Self-adaptive chip parameter trimming method based on convolution operation calculation

Through convolution operation calculation and adaptive combined search algorithm, the problem that traditional chip parameter optimization methods are difficult to find the optimal parameter combination is solved, and the precise adjustment of chip performance and rapid determination of parameter combination is realized, which improves the reliability of the chip and stability.

CN118914807APending Publication Date: 2024-11-08WUXI BISTAR SEMICON CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410934606.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Traditional chip parameter optimization methods are difficult to accurately find the optimal parameter combination, resulting in chip performance not reaching the expected goals.

Method used

Adaptive chip parameter adjustment method based on convolution operation calculation is adopted, the time domain response data is decomposed into a base function through singular value decomposition, and the data is reconstructed through convolution operation, combined with the adaptive combination search algorithm to optimize parameters, determine Optimal parameter combination.

Benefits of technology

It realizes rapid analysis of the impact of chip parameters on performance, accurately evaluates the performance of the current parameter combination, ensures that the chip performance meets the design requirements, improves the reliability and stability of the product, reduces interference from human factors, and improves the accuracy of parameter adjustment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914807A_ABST
    Figure CN118914807A_ABST
Patent Text Reader

Abstract

The invention provides an adaptive chip parameter trimming method based on convolution operation calculation, and relates to the technical field of chip trimming, the method comprises the following steps: collecting time domain response data of a chip tested under different chip parameter settings; decomposing the time domain response data of the chip into a basis function by using a singular value decomposition method, and reconstructing the time domain response data through convolution operation; performing error comparison on the reconstructed time domain response data and preset standard time domain response data, optimizing chip parameters through an adaptive combined search algorithm, and determining an optimal parameter combination; and according to the determined optimal parameter combination, chip parameters are trimmed through a trimming method. According to the method, the performance of the current parameter combination can be accurately evaluated, so that the optimal parameter combination is found, the chip parameters are accurately adjusted through a trimming method, it is ensured that the performance of the chip meets the design requirement, and reliability and stability are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of chip tuning technology, and in particular to an adaptive chip parameter tuning method based on convolution operation calculation. Background Art

[0002] In the process of chip production, the actual production conditions are often difficult to fully meet expectations, so the produced chips often have varying degrees of process deviations. These deviations are not only reflected in the chip's external dimensions, but more importantly, they will directly affect the actual values ​​of the chip's key parameters, causing significant differences between them and the design expectations. This difference not only directly weakens the performance of the chip, but may also seriously affect its stability and reliability, and even cause the chip to fail to meet the performance index requirements in specific application scenarios.

[0003] In order to overcome the problems caused by these process deviations, a trimming process is usually introduced during the mass production test phase. Trimming is a complex and delicate task that aims to accurately compensate for the impact of process deviations by adjusting the adjustable parameters inside the chip, such as voltage, current, frequency, etc. This process requires a high degree of technical precision and strict process control to ensure that the performance of the trimmed chip can be as close to the design target as possible to meet the needs of users and application scenarios.

[0004] In the existing chip design and manufacturing process, parameter optimization faces many technical defects and challenges. First, because the chip contains a large number of circuits, components and interconnect structures, these elements interact with each other to form a highly complex system. Traditional parameter optimization methods are difficult to accurately find the optimal parameter combination, which will cause the chip performance to fail to achieve the expected goal.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of this, the present invention provides an adaptive chip parameter adjustment method based on convolution operation calculation to solve the problem that the traditional parameter optimization method mentioned above is difficult to accurately find the optimal parameter combination.

[0007] In order to solve the above problems, the specific technical solutions adopted by the present invention are as follows:

[0008] An adaptive chip parameter adjustment method based on convolution operation calculation, the adaptive chip parameter adjustment method based on convolution operation calculation comprises the following steps:

[0009] S1, collect the time domain response data of the chip under different chip parameter settings;

[0010] S2. Decomposing the chip's time domain response data into basis functions using a singular value decomposition method, and reconstructing the time domain response data through a convolution operation based on the basis functions and chip parameters;

[0011] S3, comparing the reconstructed time domain response data with the preset standard time domain response data, optimizing the chip parameters through an adaptive combined search algorithm, and determining the optimal parameter combination;

[0012] S4. According to the determined optimal parameter combination, the chip parameters are adjusted by using an adjustment method.

[0013] Preferably, the step of decomposing the time domain response data of the chip into basis functions by using the singular value decomposition method, and reconstructing the time domain response data by a convolution operation according to the basis functions and the chip parameters comprises the following steps:

[0014] S21, based on the time domain response data of the chip, a signal model is established in the form of a complex exponential signal sum;

[0015] S22, discretizing the signal model to obtain discretized chip time domain response data;

[0016] S23, performing singular value decomposition on the discretized chip time domain response data and constructing basis functions;

[0017] S24, performing convolution and deconvolution calculations on the basis function and the chip parameters to obtain a signal response coefficient, and using the signal response coefficient to reconstruct the time domain response data.

[0018] Preferably, the step of establishing a signal model in the form of a complex exponential signal sum based on the time domain response data of the chip comprises the following steps:

[0019] S211, performing Fourier transform on the time domain response data of the chip to obtain a frequency domain signal;

[0020] S212, identifying peaks in the spectrum and determining the main frequency components of the frequency domain signal by analyzing the spectrum of the frequency domain signal;

[0021] S213, converting the main frequency component of the frequency domain signal back to the time domain, and performing envelope analysis on the time domain signal to determine the attenuation factor;

[0022] S214. Fitting a signal model by least square method according to the determined main frequency component and attenuation factor.

