A low-power adaptive refresh control method suitable for LPDDR memory module

Abstract

The application provides a low-power adaptive refresh control method suitable for an LPDDR memory module, relates to computer component manufacturing, and belongs to the field of dynamic random access memories.The method comprises the following steps: using an intelligent refresh data prediction model to intelligently predict a safe low-power refresh interval duration corresponding to a target LPDDR memory module in a current time interval according to multi-source basic data; and configuring the safe low-power refresh interval duration intelligently predicted in advance for the target LPDDR memory module in the current time interval.The application solves the technical problem that the configured refresh parameters are easily mismatched with the changed device temperature due to the lag of refresh parameter configuration, by adopting an artificial intelligence model customized for the target LPDDR memory module, based on the multi-source basic data screened in a targeted manner, completing the advance analysis of the future refresh parameters of the target LPDDR memory module, and performing advance configuration, so that the above technical problem is solved.

Description

Technical Field

[0001] This invention relates to the manufacture of computer components, belonging to the field of dynamic random access memory, and particularly to a low-power adaptive refresh control method suitable for LPDDR memory modules. Background Technology

[0002] Dynamic Random Access Memory (DRAM) is a common memory component in computer manufacturing, and LPDDR memory modules are one type of main computer memory. LPDDR (Low Power Double Data Rate) is a low-power memory technology designed specifically for mobile and portable devices. It is widely used in smartphones, tablets, thin and light laptops, wearable devices, and edge AI scenarios. Its core goal is to significantly reduce power consumption while ensuring high performance, thereby extending battery life.

[0003] Clearly, because LPDDR memory modules often operate within relatively confined micro-device spaces, using a higher refresh rate (shorter refresh interval) relative to the memory module's temperature, while preventing data loss, leads to wasted power consumption and further increases in memory module temperature. Conversely, using a lower refresh rate (longer refresh interval) avoids both wasted power consumption and further temperature increases, but risks data loss and data security issues. Therefore, a dynamic balance between power saving, device cooling, and data security must be maintained throughout the manufacturing and use of LPDDR memory modules.

[0004] For example, Chinese invention patent publication CN119181400A discloses a memory refresh method, a memory controller, a storage system, and a computing device. The method is applied to a memory controller, which includes a counter for counting delayed refresh commands. The method includes: when the memory controller switches to a low-power operating mode, the memory controller sends a first number of refresh commands to the memory, the first number being not less than the number of delayed refresh commands counted by the counter; after the memory controller has sent the first number of refresh commands, the memory controller sends refresh commands to the memory at a set frequency; wherein the frequency at which the memory controller sends the first number of refresh commands is greater than the set frequency. Using this method can reduce system power consumption caused by scheduling memory refresh operations.

[0005] For example, Chinese invention patent publication CN102290094A proposes a low-power circuit for reliably maintaining the self-refresh state of memory. The low-power circuit includes a first resistor, a second resistor, and a PNP transistor. The base of the PNP transistor is connected to one end of the first resistor and the output terminal of the memory controller, while the other end of the first resistor is grounded. The emitter of the PNP transistor is connected to one end of the second resistor and the clock signal input terminal of the memory, while the other end of the second resistor is grounded. The collector of the PNP transistor is grounded. The circuit proposed in this invention can stably and reliably maintain the memory in a self-refresh state, and the circuit itself does not require an additional power supply.

[0006] However, the aforementioned existing technologies only involve the logic control and circuit design for achieving low power consumption in ordinary memory, and do not address how to achieve a dynamic balance between power saving, device cooling, and data security during the manufacturing and use of LPDDR memory modules. When executing low-power adaptive refresh control of LPDDR memory modules, it is necessary to determine the corresponding safe low-power refresh interval duration of the target LPDDR memory module at the current moment based on the real-time device temperature of the target LPDDR memory module. As a result, since various operations such as temperature measurement, refresh control strategy decision-making, and refresh interval duration configuration take a certain amount of time, the configuration of the safe low-power refresh interval duration for the target LPDDR memory module is delayed, causing a mismatch between the configured refresh parameters and the changed device temperature. This can easily lead to power waste and temperature rise due to a relatively fast refresh frequency, or loss of stored content due to a relatively slow refresh frequency. Summary of the Invention

[0007] To address technical issues in related fields, this invention provides a low-power adaptive refresh control method suitable for LPDDR memory modules. Addressing the characteristic parameter that the safe refresh interval for each temperature of the LPDDR memory module corresponds to the longest refresh interval duration that prevents data loss when the device temperature of the target LPDDR memory module is the specified temperature, this invention employs an artificial intelligence model with a customized structure designed for the target LPDDR memory module. Based on selectively filtered multi-source basic data, it intelligently predicts the safe low-power refresh interval duration corresponding to the current time interval (which is considered a future time interval). This allows for the pre-configuration of the safe low-power refresh interval duration for the current time interval, preventing mismatches between the configured refresh parameters and the changed device temperature. Consequently, the power refresh rate of the target LPDDR memory module is consistently maintained at the slowest refresh rate to prevent data loss, achieving a dynamic balance between power saving, temperature maintenance, and data security.

