Micro-system SiP power consumption analysis and low-power optimization verification method and system

Abstract

The application relates to the technical field of semiconductor system-in-package design verification, and discloses a microsystem SiP power consumption analysis and low-power consumption optimization verification method and system. Working state parameters of digital modules, analog modules and radio frequency modules in a microsystem system-in-package are acquired, an electromagnetic interference transfer relationship and a heat conduction network are established to form a multidimensional power consumption coupling model; dielectric characteristic correction and electric conductance characteristic correction are formed according to module working temperatures to update the electromagnetic interference transfer relationship, power consumption distribution data are obtained by combining power consumption change gradient adjustment and iterative weight updating; a low-power consumption optimization strategy is formed according to the power consumption distribution, and the model is corrected by using test power consumption data after simulation verification and test verification. The application is suitable for application specific integrated circuit microsystem system-in-package, and improves the accuracy of power consumption analysis and low-power consumption optimization verification.

Description

Technical Field

[0001] This invention relates to the field of semiconductor system-in-package design verification technology, specifically to a method and system for power consumption analysis and low-power optimization verification of microsystems (SiP). Background Technology

[0002] Microsystem-level packaging (MSIP) integrates digital, analog, and RF modules. These modules differ in operating frequency, power supply status, power density, thermal boundary, and electromagnetic susceptibility. Conventional power analysis performs power summarization, thermal analysis, and electromagnetic analysis separately. The model uses fixed parameters and unidirectional data transfer, resulting in data breakpoints across physical domains. Module power consumption creates localized heat sources, which are transferred through the package substrate, interconnects, heat dissipation structures, and external thermal boundaries, causing variations in module operating temperature. These temperature variations alter the conductive state of the propagation path between modules and the dielectric state of the package material, thus changing the electromagnetic propagation path between modules.

[0003] The thermal model outputs the module operating temperature only for thermal state evaluation; the module operating temperature cannot be used for electromagnetic model parameter updates, resulting in interrupted data transfer between thermal and electromagnetic states. The electromagnetic model uses an initial fixed transfer coefficient throughout the power consumption analysis, failing to reflect electromagnetic propagation changes relative to a reference temperature. Repeated calculations of module power consumption and operating temperature cannot fully represent the additional power consumption resulting from temperature changes causing changes in the electromagnetic transfer state between modules. For operating states with simultaneous localized high power consumption, high-frequency coupling, and strong thermal feedback, directly using candidate power consumption to cover the previous cycle's power consumption results in significant single-cycle variations, and the fixed update amplitude causes the power consumption estimate to fluctuate repeatedly around a stable state.

[0004] In microwave and millimeter-wave operation, the internal interconnect structure of a system-in-package (SiP) exhibits distributed parameter characteristics. Through-silicon vias (TSVs), redistribution layers, package traces, bumps, and adjacent ground structures create parasitic capacitance and inductance, resulting in frequency-selective electromagnetic propagation. Local temperature changes alter the dielectric or conductive state of the propagation path, altering not only the propagation strength but also causing a frequency domain shift in the equivalent resonant position of the propagation path. When the relative position of the digital module's switching noise harmonic spectrum and the RF module's sensitive frequency band changes with temperature, using only a temperature-scalar transfer coefficient is insufficient to distinguish between overall changes in coupling strength and the shift of spectral energy towards the sensitive frequency band. When the digital module's clock harmonics shift from outside to inside the RF sensitive frequency band, even with a small change in the source module's base power consumption, the degree of interference to the target module significantly changes.

[0005] The transient large-scale switching of digital modules and the transmitting or sensitive receiving activities of RF modules have temporal positions. Two activity time windows do not overlap in the source clock domain, but the arrival time of digital switching interference at the RF module location after propagation delay through the medium may overlap, forming a short-term high-amplitude coupling state. Traditional low-power strategies mainly reduce average power consumption. In scenarios requiring the maintenance of digital module computing power or RF module operation, simply reducing voltage or shutting down functional modules impacts system performance. Existing technologies lack methods to locate high-risk time windows based on dynamic electromagnetic coupling states and adjust the digital module clock trigger position within timing margins. They also lack closed-loop mechanisms for calibrating electromagnetic interference propagation relationships and heat conduction network parameters using test data. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for power consumption analysis and low power optimization verification of microsystem SiP, which solves the technical problems in power consumption analysis of microsystem SiP, such as the disconnect between power consumption calculation, heat conduction and electromagnetic interference transmission, the inability to correct electromagnetic transmission relationships between modules due to temperature changes, fluctuations in power consumption estimates caused by fixed iteration update amplitude, and the lack of time-domain active avoidance and test data back-correction model mechanism for low power strategies.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] The method for power consumption analysis and low-power optimization verification of microsystems (SiP) includes the following steps:

[0009] Obtain the operating state parameters of heterogeneous functional modules within the system-level package of a microsystem, establish electromagnetic interference transmission relationships and heat conduction networks based on the operating state parameters, and form a multidimensional power consumption coupling model.

