transmitted through the shortest optical path undergoes optoelectronic conversion to trigger the memristor array, performs a multiply-accumulate operation in the memristor cells based on Ohm's law and Kirchhoff's law, and online adjusts the conductance value using the STDP rule to generate the final pulse sequence; An error code verification module that performs error code detection on the final pulse sequence. When the detected error rate exceeds the set threshold, switch to the standby nanowire electronic link for parity bit transmission and retransmission request, and output the verification result.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the data processing method based on the artificial intelligence chip according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the data processing method based on the artificial intelligence chip according to any one of claims 1 to 6.

Images