Mixed signal processing method and system for memory bank

Abstract

The invention discloses a mixed signal processing method and system for a memory bank, and relates to the technical field of computer system structures and signal processing, and the method comprises the following steps: an application load sensing step: extracting characteristic parameters related to future memory access intensity, and generating a quantized future work load intensity index; an electrothermal transient effect prediction step: receiving a working load intensity index, and calculating a predicted temperature rise and a predicted voltage drop about to occur at the memory interface in a future time window; a compensation strategy generation step of calculating and generating a feedforward type physical layer compensation control instruction according to the predicted temperature rise and the predicted voltage drop; a feedforward parameter adjustment step: before the predicted electrothermal transient effect actually occurs, executing a compensation control instruction, and actively adjusting physical parameters of a memory interface to offset electrothermal shock; according to the method, transient data transmission errors caused by the pulse type load are eliminated, and the operation stability of the memory subsystem under the severe working condition is remarkably enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer architecture and signal processing technology, more particularly, the present application relates to a hybrid signal processing method and system for memory bar. BACKGROUND

[0002] In the scene of data center, high performance computing, etc., the performance and stability of the memory subsystem is the key bottleneck, the high speed of the memory interface compresses the signal timing window to sub-nanosecond level, and the pulse type application load can cause severe electro-thermal transient effect in a very short time, mainly manifested as transient temperature rise and power rail collapse, which directly threatens the accuracy of data transmission.

[0003] The prior art strictly follows the principle of hierarchical decoupling, and the signal adjustment of the physical layer is passive and lagging, which cannot foresee and cope with the sub-nanosecond transient failure caused by load surge, and instead brings significant performance overhead, therefore, there is an urgent need in the art for an innovative technical solution with prediction ability, which can actively adjust the physical layer parameters for forward-looking compensation.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a hybrid signal processing method and system for memory bar, and the technical scheme is as follows:

[0006] The application load sensing step, real-time analysis of resource scheduling information of the operating system, extracts the characteristic parameters related to the future memory access intensity, and generates the quantized future work load intensity index;

[0007] The electro-thermal transient effect prediction step receives the work load intensity index, and calculates the predicted temperature rise and predicted voltage drop of the memory interface to be occurred in the future time window by using the preset predictive electro-thermal coupling model;

[0008] The compensation strategy generation step calculates and generates the feedforward physical layer compensation control instruction according to the predicted temperature rise and predicted voltage drop, and combines the preset system stability margin target;

[0009] The feedforward parameter adjustment step executes the compensation control instruction to actively adjust the physical parameters of the memory interface before the predicted electro-thermal transient effect actually occurs, so as to offset the electro-thermal impact.

[0010] Preferably, the work load intensity index is a dimensionless relative value calculated at time t, which represents the future memory access pressure, and the input parameters of the work load intensity index are normalized.

[0011] Preferably, the electrothermal transient effect prediction step includes:

[0012] The predicted temperature rise is calculated using a first preset formula, which consists of a steady-state response term and a transient response term.

[0013] The predicted voltage drop is calculated using a second preset formula, which consists of a steady-state response term and a transient response term.

[0014] The first and second preset formulas constitute the predictive electrothermal coupling model.

[0015] Preferably, the derivative term of the predictive electrothermal coupling model in the transient response term is approximated by backward difference, wherein the backward difference is obtained by subtracting the workload intensity index before the preset sampling time interval from the current workload intensity index and dividing by the sampling time interval.

[0016] Preferably, the physical layer compensation control command includes a target reference voltage adjustment and a target termination resistance adjustment, wherein the target reference voltage adjustment and the target termination resistance adjustment are obtained through a preset linear compensation strategy model.

[0017] Preferred linear compensation strategy models include:

[0018] The target reference voltage adjustment is calculated using a third preset formula, which is based on predicted temperature rise and predicted voltage drop.

[0019] The target termination resistance adjustment is calculated using the fourth preset formula, which is based on the predicted temperature rise.

[0020] Preferably, the feedforward parameter adjustment steps include:

[0021] The target reference voltage adjustment and the target termination resistor adjustment are respectively superimposed on the nominal reference voltage and nominal termination resistor of the memory interface to obtain a new reference voltage and a new termination resistor.