[0010] The module power consumption estimate is input into the multidimensional power consumption coupling model to obtain the module operating temperature and electromagnetic interference response. Based on the module operating temperature, dielectric property correction and conductivity correction are generated, and the electromagnetic interference transmission relationship is updated accordingly.

[0011] The updated electromagnetic interference transmission relationship and module operating temperature are fed back into the multidimensional power consumption coupling model to obtain candidate power consumption estimates. The iteration update weights are determined based on the gradient of the change between the candidate power consumption estimates and the power consumption estimates of the previous iteration cycle, and the module power consumption estimates are updated. This process is repeated until the convergence condition is met to obtain power consumption distribution data.

[0012] Based on the power distribution data, high-power regions and modules with abnormal power consumption are identified. Based on the identification results, low-power optimization strategies including clock gating, power domain management or working mode switching are formed and simulation verification is completed.

[0013] The low-power optimization strategy verified by simulation is tested and verified. The power consumption data obtained from the test is used to correct the multidimensional power coupling model and output the evaluation results of the optimization effect.

[0014] As a preferred technical solution of the present invention, the heterogeneous functional module includes a digital module, an analog module, and a radio frequency module; the operating state parameters of the digital module include operating frequency, logic conversion state, and switching activity state; the operating state parameters of the analog module include supply voltage, bias current, and signal gain; and the operating state parameters of the radio frequency module include transmit power, receive state, and operating frequency band; the operating state parameters are used to form a module state vector according to a unified sampling time marker, and the module state vector is input into a multidimensional power consumption coupling model.

[0015] As a preferred technical solution of the present invention, the establishment of electromagnetic interference transmission relationship includes: forming an interference source spectrum description based on the switching state of the digital module; forming a disturbance sensitivity description by combining the disturbance response of the analog module and the radio frequency module in the corresponding frequency band; forming a propagation path description by combining the physical position relationship, interconnection relationship and packaging material parameters between heterogeneous functional modules; generating initial electromagnetic interference transmission coefficients between modules according to the interference source spectrum description, disturbance sensitivity description and propagation path description, and forming an electromagnetic interference transmission matrix by the initial electromagnetic interference transmission coefficients between modules.

[0016] As a preferred technical solution of the present invention, the establishment of the heat conduction network includes: establishing hot nodes and heat transfer connections between hot nodes based on the heterogeneous functional modules, the packaging substrate and the packaging heat dissipation boundary; configuring heat network parameters formed by packaging structure data, material thermal parameters or calibration data for the hot nodes and heat transfer connections; mapping the estimated power consumption of the module to the heat input of the corresponding hot node, and obtaining the module operating temperature of each heterogeneous functional module based on the heat transfer connections between the hot nodes.

[0017] As a preferred embodiment of the present invention, the dielectric property correction and conductivity correction are generated based on the module operating temperature, including: for the inter-module transmission path in the electromagnetic interference transmission matrix, the dielectric property correction is generated based on the module operating temperature and the material parameter data, packaging structure electromagnetic simulation calibration results, or test calibration results corresponding to the inter-module transmission path; the conductivity correction is generated based on the module operating temperature and the interconnect conductivity parameter data or test calibration results corresponding to the inter-module transmission path; the dielectric property correction and conductivity correction are applied together to the corresponding initial inter-module electromagnetic interference transmission coefficient to generate a dynamic electromagnetic interference transmission matrix.

[0018] As a preferred technical solution of the present invention, determining the iterative update weight and updating the module power consumption estimate includes: normalizing the difference between the candidate power consumption estimate and the power consumption estimate of the previous iteration cycle to form a power consumption change gradient reflecting the current iteration state; decreasing the iterative update weight when the power consumption change gradient increases, and increasing the iterative update weight when the power consumption change gradient decreases; using the iterative update weight to perform a weighted update on the candidate power consumption estimate and the power consumption estimate of the previous iteration cycle; outputting power consumption distribution data when the normalized power consumption change of all heterogeneous functional modules meets the convergence threshold, and decreasing the iterative update weight and re-iterating from the most recently stable module power consumption estimate when a continuous increase in power consumption change is detected.

