Chip control method and device, equipment, medium and product

Abstract

The embodiment of the invention discloses a chip control method and device, equipment, a medium and a product, and relates to the field of chips. The method comprises the following steps: acquiring a chip parameter of a first chip; analyzing the chip parameters of the first chip through a pre-trained voltage frequency model to obtain a voltage frequency point corresponding to the first chip; and controlling the first chip to operate at the working voltage and the working frequency based on the voltage frequency point. The voltage frequency point is a personalized customized voltage frequency point for the first chip and accords with the chip parameter of the first chip, that is, the adaptation degree between the voltage frequency point and the chip parameter of the first chip is improved, the performance of the first chip is improved under the dynamically determined voltage frequency point, and the performance of the first chip is improved. The problem that the performance of the unified voltage frequency point on the first chip is poor is avoided.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of chips, and in particular to a chip control method, device, equipment, medium and product. Background Art

[0002] During the operation of the chip, it is usually necessary to control the chip to operate at a preset operating voltage and operating frequency, wherein a voltage-frequency data pair consisting of the operating voltage and the operating frequency is the voltage-frequency point when the chip is operating.

[0003] In the related art, a voltage frequency point is usually determined based on test data of multiple chips, and each chip is controlled to operate at the determined voltage frequency point.

[0004] However, the generalization capability of the above voltage-frequency point is limited, and the chip is prone to performance waste when running at the voltage-frequency point. Summary of the invention

[0005] The embodiments of the present application provide a chip control method, device, equipment, medium and product. The technical solution is as follows:

[0006] On the one hand, an embodiment of the present application provides a chip control method, which is executed by a computer device, wherein a first chip is running in the computer device, and the method includes:

[0007] Acquire a chip parameter of the first chip, where the chip parameter is used to indicate a physical characteristic of the first chip;

[0008] Analyze the chip parameters of the first chip through a voltage-frequency model to obtain a voltage-frequency point corresponding to the first chip, wherein the voltage-frequency point includes an operating voltage of the first chip and an operating frequency corresponding to the operating voltage, wherein the voltage-frequency model is a model obtained by training the chip parameters of the sample chip and the screened voltage-frequency data after screening the voltage-frequency data of the sample chip based on the stress test data;

[0009] The first chip is controlled to operate at the operating voltage and the operating frequency based on the voltage-frequency point.

[0010] On the other hand, an embodiment of the present application provides a chip control device, the device comprising:

[0011] an acquisition module, configured to acquire chip parameters of the first chip, where the chip parameters are used to indicate physical characteristics of the first chip;

[0012] an analysis module, configured to analyze chip parameters of the first chip through a voltage-frequency model to obtain voltage-frequency points corresponding to the first chip, wherein the voltage-frequency points include an operating voltage of the first chip and an operating frequency corresponding to the operating voltage, wherein the voltage-frequency model is a model obtained by training the chip parameters of the sample chip and the screened voltage-frequency data after screening the voltage-frequency data of the sample chip based on the stress test data;

[0013] The control module is configured to control the first chip to operate at the operating voltage and the operating frequency based on the voltage-frequency point.

[0014] On the other hand, an embodiment of the present application provides a computer device, which includes a processor and a memory, wherein at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement a chip control method as provided in the above-mentioned embodiment of the present application.

[0015] On the other hand, an embodiment of the present application provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement a chip control method as provided in the above-mentioned embodiment of the present application.

[0016] On the other hand, an embodiment of the present application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the chip control method provided in the above-mentioned embodiment of the present application.

[0017] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0018] By designing and training a voltage-frequency model, after analyzing the chip parameters of the first chip, a voltage-frequency point corresponding to the first chip is output. The voltage-frequency point is a voltage-frequency point customized for the first chip and conforms to the chip parameters of the first chip, that is, the compatibility between the voltage-frequency point and the chip parameters of the first chip is improved. At the dynamically determined voltage-frequency point, the performance of the first chip is improved, avoiding the problem of poor performance of the first chip at a uniform voltage-frequency point. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1is a schematic diagram of the structure of a control system of a chip provided by an exemplary embodiment of the present application;

[0021] Figure 2 is a flow chart of a chip control method provided by an exemplary embodiment of the present application;

[0022] Figure 3 is based on Figure 2 A schematic diagram of the training process of the inflection point analysis model provided by the illustrated embodiment;

[0023] Figure 4 is a flow chart of a chip control method provided by another exemplary embodiment of the present application;

[0024] Figure 5 is based on Figure 4 A schematic diagram of a training process of a voltage-frequency model provided by the illustrated embodiment;

[0025] Figure 6 is a flow chart of a data preparation phase provided by an exemplary embodiment of the present application;

[0026] Figure 7 is a flow chart of a modeling phase of a voltage-frequency model provided by an exemplary embodiment of the present application;

[0027] Figure 8 is a flow chart of a modeling phase of an inflection point model provided by an exemplary embodiment of the present application;

[0028] Fig. 9 is a schematic diagram of a voltage-frequency curve provided by an exemplary embodiment of the present application;

[0029] Fig.10 is a flow chart of the application phase provided by an exemplary embodiment of the present application;

[0030] Fig.11 is a schematic diagram of a control process of a chip in an application phase provided by another exemplary embodiment of the present application;

[0031] Fig.12 is a structural block diagram of a control device of a chip provided by an exemplary embodiment of the present application;

[0032] Fig.13 is a structural block diagram of a control device of a chip provided by another exemplary embodiment of the present application;

[0033] Fig.14 It is a structural block diagram of a computer device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0035] First, the terms involved in the embodiments of the present application are introduced.

[0036] Voltage-frequency point: refers to the data pair of operating voltage and operating frequency used to control the operation of the chip. In chip design, the voltage-frequency point is used to determine the performance and energy efficiency that the chip can achieve under different workloads and environmental conditions. Higher frequencies can provide chips with stronger processing capabilities, but they also require higher voltages to ensure operational stability, which in turn leads to increased power consumption and heat generation. Therefore, in chip design, it is necessary to select the appropriate voltage-frequency point to balance factors such as performance, power consumption, and heat dissipation.

[0037] In some embodiments, in different application scenarios of the chip, power consumption can be reduced by lowering the frequency and voltage, or by increasing the frequency and voltage when higher performance is required, thereby achieving higher energy efficiency and performance.

[0038] Chip parameters: used to indicate the physical characteristics of a chip. In some embodiments, chip parameters are physical parameters that are fixed after the chip is designed and manufactured. In some embodiments, chip parameters include parameters obtained by running the chip after it is designed and manufactured, or chip parameters include parameters configured when the chip is designed.

[0039] Illustratively, the chip parameters include operating data obtained after the chip is designed and manufactured and automatically tested on an integrated circuit automatic test equipment (ATE), wherein, in the ATE stage, a preset excitation signal is applied to a designated input end of the chip, and the output end of the chip is monitored, and the propagation speed of the chip is determined based on the time between the input and the output as the chip parameter. In some embodiments, the chip parameters are embodied as the process parameters of the chip.

[0040] Alternatively, the chip parameters include configuration parameters determined during the chip design phase, such as the number of chip cores, chip memory capacity, etc., which are not limited in the embodiments of the present application.

[0041] Voltage-frequency model: refers to a model used to predict the voltage frequency point corresponding to the chip. In the embodiment of the present application, the input data of the voltage-frequency model is the chip parameters of the chip, and the output data is the voltage-frequency relationship corresponding to the chip.

[0042] Optionally, the voltage-frequency relationship output by the voltage-frequency model includes voltage-frequency points corresponding to the chip in different working states, wherein the voltage-frequency points can be expressed in the form of a voltage-frequency curve or in the form of a voltage-frequency table.

[0043] The working state refers to the environmental state of the chip when it is running. In some embodiments, the environmental state includes voltage state, temperature state, load state, running scene state, etc. The voltage state is a state characterized by voltage data collected by the voltage detector / voltage sensor in the chip; the temperature state is a state characterized by temperature data collected by the temperature sensor in the chip; the load state is a state characterized by the load condition corresponding to the input data of the chip; the running scene state is a state characterized by the data processing flow corresponding to the input data of the chip, such as: applied to artificial intelligence (AI) model training scenarios, or applied to game program running scenarios, etc.

[0044] Inflection point analysis model: refers to a model used to predict the inflection point voltage corresponding to a chip. In some embodiments, the input data of the inflection point analysis model includes chip parameters of the chip, and the output data is the inflection point voltage corresponding to the chip. The inflection point voltage refers to the voltage at which the voltage frequency of the chip at different operating temperatures is similar.

[0045] In some embodiments, the input data of the inflection point analysis model also includes the operation scenario of the chip, and the inflection point analysis model outputs the inflection point voltage corresponding to the chip according to the chip parameters and the operation scenario of the chip.

[0046] In some embodiments, the voltage frequency curves predicted by the chip under different temperature states are obtained, and the voltage frequency curves predicted under different temperature states are sorted to determine the inflection point voltage. In some embodiments, the differences between the frequencies corresponding to the same voltage point of each voltage frequency curve are obtained, and the difference data corresponding to multiple voltage points are obtained, and the voltage point with the smallest difference data is selected as the inflection point voltage. Schematically, the difference between the frequency data corresponding to each voltage frequency curve at 700mV, 800mV, and 900mV is obtained, such as: the standard deviation between the frequency data is obtained as the difference data, and the voltage point with the smallest difference data is selected as the inflection point voltage. It is worth noting that the above voltage points are only schematic examples. In actual applications, more or fewer voltage points can also be selected for comparison of difference data. The above standard deviation difference data determination method is also a schematic example. In actual applications, the difference data can also be reflected by variance or other difference reflection methods.

[0047] In chip design, related technologies usually fix a voltage-frequency parameter or a set of voltage-frequency parameters for all chips, and then select between these voltage-frequency parameters according to their respective algorithms. These voltage-frequency parameters are often based on a small number of chips and test data, and have weak generalization capabilities. In actual applications, the characteristics of each chip are different, and using the same or the same set of parameters for all chips will result in a large performance waste.

[0048] Indicatively, the above voltage frequency point setting method has at least the following problems: 1. Fixed performance setting: The preset voltage frequency point is set based on the performance average of different chips or the standard process, without taking into account the uniqueness of each chip. This leads to potential performance loss. 2. Insufficient generalization processing: Since all chips use the same voltage frequency point, the specificity of a single chip cannot be optimized, reducing overall efficiency and performance. 3. Insufficient adjustment speed and flexibility: When it is necessary to quickly adjust chip performance to cope with sudden workloads, fixed voltage frequency points cannot provide instant and precise adjustments. 4. Inadequate performance optimization: Due to the lack of consideration of chip differences, the chip fails to perform its optimal performance in specific application scenarios or temperature conditions, especially in situations where high-performance computing is required. 5. Low energy efficiency: The above voltage frequency point setting method fails to adjust the voltage frequency according to different usage scenarios, resulting in relatively low chip energy efficiency in some cases, increasing energy consumption and heat dissipation requirements.