[0008] According to the present invention, a low-power adaptive refresh control method suitable for LPDDR memory modules is provided, the method comprising: The number of channels, operating voltage, adjustment support identifier, technical support identifier, mode support identifier, and maximum storage capacity of the target LPDDR memory module are analyzed as multiple configuration parameters of the target LPDDR memory module. The average device temperature, temperature gradient, and safe low-power refresh interval duration of the target LPDDR memory module in each of the previous time intervals before the current time interval are analyzed to form the relevant refresh information of the target LPDDR memory module in the current time interval. The convolutional neural network is trained in each iteration to obtain the convolutional neural network after each iteration and is output as the intelligent refresh data prediction model. The number of training iterations and the number of channels of the target LPDDR memory module show the same numerical trend. The intelligent refresh data prediction model uses the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. Configure the target LPDDR memory module for the current time interval with intelligent prediction of safe low-power refresh interval duration.

[0009] Compared with the prior art, the present invention has at least the following outstanding substantive features: Substantial Feature 1: Based on multiple configuration parameters of the target LPDDR memory module and its past refresh correlations within the current time interval, an AI model customized for the target LPDDR memory module is used to pre-determine the safe low-power refresh interval duration for the target LPDDR memory module within the current time interval and pre-configure the safe low-power refresh interval duration. This ensures that the power refresh rate of the target LPDDR memory module can always be kept at the slowest refresh rate to prevent loss of stored content. Through an intelligent mechanism, the power saving, temperature maintenance, and data security of the target LPDDR memory module are balanced, avoiding waste caused by relatively fast power refresh or loss of stored content caused by relatively slow power refresh. Substantial Feature Two: To intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval, different customized artificial intelligence models, namely intelligent refresh data prediction models, are designed for different LPDDR memory modules. The intelligent refresh data prediction model corresponding to the target LPDDR memory module is a convolutional neural network after completing each learning iteration. The number of learning iterations of the convolutional neural network shows the same numerical trend as the number of channels of the target LPDDR memory module. The convolutional neural network used includes a sequentially connected input layer, convolutional layer, pooling layer, and fully connected layer. The number of pooling layers is positively correlated with the maximum storage capacity of the target LPDDR memory module. The convolutional layers use the Sigmoid function as the activation function, and the pooling layers use the ReLU function as the activation function. The customized structural design of the above-mentioned intelligent refresh data prediction model ensures the reliability and stability of the intelligent prediction results. Substantial Feature Three: In each learning iteration of the convolutional neural network, the known safe low-power refresh interval duration corresponding to a certain LPDDR memory module in a certain past time interval is used as the single output data of the convolutional neural network. The occupancy duration of the time interval, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the certain LPDDR memory module, and various past refresh correlation contents of the certain LPDDR memory module in the certain past time interval are used as various input data of the convolutional neural network to complete the learning, thereby ensuring the learning effect of each learning iteration of the convolutional neural network. Substantial Feature 4: To intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval, various basic data have been specifically screened, including the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval. The thorough and comprehensive screening of the above basic data further ensures the reliability and stability of the intelligent prediction results. Substantial Feature Five: Specifically, the target LPDDR memory module's configuration parameters include the number of channels, operating voltage, adjustment support identifier, technical support identifier, mode support identifier, and maximum storage capacity. The target LPDDR memory module's past refresh correlation content for the current time interval includes the average device temperature for each of the previous time intervals, the temperature gradient values ​​for each of the previous time intervals, and the safe low-power refresh interval duration for each of the previous time intervals. This provides a customized data structure for the various basic data used for intelligent prediction. Attached Figure Description

[0010] The embodiments of the present invention will now be described with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram illustrating the working scenario of a low-power adaptive refresh control method for LPDDR memory modules according to the present invention.

[0011] Figure 2 The present invention is illustrated in Embodiment 1 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0012] Figure 3 The present invention is illustrated in Embodiment 2 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0013] Figure 4 The present invention is illustrated in Embodiment 3 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0014] Figure 5 This is a flowchart illustrating the steps of a low-power adaptive refresh control method for LPDDR memory modules according to Embodiment 4 of the present invention.

[0015] Figure 6 This is a flowchart illustrating the steps of a low-power adaptive refresh control method for LPDDR memory modules according to Embodiment 5 of the present invention. Detailed Implementation

[0016] like Figure 1 The diagram illustrates a working scenario of a low-power adaptive refresh control method for LPDDR memory modules according to the present invention. This invention relates to computer component manufacturing and belongs to the field of dynamic random access memory.