[0023] Preferably, the expression of the signal model is:

[0024]

[0025] Where K(t) represents the time domain response data of the chip at time t; R i represents the complex amplitude of the i-th complex exponential signal; b i represents the attenuation factor; m represents the number of complex exponential signals; f i Represents the main frequency component of the complex exponential signal.

[0026] Preferably, performing singular value decomposition on the discretized chip time domain response data and constructing basis functions comprises the following steps:

[0027] S231, arranging the discretized chip time domain response data in chronological order and establishing a data matrix;

[0028] S232, performing singular value decomposition on the data matrix to obtain singular values ​​and corresponding singular vectors;

[0029] S233. According to the size of the singular values, select the first m singular values ​​and their corresponding singular vectors to construct the basis function.

[0030] Preferably, the convolution and deconvolution calculation of the basis function and the chip parameters to obtain the signal response coefficient, and reconstructing the time domain response data using the signal response coefficient comprises the following steps:

[0031] S241, dividing the basis function into a plurality of subsets, and for each subset, performing parallel convolution calculation on the basis function and the chip parameters to obtain a plurality of convolution results;

[0032] S242, concatenating the obtained convolution results according to the division order to obtain a complete convolution result;

[0033] S243, performing deconvolution calculation on the complete convolution result to obtain a signal response coefficient;

[0034] S244. Reconstruct the time domain response data by weighted summation method according to the signal response coefficient and the basis function.

[0035] Preferably, the error comparison between the reconstructed time domain response data and the preset standard time domain response data, optimizing the chip parameters by an adaptive combined search algorithm, and determining the optimal parameter combination comprises the following steps:

[0036] S31, using a mean square error calculation method to calculate the error value between the reconstructed time domain response data and the standard time domain response data;

[0037] S32, using Tent chaotic mapping to initialize the population of chip parameters to obtain an initial solution set of chip parameters;

[0038] S33. Combine the adaptive harmony search algorithm and the sparrow search strategy to iteratively update the chip parameters and determine the optimal parameter combination of the chip.

[0039] Preferably, the step of initializing the population of chip parameters using Tent chaotic mapping to obtain an initial solution set of chip parameters comprises the following steps:

[0040] S321, determining chip parameters corresponding to the calculated error values ​​and having differences, and using the chip parameters as the chip parameters that need to be optimized;

[0041] S322, randomly generating an initial population within a preset range, and transforming it using the Tent mapping formula;

[0042] S323. Using the initial population processed by Tent mapping as an initial solution set of chip parameters.

[0043] Preferably, the method of combining the adaptive harmony search algorithm and the sparrow search strategy, iteratively updating the chip parameters, and determining the optimal chip parameter combination includes the following steps:

[0044] S331, presetting relevant parameters of the harmony search algorithm and the sparrow search strategy;

[0045] S332, initializing the harmony memory library and the sparrow population, and using the chip parameter initial solution set as a starting point for iterative search;

[0046] S333, generating a new parameter combination based on the harmony search algorithm and the sparrow search strategy, calculating its error value, and updating the harmony memory bank and the sparrow population;

[0047] S334, adaptively adjusting the relevant parameters of the harmony search algorithm and the sparrow search strategy according to the variation trend of the error value;

[0048] S335. Repeat steps S333 and S334 until the preset maximum number of iterations is reached, and select the parameter combination with the smallest error value from the harmony memory library as the optimal parameter combination of the chip.

[0049] Preferably, the generating of a new parameter combination based on the harmony search algorithm and the sparrow search strategy, calculating its error value, and updating the harmony memory library and the sparrow population comprises the following steps:

[0050] S3331. In each iteration, randomly select parameter values ​​in the harmony memory library, and generate a new parameter combination by combining the harmony memory library consideration rate and the pitch fine-tuning rate;

[0051] S3332, updating the position and speed of each sparrow in the sparrow population according to the sparrow search strategy, and generating a new parameter combination;

[0052] S3333, calculating the error value of each new parameter combination, and comparing it with the worst solution in the harmony memory library, if the error value of the new parameter combination is better, then replacing the worst solution in the harmony memory library;

[0053] S3334. If the parameter combination generated by the sparrow search strategy is better than any solution in the harmony memory bank, add it to the harmony memory bank.

[0054] The beneficial effects of the present invention are:

[0055] 1. The present invention utilizes convolution operation to reconstruct time domain response data, can efficiently process a large amount of test data, and quickly analyze the impact of chip parameters on performance. The adaptive combined search algorithm can quickly adjust the search direction according to the error comparison result, thereby accelerating the determination of the optimal parameter combination. The singular value decomposition method can accurately decompose the time domain response data into basis functions, which can reflect the impact of chip parameters on performance. By comparing the errors with standard time domain response data, the performance of the current parameter combination can be accurately evaluated, thereby finding the optimal parameter combination. The chip parameters can be accurately adjusted by the adjustment method, which can ensure that the chip performance meets the design requirements and improve the reliability and stability of the product. The error comparison and adaptive optimization process can reduce the interference of human factors and improve the accuracy of parameter adjustment.