[0049] Optionally, the chip involved in the embodiments of the present application includes at least one of a central processing unit (CPU) chip, a graphics processing unit (GPU) and the like.

[0050] In response to the above problems, an embodiment of the present application proposes a voltage and frequency adaptive adjustment method based on deep learning to improve the performance and energy efficiency of the chip in different application scenarios.

[0051] In an embodiment of the present application, the chip parameters of the chip are analyzed through a pre-trained voltage-frequency model, and the voltage frequency points corresponding to each chip are determined individually based on the chip parameters, so that the chip runs at a personalized customized voltage-frequency point, thereby maximizing performance.

[0052] The chip control method provided in the embodiment of the present application can be executed by a terminal, or by a server, or by a terminal and a server in cooperation. Taking the method executed by a terminal as an example, Figure 1 FIG. 1 is a schematic diagram of a control system of a chip provided by an exemplary embodiment of the present application. Figure 1As shown, the control system of the chip includes: a terminal 100, and the above-mentioned voltage-frequency model 110 is stored in the terminal 100.

[0053] The terminal 100 may be an electronic device such as a mobile phone, a tablet computer, a vehicle-mounted terminal (vehicle computer), a PC (Personal Computer), etc. The terminal 110 also stores a chip control program.

[0054] Schematically, a voltage-frequency model is stored in the terminal 100. When the terminal 100 controls the chip, the chip parameters of the chip configured inside the terminal 100 are first obtained, and the chip parameters are input into the voltage-frequency model 110, so as to obtain the voltage-frequency point of the chip through analysis of the voltage-frequency model 110, and control the chip operation through the operating voltage and operating frequency at the voltage-frequency point.

[0055] The terminal calls the voltage-frequency model according to the chip control program to analyze the chip parameters of the chip and obtains the voltage-frequency point of the chip.

[0056] Figure 2 1 is a flowchart of a chip control method provided by an exemplary embodiment of the present application, the method is executed by a computer device, and a first chip is running in the computer device. Schematically, the method can be executed by a terminal, or by a server, or by a terminal and a server in cooperation. Figure 2 As shown, the method includes the following steps.

[0057] Step 220: Obtain chip parameters of the first chip.

[0058] The chip parameters are used to indicate the physical characteristics of the first chip. In some embodiments, the chip parameters are also used to indicate the process parameters of the first chip.

[0059] In some embodiments, when the first chip is powered on, the chip parameters of the first chip are read from the storage space. Optionally, the chip parameters of the first chip are read from the storage space of the first chip.

[0060] Chip parameters are physical parameters that are fixed after the chip is designed and manufactured. Optionally, the chip parameters include parameters obtained by running the first chip after it is designed and manufactured, or the chip parameters include parameters configured when the first chip is designed, or the chip parameters include parameters obtained by running the first chip after it is designed and manufactured and parameters configured when the first chip is designed.

[0061] Illustratively, chip parameters include operational data obtained after the chip is designed and manufactured and then automatically tested in the ATE stage.

[0062] Optionally, in the test phase of the first chip, an excitation signal is input to the first chip, and transmission performance data of the excitation signal in the first chip is collected as chip parameters of the first chip. Schematically, in the ATE phase, a preset excitation signal is applied to a specified input end of the chip, and the output end of the chip is monitored, and the propagation speed of the chip is determined based on the time between the input and the output as a chip parameter. In some embodiments, the chip parameter is referred to as process data of the first chip.

[0063] Alternatively, the chip parameters include configuration parameters determined during the chip design phase, such as the number of chip cores, chip memory capacity, etc., which are not limited in the embodiments of the present application.

[0064] By testing and acquiring chip parameters in the testing phase and pre-storing the chip parameters in the chip, the efficiency of acquiring the chip parameters when acquiring the voltage-frequency curve is improved.

[0065] Optionally, the chip parameters of the first chip are obtained by reading a one-time programmable memory (eFuse) of the first chip. The eFuse can store the chip parameters of the first chip, and the chip parameters cannot be changed after being written into the eFuse, thereby ensuring the security and non-tamperability of the parameter information.

[0066] Step 240: Analyze chip parameters of the first chip using a voltage-frequency model to obtain a voltage-frequency point corresponding to the first chip.

[0067] The voltage-frequency model is a model obtained by screening the voltage-frequency data of a sample chip based on stress test data and then training the model using the chip parameters of the sample chip and the screened voltage-frequency data.

[0068] The voltage-frequency model is a model obtained by modeling the relationship between the chip parameters of the sample chip and the voltage-frequency curve. The voltage-frequency curve is a curve collected during the operation of the sample chip.

[0069] The voltage-frequency curve is obtained by obtaining multiple sets of voltage-frequency data pairs corresponding to stress test data that meet the stress test requirements, and linearly fitting the multiple sets of voltage-frequency data pairs. The voltage-frequency data pairs corresponding to the stress test data that meet the stress test requirements are obtained by collecting the stress test data and voltage-frequency data pairs of the sample chip during the operation of the sample chip, and then filtering the voltage-frequency data pairs according to the stress test data. The stress test data is used to indicate the performance of the sample chip for different operating voltages and operating frequencies during operation.

[0070] The voltage-frequency point includes the operating voltage of the first chip and the operating frequency corresponding to the operating voltage.

[0071] In some embodiments, obtaining a voltage frequency point by analyzing chip parameters using a voltage frequency model includes at least one of the following methods:

[0072] 1. Analyze the chip parameters through the pre-trained voltage-frequency model and directly output the voltage-frequency point corresponding to the first chip.

[0073] Optionally, the voltage-frequency model is used to analyze the chip parameters and output voltage-frequency points corresponding to the chip parameters, that is, the voltage-frequency model is used to build a model between the chip parameters and the voltage-frequency points.

[0074] In this solution, during the training process of the voltage-frequency model, modeling is performed using the chip parameters of the sample chip and the voltage-frequency point data of the sample chip when its performance meets the performance requirements, and the correlation between the chip parameters and the voltage-frequency points is constructed to obtain the voltage-frequency model.

[0075] In some embodiments, the operation scenario of the first chip is obtained, and the operation scenario is used to characterize the use scenario of the first chip, such as: the first chip is used for AI training, or the first chip is used for game operation, etc. Optionally, the operation scenario is also used to characterize the load of the first chip, such as: the amount of data input into the first chip per unit time. The chip parameters and the operation scenario are analyzed by the pre-trained voltage-frequency model, and the voltage-frequency point corresponding to the first chip is directly output.

[0076] 2. Analyze the chip parameters through the pre-trained voltage-frequency model, output the candidate voltage-frequency points corresponding to the first chip in the default operating state, collect operating status data used to characterize the operating status of the first chip, adjust the candidate voltage-frequency points based on the difference between the operating status data and the default operating state, and obtain the voltage-frequency points corresponding to the first chip.

[0077] In schematic form, the chip parameters are analyzed by a pre-trained voltage-frequency model, and the candidate voltage-frequency points of 800mV and 1300MHz corresponding to the first chip at an operating temperature of 50° are output. The current operating temperature of the first chip is obtained as 55°, and the candidate voltage-frequency points are adjusted with a deviation corresponding to 5° to obtain 850mV and 1380MHz.

[0078] 3. Analyze the chip parameters of the first chip through the pre-trained voltage-frequency model, and output the voltage-frequency relationship corresponding to the first chip, where the voltage-frequency relationship includes voltage-frequency points corresponding to the first chip under multiple different operating states; collect operating status data for characterizing the operating status of the first chip, and dynamically determine the voltage-frequency points in the voltage-frequency relationship corresponding to the operating status data in real time.

[0079] In some embodiments, the chip parameters and operating scenarios of the first chip are analyzed through a pre-trained voltage-frequency model, and the voltage-frequency relationship corresponding to the first chip is output. That is, the voltage-frequency model is a model trained based on the chip parameters of the sample chip, the operating scenario of the sample chip, and the voltage-frequency data of the sample chip when it is running.

[0080] Alternatively, the chip parameters of the first chip are analyzed by a pre-trained voltage-frequency model, and after the voltage-frequency relationship of the first chip is output, the operating status data and operating scenario of the first chip are collected, and the voltage-frequency points corresponding to the operating status data and the operating scenario in the voltage-frequency relationship are obtained. That is, the voltage-frequency relationship includes the correspondence between each operating status data and operating scenario and the voltage-frequency point, and the voltage-frequency relationship includes the voltage-frequency points corresponding to the first chip in multiple different operating states and operating scenarios.

[0081] In the embodiment of the present application, the correspondence between the operating status data and the voltage frequency point is taken as an example to illustrate that there is a one-to-one correspondence between the operating status data and the voltage frequency point. After obtaining the operating status data, the voltage frequency point corresponding to the operating status data is directly obtained.

[0082] In other embodiments, the voltage-frequency relationship includes voltage-frequency curves corresponding to the first chip in multiple different operating states, the voltage-frequency curves corresponding to the operating state data in the voltage-frequency relationship are obtained, and the voltage-frequency points are determined from the voltage-frequency curves; or, the voltage-frequency relationship includes voltage-frequency tables corresponding to the first chip in multiple different operating states, wherein the voltage-frequency table includes a group of voltage-frequency points corresponding to each operating state, a group of voltage-frequency points corresponding to the operating state data in the voltage-frequency relationship are obtained, and the voltage-frequency points are determined from the group of voltage-frequency points.

[0083] Optionally, when determining a voltage-frequency point from a voltage-frequency curve or a group of voltage-frequency points, the selection can be based on parameters such as the load condition, operating scenario, etc. of the first chip, or based on predetermined selection rules, or based on the load condition, operating scenario, and a combination of predetermined selection rules.

[0084] In an illustrative manner, when obtaining a voltage-frequency point corresponding to the operating status data in the voltage-frequency relationship, at least one of the following methods is included:

[0085] 3.1 Obtain inflection point status data corresponding to the first chip, the inflection point status data is used to compare with the operation status data, and determine the voltage frequency point according to the comparison result;

[0086] Then, a comparison relationship between the running status data and the inflection point status data is obtained, and a voltage-frequency point corresponding to the running status data in the voltage-frequency relationship is determined based on the comparison relationship.