[0017] The specific technical process of this invention is as follows: Technical Process A: Design different customized artificial intelligence models, i.e., intelligent refresh data prediction models, for different LPDDR memory modules. Figure 1 The target LPDDR memory module is a 2-channel LPDDR memory module; Specifically, such as Figure 1 As shown, a smart refresh data prediction model is designed for the target LPDDR memory module. Its customized structure is mainly reflected in the following aspects: Firstly, the intelligent refresh data prediction model corresponding to the target LPDDR memory module is a convolutional neural network that has completed each learning iteration. The number of learning iterations of the convolutional neural network shows the same numerical trend as the number of channels in the target LPDDR memory module. Secondly, the convolutional neural network used includes sequentially connected input layers, convolutional layers, pooling layers, and fully connected layers. The number of pooling layers is positively correlated with the maximum storage capacity of the target LPDDR memory module. Thirdly: In the convolutional neural network in use, the convolutional layer uses the Sigmoid function as the activation function, and the pooling layer uses the ReLU function as the activation function; Fourthly: In each learning process performed on the convolutional neural network, the safe low-power refresh interval duration corresponding to a known LPDDR memory module in a certain past time interval is used as the single output data of the convolutional neural network. The occupancy duration of the time interval, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the LPDDR memory module, and various past refresh correlation contents of the LPDDR memory module in the certain past time interval are used as various input data of the convolutional neural network to complete the learning process, thereby ensuring the learning effect of each learning process of the convolutional neural network. In this way, the customized structural design of the above-mentioned intelligent refresh data prediction model ensures the reliability and stability of the intelligent prediction results. Technical Process B: To intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval, various basic data were specifically selected; Specifically, such as Figure 1 As shown, the basic data includes the duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval. Figure 1 In the data, the two auxiliary data points are the duration of occupation of the time interval and the safe refresh interval duration corresponding to zero degrees; More specifically, the target LPDDR memory module has several configuration parameters, including the number of channels, operating voltage, adjustment support identifier, technical support identifier, mode support identifier, and maximum storage capacity. The target LPDDR memory module's past refresh correlation content in the current time interval includes the average device temperature, temperature gradient values, and safe low-power refresh interval durations for each of the previous time intervals. This provides a customized data structure for the various basic data used for intelligent prediction. In this way, the reliability and stability of the intelligent prediction results are further guaranteed through the thorough and comprehensive screening of the above-mentioned basic data. Technical Path C: Utilizing a smart refresh data prediction model designed for the target LPDDR memory module based on Technical Path A, and using the specific basic data selected in Technical Path B, intelligently predicts the safe low-power refresh interval duration for the target LPDDR memory module within the current time interval. Figure 1 As shown; Specifically, the safe low-power refresh interval duration corresponding to each time interval is the safe refresh interval duration corresponding to the lowest device temperature of the target LPDDR memory module within the time interval, and the safe refresh interval duration corresponding to each temperature is the longest refresh interval duration to avoid data loss of the target LPDDR memory module when the device temperature of the target LPDDR memory module is the specified temperature. Specifically, the current time interval starts at the current moment, and the current time interval and all previous time intervals before the current time interval form a complete time segment on the time axis, with each time interval occupying an equal duration; Therefore, the current time interval is actually a future time interval, and the intelligently predicted safe low-power refresh interval is actually the refresh data that the target LPDDR memory module needs to refresh in the future while taking into account power saving, temperature maintenance and data security. Technical Path D: At the current moment, i.e., at the beginning of the current time interval, the safe low-power refresh interval duration predicted by Technical Path C is pre-configured for use by the target LPDDR memory module within the current time interval, such as... Figure 1 As shown; Thus, after the intelligent prediction of the future low-power adaptive refresh data of the target LPDDR memory module is completed in technical process C, the advance configuration of the future low-power adaptive refresh data of the target LPDDR memory module is completed in technical process D. Therefore, through the coordinated operation of the above-mentioned technical processes, this invention addresses the technical problem of mismatch between configured refresh parameters and changed device temperatures caused by delayed refresh parameter configuration. It addresses this issue by using a custom-designed artificial intelligence model to pre-determine the safe low-power refresh interval duration for the target LPDDR memory module within the current time interval, based on multiple configuration parameters and past refresh-related data of the target LPDDR memory module in the current time interval. This pre-configures the safe low-power refresh interval duration, ensuring that the power refresh rate of the target LPDDR memory module remains at the slowest refresh rate to prevent data loss. This intelligent mechanism achieves a balance between power saving, temperature maintenance, and data security for the target LPDDR memory module, avoiding waste from relatively fast power refreshes or data loss from relatively slow power refreshes.

[0018] The key points of this invention are: the directional design of artificial intelligence models with different customized structures for different LPDDR memory modules, namely intelligent refresh data prediction models; the full and comprehensive screening of multi-source basic data and the customized data structure design; the targeted design of each learning process of the convolutional neural network; and the advance analysis and configuration of low-power adaptive refresh data by utilizing the characteristic parameter of the safe refresh interval duration corresponding to each temperature of the LPDDR memory module.

[0019] The following will describe in detail an embodiment of a low-power adaptive refresh control method for LPDDR memory modules according to the present invention.