[0056] 2. The present invention can efficiently extract key information from a large amount of discrete chip time domain response data through the singular value decomposition method, construct a basis function that can reflect the chip performance characteristics, establish a signal model in the form of a complex exponential signal, and accurately identify the main frequency components and attenuation factors of the signal through Fourier transform, spectrum analysis, envelope analysis and other means, thereby ensuring the accuracy and reliability of the signal model. The basis function is convolved and deconvolved with the chip parameters, and the signal response coefficient can be flexibly adjusted to achieve accurate reconstruction of the time domain response data. By dividing the basis function into several subsets and performing parallel convolution calculations, the computing efficiency can be significantly improved. This parallel processing method makes full use of computing resources, speeds up processing, and makes the entire processing process more efficient.

[0057] 3. The present invention utilizes a mean square error calculation method to accurately measure the difference between the reconstructed time domain response data and the standard time domain response data, thereby accurately evaluating the performance of the current chip parameters. The Tent chaotic mapping is used to initialize the population of chip parameters, which can generate an initial solution set with uniform distribution and good diversity. The adaptive harmonic search algorithm and the sparrow search strategy are combined to iteratively update the chip parameters, which can fully utilize the advantages of the two search strategies to achieve efficient global and local searches, and select the parameter combination with the smallest error value from the harmonic memory library as the optimal parameter combination of the chip, which can ensure that the parameter combination finally output has the best performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0059] Figure 1 The present invention is a flowchart of an adaptive chip parameter adjustment method based on convolution operation calculation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0061] According to an embodiment of the present invention, a method for adaptive chip parameter adjustment based on convolution operation calculation is provided.

[0062] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, the adaptive chip parameter adjustment method based on convolution operation calculation comprises the following steps:

[0063] S1, collect the time domain response data of the chip under different chip parameter settings;

[0064] It should be noted that collecting the time domain response data of the chip under different chip parameter settings includes the following steps:

[0065] Determine the chip parameter range that needs to be tested based on chip design and application requirements. For example, parameters may include operating frequency, voltage, temperature, etc.

[0066] Build a chip testing platform and configure testing instruments;

[0067] Set the chip parameters according to the predetermined parameter range; set one parameter combination each time and ensure that other parameters remain unchanged to facilitate comparative analysis;

[0068] Test each parameter combination and record the chip’s time domain response data;

[0069] The collected time domain response data are classified and stored, and the chip parameter settings corresponding to each group of data are indicated.

[0070] S2. Decomposing the chip's time domain response data into basis functions using a singular value decomposition method, and reconstructing the time domain response data through a convolution operation based on the basis functions and chip parameters;

[0071] As a preferred embodiment, the method of decomposing the time domain response data of the chip into basis functions by using the singular value decomposition method, and reconstructing the time domain response data by convolution operation according to the basis functions and chip parameters includes the following steps:

[0072] S21, based on the time domain response data of the chip, a signal model is established in the form of a complex exponential signal sum;

[0073] As a preferred embodiment, the signal model is established in the form of a complex exponential signal sum based on the time domain response data of the chip, and includes the following steps:

[0074] S211, performing Fourier transform on the time domain response data of the chip to obtain a frequency domain signal;

[0075] It should be noted that performing Fourier transform on the time domain response data of the chip to obtain the frequency domain signal includes the following steps:

[0076] Before Fourier transform, the time domain response data is preprocessed, including removing the DC component and window function processing (such as Hanning window, Hamming window, etc.) to reduce spectrum leakage and sidelobe effects;

[0077] Perform fast Fourier transform on the preprocessed time domain response data to generate a complex array representing the frequency domain signal;

[0078] From the Fast Fourier Transform results, the spectrum can be calculated, which is a graphical representation of the amplitude and phase of the frequency domain signal as a function of frequency.

[0079] S212, identifying peaks in the spectrum and determining the main frequency components of the frequency domain signal by analyzing the spectrum of the frequency domain signal;

[0080] It should be noted that, by analyzing the spectrum of the frequency domain signal, identifying the peak in the spectrum and determining the main frequency component of the frequency domain signal includes the following steps:

[0081] Extract the amplitude information of the spectrum from the result of Fourier transform. The amplitude reflects the energy distribution of the signal at different frequencies.

[0082] Match the spectrum amplitude with the frequency and draw a spectrum graph. The spectrum graph usually uses frequency as the horizontal axis and amplitude as the vertical axis to intuitively show the distribution of the signal in the frequency domain;

[0083] In the spectrum graph, the peaks usually correspond to the main frequency components of the signal. The peak points appear as local maxima in amplitude. Peak detection algorithms in programming languages ​​(such as finding local maxima) can be used to automatically identify these peaks.

[0084] After the peaks are identified, the frequencies corresponding to these peaks are determined as the main frequency components of the signal. The main frequency components are usually the frequencies corresponding to the peaks with the largest amplitude or the first few largest peaks; based on the signal characteristics and analysis requirements, one or more main frequency components can be selected for further analysis.

[0085] S213, converting the main frequency component of the frequency domain signal back to the time domain, and performing envelope analysis on the time domain signal to determine the attenuation factor;

[0086] It should be noted that the frequency domain signal containing only the main frequency component is converted back to the time domain through the inverse Fourier transform (IFT), and the envelope detection method is applied to the time domain signal, which usually involves techniques such as smoothing filtering or Hilbert transform to extract the amplitude envelope of the signal. The attenuation factor can be estimated by observing the speed at which the envelope decreases over time.

[0087] If the envelope decreases linearly with time, the decay factor can be estimated by fitting a straight line and calculating the slope.