[0087] Schematically, the operating status data includes the operating temperature, then the inflection point temperature corresponding to the first chip is obtained, the inflection point temperature is a preset temperature, and the voltage frequency point corresponding to the operating status data in the voltage-frequency relationship is determined based on the first comparison relationship between the operating temperature and the inflection point temperature. Schematically, the voltage-frequency relationship includes a first voltage-frequency point corresponding to a low temperature and a second voltage-frequency point corresponding to a high temperature, then when the current temperature is greater than the inflection point temperature, the second voltage-frequency point corresponding to the high temperature is obtained; when the current temperature is lower than the inflection point temperature, the first voltage-frequency point corresponding to the low temperature is obtained.

[0088] Optionally, when the current temperature is equal to the inflection point temperature, the first voltage-frequency point or the second voltage-frequency point is used, which is not limited in this embodiment.

[0089] By analyzing the working characteristics of the first chip at different operating temperatures, the turning point of the operating temperature on the performance is determined, and the temperature turning point, that is, the judgment of the inflection point temperature, is introduced, so that the chip can work in the best state at different operating temperatures.

[0090] Indicatively, the operating status data includes the current voltage, then the inflection point voltage corresponding to the third chip is obtained, the inflection point voltage is a voltage at which the voltage frequency of the first chip at different operating temperatures is similar, and the voltage frequency point corresponding to the operating status data in the voltage-frequency relationship is determined based on the second comparison relationship between the current voltage and the inflection point voltage. Indicatively, the voltage-frequency relationship includes a third voltage-frequency point corresponding to the low voltage and a fourth voltage-frequency point corresponding to the high voltage, then when the current voltage is greater than the inflection point voltage, the fourth voltage-frequency point corresponding to the high voltage is obtained; when the current voltage is lower than the inflection point voltage, the third voltage-frequency point corresponding to the low voltage is obtained.

[0091] Optionally, when the current voltage is equal to the inflection point voltage, the third voltage-frequency point or the fourth voltage-frequency point is used, which is not limited in this embodiment.

[0092] By analyzing the working characteristics of the first chip at different voltages, the turning point of the voltage's impact on performance is determined, and the voltage turning point, that is, the judgment of the inflection point voltage, is introduced, so that the chip can work in the best state under different voltage conditions.

[0093] Schematically, the operating status data includes the current voltage and the operating temperature, and the corresponding voltage-frequency point is selected according to the current voltage and the operating temperature.

[0094] That is, obtain the inflection point voltage and inflection point temperature corresponding to the first chip, the inflection point voltage is a voltage with similar voltage-frequency performance of the first chip at different operating temperatures, and the inflection point temperature is a pre-set temperature; obtain a first comparison relationship between the operating temperature and the inflection point temperature, the first comparison relationship includes the operating temperature being greater than the inflection point temperature or the operating temperature being less than the inflection point temperature; obtain a second comparison relationship between the current voltage and the inflection point voltage, the second comparison relationship includes the current voltage being greater than the inflection point voltage or the current voltage being less than the inflection point voltage; determine the voltage-frequency point in the voltage-frequency relationship corresponding to the operating status data based on the first comparison relationship and the second comparison relationship.

[0095] Optionally, when the operating temperature is greater than the inflection point temperature and the current voltage is greater than the inflection point voltage, a first voltage frequency point is determined from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, and the first voltage frequency point is a frequency point corresponding to the high temperature and high pressure obtained by pre-fitting in the voltage-frequency relationship;

[0096] When the operating temperature is greater than the inflection point temperature and the current voltage is less than the inflection point voltage, a second voltage frequency point is determined from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, and the second voltage frequency point is a frequency point corresponding to high temperature and low voltage obtained by pre-fitting in the voltage-frequency relationship;

[0097] When the operating temperature is lower than the inflection point temperature and the current voltage is higher than the inflection point voltage, a third voltage frequency point is determined from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, and the third voltage frequency point is a frequency point corresponding to the low temperature and high pressure obtained by pre-fitting in the voltage-frequency relationship;

[0098] When the operating temperature is lower than the inflection point temperature and the current voltage is lower than the inflection point voltage, a fourth voltage-frequency point is determined from the voltage-frequency relationship as the voltage-frequency point corresponding to the operating status data, and the fourth voltage-frequency point is a frequency point corresponding to the low temperature and low pressure obtained by pre-fitting in the voltage-frequency relationship.

[0099] The first voltage frequency point, the second voltage frequency point, the third voltage frequency point and the fourth frequency point are voltage frequency points determined according to a pre-selected high temperature voltage curve and a pre-selected low temperature voltage curve. The high temperature voltage curve and the pre-selected low temperature voltage curve are determined according to the operation scenario and load conditions of the first chip.

[0100] By obtaining the current voltage, operating temperature, inflection point voltage and inflection point temperature, a comparison is made between the current voltage and the inflection point voltage, and a comparison is made between the operating temperature and the inflection point temperature, so that a voltage frequency point is selected from the voltage-frequency relationship according to the comparison result. There is no need to store the voltage-frequency curves corresponding to all temperatures, but only the voltage frequency points corresponding to high temperature and high pressure, high temperature and low pressure, low temperature and high pressure, and low temperature and low pressure, respectively, need to be stored. The voltage frequency point is adaptively selected according to the comparison result, thereby improving the efficiency of determining the voltage frequency point.

[0101] In some embodiments, when the operating temperature is equal to the inflection point temperature and the current voltage is not equal to the inflection point voltage, after the high voltage / low voltage is determined based on the second comparison relationship, the first comparison relationship is arbitrarily selected to correspond to the high temperature or low temperature; similarly, when the operating temperature is not equal to the inflection point temperature and the current voltage is equal to the inflection point voltage, after the high temperature / low voltage is determined based on the first comparison relationship, the second comparison relationship is arbitrarily selected to correspond to the high voltage or low voltage; similarly, when the operating temperature is equal to the inflection point temperature and the current voltage is equal to the inflection point voltage, the voltage frequency point is arbitrarily determined from high temperature and high pressure, high temperature and low pressure, low temperature and high pressure, and low temperature and low pressure, or a pre-set default voltage frequency point is determined, and this embodiment is not limited to this.

[0102] Indicatively, under the conservative strategy, the inflection point voltage measurement needs to save the voltage frequency curve or voltage frequency table corresponding to the lowest temperature and the highest temperature respectively, that is, each chip saves two different voltage frequency curves or voltage frequency tables. Taking the voltage frequency curve as an example, when the operating temperature is higher than the inflection point temperature and the current voltage is higher than the inflection point voltage, select the voltage frequency point with the largest voltage in the voltage frequency curve of the highest temperature; when the operating temperature is higher than the inflection point temperature and the current voltage is lower than the inflection point voltage, select the voltage frequency point with the smallest voltage in the voltage frequency curve of the highest temperature; when the operating temperature is lower than the inflection point temperature and the current voltage is higher than the inflection point voltage, select the voltage frequency point with the largest voltage in the voltage frequency curve of the lowest temperature; when the operating temperature is lower than the inflection point temperature and the current voltage is lower than the inflection point voltage, select the voltage frequency point with the smallest voltage in the voltage frequency curve of the lowest temperature. It is worth noting that the selection of the above voltage frequency curve is only an illustrative example, and the embodiment of the present application can also select voltage frequency curves at other temperatures to determine the voltage frequency points.

[0103] The inflection point temperature is a preset temperature, and the preset inflection point temperature is obtained, such as: the operating temperature range of the first chip is 0° to 100°, and 50° is used as the inflection point temperature; or, the developer uses 55° as the inflection point temperature according to experience and temperature deviation.

[0104] The inflection point voltage is a voltage obtained based on chip parameter fitting. The chip parameters of the first chip are analyzed by a pre-trained inflection point analysis model to obtain the inflection point voltage corresponding to the first chip.

[0105] The inflection point analysis model is a model obtained by modeling the chip parameters of the sample chip and the inflection point voltage of the sample chip, wherein the inflection point voltage of the sample chip is obtained by inputting the chip parameters of the sample chip into a pre-trained voltage-frequency model, outputting voltage-frequency data corresponding to multiple chip parameters under different operating states, and performing linear fitting on the voltage-frequency data corresponding to each chip parameter under different operating states.

[0106] The inflection point voltage is obtained by analyzing the voltage-frequency data through linear fitting of the Monte Carlo model to obtain the voltage-frequency curves corresponding to each chip parameter under different operating conditions. The voltage is then analyzed for the voltage-frequency curves corresponding to each chip parameter under different operating conditions. The frequency similarity corresponding to the inflection point voltage on the voltage-frequency curve under different operating conditions meets the similarity requirements.

[0107] Before the inflection point analysis model is applied, it is first necessary to model and train the inflection point analysis model. In some embodiments, the chip parameters of multiple sample chips are obtained, the chip parameters of the multiple sample chips are input into the pre-trained voltage-frequency model, the voltage-frequency data corresponding to the multiple chip parameters under different operating states are output, the voltage-frequency data corresponding to each chip parameter under different operating states are linearly fitted, the inflection point voltage corresponding to each chip parameter is obtained, and the inflection point voltage corresponding to each chip parameter is modeled to obtain the inflection point analysis model.

[0108] When determining the inflection point voltage by linear fitting, the voltage-frequency data is analyzed by linear fitting of the Monte Carlo model to obtain the voltage-frequency curves corresponding to each chip parameter under different operating conditions; the voltage-frequency curves corresponding to each chip parameter under different operating conditions are analyzed to obtain the inflection point voltages corresponding to each chip parameter, wherein the frequency similarity corresponding to the inflection point voltage on the voltage-frequency curve under different operating conditions meets the similarity requirement.

[0109] In other embodiments, the chip parameters and operation scenarios of multiple sample chips are obtained, and the chip parameters and operation scenarios of multiple sample chips are input into the pre-trained voltage-frequency model, and the voltage-frequency data corresponding to the multiple chip parameters under different operation states are output, and the voltage-frequency data corresponding to each chip parameter under different operation states are linearly fitted to obtain the inflection point voltage corresponding to each chip parameter, and each chip parameter, operation scenario and corresponding inflection point voltage are modeled to obtain an inflection point analysis model. When the inflection point analysis model is applied, the chip parameters and operation scenarios of the first chip (such as: the chip usage scenario or the chip load condition) are input, and the inflection point voltage of the first chip is output.

[0110] When determining the inflection point voltage by linear fitting, the voltage frequency data is analyzed by linear fitting of the Monte Carlo model to obtain the voltage-frequency curves corresponding to each chip parameter and operating scenario under different operating conditions; the voltage-frequency curves corresponding to each chip parameter and operating scenario under different operating conditions are analyzed to obtain the inflection point voltages corresponding to each chip parameter and operating scenario, wherein the frequency similarity corresponding to the inflection point voltage on the voltage-frequency curve under different operating conditions meets the similarity requirement.