[0020] Example 1 Figure 2 The present invention is illustrated in Embodiment 1 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0021] like Figure 2 As shown, the low-power adaptive refresh control method for LPDDR memory modules includes the following specific steps: Step S201: Analyze the number of channels, operating voltage, adjustment support identifier, technical support identifier, mode support identifier, and maximum storage capacity of the target LPDDR memory module as multiple configuration parameters of the target LPDDR memory module; Specifically, multiple parameter parsing units can be selected to parse the number of channels, operating voltage, adjustment support identifier, technical support identifier, mode support identifier, and maximum storage capacity of the target LPDDR memory module respectively. Step S202: Analyze the average device temperature, temperature gradient values, and safe low-power refresh interval duration of the target LPDDR memory module in each of the previous time intervals before the current time interval, respectively, as the past refresh association content of the target LPDDR memory module in the current time interval. For example, the current time interval starts at the current time. If the starting time is 5:00 PM, the current time interval is from 5:00 PM to 5:05 PM. Obviously, the current time interval is a future time segment. At the same time, the duration of each time interval is 5 minutes. The previous time intervals before the current time interval are 4:55 PM to 5:00 PM, 4:50 PM to 4:55 PM, 4:45 PM to 4:50 PM, 4:40 PM to 4:45 PM, 4:35 PM to 4:40 PM, 4:30 PM to 4:35 PM, 4:25 PM to 4:30 PM, and 4:20 PM to 4:25 PM, for a total of 8 previous time intervals. For example, each time interval is a whole minute. Specifically, for the past time interval from 16:55 to 17:00, the time intervals are 16:56, 16:57, 16:58 and 16:59, a total of 4 times. Obviously, the above 8 past time intervals and the 4 evenly spaced moments in each time interval are examples. The actual values ​​of the two can be dynamically changed according to the specific scenario. For example, the number of evenly spaced moments in each time interval can be proportional to the number of channels of the target LPDDR memory module, while the number of each past time interval before the current time interval can be proportional to the maximum storage capacity of the target LPDDR memory module. Step S203: Perform each learning iteration on the convolutional neural network to obtain the convolutional neural network after each learning iteration and output it as the intelligent refresh data prediction model. The number of learning iterations and the number of channels of the target LPDDR memory module show the same numerical change trend. For example, the number of learning iterations showing the same numerical trend as the number of channels in the target LPDDR memory module includes: 500 learning iterations for a target LPDDR memory module with 1 channel, 1000 learning iterations for a target LPDDR memory module with 2 channels, 1500 learning iterations for a target LPDDR memory module with 3 channels, and so on. Step S204: Using the intelligent refresh data prediction model, based on the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, the model intelligently predicts the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. Here, since the current time interval is actually a future time interval, the intelligently predicted safe low-power refresh interval is actually the refresh data that the target LPDDR memory module needs to refresh in the future while taking into account power saving, temperature maintenance and data security. Step S205: Configure the intelligently predicted safe low-power refresh interval duration for the target LPDDR memory module in the current time interval in advance; Therefore, after completing the intelligent prediction of the future low-power adaptive refresh data of the target LPDDR memory module, the present invention can perform the advance configuration of the future low-power adaptive refresh data of the target LPDDR memory module, thereby avoiding the mismatch between refresh data and the ambient temperature, so that the power refresh rate of the target LPDDR memory module can always be kept at the slowest refresh rate to maintain the storage content without loss. Through the intelligent mechanism, the power saving, temperature maintenance and data security of the target LPDDR memory module are achieved, avoiding the waste of relatively fast power refresh or the loss of storage content due to relatively slow power refresh. Among them, the adjustment support identifier of the target LPDDR memory module is used to indicate whether the target LPDDR memory module supports DVFS, i.e., dynamic voltage frequency adjustment; the technical support identifier of the target LPDDR memory module is used to indicate whether the target LPDDR memory module supports PAUSE technology, i.e., pause clock signal technology; and the mode support identifier of the target LPDDR memory module is used to indicate whether the target LPDDR memory module supports power management function based on deep sleep mode. Specifically, the adjustment support identifier, technical support identifier, and mode support identifier of the target LPDDR memory module can all use different values ​​to represent different support states; The temperature gradient value corresponding to each time interval is the standard deviation value of each device temperature corresponding to each time interval at a uniform interval within the time interval, and the average device temperature corresponding to each time interval is the arithmetic mean of each device temperature corresponding to each time interval at a uniform interval within the time interval. For example, for the past time interval from 16:55 to 17:00, the evenly spaced times within it are 16:56, 16:57, 16:58 and 16:59, a total of 4 times. Thus, the average equipment temperature corresponding to the past time interval is the arithmetic mean of the four equipment temperatures corresponding to the evenly spaced times within the past time interval, namely 16:56, 16:57, 16:58 and 16:59. The safe low-power refresh interval duration for each time interval is the safe refresh interval duration corresponding to the lowest device temperature of the target LPDDR memory module within the time interval. Among them, the safe refresh interval duration corresponding to each temperature is the longest refresh interval duration to avoid data loss in the target LPDDR memory module when the device temperature of the target LPDDR memory module is the specified temperature; The current time interval starts at the current moment. The current time interval and all previous time intervals before the current time interval form a complete time segment on the time axis, and the duration of each time interval is equal. For example, the current time interval is from 5:00 PM to 5:05 PM. The previous time intervals before the current time interval are 4:55 PM to 5:00 PM, 4:50 PM to 4:55 PM, 4:45 PM to 4:50 PM, 4:40 PM to 4:45 PM, 4:35 PM to 4:40 PM, 4:30 PM to 4:35 PM, 4:25 PM to 4:30 PM, and 4:20 PM to 4:25 PM, a total of 8 previous time intervals. The current time interval and the previous time intervals before the current time interval form a complete time segment on the time axis, that is, from 4:20 PM to 5:05 PM. At the same time, for the current time interval and all previous time intervals, the duration of each time interval is equal, for example, it can be 5 minutes. Among them, performing each learning iteration on the convolutional neural network to obtain the convolutional neural network after each learning iteration and outputting it as the intelligent refresh data prediction model, the number of learning iterations and the number of channels of the target LPDDR memory module showing the same numerical change trend include: the number of pooling layers of the convolutional neural network is positively correlated with the maximum storage capacity of the target LPDDR memory module; For example, the positive correlation between the number of pooling layers in the convolutional neural network and the maximum storage capacity of the target LPDDR memory module includes: when the maximum storage capacity of the target LPDDR memory module is 16GB, the number of pooling layers in the convolutional neural network is 1; when the maximum storage capacity of the target LPDDR memory module is 32GB, the number of pooling layers in the convolutional neural network is 2; when the maximum storage capacity of the target LPDDR memory module is 64GB, the number of pooling layers in the convolutional neural network is 3; when the maximum storage capacity of the target LPDDR memory module is 128GB, the number of pooling layers in the convolutional neural network is 4, and so on. The process of performing learning iterations on the convolutional neural network to obtain a convolutional neural network after each learning iteration and outputting it as the intelligent refresh data prediction model, with the number of learning iterations showing the same numerical change trend as the number of channels in the target LPDDR memory module, also includes: the convolutional neural network comprising a sequentially connected input layer, convolutional layer, pooling layer, and fully connected layer, wherein the convolutional layer uses the Sigmoid function as the activation function and the pooling layer uses the ReLU function as the activation function; In each learning process performed on the convolutional neural network, the safe low-power refresh interval duration corresponding to a known LPDDR memory module in a certain past time interval is used as the single output data of the convolutional neural network, and the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the LPDDR memory module, and various past refresh correlation contents of the LPDDR memory module in the certain past time interval are used as various input data of the convolutional neural network to complete this learning process. For example, you can choose to use the MATLAB toolbox to test and simulate each learning process performed on the convolutional neural network.