[0088] If the envelope decreases nonlinearly with time (such as exponential decay), a more complex model (such as an exponential function) needs to be used to fit the envelope and determine the decay constant.

[0089] S214. Fitting a signal model by least square method according to the determined main frequency component and attenuation factor.

[0090] It should be noted that the model parameters are first determined, the main frequency components and the corresponding attenuation factors are obtained from the spectrum analysis, and the initial estimated complex amplitude can be obtained from the amplitude and phase information of the spectrum.

[0091] The objective function, i.e., the sum of squares of fitting errors, is constructed, and parameter optimization is performed with the purpose of minimizing the objective function. The model parameters are optimized using the least squares method to minimize the error between the model output and the actual time domain response data.

[0092] Specifically, the expression of the signal model is:

[0093]

[0094] Where K(t) represents the time domain response data of the chip at time t; R i represents the complex amplitude of the i-th complex exponential signal; b i represents the attenuation factor; m represents the number of complex exponential signals; f i Represents the main frequency component of the complex exponential signal.

[0095] S22, discretizing the signal model to obtain discretized chip time domain response data;

[0096] It should be noted that discretizing the signal model to obtain discretized chip time domain response data includes the following steps:

[0097] According to the Nyquist sampling theorem, choose a suitable sampling interval (time step) to ensure that the sampling frequency is at least twice the highest frequency of the signal.

[0098] According to the sampling interval, a discrete time series is generated, and the signal model is calculated at discrete time points to obtain the discrete chip time domain response data.

[0099] S23, performing singular value decomposition on the discretized chip time domain response data and constructing basis functions;

[0100] As a preferred implementation, the singular value decomposition of the discretized chip time domain response data and the construction of basis functions include the following steps:

[0101] S231, arranging the discretized chip time domain response data in chronological order and establishing a data matrix;

[0102] S232, performing singular value decomposition on the data matrix to obtain singular values ​​and corresponding singular vectors;

[0103] S233. According to the size of the singular values, select the first m singular values ​​and their corresponding singular vectors to construct the basis function.

[0104] For example, the specific steps of performing singular value decomposition on the discretized chip time domain response data and constructing the basis function are as follows:

[0105] Arrange multiple time-domain response data sequences in chronological order to form a data matrix X; assume that each time-domain response data is a row, there are n rows in total, and each row contains p time points.

[0106] Perform singular value SVD decomposition on the data matrix X: X = UΣV T ;

[0107] Among them, U is an n×n orthogonal matrix, Σ is an n×p diagonal matrix, V is a p×p orthogonal matrix, and T represents transpose.

[0108] Sort the singular values ​​in Σ according to their sizes, and select the first m singular values ​​σ1, σ2, …, σm and their corresponding singular vectors u1, u2, …, um and v1, v2, …, vm.

[0109] New basis functions are constructed based on the first m singular vectors u1,u2,…,um and v1,v2,…,vm. These basis functions can be used to reconstruct the original data or perform dimensionality reduction analysis.

[0110] S24, performing convolution and deconvolution calculations on the basis function and the chip parameters to obtain a signal response coefficient, and using the signal response coefficient to reconstruct the time domain response data.

[0111] As a preferred implementation, the convolution and deconvolution calculation of the basis function and the chip parameters to obtain the signal response coefficient, and the reconstruction of the time domain response data using the signal response coefficient includes the following steps:

[0112] S241, dividing the basis function into a plurality of subsets, and for each subset, performing parallel convolution calculation on the basis function and the chip parameters to obtain a plurality of convolution results;

[0113] It should be noted that the convolution calculation involves multiplying each basis function with the corresponding chip parameter and accumulating or integrating them in time to produce the convolution result.

[0114] S242, concatenating the obtained convolution results according to the division order to obtain a complete convolution result;

[0115] S243, performing deconvolution calculation on the complete convolution result to obtain a signal response coefficient;

[0116] S244. Reconstruct the time domain response data by weighted summation method according to the signal response coefficient and the basis function.

[0117] For example, the specific implementation steps of convolving and deconvolving the basis function with the chip parameters to obtain the signal response coefficient and reconstructing the time domain response data using the signal response coefficient are as follows:

[0118] Divide the basis functions into several subsets. Assume that the basis function is B = [b1, b2, ..., bn], which can be divided into k subsets, each of which contains n / k basis functions.

[0119] For each subset Bi, convolution calculation is performed with the chip parameter P, and the convolution result obtained is Ci: Ci = Bi*P;

[0120] Concatenate the convolution results Ci of all subsets in order to obtain the complete convolution result C: C = [C1, C2, ..., Ck];

[0121] The complete convolution result C is deconvolved to extract the signal response coefficient H. H can be obtained by using a deconvolution algorithm or solving a least squares problem.

[0122] The signal response coefficient H is used to perform weighted summation on the basis function B to reconstruct the time domain response data.

[0123] S3, comparing the reconstructed time domain response data with the preset standard time domain response data, optimizing the chip parameters through an adaptive combined search algorithm, and determining the optimal parameter combination;

[0124] As a preferred implementation, the error comparison between the reconstructed time domain response data and the preset standard time domain response data, optimizing the chip parameters by an adaptive combined search algorithm, and determining the optimal parameter combination includes the following steps:

[0125] S31, using a mean square error calculation method to calculate the error value between the reconstructed time domain response data and the standard time domain response data;

[0126] It should be noted that, for the reconstructed time domain response data and the standard time domain response data, the difference between them is calculated, and then the sum of squared errors is accumulated and divided by the total number of data (ie, sequence length) to obtain the mean square error.