[0111] For illustration, we will take the chip parameter analysis as an example. Please refer to Figure 3 , which shows a schematic diagram of the training process of an inflection point analysis model provided by an exemplary embodiment of the present application. Figure 3 As shown, firstly, chip parameters 310 of a plurality of sample chips are obtained, and the chip parameters 310 are input into a pre-trained voltage-frequency model 320, and the voltage-frequency data corresponding to each chip parameter at different operating temperatures are output.

[0112] Linear fitting is performed on the voltage-frequency data corresponding to each chip parameter at different operating temperatures, and the voltage-frequency curves 330 corresponding to each chip parameter at different operating temperatures are obtained through the Monte Carlo model. The voltage-frequency curves 330 corresponding to each chip parameter at different operating temperatures are analyzed to obtain the inflection point voltages 340 corresponding to each chip parameter. Indicatively, for the voltage-frequency curves corresponding to chip parameter A at different operating temperatures, the differences between the frequency data corresponding to the multiple voltage-frequency curves at each voltage are collected, such as obtaining the standard deviation between the frequency data as the difference data, and selecting the voltage point with the smallest difference data as the inflection point voltage corresponding to the chip parameter A.

[0113] After obtaining the inflection point voltage 340 corresponding to each chip parameter, the chip parameters and the inflection point voltage are modeled to obtain an inflection point analysis model 350. The inflection point analysis model 350 is a model obtained by modeling the chip parameters and the inflection point voltage, and is used to output the corresponding inflection point voltage according to the chip parameters.

[0114] By using the inflection point model obtained through pre-training, only the chip parameters need to be input to obtain the inflection point voltage corresponding to the chip parameters, thereby improving the efficiency of confirming the inflection point voltage and avoiding the need to re-fit the voltage-frequency curve corresponding to each temperature when determining the inflection point voltage, thereby reducing the time consumption for confirming the inflection point voltage.

[0115] 3.2 The operating status data includes the operating temperature. After obtaining the voltage-frequency curve or voltage-frequency table corresponding to the operating temperature, the voltage-frequency point is determined from the voltage-frequency curve or voltage-frequency table based on the load condition or operating scenario of the first chip.

[0116] In this embodiment, the voltage-frequency point is determined based on the load condition of the first chip as an example. Based on the operating temperature of the first chip, a voltage-frequency curve corresponding to the operating temperature is obtained. The load data of the first chip is obtained, that is, the description information of the input data of the first chip is obtained as the load data, such as: the amount of data input by the first chip in a historical time period. If the amount of data is greater than or equal to the preset threshold, the highest operating voltage and the corresponding operating frequency in the voltage-frequency curve corresponding to the operating temperature are selected; if the amount of data is less than the preset threshold, the lowest operating voltage and the corresponding operating frequency in the voltage-frequency curve corresponding to the operating temperature are selected.

[0117] In some embodiments, after the first chip is running, the voltage-frequency curve corresponding to some temperatures of the first chip within the operating temperature range is first stored, and when the operating temperature of the first chip is updated and is not within the voltage-frequency curve range corresponding to some temperatures, the stored voltage-frequency curve is updated. Schematically, after the first chip is powered on, the operating temperature is collected to be 10°, then the voltage-frequency curves corresponding to 8°, 9°, 10°, 11°, and 12° are stored respectively, and when the operating temperature of the first chip is updated to 13°, the stored voltage-frequency curve is updated, the chip parameters of the first chip are re-analyzed through the voltage-frequency model, and the voltage-frequency curves corresponding to 11°, 12°, 13°, 14°, and 15° are stored respectively.

[0118] It should be noted that the above storage method of the voltage-frequency curve is only a schematic example.

[0119] By acquiring operating data, a voltage frequency point corresponding to the operating status data is determined from multiple voltage frequency points of the first chip, so that the voltage frequency point can not only adapt to the first chip itself, but also adapt to the operating status data of the first chip that changes dynamically during operation, thereby improving the adaptability of the voltage frequency point, improving the accuracy of the voltage frequency point, improving the performance of the first chip, and avoiding waste of resources of the first chip.

[0120] In the above embodiment, the operating status data is acquired through sensors, that is, the sensor data acquired by the sensor in the first chip is obtained as the operating status data, wherein, when the operating status data includes the operating temperature, the sensor includes a temperature sensor; when the operating status data includes the current voltage, the sensor includes a voltage sensor.

[0121] Step 260: Control the first chip to operate at an operating voltage and an operating frequency based on the voltage-frequency point.

[0122] The first chip is controlled to operate at an operating voltage and an operating frequency within the voltage-frequency point, thereby improving the adaptability between the voltage-frequency point and the first chip.

[0123] Optionally, after driving the first chip to operate at an operating voltage and an operating frequency, it is also necessary to update the operating status data characterizing the operating status of the first chip in real time according to a preset period, and obtain the voltage-frequency point corresponding to the updated operating status data in the voltage-frequency relationship, and control the operation of the first chip based on the voltage-frequency point corresponding to the updated operating status data.

[0124] The operating status data is updated in real time at a preset period, and the voltage frequency point is updated in real time according to the operating status data, so that when the operating status of the first chip changes, the voltage frequency point can also be adjusted in time to adapt to the change of the operating status, thereby improving the adaptability of the voltage frequency point.

[0125] To summarize, the method provided in the embodiment of the present application designs and trains a voltage-frequency model, analyzes the chip parameters of the first chip, and outputs a voltage-frequency point corresponding to the first chip. The voltage-frequency point is a voltage-frequency point customized for the first chip and conforms to the chip parameters of the first chip, that is, the compatibility between the voltage-frequency point and the chip parameters of the first chip is improved. At the dynamically determined voltage-frequency point, the performance of the first chip is improved, and the problem of poor performance of a uniform voltage-frequency point on the first chip is avoided.

[0126] In some embodiments, the voltage-frequency model needs to be trained before being applied. Figure 4 FIG. 1 is a flowchart of a chip control method provided by another exemplary embodiment of the present application, and the method is executed by a computer device. Schematically, the method can be executed by a terminal, or by a server, or by a terminal and a server in cooperation. Figure 4 As shown, before the above step 240, the following steps are also included.

[0127] Step 420: Obtain chip parameters of the sample chip.

[0128] Optionally, the process of acquiring the chip parameters of the sample chip may refer to the process of acquiring the chip parameters of the first chip, which will not be described in detail in this embodiment.

[0129] In some embodiments, the chip parameters of all sample chips are acquired in advance during the data collection phase; or, when a sample chip is used for a running test, the chip parameters of the sample chip are acquired.

[0130] Step 440 : collecting a voltage-frequency curve of the sample chip during operation of the sample chip.

[0131] In some embodiments, during the operation of the sample chip, stress test data of the sample chip is collected, and the stress test data is used to indicate the performance of the sample chip under different operating voltages and operating frequencies during the operation. During the operation of the sample chip, voltage-frequency data pairs are collected, and the voltage-frequency data pairs include the correspondence between the operating voltage and the operating frequency during the operation of the sample chip.

[0132] There is a corresponding relationship between the voltage-frequency data pair and the stress test data, that is, after the data to be processed is input to the sample chip, the stress test data and the voltage-frequency data pair collected at the same time are regarded as a set of corresponding data, indicating the performance of the sample chip under the voltage-frequency data pair. Multiple sets of voltage-frequency data pairs that meet the stress test requirements are obtained, and the multiple sets of voltage-frequency data pairs are linearly fitted to obtain a voltage-frequency curve.

[0133] In schematic form, after the data to be processed is input into the sample chip, the stress test data corresponding to the sample chip at different voltage frequencies are collected, and the voltage frequencies corresponding to the stress test data whose stress test results meet the stress test requirements are screened, such as: the voltage frequency data corresponding to the stress test data whose delay is less than the delay duration threshold, and the screened voltage frequency data are fitted to obtain a voltage frequency curve.

[0134] In some embodiments, the voltage-frequency curve of the sample chip includes voltage-frequency curves corresponding to the sample chip at different operating temperatures.

[0135] Collect the voltage and frequency data corresponding to the sample chip at different operating temperatures during operation, and collect the stress test data of the sample chip during the operation of the sample chip. There is a corresponding relationship between the operating temperature, voltage and frequency data, and stress test data. That is, after inputting the data to be processed into the sample chip, the stress test data, temperature data, and voltage and frequency data collected at the same time are used as a set of corresponding data, indicating the performance of the sample chip at the voltage and frequency data and operating temperature. Obtain multiple sets of voltage and frequency data and temperature data that meet the stress test requirements, and linearly fit the multiple sets of voltage and frequency data and temperature data to obtain the voltage and frequency curve of the sample chip at each operating temperature.

[0136] In schematic form, after inputting the data to be processed into the sample chip, the stress test data corresponding to the sample chip at different operating temperatures and different voltage frequencies are collected, and the voltage frequencies corresponding to the stress test data whose stress test results meet the stress test requirements are screened, such as: the operating temperature and voltage frequency data corresponding to the stress test data whose delay is less than the delay time threshold, and fitting the screened voltage and frequency data to obtain the voltage and frequency curves corresponding to each operating temperature.

[0137] The data corresponding to the same chip parameters are used to construct a voltage-frequency curve corresponding to the chip parameters.

[0138] In some embodiments, when collecting the voltage-frequency curve of a sample chip, constraints such as chip load or chip specifications may be added. Schematically, the chip load is taken as an example to illustrate, that is, the operating temperature, voltage-frequency data, and stress measurement data of the sample chip under a specific load are obtained, thereby obtaining the voltage-frequency curves corresponding to each temperature of the sample chip under the specific load.

[0139] Step 460 , modeling the relationship between the chip parameters of the sample chip and the voltage-frequency curve to obtain a voltage-frequency model.

[0140] The voltage-frequency model is used to construct the relationship between chip parameters and voltage-frequency curves; optionally, the voltage-frequency model is used to construct voltage-frequency curves corresponding to chip parameters at various operating temperatures; optionally, the voltage-frequency model is used to construct voltage-frequency curves corresponding to chip parameters at various operating temperatures under specific load environments.

[0141] In some embodiments, the relationship between the chip parameters of the sample chip and the voltage-frequency curve is modeled to obtain a candidate voltage-frequency model, and the candidate voltage-frequency model is subjected to deviation correction to obtain a voltage-frequency model.

[0142] That is, the voltage-frequency model models the relationship between the chip parameters of the sample chip and the voltage-frequency curve. After the candidate voltage-frequency model is obtained, the candidate voltage-frequency model is bias-corrected to expand the coverage of the voltage-frequency curve predicted by the candidate voltage-frequency model to obtain the model.