[0022] Preferably, the activation condition for the compensation strategy generation step is: when the absolute value of the predicted temperature rise is greater than the preset temperature threshold, or when the absolute value of the predicted voltage drop is greater than the preset voltage threshold.

[0023] A mixed-signal processing system for memory modules, comprising:

[0024] The application load awareness module is used to analyze the resource scheduling information of the operating system in real time, extract feature parameters related to future memory access intensity, and generate a quantitative indicator of future workload intensity.

[0025] The predictive electrothermal modeling module receives workload intensity indicators and uses a preset predictive electrothermal coupling model to calculate the predicted temperature rise and predicted voltage drop of the memory interface within a future time window.

[0026] The hybrid signal compensation control module is used to calculate and generate feedforward physical layer compensation control commands based on the predicted temperature rise and predicted voltage drop, combined with the preset system stability margin target.

[0027] The dynamic signal parameter adjustment module is used to execute compensation control instructions before the predicted electrothermal transient effect actually occurs, and actively adjust the physical parameters of the memory interface to counteract the electrothermal shock.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. This invention proactively predicts temperature rise and voltage drop caused by workload changes and actively compensates for parameters before the actual occurrence of electrothermal shock, thereby eliminating transient data transmission errors caused by pulsed loads and significantly enhancing the operational stability and reliability of the memory subsystem under harsh operating conditions.

[0030] 2. By pre-emptively offsetting the electrothermal transient effect, this invention ensures signal integrity, enabling the memory subsystem to continue operating at peak performance even under extreme loads, significantly improving data throughput.

[0031] 3. In this invention, the compensation strategy sets a precise activation threshold and only activates when the predicted temperature rise or voltage drop reaches a level that threatens system stability. This intelligent mechanism of on-demand adjustment avoids unnecessary frequent adjustments when the system experiences slight fluctuations, reduces control overhead and energy consumption, and improves the overall operating efficiency of the system.

[0032] 4. This invention directly links the task scheduling information of the upper-level operating system with the physical parameter adjustment of the underlying hardware, breaking down the traditional barrier of software and hardware separation, creating an efficient collaborative optimization path from software to hardware, and realizing deep integration and optimization of system resources. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention.

[0034] Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example 1

[0037] Please see Figure 1 A mixed-signal processing method for memory modules, comprising:

[0038] The application load awareness step analyzes the operating system's resource scheduling information in real time, extracts feature parameters related to future memory access intensity, and generates a quantitative indicator of future workload intensity.

[0039] The electrothermal transient effect prediction step receives the workload intensity index and uses a preset predictive electrothermal coupling model to calculate the predicted temperature rise and predicted voltage drop of the memory interface within a future time window.

[0040] The compensation strategy generation step calculates and generates feedforward physical layer compensation control commands based on the predicted temperature rise and predicted voltage drop, combined with the preset system stability margin target.

[0041] The feedforward parameter adjustment step executes compensation control instructions before the predicted electrothermal transient effect actually occurs, actively adjusting the physical parameters of the memory interface to counteract the electrothermal shock.

[0042] This method demonstrates outstanding performance in data centers and high-performance computing environments. Transient temperature rise and voltage drop at the memory interface pose a direct threat to the accuracy of data transmission. The core of this method lies in innovating the traditional passive and lagging signal adjustment mechanism into a proactive and forward-looking response strategy. It aims to eliminate sub-nanosecond transient failures caused by pulsed application loads, thereby significantly improving the stability and data throughput of the memory subsystem under harsh operating conditions without compromising system performance.

[0043] Specifically, this method accurately captures the operating system's resource scheduling information by applying a load-aware step. It generates a quantifiable indicator of future workload intensity by analyzing the scheduling priority of operating system tasks, scheduling queue status, historical memory access patterns, and the expected behavior of current tasks. This information, as the core characteristic parameter of future memory access intensity, can reflect potential resource demands more proactively than simple hardware performance counters. The generation of this indicator provides a solid data foundation for subsequent physical layer predictions, ensuring that the system can accurately predict potential electrothermal transient effects.

[0044] Following this, the electrothermal transient effect prediction step takes over this workload intensity index and, with the help of a finely pre-set predictive electrothermal coupling model, accurately estimates the temperature rise and voltage drop faced by the memory interface within a very short future time window. The introduction of this model enables the system to effectively avoid the inherent response delay of traditional physical sensors, thereby achieving microsecond-level early warning and response before sub-nanosecond-level timing risks arrive.