[0127] S32, using Tent chaotic mapping to initialize the population of chip parameters to obtain an initial solution set of chip parameters;

[0128] As a preferred implementation, the process of initializing the population of chip parameters using Tent chaotic mapping to obtain an initial solution set of chip parameters includes the following steps:

[0129] S321, determining chip parameters corresponding to the calculated error values ​​and having differences, and using the chip parameters as the chip parameters that need to be optimized;

[0130] It should be noted that, according to the calculated error value, determining the chip parameter corresponding to the error value with difference, and using it as the chip parameter to be optimized includes the following steps:

[0131] For each chip parameter, sensitivity analysis can be performed by changing one parameter while keeping other parameters unchanged and observing the change in error value to understand the extent of their influence on the error;

[0132] Based on the results of the sensitivity analysis, determine which parameters have the greatest or most significant impact on the error; these parameters will be the focus of optimization;

[0133] Based on the sensitivity analysis results, select those chip parameters that have a significant impact on the error as the parameters that need to be optimized, and set a reasonable optimization range for each selected parameter. This range should be based on the physical limitations of the parameter, engineering experience, or previous experimental results.

[0134] S322, randomly generating an initial population within a preset range, and transforming it using the Tent mapping formula;

[0135] It should be noted that a numerical value is randomly generated within the preset range of each parameter as a parameter value of the first individual in the population, and this process is repeated until all parameter values ​​of the first individual in the population are generated.

[0136] Tent mapping is a chaotic mapping method that can generate chaotic sequences. Chaotic sequences have randomness, ergodicity and regularity, and are suitable for population initialization in optimization algorithms.

[0137] For each individual in the initial population, its parameter value is used as the initial state of the Tent mapping, and the Tent mapping formula is iterated multiple times to generate a chaotic sequence. The formula for the Tent mapping is:

[0138]

[0139] In the formula, Z h+1 Represents the state variable of the next iteration step. The next value calculated by the Tent mapping formula; Z h represents the state variable of the current iteration step, which is a number in the interval [0,1] and is used to represent the value of the current chaotic map; β represents the cutoff point parameter of the Tent map; it usually takes a value in the range of (0,1) and is used to divide the current state variable Z h Different calculation branches of , h represents the index of the current iteration step.

[0140] S323. Using the initial population processed by Tent mapping as an initial solution set of chip parameters.

[0141] It should be noted that after the Tent mapping transformation, the obtained parameter values ​​may still be in the range of [0,1]. Therefore, these values ​​need to be mapped to the actual value range of the chip parameters. This is usually achieved through linear transformation or other appropriate mapping methods. After the above processing, the parameter values ​​of each individual have been mapped to the actual value range of the chip parameters, and these individuals constitute the initial solution set of the chip parameters.

[0142] S33. Combine the adaptive harmony search algorithm and the sparrow search strategy to iteratively update the chip parameters and determine the optimal parameter combination of the chip.

[0143] As a preferred implementation, the method of combining the adaptive harmony search algorithm and the sparrow search strategy, iteratively updating the chip parameters, and determining the optimal chip parameter combination includes the following steps:

[0144] S331, pre-set relevant parameters of the harmony search algorithm and the sparrow search strategy; the relevant parameters include key parameters such as the harmony memory size, the harmony memory consideration rate, the pitch fine-tuning rate, the sparrow population size and the number of iterations.

[0145] It should be noted that the harmonic memory size (HMS): The size of the harmonic memory determines the number of solutions that the algorithm can retain during the search process. A larger HMS can improve the diversity of the search, but may also lead to increased computational costs. In general, the setting of HMS should be determined based on the size and complexity of the problem, and is usually set to a small proportion of the size of the solution space of the optimization problem.

[0146] Harmonic Memory Consideration Rate (HMCR): HMCR is the probability of retaining a solution component from the harmonic memory when generating a new solution. A larger HMCR value can promote the algorithm to search on the basis of existing solutions, thereby accelerating convergence; while a smaller HMCR value may increase the diversity of the search, but may also lead to slower convergence. The setting of HMCR should be determined according to the characteristics and requirements of the problem, and usually needs to be adjusted in multiple experiments to find the optimal value.

[0147] Pitch Fine Adjustment Rate (PAR): PAR is the probability of fine-tuning the solution components in the harmony memory library. An appropriate PAR value can improve the local search ability of the algorithm, but an excessively large PAR value may cause the algorithm to linger near the local optimal solution and be unable to jump out. The setting of PAR should also be determined according to the characteristics and requirements of the problem, and generally needs to be adjusted in multiple experiments.

[0148] Maximum number of iterations: The maximum number of iterations is one of the termination conditions for the algorithm. Setting an appropriate maximum number of iterations can ensure that the algorithm finds a satisfactory solution in a reasonable time. The number of iterations should be determined based on the size and complexity of the problem and the limitations of computing resources.

[0149] Parameter settings for the sparrow search strategy:

[0150] Sparrow population size: Population size refers to the size of the flock of birds in each generation of the sparrow algorithm. Population size is an important hyperparameter that directly affects the search ability of the sparrow algorithm. Generally speaking, the larger the population size, the stronger the search ability, but it also increases the time and space complexity of the algorithm. Therefore, it is necessary to balance the search ability and computational cost when setting the population size.