[0143] Deviation correction refers to expanding the coverage of the voltage frequency curve predicted by the candidate voltage frequency model. Schematically, the candidate voltage frequency model is subjected to deviation correction based on the load difference deviation to obtain a voltage frequency model, such as: in the training, the load data of the sample chip is input from 1.5 to 3.0, and in the application process, the load of the chip reaches 3.2, then the voltage frequency curve covered by the candidate voltage frequency model based on the load difference deviation is offset in the direction of increasing the voltage, and the voltage frequency model is obtained, so that the increased voltage frequency curve can adapt to the maximum load of 3.2; or, the candidate piezoelectric frequency model is subjected to deviation correction based on the edge range deviation of the chip operating temperature to obtain a voltage frequency model, schematically, the candidate voltage frequency model is subjected to deviation correction based on the edge range deviation of the operating temperature to obtain a voltage frequency model, such as: in the training, the operating temperature of the sample chip is from 0° to 90°, and in the application process, the operating temperature of the chip reaches 100°, then the voltage frequency curve covered by the candidate voltage frequency model based on the edge range deviation of the operating temperature is stretched in the direction of increasing the voltage, and the voltage frequency model is obtained, so that the increased voltage frequency curve can adapt to the maximum operating temperature of 100°. The process of performing deviation correction on the candidate voltage-frequency model to obtain the voltage-frequency model can be performed by the developer, or can be automatically generated by an algorithm after analyzing the parameter difference.

[0144] Figure 5 FIG. 1 is a schematic diagram of a training process of a voltage-frequency model provided by an exemplary embodiment of the present application. Figure 5 As shown, during the training process of the voltage-frequency model, the chip parameters of the sample chip and the key parameters of the sample chip during operation are collected, such as Figure 5 In the illustrated embodiment, the operating temperature data 511 of the sample chip is collected by the collector 501, the chip parameters 512 of the sample chip are collected by the collector 502, and the voltage frequency data 513 of the sample chip is collected by the collector 503. The collected operating temperature data 511, chip parameters 512, and voltage frequency data 513 are corrected 510 to obtain the corrected operating temperature data 511, chip parameters 512, and voltage frequency data 513.

[0145] Given an operation scenario 520 of the sample chip, such as a specific load or a specific specification, load data 530 corresponding to the operation scenario 520 is input to the sample chip under the operation scenario 520, and performance data of the sample chip under the load data 530 is collected, and data processing 540 is performed on the performance data, such as data cleaning and data correction, to obtain voltage-frequency curves 550 corresponding to the sample chip at different operation temperatures. Through the chip parameters 512 and the voltage-frequency curves 550 corresponding to the sample chip at different operation temperatures, the relationship between the chip parameters 512 of the sample chip and the voltage-frequency curve 550 is modeled to obtain a candidate voltage-frequency model 560.

[0146] The candidate voltage-frequency model 560 is subjected to deviation correction to obtain a voltage-frequency model 570 .

[0147] By collecting the chip parameters of the sample chip and the voltage-frequency curve during the operation of the sample chip as samples, a voltage-frequency model is constructed and trained, which improves the efficiency of the voltage-frequency model in outputting the voltage-frequency curve and improves the analysis ability of the chip parameters.

[0148] To summarize, the method provided in the embodiment of the present application designs and trains a voltage-frequency model, analyzes the chip parameters of the first chip, and outputs a voltage-frequency point corresponding to the first chip. The voltage-frequency point is a voltage-frequency point customized for the first chip and conforms to the chip parameters of the first chip, that is, the compatibility between the voltage-frequency point and the chip parameters of the first chip is improved. At the dynamically determined voltage-frequency point, the performance of the first chip is improved, and the problem of poor performance of a uniform voltage-frequency point on the first chip is avoided.

[0149] When analyzing the voltage frequency point of the first chip according to the voltage frequency model, the chip control method provided in the embodiment of the present application is mainly through constructing and training the voltage frequency model and the inflection point analysis model, and after the first chip is powered on, determining the voltage frequency point of the first chip through the voltage frequency model and the inflection point analysis model.

[0150] The embodiment of the present application mainly includes the following stages: 1. Data preparation stage; 2. Modeling stage of voltage-frequency model; 3. Modeling stage of inflection point analysis model; 4. Application stage. The above stages are described in detail below.

[0151] 1. Data preparation stage

[0152] Indicatively, Figure 6 As shown in the figure, the data preparation stage mainly includes the following steps:

[0153] Step 601, a sensor for monitoring data is set at a core position according to the structure of the chip.

[0154] In the embodiment of the present application, the sensor is implemented as a temperature sensor as an example for explanation, and the temperature sensor is set at the core position of the chip structure, and the core position refers to the position with the highest temperature in the chip structure. In some embodiments, a plurality of temperature sensors are set in the chip structure.

[0155] Schematically, temperature sensors are arranged at locations where computing units are densely distributed in the chip structure.

[0156] In step 601, the setting position of the temperature sensor is an initial candidate position, and the initial candidate position is a position to be corrected.

[0157] Accurate modeling requires obtaining the physical parameters of the chip itself in advance, relying on the data returned by dedicated sensors inside the chip, and the sensors need to be calibrated and fitted.

[0158] After the temperature sensor collects data, it sends the collected temperature data to the temperature detection unit. Temperature monitoring unit: obtains the operating temperature of the chip through the sensor inside the chip, and sends the operating temperature to the system management unit. The system management unit is a system management control unit (SMC) composed of the CPU, which is responsible for reading the physical parameters of the chip and the data returned by the temperature monitoring unit, and setting the corresponding operating voltage and operating frequency according to the preset voltage-frequency model.

[0159] Step 602: According to the working scenario and design requirements of the chip, a load is designed to perform a simulation stress test on the chip to obtain a thermal distribution map of the chip.

[0160] That is, according to the working scenarios and design requirements of the chip, load data is input into the chip to control the operation of the chip, and a thermal distribution map of the chip is generated based on the temperature data collected by the sensor.

[0161] Illustratively, the chip design requirements indicate that the chip is used for AI computing, so training sample data in the AI ​​training process is input into the chip, the chip is controlled to perform AI training computing, and temperature data collected by the temperature sensor is collected.

[0162] Since a plurality of temperature sensors are arranged in the chip, the temperature data respectively collected by each temperature sensor is collected, and the current thermal distribution map of the chip, that is, the thermal map, is constructed based on the temperature data respectively collected by each temperature sensor.

[0163] Step 603: adjusting the location of the sensor in the chip according to the thermal distribution map.

[0164] Based on the thermal distribution map, the chip structure is analyzed to determine whether the current sensor location belongs to the core area with higher heat. By setting the sensor to the core area with higher heat, the deviation between the collected temperature data and the actual temperature data is reduced.

[0165] Step 604 , selecting different sample chips with different process ranges according to chip manufacturing process and process requirements, so as to improve the process coverage of the sample chips.

[0166] In schematic form, the chip parameters of each sample chip are collected, and based on the process design range of each chip, such as the number of chip cores, the chip memory size, etc., sample chips corresponding to each chip parameter are selected as much as possible within the chip parameter range corresponding to the process design as chips for training the voltage-frequency model, so as to improve the chip parameter range coverage of the sample chips.

[0167] 2. Modeling stage of voltage-frequency model

[0168] Indicatively, Figure 7 As shown, the modeling stage of the voltage-frequency model mainly includes the following steps.

[0169] Step 701, conduct a large number of functional and stress tests on chips of different processes in specific scenarios, and process the collected data.

[0170] Indicatively, chips of different processes are stress tested in specific scenarios corresponding to chip design, and chip parameters, operating temperature data during chip operation, and voltage and frequency data during chip operation are collected. In addition, stress test data during chip operation is collected to characterize the performance of the chip in specific scenarios.

[0171] After collecting the above data, data processing is performed on the collected data. The data processing process includes data cleaning, data correction, etc. Indicatively, the collected data is deduplicated, and the stress test data in the collected data that does not meet the stress test requirements is screened, etc. This embodiment is not limited to this.

[0172] Indicative, reference Figure 5 In the illustrated process, the operating temperature data 511, chip parameters 512, and voltage-frequency data 513 of the sample chip are collected, and the collected operating temperature data 511, chip parameters 512, and voltage-frequency data 513 are corrected 510.

[0173] Given an operating scenario 520 for a given sample chip, such as a specific load or a specific specification, load data 530 corresponding to the operating scenario 520 is input into the sample chip under the operating scenario 520, and performance data of the sample chip under the load data 530 is collected, and data processing 540 is performed on the performance data, such as data cleaning, data correction, etc., to obtain voltage-frequency curves 550 corresponding to the sample chip at different operating temperatures.

[0174] Step 702, construct the processed data into a specific data set, input it into a deep learning model for training and modeling, and obtain a candidate voltage-frequency model.

[0175] The above operating temperature data, chip parameters, voltage-frequency data and voltage-frequency curve are stored as a set of data in a data set as sample data for constructing a voltage-frequency model.

[0176] The data stored in the data set is input into the deep learning model for training to obtain a candidate voltage-frequency model. The deep learning model can be implemented as a mathematical model or an AI model, which is not limited in the embodiments of the present application.

[0177] The candidate voltage-frequency model is a model trained based on the chip parameters of the above-mentioned sample chip and the data collected during the operation process. However, there is a problem that the data such as the chip parameters or operating temperature of the sample chip cannot be covered by the boundaries, that is, the operating temperature of the chip in the sample chip is reflected as a maximum of 90°, but in actual applications there is an operating temperature of 100°. Therefore, the candidate voltage-frequency model is the model to be corrected.

[0178] Step 703: Add boundary parameters to the candidate voltage-frequency model to obtain a voltage-frequency model.

[0179] The candidate voltage-frequency model is subjected to deviation correction to obtain a voltage-frequency model. Deviation correction refers to expanding the coverage of the voltage-frequency curve predicted by the candidate voltage-frequency model. Schematically, the candidate voltage-frequency model is subjected to deviation correction based on the load difference deviation to obtain a voltage-frequency model; or, the candidate piezoelectric frequency model is subjected to deviation correction based on the edge range deviation of the chip operating temperature to obtain a voltage-frequency model.

[0180] 3. Modeling stage of inflection point analysis model

[0181] Indicatively, Figure 8 As shown, the modeling phase of the inflection point analysis model mainly includes the following steps.

[0182] Step 801, under chip manufacturing process and process conditions, generate a voltage-frequency model obtained by chip parameter input training, and voltage-frequency curves corresponding to the chip parameters at different operating temperatures.