[0045] Based on the above prediction results, the compensation strategy generation step, according to the predicted temperature rise and voltage drop, and in combination with the established system stability margin target, accurately derives the feedforward physical layer compensation control commands; these commands constitute the basis for the system's active intervention, aiming to accurately offset the impending electrothermal shock.

[0046] Ultimately, in the feedforward parameter adjustment step, the aforementioned compensation control instructions are executed in advance to proactively adjust the key physical parameters of the memory interface before the predicted electrothermal transient effect actually occurs. This proactive adjustment changes the traditional post-event remediation approach, avoids data transmission errors, maximizes the performance of the memory subsystem under extreme loads, and enables it to continue to operate stably under more stringent conditions.

[0047] Example 2

[0048] The workload intensity index is a dimensionless relative value that is calculated at time t and represents the future memory access pressure. The input parameters of the workload intensity index have all been normalized.

[0049] The workload intensity index is designed to transform the "application intent" of upper-layer applications, which cannot be directly measured, into a precise quantitative indicator that can be physically predicted. This index is calculated comprehensively at time t, and its core design lies in the fact that operating system task scheduling information can reflect future resource demands more proactively than hardware performance counters. This index is a dimensionless relative value that characterizes future memory access pressure, and all its input parameters have undergone rigorous normalization. This dimensionless processing ensures that input data from different sources and with different dimensions can be uniformly integrated, thereby improving the model's generalization ability.

[0050] The workload intensity index is calculated using the following formula:

[0051]

[0052] (I(t): Workload intensity index at time t, a dimensionless relative value representing future memory access pressure; N: Total number of tasks; p: Task index; w) p : Task static priority weight; P p The static priority of task p can usually be obtained directly from the operating system's task management module and then normalized; wq : Scheduling queue length weight; Q p (t): The normalized length of the scheduling queue for task p at time t, which can be obtained by monitoring the task queue status of the operating system scheduler and normalized to its maximum value or dynamic range; w m : Memory behavior weights; M p (t): The normalized representation of the recent memory behavior of task p at time t, which can be obtained by statistically analyzing the number of memory accesses, memory bandwidth requirements or cache hit rate of the task in a short time window in the past and then normalizing them.

[0053] This formula is rooted in a deep understanding of system response: the operating system's task scheduling information can predict future resource needs earlier and more accurately; it transforms abstract "application intent" into actionable physical predictive inputs, enabling cross-level linkage from the software to the hardware level.

[0054] The formula is clearly structured, including core parameters such as task static priority, scheduling queue length, and recent memory behavior. It comprehensively reflects future memory access pressure through weighted summation. All parameters are normalized to ensure the relativity and comparability of the results. Its application lies in serving as the sole input to the subsequent electrothermal transient prediction model, realizing strong correlation between steps. Its technical effect is to provide accurate and forward-looking load assessment, providing a quantitative basis for the system to perform predictive avoidance.

[0055] The weighting coefficients in the workload intensity index are not set arbitrarily, but determined through an offline calibration process. On a controlled test bench, excitation signals with known intensity and rate of change are applied, and operating system scheduling data, actual power consumption, temperature rise, and voltage drop data of the memory interface are collected simultaneously. Mathematical fitting tools such as multiple linear regression and least squares method are used to solve for a set of parameter values ​​that minimize the error between the model's predicted values ​​and the actual measured values. This set of calibrated parameter values ​​is then embedded in the memory controller firmware or system driver, serving as the basis for the online runtime model. This rigorous determination method ensures a close fit between the model parameters and the specific hardware platform, thereby improving the accuracy of predictions and the stability of the system.

[0056] Example 3

[0057] The steps for predicting electrothermal transient effects include:

[0058] The predicted temperature rise is calculated using a first preset formula, which consists of a steady-state response term and a transient response term.

[0059] The predicted voltage drop is calculated using a second preset formula, which consists of a steady-state response term and a transient response term.

[0060] The first and second preset formulas constitute the predictive electrothermal coupling model;

[0061] The derivative of the predictive electrothermal coupling model in the transient response term is approximated by backward difference, which is obtained by subtracting the workload intensity index before the preset sampling time interval from the current workload intensity index and dividing by the sampling time interval.