[0022] Example 2 Figure 3 The present invention is illustrated in Embodiment 2 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0023] like Figure 3 As shown, with Figure 2Unlike the previous embodiment, in the low-power adaptive refresh control method for LPDDR memory modules, before parsing the average device temperature, temperature gradient values, and safe low-power refresh interval durations corresponding to each of the previous time intervals of the target LPDDR memory module in the current time interval, and using these as the associated content of the target LPDDR memory module's past refreshes in the current time interval, i.e., before step S202, the method further includes: Step S206: Measure the device temperature of the target LPDDR memory module in a time-sharing manner. The time-sharing measurement results are used to analyze the average device temperature of the target LPDDR memory module in each of the previous time intervals before the current time interval. Specifically, a contact-type temperature sensor can be selected to provide basic temperature data for the analysis of the average device temperature of the target LPDDR memory module in each of the previous time intervals before the current time interval. The time-division measurement results are also used to analyze the temperature gradient values ​​corresponding to each of the previous time intervals before the current time interval.

[0024] Example 3 Figure 4 The present invention is illustrated in Embodiment 3 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0025] like Figure 4 As shown, with Figure 2 Unlike the previous embodiment, in the low-power adaptive refresh control method for LPDDR memory modules, after using an intelligent refresh data prediction model to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval based on the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, i.e. after step S204, the method further includes: Step S207: When the safe refresh interval corresponding to the device temperature of the target LPDDR memory module is less than the safe low-power refresh interval corresponding to the target LPDDR memory module in the current time interval at a certain moment within the current time interval, a refresh speed alarm signal is issued. When the safe refresh interval corresponding to the device temperature of the target LPDDR memory module is greater than or equal to the safe low-power refresh interval corresponding to the target LPDDR memory module in the current time interval at each evenly spaced moment within the current time interval, a refresh speed safety signal is issued. Specifically, the refresh rate alarm signal can be used to trigger a field alarm device to perform an alarm operation corresponding to an excessive refresh rate of the equipment temperature. The field alarm device can be an optical alarm device or an acoustic alarm device.