[0183] In some embodiments, the sample chips used for training the voltage-frequency model and the training inflection point analysis model are sample chips in the same data set, or the sample chips used for training the voltage-frequency model and the training inflection point analysis model are different sample chips in different data sets, which is not limited in the embodiments of the present application.

[0184] Obtain the chip parameters of each sample chip and input them into the above Figure 7 In the trained voltage-frequency model, the voltage-frequency curves corresponding to the chip parameters at different operating temperatures are output.

[0185] Indicatively, Fig. 9 As shown, it shows a schematic diagram of a voltage-frequency curve provided by an exemplary embodiment of the present application, which includes chip parameters 900, and the voltage-frequency curve includes a voltage-frequency curve 910 of the chip of the chip parameters 900 when the operating temperature is 0° and a voltage-frequency curve 920 when the operating temperature is 100°.

[0186] Step 802: Based on the Monte Carlo model, the voltage frequency curves of different sample chips at various sampling temperatures are fitted to obtain the inflection point voltages of different sample chips.

[0187] Monte Carlo model is a statistical method that simulates or understands the probabilistic process of complex systems or mathematical problems through random sampling. Monte Carlo model can be applied in various fields, including but not limited to physics, engineering, finance and economics, and biostatistics.

[0188] In the Monte Carlo model, probability problems are calculated or simulated by repeated random sampling to obtain numerical solutions to the problem.

[0189] In an embodiment of the present application, the voltage-frequency curves of different sample chips at each sampling temperature are fitted by a Monte Carlo model, and the differences between the operating frequencies corresponding to different temperatures under each operating voltage are sampled, and the operating voltage with the smallest operating frequency difference is selected as the inflection point voltage.

[0190] Indicative, for Fig. 9 The chip parameter 900 shown corresponds to voltage-frequency curves at different operating temperatures, and the differences between the frequency data corresponding to the multiple voltage-frequency curves at each operating voltage are collected, such as: obtaining the standard deviation between the frequency data as the difference data, and selecting the voltage point with the smallest difference data as the inflection point voltage corresponding to the chip parameter 900, so as to obtain the standard deviation between the frequency data as the difference data, and then obtaining ... Fig. 9 Taking the curve shown as an example, the inflection point voltage is about 940mV.

[0191] Step 803 , modeling the chip parameters and inflection point voltage of the sample chip to obtain an inflection point analysis model.

[0192] Construct a mathematical relationship model between chip parameters and inflection point voltage. After obtaining the inflection point voltage corresponding to each chip parameter, the chip parameters and the inflection point voltage are modeled to obtain an inflection point analysis model. The inflection point analysis model is a model obtained by modeling the chip parameters and the inflection point voltage, and is used to output the corresponding inflection point voltage according to the chip parameters.

[0193] After obtaining the above voltage-frequency model and inflection point analysis model, the voltage-frequency model and the inflection point analysis model are implanted into the firmware code of the chip, wherein the physical parameters of the chip are obtained by reading the eFuse.

[0194] 4. Application Phase

[0195] Indicatively, Fig.10 As shown, the application phase mainly includes the following steps.

[0196] Step 1001: After the system management unit is powered on, the chip parameters of the first chip are obtained, and the voltage-frequency relationship of the chip is generated according to the voltage-frequency model for storage.

[0197] The voltage-frequency relationship may be implemented as a voltage-frequency curve, or as a voltage-frequency table obtained by sampling the voltage-frequency curve.

[0198] Different voltage-frequency curves are selected according to different scenarios and expected temperature ranges. For example, when a conservative strategy is adopted, the voltage-frequency tables for the lowest temperature and the highest temperature are saved separately, that is, each chip saves two different voltage-frequency tables.

[0199] Step 1002, when the first chip is working, the chip parameters are analyzed according to the inflection point model to obtain the inflection point voltage, and the preset inflection point temperature is obtained, and the operating temperature and current voltage returned by the sensor are obtained.

[0200] Among them, it is determined whether the current voltage is higher than the inflection point voltage and whether the operating temperature is higher than the inflection point temperature in the current working scenario of the first chip.

[0201] Optionally, when the operating temperature is greater than the inflection point temperature and the current voltage is greater than the inflection point voltage, a first voltage frequency point is determined from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, and the first voltage frequency point is a frequency point corresponding to the high temperature and high pressure obtained by pre-fitting in the voltage-frequency relationship;

[0202] When the operating temperature is greater than the inflection point temperature and the current voltage is less than the inflection point voltage, a second voltage frequency point is determined from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, and the second voltage frequency point is a frequency point corresponding to high temperature and low voltage obtained by pre-fitting in the voltage-frequency relationship;

[0203] When the operating temperature is lower than the inflection point temperature and the current voltage is higher than the inflection point voltage, a third voltage frequency point is determined from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, and the third voltage frequency point is a frequency point corresponding to the low temperature and high pressure obtained by pre-fitting in the voltage-frequency relationship;

[0204] When the operating temperature is lower than the inflection point temperature and the current voltage is lower than the inflection point voltage, a fourth voltage-frequency point is determined from the voltage-frequency relationship as the voltage-frequency point corresponding to the operating status data, and the fourth voltage-frequency point is a frequency point corresponding to the low temperature and low pressure obtained by pre-fitting in the voltage-frequency relationship.

[0205] Step 1003: if the operating temperature of the first chip changes in the working scenario, the voltage frequency point is updated.

[0206] The sensor parameters are periodically acquired and the above steps 1001 to 1003 are repeated.

[0207] Optionally, the operating status data characterizing the operating status of the first chip is updated in real time according to a preset period, and the voltage-frequency point corresponding to the updated operating status data in the voltage-frequency relationship is obtained, and the operation of the first chip is controlled based on the voltage-frequency point corresponding to the updated operating status data.

[0208] In some embodiments, under the conservative strategy, the voltage-frequency tables of the lowest temperature and the highest temperature are saved respectively, and the voltage frequency point is directly updated from the voltage-frequency table according to the updated temperature data; under the load strategy, the voltage-frequency relationship corresponding to some of the sampling temperatures in all the sampling temperatures is selected for storage, such as: within the temperature range of 0° to 100°, if the current operating temperature of the first chip is 10°, the voltage-frequency relationship within the range of 0° to 50° is stored; and the voltage-frequency curve corresponding to the current operating temperature is selected from the stored voltage-frequency relationship, thereby selecting the voltage-frequency point. If the operating temperature of the first chip is increased to 55°, the stored voltage-frequency relationship within the range of 0° to 50° is deleted, and the voltage-frequency relationship within the range of 51° to 100° is stored, and the voltage-frequency curve corresponding to the current operating temperature is selected from the stored voltage-frequency relationship, thereby selecting the voltage-frequency point.

[0209] The above storage method of the voltage-frequency relationship is only an illustrative example and is not limited in this embodiment.

[0210] Fig.11FIG. 1 is a schematic diagram of a control process of a chip in an application phase provided by another exemplary embodiment of the present application. Fig.11 As shown, the application stage includes: obtaining a voltage-frequency model 1101 and an inflection point model 1102, obtaining chip parameters 1103 of the first chip, and storing the voltage-frequency model 1101, the inflection point model 1102 and the chip parameters 1103 of the first chip in a storage area.

[0211] During the operation of the first chip, the chip parameters 1103 are analyzed by the voltage-frequency model 1101 to obtain the voltage-frequency curves corresponding to the first chip at different temperatures, and the chip parameters 1103 are analyzed by the inflection point model 1102 to obtain the inflection point voltage of the first chip.

[0212] If the current voltage of the first chip is less than the inflection point voltage and the operating temperature is less than the inflection point temperature, the voltage frequency point is determined according to the first voltage frequency curve (such as the portion of the voltage frequency curve that is less than the inflection point voltage at 0°); if the current voltage of the first chip is less than the inflection point voltage and the operating temperature is greater than the inflection point temperature, the voltage frequency point is determined according to the second voltage frequency curve (such as the portion of the voltage frequency curve that is less than the inflection point voltage at 100°); if the current voltage of the first chip is greater than the inflection point voltage and the operating temperature is less than the inflection point temperature, the voltage frequency point is determined according to the first voltage frequency curve (such as the portion of the voltage frequency curve that is greater than the inflection point voltage at 0°); if the current voltage of the first chip is greater than the inflection point voltage and the operating temperature is greater than the inflection point temperature, the voltage frequency point is determined according to the first voltage frequency curve (such as the portion of the voltage frequency curve that is greater than the inflection point voltage at 100°).

[0213] After determining the new voltage-frequency point 1140, the first chip is controlled to operate at the operating voltage 1141 and the operating frequency 1142 of the new voltage-frequency point. During the continuous operation of the first chip, the temperature data and the voltage data of the first chip are continuously monitored and collected.

[0214] To summarize, the method provided in the embodiment of the present application designs and trains a voltage-frequency model, analyzes the chip parameters of the first chip, and outputs a voltage-frequency point corresponding to the first chip. The voltage-frequency point is a voltage-frequency point customized for the first chip and conforms to the chip parameters of the first chip, that is, the compatibility between the voltage-frequency point and the chip parameters of the first chip is improved. At the dynamically determined voltage-frequency point, the performance of the first chip is improved, and the problem of poor performance of a uniform voltage-frequency point on the first chip is avoided.

[0215] The method provided in the embodiment of the present application provides a unique voltage frequency definition method, and obtains the operating characteristic parameters of the chip through scientific testing methods and data modeling, so that each chip can operate at the optimal performance point.

[0216] The method provided in the embodiment of the present application is adaptable to a variety of different scenarios, analyzes the working characteristics of the chip at different temperatures, finds the turning point of the impact of temperature on performance, and introduces the judgment of the turning point, so that the chip can work in the best state at different temperatures.

[0217] The method provided in the embodiment of the present application has at least the following beneficial effects:

[0218] 1. Flexible voltage and frequency setting: The voltage and frequency setting method in the related art relies on preset fixed points, which are based on average values ​​or standard process presets. However, this method does not take into account the uniqueness of each chip, resulting in the inability to achieve optimal performance for each chip, thereby causing performance loss. In contrast, the method provided in the embodiment of the present application adopts a flexible adjustment strategy that can be personalized according to the specific characteristics and performance potential of each chip. It not only improves the overall performance of the chip, but also ensures maximum energy efficiency because each chip can operate at its optimal voltage and frequency point. Through real-time analysis and adjustment, our method can accurately match the unique needs of each chip, thereby optimizing performance, reducing energy consumption, and enhancing the market competitiveness of the product.

[0219] 2. Adjustment speed and flexibility: In related technologies, fixed voltage and frequency points cannot quickly adapt to sudden changes in workload. The method provided in the embodiment of the present application adopts dynamic adjustment technology to match the optimal voltage and frequency for different scenarios and states, which can instantly respond to changes in chip performance requirements and quickly adapt to the increase or decrease of workload.