[0062] The core of the electrothermal transient effect prediction step lies in accurately transforming the abstract workload intensity index into a specific physical prediction quantity. This step utilizes a pre-set predictive electrothermal coupling model, which is rooted in the first-order linear approximation theory of system response. It cleverly simplifies the complex nonlinear electrothermal physical process into the sum of the steady-state and transient terms that have the most significant impact on system behavior. The technical motivation of this model is to provide a forward-looking and quantitative physical state prediction, thereby enabling early identification and proactive intervention of potential risks to memory interfaces.

[0063] The calculation of predicted temperature rise is completed through the first preset formula, while the calculation of predicted voltage drop depends on the second preset formula; these two formulas are structurally parallel and together constitute the core prediction engine of the entire predictive electrothermal coupling model.

[0064] The first preset formula calculates and predicts the temperature rise:

[0065]

[0066] (ΔT: predicted temperature rise within the future time window; t: current time; δ: prediction time window; κ: prediction time window) T : Basic electrothermal conversion coefficient; I(t): Work load intensity index at time t; ζ T : Transient response coefficient for temperature rise; dI(t) / dt: Time derivative of the workload intensity index;

[0067] The second preset formula calculates the predicted voltage drop:

[0068]

[0069] (V d : Predicted voltage drop within the future time window; t: Current time; δ: Predicted time window; κ V : Basic electrical conversion factor; I(t): Work load intensity index at time t; ζ V : Voltage transient response coefficient; dI(t) / dt: time derivative of the workload intensity index;

[0070] The above formulas all contain two key parts: the first term is the steady-state response term (κ·I(t)), which reflects the continuous effect of the current load; the second term is the transient response term (ζ·dI(t) / dt), which captures the rate effect of load change. To ensure practical feasibility, the derivative term in the transient response term is approximated using backward difference, specifically in the following form:

[0071]

[0072] (Δt s (Sampling time interval);

[0073] The approximate calculation method for this derivative term originates from the backward difference principle in numerical analysis, which transforms the theoretical derivative into a quantifiable, actual calculated value based on discrete sampled data, thereby making real-time prediction possible.

[0074] The application of these formulas is to transform abstract workload intensity indicators into predicted temperature rise and predicted voltage drop with clear physical dimensions, providing accurate input for the generation of subsequent compensation strategies; its technical effect is significant: through accurate prediction of temperature rise and voltage drop, the system can take intervention measures in advance before the occurrence of electrothermal shock, avoid the inherent delay of traditional passive adjustment mechanisms, significantly improve memory signal integrity, and ensure stable operation of the system under high load.

[0075] All parameters in the predictive electrothermal coupling model (κ) T ,κ V ,ζ T ,ζ V These are all empirical parameters tightly bound to specific hardware platforms; they are determined through a one-time offline calibration process. On a controlled test bench, a series of excitation signals with known intensity and rate of change are applied, and high-precision measurement equipment is used to synchronously collect data on the actual power consumption, temperature rise, and voltage drop of the memory interface. Then, mathematical fitting tools such as multiple linear regression and least squares are used to analyze a large amount of collected data, thereby solving for a set of parameter values ​​that minimize the error between the model's predicted values ​​and the actual measured values. These calibrated parameters, which are fixed in the memory controller firmware or system driver, are the key to accurate predictions when the model is running online.

[0076] Example 4

[0077] The physical layer compensation control command includes the target reference voltage adjustment amount and the target termination resistance adjustment amount, which are obtained through a preset linear compensation strategy model.

[0078] Linear compensation strategy models include:

[0079] The target reference voltage adjustment is calculated using a third preset formula, which is based on predicted temperature rise and predicted voltage drop.

[0080] The target termination resistance adjustment is calculated using the fourth preset formula, which is based on the predicted temperature rise.

[0081] After obtaining accurate physical predictions, the compensation strategy generation step transforms these predictions into precise electrical compensation instructions, aiming to preemptively offset potential electrothermal shocks. This process relies on a pre-defined linear compensation strategy model, which is rooted in the physical characteristics of semiconductor devices: within the normal operating range, the amount of compensation required to offset the effects of temperature and voltage variations on signal timing is approximately linearly related to the physical offset. The technical motivation of this model is to provide a stable and repeatable mechanism that transforms complex physical effects into operable electrical parameter adjustments.