[0026] Example 4 Figure 5 This is a flowchart illustrating the steps of a low-power adaptive refresh control method for LPDDR memory modules according to Embodiment 4 of the present invention.

[0027] like Figure 5 As shown, with Figure 2 Unlike the previous embodiment, in the low-power adaptive refresh control method for LPDDR memory modules, after using an intelligent refresh data prediction model to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval based on the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, i.e. after step S204, the method further includes: Step S208: When the safe refresh interval time corresponding to the device temperature of the target LPDDR memory module is less than the safe low-power refresh interval time corresponding to the target LPDDR memory module in the current time interval at a certain moment within the current time interval, an emergency cooling operation is performed; when the safe refresh interval time corresponding to the device temperature of the target LPDDR memory module is greater than or equal to the safe low-power refresh interval time corresponding to the target LPDDR memory module in the current time interval at evenly spaced moments within the current time interval, the emergency cooling operation is temporarily suspended. Specifically, a fan array can be used to perform an emergency cooling operation when the safe refresh interval corresponding to the device temperature of the target LPDDR memory module at a certain moment in the current time interval is less than the safe low-power refresh interval corresponding to the target LPDDR memory module in the current time interval.

[0028] Example 5 Figure 6 The present invention is illustrated in Embodiment 5 of the present invention as a flowchart of a low-power adaptive refresh control method for LPDDR memory modules.

[0029] like Figure 6 As shown, with Figure 2Unlike the previous embodiment, in the low-power adaptive refresh control method for LPDDR memory modules, after using an intelligent refresh data prediction model to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval based on the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, i.e. after step S204, the method further includes: Step S209: Receive the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval according to the intelligent prediction, and package the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval into a network data packet and wirelessly transmit it to the remote computer component monitoring server. The process of receiving the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval based on intelligent prediction, and then wirelessly transmitting the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval to a network data packet to a remote computer component monitoring server includes: the network data packet being an IP data packet; For example, the process of receiving the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval based on intelligent prediction, and then wirelessly transmitting the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval into a network data packet to a remote computer component monitoring server also includes: the remote computer component monitoring server can be a blockchain service network element, a cloud computing service network element, or a big data service network element.

[0030] Next, the various method embodiments of the present invention will be described in detail.

[0031] In a low-power adaptive refresh control method for LPDDR memory modules according to various method embodiments of the present invention: The average device temperature corresponding to each time interval is the arithmetic mean of the device temperatures corresponding to each time point at a uniform interval within the time interval, including: the number of selected time points is proportional to the number of channels of the target LPDDR memory module; For example, the number of selected moments is proportional to the number of channels in the target LPDDR memory module, including: when the number of channels in the target LPDDR memory module is 1, the number of selected moments is 4; when the number of channels in the target LPDDR memory module is 2, the number of selected moments is 8; when the number of channels in the target LPDDR memory module is 3, the number of selected moments is 12, and so on. The temperature gradient value corresponding to each time interval is the standard deviation value of each device temperature corresponding to each time point at a uniform interval within the time interval, including: selecting a GAL chip to calculate the standard deviation value of each device temperature corresponding to each time point at a uniform interval within the time interval.

[0032] In a low-power adaptive refresh control method for LPDDR memory modules according to various method embodiments of the present invention: The numerical mapping function is selected to represent the positive correlation between the number of pooling layers in the convolutional neural network and the maximum storage capacity of the target LPDDR memory module. For example, one can choose to use numerical simulation mode to perform tests and simulations on the numerical mapping relationships of numerical mapping functions; The selection of a numerical mapping function to represent a positive correlation between the number of pooling layers in the convolutional neural network and the maximum storage capacity of the target LPDDR memory module includes: in the numerical mapping function, the maximum storage capacity of the target LPDDR memory module is the input parameter of the numerical mapping function; The selection of a numerical mapping function to represent the positive correlation between the number of pooling layers of the convolutional neural network and the maximum storage capacity of the target LPDDR memory module further includes: the number of pooling layers of the convolutional neural network that is positively correlated with the maximum storage capacity of the target LPDDR memory module is the output parameter of the numerical mapping function.

[0033] And in a low-power adaptive refresh control method for LPDDR memory modules according to various method embodiments of the present invention: The intelligent refresh data prediction model uses the time interval occupancy duration, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. This includes inputting the time interval occupancy duration, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval into the intelligent refresh data prediction model in parallel. For example, you can choose to use PAL devices to complete the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval as parallel inputs to the intelligent refresh data prediction model; The intelligent refresh data prediction model uses the time interval occupancy duration, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. This also includes: running the intelligent refresh data prediction model to obtain the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval output by the intelligent refresh data prediction model. Among them, the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval, the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the target LPDDR memory module, and various past refresh related contents of the target LPDDR memory module in the current time interval are all in the form of normalized numerical processing. For example, the safe low-power refresh interval duration, the time interval occupancy duration, the safe refresh interval duration corresponding to zero degrees, the various configuration parameters of the target LPDDR memory module, and the various past refresh related contents of the target LPDDR memory module in the current time interval are all in a normalized numerical representation, including: the normalized numerical representation is a binary numerical representation after binary numerical conversion.