[0220] 3. Extreme performance: In related technologies, due to the lack of attention to the differences between single chips, it is often impossible to achieve the highest performance of the chip in specific applications or extreme temperature conditions. The method provided in the embodiment of the present application analyzes the performance of each chip under different temperature and load conditions in detail, automatically adjusts the voltage frequency, optimizes the performance, and ensures that the chip can maintain better performance under high load, especially in high-performance computing scenarios, thereby improving computing efficiency and reliability.

[0221] Fig.12 is a schematic diagram of the structure of a chip control device provided by an exemplary embodiment of the present application. Fig.12 As shown, the device comprises:

[0222] The acquisition module 1210 is configured to acquire chip parameters of the first chip, where the chip parameters are used to indicate physical characteristics of the first chip;

[0223] The analysis module 1220 is configured to analyze the chip parameters of the first chip through a voltage-frequency model to obtain a voltage-frequency point corresponding to the first chip, wherein the voltage-frequency point includes an operating voltage of the first chip and an operating frequency corresponding to the operating voltage, wherein the voltage-frequency model is a model obtained by training the chip parameters of the sample chip and the screened voltage-frequency data after screening the voltage-frequency data of the sample chip based on the stress test data;

[0224] The control module 1230 is configured to control the first chip to operate at the operating voltage and the operating frequency based on the voltage-frequency point.

[0225] In an optional embodiment, the analysis module 1220 is further configured to analyze chip parameters of the first chip through a pre-trained voltage-frequency model to obtain a voltage-frequency relationship corresponding to the first chip, wherein the voltage-frequency relationship includes voltage-frequency points corresponding to the first chip in a plurality of different operating states;

[0226] like Fig.13 As shown, the device also includes:

[0227] The acquisition module 1240 is configured to acquire operation status data for representing the operation status of the first chip;

[0228] The acquisition module 1210 is further configured to dynamically determine in real time a voltage-frequency point in the voltage-frequency relationship corresponding to the operating status data.

[0229] In an optional embodiment, the acquisition module 1210 is also configured to acquire inflection point status data corresponding to the first chip, the inflection point status data is used to compare with the operating status data, and determine the voltage frequency point based on the comparison result; acquire a comparison relationship between the operating status data and the inflection point status data; and determine the voltage frequency point in the voltage-frequency relationship corresponding to the operating status data based on the comparison relationship.

[0230] In an optional embodiment, the operating status data includes operating temperature;

[0231] The acquisition module 1210 is further configured to acquire an inflection point temperature corresponding to the first chip, where the inflection point temperature is a preset temperature; and determine a voltage frequency point in the voltage-frequency relationship corresponding to the operating status data based on a first comparison relationship between the operating temperature and the inflection point temperature.

[0232] In an optional embodiment, the operating status data includes a current voltage;

[0233] The acquisition module 1210 is also configured to acquire the inflection point voltage corresponding to the first chip, where the inflection point voltage is a voltage at which the voltage-frequency performance of the first chip is similar at different operating temperatures; and determine the voltage frequency point in the voltage-frequency relationship corresponding to the operating status data based on a second comparison relationship between the current voltage and the inflection point voltage.

[0234] In an optional embodiment, the acquisition module 1210 is further configured to analyze the chip parameters of the first chip through a pre-trained inflection point analysis model to obtain the inflection point voltage corresponding to the first chip.

[0235] In an optional embodiment, the acquisition module 1210 is further configured to acquire chip parameters of multiple sample chips; input the chip parameters of the multiple sample chips into the pre-trained voltage-frequency model, and output voltage-frequency data corresponding to the multiple chip parameters under different operating states; perform linear fitting on the voltage-frequency data corresponding to each chip parameter under different operating states to obtain the inflection point voltage corresponding to each chip parameter; and model each chip parameter and the inflection point voltage corresponding to the chip parameter to obtain the inflection point analysis model.

[0236] In an optional embodiment, the acquisition module 1210 is further configured to analyze the voltage-frequency data through Monte Carlo model linear fitting to obtain voltage-frequency curves corresponding to each chip parameter under different operating conditions; analyze the voltage-frequency curves corresponding to each chip parameter under different operating conditions to obtain the inflection point voltage corresponding to each chip parameter, wherein the frequency similarity corresponding to the inflection point voltage on the voltage-frequency curve under different operating conditions meets the similarity requirement.

[0237] In an optional embodiment, the acquisition module 1210 is further configured to determine a first voltage frequency point from the voltage-frequency relationship as a voltage frequency point corresponding to the operating state data when the operating temperature is greater than the inflection point temperature and the current voltage is greater than the inflection point voltage, wherein the first voltage frequency point is a frequency point corresponding to high temperature and high pressure obtained by pre-fitting in the voltage-frequency relationship;

[0238] The acquisition module 1210 is further configured to determine, when the operating temperature is greater than the inflection point temperature and the current voltage is less than the inflection point voltage, a second voltage frequency point from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, wherein the second voltage frequency point is a frequency point corresponding to high temperature and low voltage obtained by pre-fitting in the voltage-frequency relationship;

[0239] The acquisition module 1210 is further configured to determine, when the operating temperature is less than the inflection point temperature and the current voltage is greater than the inflection point voltage, a third voltage frequency point from the voltage-frequency relationship as the voltage frequency point corresponding to the operating state data, wherein the third voltage frequency point is a frequency point corresponding to the low temperature and high pressure obtained by pre-fitting in the voltage-frequency relationship;

[0240] The acquisition module 1210 is also configured to determine a fourth voltage frequency point from the voltage-frequency relationship as the voltage frequency point corresponding to the operating status data when the operating temperature is lower than the inflection point temperature and the current voltage is lower than the inflection point voltage, the fourth voltage frequency point being a frequency point corresponding to the low temperature and low pressure pre-fitted in the voltage-frequency relationship.

[0241] In an optional embodiment, the acquisition module 1210 is further configured to update the operation status data representing the operation status of the first chip in real time according to a preset period;

[0242] The acquisition module 1210 is further configured to acquire a voltage-frequency point in the voltage-frequency relationship corresponding to the updated operating status data;

[0243] The control module 1230 is further configured to control the operation of the first chip based on the voltage-frequency point corresponding to the updated operation status data.

[0244] In an optional embodiment, the acquisition module 1210 is further configured to acquire sensor data collected by a sensor in the first chip as the operating status data;

[0245] Wherein, when the operating status data includes an operating temperature, the sensor includes a temperature sensor; when the operating status data includes a current voltage, the sensor includes a voltage sensor.

[0246] In an optional embodiment, the voltage-frequency relationship includes voltage-frequency points corresponding to the first chip in a plurality of different operating states and operating scenarios, respectively, and the operating scenario is used to characterize a usage scenario of the first chip;

[0247] The acquisition module 1210 is further configured to dynamically determine in real time that the voltage-frequency relationship ends at a voltage-frequency point corresponding to the operating status data and the operating scenario.

[0248] In an optional embodiment, the acquisition module 1210 is further configured to acquire an operation scenario of the first chip;

[0249] The analysis module 1220 is further configured to analyze chip parameters and operation scenarios of the first chip through the voltage-frequency model to obtain voltage frequency points corresponding to the first chip.

[0250] In an optional embodiment, the acquisition module 1210 is further configured to acquire chip parameters of the sample chip;

[0251] The acquisition module 1210 is further configured to collect a voltage-frequency curve of the sample chip during operation of the sample chip; and to model the relationship between the chip parameters of the sample chip and the voltage-frequency curve to obtain the voltage-frequency model.

[0252] In an optional embodiment, the acquisition module 1210 is further configured to collect stress test data of the sample chip during operation of the sample chip, wherein the stress test data is used to indicate the performance of the sample chip at different operating voltages and operating frequencies during operation;

[0253] The acquisition module 1210 is further configured to collect voltage-frequency data pairs during the operation of the sample chip, wherein the voltage-frequency data pairs include a correspondence between an operating voltage and an operating frequency during the operation of the sample chip, wherein there is a correspondence between the voltage-frequency data pairs and the pressure measurement data;

[0254] The acquisition module 1210 is further configured to acquire multiple sets of voltage-frequency data pairs corresponding to stress testing data that meet stress testing requirements; and perform linear fitting on the multiple sets of voltage-frequency data pairs to obtain the voltage-frequency curve.

[0255] In an optional embodiment, the acquisition module 1210 is further configured to model the relationship between the chip parameters of the sample chip and the voltage frequency curve to obtain a candidate voltage frequency model; perform deviation correction on the candidate voltage frequency model to expand the coverage of the voltage frequency curve predicted by the candidate voltage frequency model to obtain the voltage frequency model.

[0256] In an optional embodiment, the device further comprises:

[0257] An input module, configured to input a stimulus signal to the first chip during a test phase of the first chip;

[0258] The acquisition module 1210 is further configured to collect transmission performance data of the excitation signal in the first chip as chip parameters of the first chip; and store the chip parameters in a storage space of the first chip;

[0259] The acquisition module 1210 is further configured to read chip parameters of the first chip from the storage space when the first chip is powered on.

[0260] To summarize, the device provided in the embodiment of the present application designs and trains a voltage-frequency model, analyzes the chip parameters of the first chip, and outputs a voltage-frequency point corresponding to the first chip. The voltage-frequency point is a voltage-frequency point customized for the first chip and conforms to the chip parameters of the first chip, that is, the compatibility between the voltage-frequency point and the chip parameters of the first chip is improved. At the dynamically determined voltage-frequency point, the performance of the first chip is improved, and the problem of poor performance of a uniform voltage-frequency point on the first chip is avoided.

[0261] It should be noted that the control device of the chip provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the control device of the chip provided in the above embodiment is based on the same concept as the control method embodiment of the chip. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0262] The present application also provides a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operation method of the model prediction task provided by each of the above method embodiments. It should be noted that the computer device can be as follows Fig.14 Computer equipment provided.

[0263] Please refer to Fig.14 , which shows a schematic diagram of the structure of a computer device provided by an exemplary embodiment of the present application. Specifically, the computer device 1400 includes a central processing unit (CPU) 1401, a system memory 1404 including a random access memory (RAM) 1402 and a read-only memory (ROM) 1403, and a system bus 1405 connecting the system memory 1404 and the central processing unit 1401. The computer device 1400 also includes a basic input / output system (I / O system) 1406 that helps transmit information between various devices in the computer, and a large-capacity storage device 1407 for storing an operating system 1413, application programs 1414 and other program modules 1415.