[0082] The core components of the physical layer compensation control command include the target reference voltage adjustment and the target termination resistance adjustment; these adjustments are precisely calculated using the aforementioned linear compensation strategy model.

[0083] The third preset formula calculates the target reference voltage adjustment:

[0084] ΔV r =C VT ·ΔT+C VV ·V d

[0085] (ΔV r : Target reference voltage adjustment; C VT Temperature rise compensation coefficient for voltage; ΔT: predicted temperature rise, °C VV Voltage compensation coefficient; V d (Predicted voltage drop, calculated using the second preset formula);

[0086] The formula is clearly structured and intuitively transforms the predicted temperature rise and predicted voltage drop into the adjustment requirements of the reference voltage through their respective compensation coefficients. Its application is to guide the voltage adjustment of the memory interface to deal with signal integrity issues caused by electrothermal transient effects. Its technical effect is to effectively maintain the signal level margin and avoid data transmission errors by actively adjusting the reference voltage.

[0087] The fourth preset formula calculates the target termination resistance adjustment amount:

[0088] ΔR=C RT ·ΔT

[0089] (ΔR: Target termination resistor adjustment; C) RT : Temperature rise compensation coefficient for resistance; ΔT: Predicted temperature rise);

[0090] This formula is relatively simple in structure. It mainly adjusts the termination resistor based on the predicted temperature rise to optimize signal reflection and impedance matching. Its application is to improve the reflection characteristics of the signal on the transmission line by adjusting the termination resistor, thus ensuring the integrity of the signal waveform. Its technical effect is to effectively suppress impedance mismatch caused by temperature changes and further improve the signal transmission quality.

[0091] All parameters C in the linear compensation strategy model VT C VV C RT All parameters are empirical, and their determination is also achieved through offline calibration. In a controlled test environment, known temperature and voltage offsets are applied, and the required reference voltage and termination resistance compensation are measured simultaneously. Mathematical fitting tools are used to analyze the data, thereby solving for a set of parameter values ​​that minimize the error between the model prediction and the actual measurement. These calibrated parameter values ​​are embedded in the memory controller firmware or system driver, ensuring the accuracy and stability of the compensation strategy during online operation. The stability and repeatability of the compensation operation are guaranteed, avoiding the accumulation of errors that may be caused by continuous adjustments.

[0092] Example 5

[0093] The feedforward parameter adjustment steps include:

[0094] The target reference voltage adjustment and the target termination resistor adjustment are respectively superimposed on the nominal reference voltage and nominal termination resistor of the memory interface to obtain a new reference voltage and a new termination resistor.

[0095] The feedforward parameter adjustment step is the key execution link for this method to achieve predictive avoidance. In this step, the target reference voltage adjustment and target termination resistance adjustment calculated in the aforementioned compensation strategy generation step are precisely applied. The theoretical calculation is transformed into actual physical parameter intervention, thereby enabling proactive adjustment of the physical parameters of the memory interface before the electrothermal transient effect actually manifests.

[0096] This step involves directly adding the calculated adjustment amount to the nominal reference value of the memory interface; specifically, the new reference voltage V r,n The calculation method is as follows:

[0097] V r,n =V r,0 +ΔV r

[0098] (V r,n New reference voltage; V r,0 : Nominal reference voltage of the memory interface; V r :Target reference voltage adjustment amount;

[0099] New termination resistor (R) t,n The calculation method for ) is as follows:

[0100] R t,n =R t,0 +ΔR

[0101] (R t,n New termination resistor; R t,0 : Nominal termination resistor of memory interface; ΔR: Target termination resistor adjustment amount;

[0102] This adjustment method based on a fixed reference value effectively avoids the accumulation of errors that may be caused by continuous adjustments, thereby ensuring the stability and repeatability of system operation. By actively adjusting these key physical parameters before the occurrence of electrothermal shock, the memory subsystem can preemptively offset the impending signal distortion, ensuring the integrity of data transmission. This greatly improves the system's performance under high load, eliminating the need to rely on traditional performance-sacrificing methods such as thermal throttling or ECC retransmission, thus enabling continuous and stable operation under more stringent conditions and significantly improving the overall data throughput of the system.