[0151] Iteration number: For the sparrow search strategy, the iteration number is also one of the termination conditions of the algorithm. Setting an appropriate number of iterations can ensure that the algorithm finds a satisfactory solution in a reasonable time. The number of iterations should be determined based on the scale and complexity of the problem and the limitations of computing resources.

[0152] S332, initializing the harmony memory library and the sparrow population, and using the chip parameter initial solution set as a starting point for iterative search;

[0153] It should be noted that HMS initial solutions are randomly generated within the range of chip parameter values. Each solution is a vector containing the values ​​of all chip parameters. Each initial solution is evaluated, and an objective function is usually used to calculate the quality of the solution (such as cost, performance, etc.). The evaluated initial solutions are sorted by quality (optional) and stored in the harmony memory library.

[0154] Similarly, each individual in the sparrow population is randomly generated within the range of chip parameter values. Each individual is also a vector containing all chip parameter values. Each sparrow individual is evaluated and its fitness value is calculated. According to the needs of the sparrow search strategy, other state information is initialized for each sparrow individual.

[0155] S333, generating a new parameter combination based on the harmony search algorithm and the sparrow search strategy, calculating its error value, and updating the harmony memory bank and the sparrow population;

[0156] As a preferred implementation, the generation of a new parameter combination based on the harmony search algorithm and the sparrow search strategy, the calculation of its error value, and the updating of the harmony memory library and the sparrow population include the following steps:

[0157] S3331. In each iteration, randomly select parameter values ​​in the harmony memory library, and generate a new parameter combination by combining the harmony memory library consideration rate and the pitch fine-tuning rate;

[0158] It should be noted that, in each iteration, randomly selecting parameter values ​​in the harmony memory library and generating a new parameter combination by combining the harmony memory library consideration rate and the pitch fine-tuning rate includes the following steps:

[0159] Randomly select one or more parameter combinations (solutions) from the harmony memory bank;

[0160] The harmonic memory consideration rate (HMCR) determines whether to select parameter components from the harmonic memory to form a new solution. If the HMCR condition is met, it is selected from the harmonic memory; otherwise, the parameter components are randomly generated within the parameter value range;

[0161] For the parameter component selected from the harmony memory library, whether to fine-tune it is determined according to the pitch fine-tuning rate (PAR); if the PAR condition is met, the parameter component is fine-tuned; otherwise, it remains unchanged;

[0162] The selected and / or fine-tuned parameter components are combined into a new parameter combination.

[0163] S3332, updating the position and speed of each sparrow in the sparrow population according to the sparrow search strategy, and generating a new parameter combination;

[0164] It should be noted that updating the position and speed of each sparrow in the sparrow population according to the sparrow search strategy and generating a new parameter combination includes the following steps:

[0165] Update the position and velocity of each sparrow according to the rules of the sparrow's search strategy; this usually involves different behaviors for leaders, followers, and scouts;

[0166] Generate a new parameter combination based on the updated position and velocity.

[0167] S3333, calculating the error value of each new parameter combination, and comparing it with the worst solution in the harmony memory library, if the error value of the new parameter combination is better, then replacing the worst solution in the harmony memory library;

[0168] It should be noted that for a new parameter combination generated by the harmony search algorithm or the mahjong search strategy, a preset objective function or error function is required to calculate its corresponding error value or objective function value;

[0169] The harmony memory is a collection of good solutions found in the historical search process. In each iteration, the harmony memory needs to be traversed to find the parameter combination with the maximum error value (for minimization problems) or the minimum error value (for maximization problems), that is, the current worst solution.

[0170] The error value of the new parameter combination is compared with the error value of the worst solution in the harmony memory. If the error value of the new parameter combination is better (i.e., for the minimization problem, the new error value is less than the error value of the worst solution; for the maximization problem, the new error value is greater than the error value of the worst solution), then the worst solution in the harmony memory needs to be replaced with the new parameter combination. In this way, the harmony memory can be kept updated and always contains the better solution currently found.

[0171] S3334. If the parameter combination generated by the sparrow search strategy is better than any solution in the harmony memory bank, add it to the harmony memory bank.

[0172] It should be noted that, for the new parameter combination generated by the sparrow search strategy, the error value of the new parameter combination generated by the sparrow search strategy is compared with the error values ​​of all solutions in the harmonic memory bank. If the error value of the new parameter combination is better than all solutions in the harmonic memory bank (that is, for the minimization problem, the new error value is smaller; for the maximization problem, the new error value is larger).

[0173] S334, adaptively adjusting the relevant parameters of the harmony search algorithm and the sparrow search strategy according to the variation trend of the error value;

[0174] It should be noted that the parameter adjustment of the harmony search algorithm (HS) includes:

[0175] Harmonic Memory Consideration Rate (HMCR): HMCR controls the probability of selecting parameters from the harmonic memory when a new solution is generated. If the error value does not improve significantly in consecutive generations, you can consider increasing HMCR to rely more on the high-quality solutions in the harmonic memory; on the contrary, if the error value decreases rapidly, you can reduce HMCR to increase the diversity of the search.

[0176] Pitch Fine-tuning Rate (PAR): PAR controls the degree of parameter fine-tuning in the new solution. If the error value is stagnant, you can reduce PAR to increase the search step size and speed up the jump out of the local optimum; if the error value fluctuates greatly, increasing PAR can make the search more detailed and facilitate fine-tuning near the high-quality solution.