[0264] The basic input / output system 1406 includes a display 1408 for displaying information and an input device 1409 such as a mouse and a keyboard for user inputting information. The display 1408 and the input device 1409 are connected to the central processing unit 1401 through an input / output controller 1410 connected to the system bus 1405. The basic input / output system 1406 may also include an input / output controller 1410 for receiving and processing inputs from a plurality of other devices such as a keyboard, a mouse, or an electronic stylus. Similarly, the input / output controller 1410 also provides output to a display screen, a printer, or other types of output devices.

[0265] The mass storage device 1407 is connected to the central processing unit 1401 through a mass storage controller (not shown) connected to the system bus 1405. The mass storage device 1407 and its associated computer readable media provide non-volatile storage for the computer device 1400. That is, the mass storage device 1407 may include a computer readable medium (not shown) such as a hard disk or a CD-ROM drive.

[0266] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technologies, CD-ROM, DVD or other optical storage, tape cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that the computer storage media is not limited to the above. The above-mentioned system memory 1404 and mass storage device 1407 can be collectively referred to as memory.

[0267] The memory stores one or more programs, and the one or more programs are configured to be executed by one or more central processing units 1401. The one or more programs contain instructions for implementing the operating method of the above-mentioned model prediction task or the translation method based on the machine translation model. The central processing unit 1401 executes the one or more programs to implement the operating method of the model prediction task or the translation method based on the machine translation model provided by the above-mentioned various method embodiments.

[0268] According to various embodiments of the present application, the computer device 1400 can also be connected to a remote computer on the network through a network such as the Internet. That is, the computer device 1400 can be connected to the network 1412 through the network interface unit 1411 connected to the system bus 1405, or the network interface unit 1411 can be used to connect to other types of networks or remote computer systems (not shown).

[0269] The memory also includes one or more programs, which are stored in the memory, and the one or more programs include steps executed by a computer device in the operating method for performing the model prediction task provided in the embodiment of the present application.

[0270] On the other hand, an embodiment of the present application provides a computer device, which includes a processor and a memory, wherein at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement a chip control method as provided in the above-mentioned embodiment of the present application.

[0271] On the other hand, an embodiment of the present application provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement a chip control method as provided in the above-mentioned embodiment of the present application.

[0272] On the other hand, an embodiment of the present application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the chip control method provided in the above-mentioned embodiment of the present application.

[0273] On the other hand, an embodiment of the present application provides a computer device, the computer device comprising the processor as described above. Optionally, the processor is a GPU. The computer device may be at least one of a portable computer, a desktop computer, a server, a server cluster, an artificial intelligence (AI) computing cluster, and a cloud computing cluster. Among them, the AI ​​computing cluster may also be referred to as an intelligent computing cluster or a smart computing cluster.

[0274] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A chip control method, characterized in that: The method comprises: Acquire a chip parameter of the first chip, where the chip parameter is used to indicate a physical characteristic of the first chip; Analyze the chip parameters of the first chip through the voltage-frequency model to obtain a voltage-frequency point corresponding to the first chip, wherein the voltage-frequency point includes an operating voltage of the first chip and an operating frequency corresponding to the operating voltage, wherein the voltage-frequency model is a model obtained by training the chip parameters of the sample chip and the screened voltage-frequency data after screening the voltage-frequency data of the sample chip based on the stress test data; The first chip is controlled to operate at the operating voltage and the operating frequency based on the voltage-frequency point.

2. The method according to claim 1, characterized in that The step of analyzing chip parameters of the first chip by using a voltage-frequency model to obtain a voltage frequency point corresponding to the first chip includes: Analyzing chip parameters of the first chip through a pre-trained voltage-frequency model to obtain a voltage-frequency relationship corresponding to the first chip, wherein the voltage-frequency relationship includes voltage-frequency points corresponding to the first chip under a plurality of different operating states; Collecting operation status data for characterizing the operation status of the first chip; A voltage-frequency point in the voltage-frequency relationship corresponding to the operating status data is determined dynamically in real time.

3. The method according to claim 2, characterized in that The obtaining of a voltage-frequency point corresponding to the operating status data in the voltage-frequency relationship includes: Acquire inflection point status data corresponding to the first chip, wherein the inflection point status data is used to compare with the operation status data, and determine the voltage frequency point according to the comparison result; Acquire a comparison relationship between the running status data and the inflection point status data; A voltage-frequency point corresponding to the operating status data in the voltage-frequency relationship is determined based on the comparison relationship.

4. The method according to claim 3, characterized in that The operating status data includes operating temperature; The obtaining the inflection point state data corresponding to the first chip includes: Acquire an inflection point temperature corresponding to the first chip, where the inflection point temperature is a preset temperature; The determining, based on the comparison relationship, a voltage-frequency point in the voltage-frequency relationship corresponding to the operating status data comprises: A voltage-frequency point corresponding to the operating status data in the voltage-frequency relationship is determined based on a first comparison relationship between the operating temperature and the inflection point temperature.

5. The method according to claim 3, characterized in that: The operating status data includes current voltage; The obtaining the inflection point state data corresponding to the first chip includes: Obtaining an inflection point voltage corresponding to the first chip, where the inflection point voltage is a voltage at which the voltage frequency of the first chip has similar performance at different operating temperatures; The determining, based on the comparison relationship, a voltage-frequency point in the voltage-frequency relationship corresponding to the operating status data comprises: A voltage-frequency point corresponding to the operating status data in the voltage-frequency relationship is determined based on a second comparison relationship between the current voltage and the inflection point voltage.

6. The method according to claim 3, characterized in that The inflection point state data includes an inflection point voltage; The obtaining the inflection point state data corresponding to the first chip includes: The chip parameters of the first chip are analyzed by using a pre-trained inflection point analysis model to obtain the inflection point voltage corresponding to the first chip.

7. The method according to claim 6, characterized in that The inflection point analysis model is a model obtained by modeling the chip parameters of the sample chip and the inflection point voltage of the sample chip, wherein the inflection point voltage of the sample chip is obtained by inputting the chip parameters of the sample chip into a pre-trained voltage-frequency model, outputting voltage-frequency data corresponding to multiple chip parameters under different operating states, and performing linear fitting on the voltage-frequency data corresponding to each chip parameter under different operating states.

8. The method according to claim 7, characterized in that The inflection point voltage is obtained by analyzing the voltage-frequency data through linear fitting of the Monte Carlo model to obtain the voltage-frequency curves corresponding to each chip parameter under different operating states, and then analyzing the voltage-frequency curves corresponding to each chip parameter under different operating states, wherein the frequency similarity corresponding to the inflection point voltage on the voltage-frequency curve under different operating states meets the similarity requirement.

9. The method according to claim 2, characterized in that: The method further comprises: updating the operation status data representing the operation status of the first chip in real time according to a preset period; Acquire a voltage-frequency point in the voltage-frequency relationship corresponding to the updated operating status data; The operation of the first chip is controlled based on the voltage-frequency point corresponding to the updated operation status data.

10. The method according to claim 2, characterized in that The collecting of the operating status data for characterizing the operating status of the first chip includes: Acquire sensor data collected by the sensor in the first chip as the operating status data; Wherein, when the operating status data includes an operating temperature, the sensor includes a temperature sensor; when the operating status data includes a current voltage, the sensor includes a voltage sensor.

11. The method according to claim 2, characterized in that The voltage-frequency relationship includes voltage-frequency points corresponding to the first chip in a plurality of different operating states and operating scenarios, wherein the operating scenario is used to characterize a usage scenario of the first chip; The real-time dynamic determination of a voltage-frequency point corresponding to the operating status data in the voltage-frequency relationship includes: The voltage-frequency relationship is dynamically determined in real time to obtain a voltage-frequency point corresponding to the operating status data and the operating scenario.

12. The method according to any one of claims 1 to 11, characterized in that: The method further comprises: Obtaining an operating scenario of the first chip; The step of analyzing chip parameters of the first chip by using a voltage-frequency model to obtain a voltage frequency point corresponding to the first chip includes: The chip parameters and operation scenario of the first chip are analyzed by the voltage-frequency model to obtain a voltage-frequency point corresponding to the first chip.

13. The method according to any one of claims 1 to 11, characterized in that: The voltage-frequency model is a model obtained by modeling the relationship between the chip parameters of the sample chip and the voltage-frequency curve, and the voltage-frequency curve is a curve collected during the operation of the sample chip.

14. The method according to claim 13, characterized in that The voltage-frequency curve is a curve obtained by obtaining multiple groups of voltage-frequency data pairs corresponding to stress test data that meet the stress test requirements, and linearly fitting the multiple groups of voltage-frequency data pairs, wherein the voltage-frequency data pairs corresponding to the stress test data that meet the stress test requirements are data obtained by screening the voltage-frequency data pairs according to the stress test data after collecting the stress test data and the voltage-frequency data pairs of the sample chip during the operation of the sample chip, and the stress test data is used to indicate the performance of the sample chip for different operating voltages and operating frequencies during operation.

15. The method according to claim 13, characterized in that The voltage frequency model is a model of the relationship between the chip parameters of the sample chip and the voltage frequency curve. After obtaining the candidate voltage frequency model, the candidate voltage frequency model is subjected to deviation correction to expand the coverage of the voltage frequency curve predicted by the candidate voltage frequency model to obtain the model.

16. The method according to any one of claims 1 to 11, characterized in that: The method further comprises: During a test phase of the first chip, inputting an excitation signal to the first chip; collecting transmission performance data of the excitation signal in the first chip as chip parameters of the first chip; Storing the chip parameters in the storage space of the first chip; The obtaining chip parameters of the first chip includes: When the first chip is powered on, the chip parameters of the first chip are read from the storage space.

17. A chip control device, characterized in that: The device comprises: an acquisition module, configured to acquire chip parameters of the first chip, where the chip parameters are used to indicate physical characteristics of the first chip; an analysis module, configured to analyze chip parameters of the first chip through a voltage-frequency model to obtain voltage-frequency points corresponding to the first chip, wherein the voltage-frequency points include an operating voltage of the first chip and an operating frequency corresponding to the operating voltage, wherein the voltage-frequency model is a model obtained by training the chip parameters of the sample chip and the screened voltage-frequency data after screening the voltage-frequency data of the sample chip based on the stress test data; The control module is configured to control the first chip to operate at the operating voltage and the operating frequency based on the voltage-frequency point.

18. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the chip control method according to any one of claims 1 to 16.

19. A computer-readable storage medium, characterized in that: The readable storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the chip control method according to any one of claims 1 to 16.

20. A computer program product, characterized in that It comprises a computer program or an instruction, which, when executed by a processor, implements a control method for a chip as claimed in any one of claims 1 to 16.

Images