[0103] Example 6

[0104] The activation condition for the compensation strategy generation step is: when the absolute value of the predicted temperature rise is greater than the preset temperature threshold, or when the absolute value of the predicted voltage drop is greater than the preset voltage threshold.

[0105] To avoid over-adjustment of the system and unnecessary energy consumption, the compensation strategy generation step is not continuous, but is precisely set with activation conditions; this enables intelligent on-demand compensation, initiating complex calculation and adjustment processes only when necessary, thereby optimizing system resource utilization.

[0106] The activation condition clearly stipulates that the above compensation calculation and execution will only be initiated when the absolute value of the predicted temperature rise is greater than the preset temperature threshold, or when the absolute value of the predicted voltage drop is greater than the preset voltage threshold.

[0107] The sources and methods for determining these thresholds are not arbitrary, but strictly based on the operating margins of memory chips specified in industry standards such as JEDEC. Specifically, these thresholds are obtained by subtracting a safety margin from the maximum permissible temperature rise or voltage drop specified in industry standards. This setting logic ensures that the compensation mechanism is only activated when there is a real risk to signal integrity, while reserving sufficient safety margins.

[0108] This intelligent startup mechanism avoids frequent adjustments to the system during minor fluctuations, reducing control overhead and potential system instability. At the same time, it ensures that the system can quickly and decisively initiate proactive compensation before a real electrothermal shock occurs, thereby maximizing the performance of the memory subsystem while ensuring long-term stable operation of the system and effectively preventing waste of system resources caused by unnecessary triggering.

[0109] Example 7

[0110] Please see Figure 2 A mixed-signal processing system for memory modules, comprising:

[0111] The application load awareness module is used to analyze the resource scheduling information of the operating system in real time, extract feature parameters related to future memory access intensity, and generate a quantitative indicator of future workload intensity.

[0112] The predictive electrothermal modeling module receives workload intensity indicators and uses a preset predictive electrothermal coupling model to calculate the predicted temperature rise and predicted voltage drop of the memory interface within a future time window.

[0113] The hybrid signal compensation control module is used to calculate and generate feedforward physical layer compensation control commands based on the predicted temperature rise and predicted voltage drop, combined with the preset system stability margin target.

[0114] The dynamic signal parameter adjustment module is used to execute compensation control instructions before the predicted electrothermal transient effect actually occurs, and actively adjust the physical parameters of the memory interface to counteract the electrothermal shock.

[0115] This system serves as a concrete implementation of the aforementioned methods. Its sophisticated architecture is designed to maintain excellent memory signal integrity under high load conditions. This system can dynamically and deeply integrate upper-layer application scheduling information with lower-layer physical layer signals, providing a revolutionary solution for scenarios with stringent requirements for memory performance and stability, such as data centers and high-performance computing.

[0116] The application load awareness module is the system's perception center, deployed at the operating system or virtual machine manager layer. This module continuously and in real time parses the resource scheduling information of the operating system or virtual machine manager, which contains key characteristic parameters of future memory access intensity. This module accurately extracts the core elements from these complex data streams and transforms them into quantitative indicators of future workload intensity, laying the foundation for the predictive capabilities of the entire system.

[0117] The tightly integrated predictive electrothermal modeling module receives the workload intensity index generated by the application load sensing module. This module embeds a precisely calibrated predictive electrothermal coupling model, which can accurately calculate the predicted temperature rise and voltage drop that the memory interface will face within a very short future time window. The introduction of this module enables the system to break through the inherent delay bottleneck of traditional physical sensors, achieve early warning of potential electrothermal shocks, and elevate the protection of memory signal integrity from traditional post-event remediation to a new level of pre-event prevention.

[0118] Based on the above prediction results, the hybrid signal compensation control module accurately calculates and generates feedforward physical layer compensation control commands based on the predicted temperature rise and voltage drop, combined with the preset system stability margin target. These commands carry the strategy of active intervention of the system, which aims to accurately offset the impending electrothermal shock. The intelligent decision-making capability of this module ensures the accuracy and effectiveness of the compensation strategy, thereby maintaining signal integrity to the greatest extent.