[0034] In addition, the following technical content can be cited to further highlight the essential features of the present invention: Performing each learning iteration on the convolutional neural network to obtain the convolutional neural network after each learning iteration and outputting it as the intelligent refresh data prediction model, with the number of learning iterations showing the same numerical change trend as the number of channels in the target LPDDR memory module, also includes: using a quantity change curve to represent the numerical change trend of the number of channels in the target LPDDR memory module, and using a number change curve to represent the numerical change trend of the number of learning iterations. Among them, performing each learning operation on the convolutional neural network to obtain the convolutional neural network after each learning operation and outputting it as the intelligent refresh data prediction model, and the number of learning operations and the number of channels of the target LPDDR memory module showing the same numerical change trend, also includes: normalizing the curve length of the number change curve and the number change curve to obtain the first normalized curve and the second normalized curve, and the first normalized curve and the second normalized curve completely overlap. Specifically, the complete overlap of the first normalized curve and the second normalized curve means that the curve lengths of the first normalized curve and the second normalized curve are equal, and the curvatures at uniformly spaced points are also equal. For example, a programmable logic device is selected to implement the normalization processing of the curve lengths of the quantity change curve and the number change curve. The programmable logic device is a CPLD chip designed in VHDL language.

[0035] The foregoing description of exemplary embodiments of the invention is provided for illustrative and descriptive purposes. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations will obviously be apparent to those skilled in the art. Exemplary embodiments were chosen and described in order to best illustrate the principles of the invention and its practical application, thereby enabling others skilled in the art to understand the various embodiments and variations of the invention suitable for the contemplated particular purpose. The scope of the invention is intended to be defined by the appended claims and their equivalents.

Claims

1. A low-power adaptive refresh control method suitable for LPDDR memory modules, characterized in that, The method includes: The number of channels, operating voltage, adjustment support identifier, technical support identifier, mode support identifier, and maximum storage capacity of the target LPDDR memory module are analyzed as multiple configuration parameters of the target LPDDR memory module. The average device temperature, temperature gradient, and safe low-power refresh interval duration of the target LPDDR memory module in each of the previous time intervals before the current time interval are analyzed to form the relevant refresh information of the target LPDDR memory module in the current time interval. The convolutional neural network is trained in each iteration to obtain the convolutional neural network after each iteration and is output as the intelligent refresh data prediction model. The number of training iterations and the number of channels of the target LPDDR memory module show the same numerical trend. The intelligent refresh data prediction model uses the occupation duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. Configure the target LPDDR memory module for the current time interval with intelligent prediction of safe low-power refresh interval duration.

2. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 1, characterized in that: The adjustment support identifier of the target LPDDR memory module is used to indicate whether the target LPDDR memory module supports DVFS, i.e., dynamic voltage and frequency adjustment; the technical support identifier of the target LPDDR memory module is used to indicate whether the target LPDDR memory module supports PAUSE technology, i.e., pause clock signal technology; and the mode support identifier of the target LPDDR memory module is used to indicate whether the target LPDDR memory module supports power management function based on deep sleep mode. The temperature gradient value corresponding to each time interval is the standard deviation value of each device temperature corresponding to each time interval at a uniform interval within the time interval, and the average device temperature corresponding to each time interval is the arithmetic mean of each device temperature corresponding to each time interval at a uniform interval within the time interval. The safe low-power refresh interval duration for each time interval is the safe refresh interval duration corresponding to the lowest device temperature of the target LPDDR memory module within the time interval. Among them, the safe refresh interval duration corresponding to each temperature is the longest refresh interval duration to avoid data loss in the target LPDDR memory module when the device temperature of the target LPDDR memory module is the specified temperature; The current time interval starts at the current moment. The current time interval and all previous time intervals before the current time interval form a complete time segment on the time axis, and each time interval occupies an equal duration.

3. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 2, characterized in that: The convolutional neural network is subjected to learning iterations to obtain a convolutional neural network after each learning iteration, which is then output as an intelligent refresh data prediction model. The number of learning iterations and the number of channels in the target LPDDR memory module exhibit the same numerical trend, including: the number of pooling layers in the convolutional neural network is positively correlated with the maximum storage capacity of the target LPDDR memory module. The process of performing learning iterations on the convolutional neural network to obtain a convolutional neural network after each learning iteration and outputting it as the intelligent refresh data prediction model, with the number of learning iterations showing the same numerical change trend as the number of channels in the target LPDDR memory module, also includes: the convolutional neural network comprising a sequentially connected input layer, convolutional layer, pooling layer, and fully connected layer, wherein the convolutional layer uses the Sigmoid function as the activation function and the pooling layer uses the ReLU function as the activation function; In each learning iteration of the convolutional neural network, the safe low-power refresh interval duration corresponding to a known LPDDR memory module in a certain past time interval is used as the single output data of the convolutional neural network. The occupancy duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the LPDDR memory module, and various past refresh correlation contents of the LPDDR memory module in the certain past time interval are used as the various input data of the convolutional neural network to complete this learning iteration.

4. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that, Before analyzing the average device temperature, temperature gradient values, and safe low-power refresh interval duration of the target LPDDR memory module in each previous time interval before the current time interval, as well as the corresponding values ​​for each previous time interval, to obtain the relevant information about the target LPDDR memory module's past refreshes in the current time interval, the method further includes: The device temperature of the target LPDDR memory module is measured in time-sharing mode. The time-sharing measurement results are used to analyze the average device temperature of the target LPDDR memory module in each of the previous time intervals before the current time interval. The time-division measurement results are also used to analyze the temperature gradient values ​​corresponding to each of the previous time intervals before the current time interval.

5. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that, After using the intelligent refresh data prediction model to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval based on the occupied duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, the method further includes: If, at a certain moment within the current time interval, the safe refresh interval corresponding to the device temperature of the target LPDDR memory module is less than the safe low-power refresh interval corresponding to the target LPDDR memory module within the current time interval, a refresh speed alarm signal is issued. If, at each evenly spaced moment within the current time interval, the safe refresh interval corresponding to the device temperature of the target LPDDR memory module is greater than or equal to the safe low-power refresh interval corresponding to the target LPDDR memory module within the current time interval, a refresh speed safety signal is issued.

6. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that, After using the intelligent refresh data prediction model to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval based on the occupied duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, the method further includes: If, at a certain moment within the current time interval, the safe refresh interval corresponding to the device temperature of the target LPDDR memory module is less than the safe low-power refresh interval corresponding to the target LPDDR memory module within the current time interval, an emergency cooling operation is performed. If, at any time interval within the current time interval, the safe refresh interval corresponding to the device temperature of the target LPDDR memory module is greater than or equal to the safe low-power refresh interval corresponding to the target LPDDR memory module within the current time interval, the emergency cooling operation is temporarily suspended.

7. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that, After using the intelligent refresh data prediction model to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval based on the occupied duration of the time interval, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval, the method further includes: The system receives the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval based on intelligent prediction, and then packages the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval into a network data packet and wirelessly transmits it to the remote computer component monitoring server. The process of receiving the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval based on intelligent prediction, and then wirelessly transmitting the safe low-power refresh interval duration corresponding to the target LPDDR memory module in the current time interval into a network data packet to a remote computer component monitoring server includes: the network data packet being an IP data packet.

8. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that: The average device temperature corresponding to each time interval is the arithmetic mean of the device temperatures corresponding to each time point at a uniform interval within the time interval, including: the number of selected time points is proportional to the number of channels of the target LPDDR memory module; The temperature gradient value corresponding to each time interval is the standard deviation value of each device temperature corresponding to each time point at a uniform interval within the time interval, including: selecting a GAL chip to calculate the standard deviation value of each device temperature corresponding to each time point at a uniform interval within the time interval.

9. The low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that: The numerical mapping function is selected to represent the positive correlation between the number of pooling layers in the convolutional neural network and the maximum storage capacity of the target LPDDR memory module. The selection of a numerical mapping function to represent a positive correlation between the number of pooling layers in the convolutional neural network and the maximum storage capacity of the target LPDDR memory module includes: in the numerical mapping function, the maximum storage capacity of the target LPDDR memory module is the input parameter of the numerical mapping function; The selection of a numerical mapping function to represent the positive correlation between the number of pooling layers of the convolutional neural network and the maximum storage capacity of the target LPDDR memory module further includes: the number of pooling layers of the convolutional neural network that is positively correlated with the maximum storage capacity of the target LPDDR memory module is the output parameter of the numerical mapping function.

10. A low-power adaptive refresh control method for LPDDR memory modules as described in claim 3, characterized in that: The intelligent refresh data prediction model uses the time interval occupancy duration, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. This includes inputting the time interval occupancy duration, the safe refresh interval duration corresponding to zero degrees, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval into the intelligent refresh data prediction model in parallel. The intelligent refresh data prediction model uses the time interval occupancy duration, the safe refresh interval duration corresponding to zero degree, multiple configuration parameters of the target LPDDR memory module, and various past refresh-related contents of the target LPDDR memory module in the current time interval to intelligently predict the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval. This also includes: running the intelligent refresh data prediction model to obtain the safe low-power refresh interval duration of the target LPDDR memory module in the current time interval output by the intelligent refresh data prediction model. Among them, the safe low-power refresh interval duration, the time interval occupation duration, the safe refresh interval duration corresponding to zero degree, the multiple configuration parameters of the target LPDDR memory module, and the various past refresh-related contents of the target LPDDR memory module in the current time interval are all in the form of normalized numerical processing.

Images