[0177] The parameter adjustments of the Sparrow Search Strategy (SSA) include:

[0178] Leader ratio and follower ratio: The sparrows in SSA are divided into leaders, followers, and scouts. Leaders are responsible for exploring new solution spaces, followers follow leaders to search, and scouts are used to jump out of local optima. According to the changing trend of the error value, the ratio of leaders, followers, and scouts can be dynamically adjusted. For example, when the error value decreases slowly, the ratio of leaders is increased to enhance the exploration ability; when the error value fluctuates greatly, the ratio of followers is increased to improve the stability of the search.

[0179] Search step size: Similar to PAR in harmony search, the search step size in SSA can also be adjusted according to the trend of the error value. If the error value is stagnant, the search step size can be increased to speed up the search; if the error value is close to the optimal solution, reducing the search step size can improve the accuracy of the search.

[0180] In addition, the adaptive strategy can be implemented by setting a certain monitoring cycle, such as triggering the adaptive adjustment of parameters after a certain number of iterations or when the change in the error value is less than a certain threshold. When adjusting parameters, statistical analysis can be performed based on historical data (such as the change in error values ​​of the last few generations) to determine the direction and magnitude of parameter adjustment.

[0181] S335. Repeat steps S333 and S334 until the preset maximum number of iterations is reached, and select the parameter combination with the smallest error value from the harmony memory library as the optimal parameter combination of the chip.

[0182] S4. According to the determined optimal parameter combination, the chip parameters are adjusted by using an adjustment method.

[0183] It should be noted that, according to the previously determined optimal parameter combination, the relevant parameters of the trimming equipment, such as voltage, current, temperature, etc., are set, and the appropriate trimming method is selected according to the chip type, design requirements and actual parameter deviation. Common trimming methods include electrical trimming, laser trimming, Zener trimming, etc.

[0184] For electrical trimming, fuses can be blown or adjusted by applying current or voltage at specific locations on the chip, thereby changing circuit parameters;

[0185] For laser trimming, a laser beam can be used to directly ablate or cut specific areas on the chip to adjust circuit parameters;

[0186] For Zener trimming, the characteristics of the reverse breakdown diode can be used to remove resistance or adjust circuit parameters.

[0187] After determining the trimming method and parameters, implement the trimming operation according to the set steps and processes. During the trimming process, it is necessary to closely monitor the operating status of the trimming equipment and the changes in chip parameters to ensure the accuracy and effectiveness of the trimming operation.

[0188] In summary, with the aid of the above-mentioned technical scheme of the present invention, the present invention utilizes convolution operation to reconstruct time domain response data, can efficiently process a large amount of test data, quickly analyze the influence of chip parameters on performance, and can quickly adjust the search direction according to the error comparison result through the adaptive combined search algorithm, thereby accelerating the determination of the optimal parameter combination, and the singular value decomposition method can accurately decompose the time domain response data into basis functions, which can reflect the influence of chip parameters on performance, and can accurately evaluate the performance of the current parameter combination by comparing the error with the standard time domain response data, thereby finding the optimal parameter combination, and accurately adjusting the chip parameters by the adjustment method, which can ensure that the chip performance meets the design requirements and improve the reliability and stability of the product, and the error comparison and adaptive optimization process can reduce the interference of human factors and improve the accuracy of parameter adjustment; the present invention can efficiently extract key information from a large amount of discrete chip time domain response data through the singular value decomposition method, construct a basis function that can reflect the performance characteristics of the chip, establish a signal model in the form of a complex exponential signal and, and through Fourier transform, spectrum analysis, envelope The main frequency components and attenuation factors of the signal are accurately identified by means of analysis, thereby ensuring the accuracy and reliability of the signal model. The basis function is convolved and deconvolved with the chip parameters, so that the signal response coefficient can be flexibly adjusted to realize the accurate reconstruction of the time domain response data. By dividing the basis function into several subsets and performing parallel convolution calculations, the calculation efficiency can be significantly improved. This parallel processing method makes full use of the calculation resources, speeds up the processing speed, and makes the entire processing process more efficient. The present invention uses a mean square error calculation method to accurately measure the difference between the reconstructed time domain response data and the standard time domain response data, thereby accurately evaluating the performance of the current chip parameters. The population of chip parameters is initialized by using Tent chaotic mapping, so that an initial solution set with uniform distribution and good diversity can be generated. The chip parameters are iteratively updated by combining the adaptive harmony search algorithm and the sparrow search strategy, so that the advantages of the two search strategies can be fully utilized to realize efficient global search and local search. The parameter combination with the smallest error value is selected from the harmony memory library as the optimal parameter combination of the chip, which can ensure that the parameter combination finally output has the best performance.

[0189] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program code.

[0190] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An adaptive chip parameter adjustment method based on convolution operation calculation, characterized in that: The adaptive chip parameter adjustment method based on convolution operation calculation includes the following steps: S1, collect the time domain response data of the chip under different chip parameter settings; S2. Decomposing the chip's time domain response data into basis functions using a singular value decomposition method, and reconstructing the time domain response data through a convolution operation based on the basis functions and chip parameters; S3, comparing the reconstructed time domain response data with the preset standard time domain response data, optimizing the chip parameters through an adaptive combined search algorithm, and determining the optimal parameter combination; S4. According to the determined optimal parameter combination, the chip parameters are adjusted by using an adjustment method.

2. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 1, characterized in that: The method of decomposing the time domain response data of the chip into basis functions by using the singular value decomposition method, and reconstructing the time domain response data by convolution operation according to the basis functions and chip parameters comprises the following steps: S21, based on the time domain response data of the chip, a signal model is established in the form of a complex exponential signal sum; S22, discretizing the signal model to obtain discretized chip time domain response data; S23, performing singular value decomposition on the discretized chip time domain response data and constructing basis functions; S24, performing convolution and deconvolution calculations on the basis function and the chip parameters to obtain a signal response coefficient, and using the signal response coefficient to reconstruct the time domain response data.

3. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 2, characterized in that: The method of establishing a signal model in the form of a complex exponential signal and based on the time domain response data of the chip includes the following steps: S211, performing Fourier transform on the time domain response data of the chip to obtain a frequency domain signal; S212, identifying peaks in the spectrum and determining the main frequency components of the frequency domain signal by analyzing the spectrum of the frequency domain signal; S213, converting the main frequency component of the frequency domain signal back to the time domain, and performing envelope analysis on the time domain signal to determine the attenuation factor; S214. Fitting a signal model by least square method according to the determined main frequency component and attenuation factor.

4. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 2, characterized in that: The expression of the signal model is: Where K(t) represents the time domain response data of the chip at time t; R i represents the complex amplitude of the i-th complex exponential signal; b i represents the attenuation factor; m represents the number of complex exponential signals; f i Represents the main frequency component of the complex exponential signal.

5. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 2, characterized in that: The process of performing singular value decomposition on the discretized chip time domain response data and constructing basis functions comprises the following steps: S231, arranging the discretized chip time domain response data in chronological order and establishing a data matrix; S232, performing singular value decomposition on the data matrix to obtain singular values ​​and corresponding singular vectors; S233. According to the size of the singular values, select the first m singular values ​​and their corresponding singular vectors to construct the basis function.

6. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 5, characterized in that: The convolution and deconvolution calculation of the basis function and the chip parameters to obtain the signal response coefficient, and the time domain response data reconstructed by using the signal response coefficient comprises the following steps: S241, dividing the basis function into a plurality of subsets, and for each subset, performing parallel convolution calculation on the basis function and the chip parameters to obtain a plurality of convolution results; S242, concatenating the obtained convolution results according to the division order to obtain a complete convolution result; S243, performing deconvolution calculation on the complete convolution result to obtain a signal response coefficient; S244. Reconstruct the time domain response data by weighted summation method according to the signal response coefficient and the basis function.

7. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 1, characterized in that: The error comparison between the reconstructed time domain response data and the preset standard time domain response data, optimizing the chip parameters by an adaptive combined search algorithm, and determining the optimal parameter combination comprises the following steps: S31, using a mean square error calculation method to calculate the error value between the reconstructed time domain response data and the standard time domain response data; S32, using Tent chaotic mapping to initialize the population of chip parameters to obtain an initial solution set of chip parameters; S33. Combine the adaptive harmony search algorithm and the sparrow search strategy to iteratively update the chip parameters and determine the optimal parameter combination of the chip.

8. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 7, characterized in that: The method of initializing the population of chip parameters by using Tent chaotic mapping to obtain an initial solution set of chip parameters comprises the following steps: S321, determining chip parameters corresponding to the calculated error values ​​and having differences, and using the chip parameters as the chip parameters that need to be optimized; S322, randomly generating an initial population within a preset range, and transforming it using the Tent mapping formula; S323. Using the initial population processed by Tent mapping as an initial solution set of chip parameters.

9. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 7, characterized in that: The method of combining the adaptive harmony search algorithm and the sparrow search strategy, iteratively updating the chip parameters, and determining the optimal chip parameter combination includes the following steps: S331, presetting relevant parameters of harmony search algorithm and sparrow search strategy; S332, initializing the harmony memory library and the sparrow population, and using the chip parameter initial solution set as a starting point for iterative search; S333, generating a new parameter combination based on the harmony search algorithm and the sparrow search strategy, calculating its error value, and updating the harmony memory bank and the sparrow population; S334, adaptively adjusting relevant parameters of the harmony search algorithm and the sparrow search strategy according to the variation trend of the error value; S335. Repeat steps S333 and S334 until the preset maximum number of iterations is reached, and select the parameter combination with the smallest error value from the harmony memory library as the optimal parameter combination of the chip.

10. The method for adaptive chip parameter adjustment based on convolution operation calculation according to claim 9, characterized in that: The method of generating a new parameter combination based on the harmony search algorithm and the sparrow search strategy, calculating its error value, and updating the harmony memory library and the sparrow population includes the following steps: S3331. In each iteration, randomly select parameter values ​​in the harmony memory library, and generate a new parameter combination by combining the harmony memory library consideration rate and the pitch fine-tuning rate; S3332, updating the position and speed of each sparrow in the sparrow population according to the sparrow search strategy, and generating a new parameter combination; S3333, calculating the error value of each new parameter combination, and comparing it with the worst solution in the harmony memory library, if the error value of the new parameter combination is better, then replacing the worst solution in the harmony memory library; S3334. If the parameter combination generated by the sparrow search strategy is better than any solution in the harmony memory bank, add it to the harmony memory bank.

Citation Information

Cited By

  • Dynamic adaptability multi-parameter trimming optimization method and system

    CN120908647A

  • Dynamic adaptive multi-parameter tuning optimization method and system

    CN120908647B