[0119] Before the predicted electrothermal transient effect actually occurs, the dynamic signal parameter adjustment module decisively executes the compensation control instructions generated by the hybrid signal compensation control module. This module proactively adjusts key physical parameters of the memory interface, such as reference voltage and termination resistance, to effectively offset the impending electrothermal shock. The collaborative optimization effect of this module breaks down the design barriers of hardware and software separation in traditional architectures, creating a direct and efficient optimization path from application scheduling to physical layer parameters. This significantly improves the stability and data throughput of the memory subsystem under extreme conditions, opening up new avenues for the design of future high-performance computing systems.

[0120] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A mixed-signal processing method for memory modules, characterized in that, include: The application load awareness step analyzes the resource scheduling information of the operating system in real time, extracts feature parameters related to future memory access intensity, and generates a quantitative indicator of future workload intensity. The electrothermal transient effect prediction step receives the workload intensity index and uses a preset predictive electrothermal coupling model to calculate the predicted temperature rise and predicted voltage drop of the memory interface within a future time window. The compensation strategy generation step calculates and generates feedforward physical layer compensation control commands based on the predicted temperature rise and predicted voltage drop, combined with the preset system stability margin target. The feedforward parameter adjustment step executes compensation control instructions before the predicted electrothermal transient effect actually occurs, actively adjusting the physical parameters of the memory interface to counteract the electrothermal shock.

2. The mixed-signal processing method for memory modules according to claim 1, characterized in that, The workload intensity index is a dimensionless relative value that is calculated at time t and represents the future memory access pressure. The input parameters of the workload intensity index have all been normalized.

3. The mixed-signal processing method for memory modules according to claim 1, characterized in that, The steps for predicting electrothermal transient effects include: The predicted temperature rise is calculated using a first preset formula, which consists of a steady-state response term and a transient response term. The predicted voltage drop is calculated using a second preset formula, which consists of a steady-state response term and a transient response term. The first and second preset formulas constitute the predictive electrothermal coupling model.

4. The mixed-signal processing method for memory modules according to claim 3, characterized in that, The derivative of the predictive electrothermal coupling model in the transient response term is approximated by backward difference, which is obtained by subtracting the workload intensity index before the preset sampling time interval from the current workload intensity index and dividing by the sampling time interval.

5. The mixed-signal processing method for memory modules according to claim 1, characterized in that, The physical layer compensation control commands include target reference voltage adjustment and target termination resistance adjustment, which are obtained through a preset linear compensation strategy model.

6. The mixed-signal processing method for memory modules according to claim 5, characterized in that, Linear compensation strategy models include: The target reference voltage adjustment is calculated using a third preset formula, which is based on predicted temperature rise and predicted voltage drop. The target termination resistance adjustment is calculated using a fourth preset formula, which is based on the predicted temperature rise.

7. A mixed-signal processing method for memory modules according to claim 5, characterized in that, The feedforward parameter adjustment steps include: The target reference voltage adjustment and the target termination resistor adjustment are respectively superimposed on the nominal reference voltage and nominal termination resistor of the memory interface to obtain a new reference voltage and a new termination resistor.

8. The mixed-signal processing method for memory modules according to claim 1, characterized in that, The conditions for starting the compensation strategy generation step are: when the absolute value of the predicted temperature rise is greater than the preset temperature threshold, or when the absolute value of the predicted voltage drop is greater than the preset voltage threshold.

9. A mixed-signal processing system for memory modules, applied to a mixed-signal processing method for memory modules according to any one of claims 1 to 8, characterized in that, include: The application load awareness module is used to analyze the resource scheduling information of the operating system in real time, extract feature parameters related to future memory access intensity, and generate a quantitative indicator of future workload intensity. The predictive electrothermal modeling module receives workload intensity indicators and uses a preset predictive electrothermal coupling model to calculate the predicted temperature rise and predicted voltage drop of the memory interface within a future time window. The hybrid signal compensation control module is used to calculate and generate feedforward physical layer compensation control commands based on the predicted temperature rise and predicted voltage drop, combined with the preset system stability margin target. The dynamic signal parameter adjustment module is used to execute compensation control instructions before the predicted electrothermal transient effect actually occurs, and actively adjust the physical parameters of the memory interface to counteract the electrothermal shock.

Images