[0019] As a preferred technical solution of the present invention, identifying high power consumption regions and power consumption abnormal modules includes: determining the relative power consumption levels between heterogeneous functional modules based on power consumption distribution data; forming a high power consumption module set by combining the spatial location and relative power consumption levels of the heterogeneous functional modules; forming a high power consumption region based on the spatial adjacency relationship of the high power consumption module set; identifying at least one power consumption abnormal state among power consumption surge, sustained high power consumption, or periodic abnormality using power consumption distribution data from continuous analysis periods, and marking the corresponding heterogeneous functional module as a power consumption abnormal module.

[0020] As a preferred technical solution of the present invention, before forming a low-power optimization strategy, the power distribution data of the continuous analysis cycle and the current state of the multidimensional power coupling model are combined as prediction input. The power change of the subsequent analysis cycle is calculated according to the thermal feedback relationship and dynamic electromagnetic interference transmission relationship of the multidimensional power coupling model. When the calculation result meets the power anomaly judgment condition, the power warning information of the corresponding heterogeneous functional module is output, and the power warning information is used as the strategy object input of the low-power optimization strategy.

[0021] As a preferred technical solution of the present invention, the multidimensional power consumption coupling model is corrected using the power consumption data obtained from the test, including: aligning the test power consumption data under the same working state with the power consumption distribution data output by the multidimensional power consumption coupling model to form model residuals; performing joint parameter fitting on the inter-module transfer coefficients in the electromagnetic interference transfer matrix and the thermal network parameters in the heat conduction network based on the model residuals; updating the multidimensional power consumption coupling model with the fitted inter-module transfer coefficients and thermal network parameters, and using the updated multidimensional power consumption coupling model for subsequent analysis cycles.

[0022] As a preferred technical solution of the present invention, a low-power optimization strategy is formed, including: determining the digital module or digital sub-functional unit whose clock can be turned off and the clock gating trigger state based on the functional state of the high-power region and the power-abnormal module; determining the adjustable power domain and power domain state transition rules based on the power supply correlation and power distribution of heterogeneous functional modules; forming the working mode switching conditions based on the power-abnormal state, functional load and timing constraints; and forming a strategy candidate set by at least one of the clock gating strategy, power domain management strategy and working mode switching strategy.

[0023] As a preferred technical solution of the present invention, when performing simulation verification on the strategy candidate set, the control state corresponding to the strategy candidate set is input into the multi-dimensional power consumption coupling model and the packaging performance simulation model to obtain the optimized power consumption distribution, module operating temperature and timing performance results; when the power consumption constraint, thermal constraint and timing performance constraint are satisfied at the same time, the corresponding low power consumption optimization strategy is output to the test verification; when any constraint is not satisfied, the corresponding control state is adjusted according to the over-limit result and the simulation verification is performed again.

[0024] As a preferred technical solution of the present invention, the microsystem system-in-package is one of the following: mobile terminal RF front-end system-in-package, automotive millimeter-wave radar system-in-package, automotive infotainment system-in-package, or IoT communication gateway system-in-package; the module power consumption, module operating temperature, and at least one timing performance parameter corresponding to the collected power consumption distribution data are tested and verified, and the optimization effect evaluation result is generated based on the data before and after optimization.

[0025] As a preferred embodiment of the present invention, the dielectric property correction and the conductivity property correction are applied together to the corresponding initial electromagnetic interference transmission coefficient between modules to generate a dynamic electromagnetic interference transmission matrix, including:

[0026] The equivalent capacitance change rate of the electromagnetic interference propagation path between modules is obtained based on the dielectric property correction amount. Based on the initial resonant frequency at the reference temperature and the equivalent capacitance change rate, the resonant frequency at the current temperature is determined according to the inverse relationship between the resonant frequency and the square root of the equivalent capacitance. The difference between the resonant frequency at the current temperature and the initial resonant frequency is determined as the signed resonant frequency offset. When the resonant frequency moves towards a lower frequency, the signed resonant frequency offset is negative.

[0027] The frequency coordinates of the interference source spectrum description are shifted according to the signed resonant frequency offset to obtain the offset interference source spectrum description, and the disturbance sensitivity description is normalized into the disturbance sensitivity weighting function.

[0028] Calculate the integral of the spectral overlap between the offset interference source spectral description and the interference sensitivity weighting function in the target frequency band. The integral of the spectral overlap satisfies:

[0029] ;

[0030] in, This is the integral of the spectral overlap, with the dimension of power; This represents the lower limit of the target frequency band, in Hz. This represents the upper frequency limit of the target frequency band, in Hz. This is a frequency variable, with the unit being Hz; This is the signed resonant frequency offset, in Hz. The power spectral density of the offset interference source is expressed in units of power per hertz. The disturbance sensitivity weighting function is dimensionless.

[0031] The spectral overlap integral and the reference spectral overlap integral under the reference temperature state are normalized to form a dimensionless frequency domain coupling coefficient. The conductivity characteristic correction is converted into a dimensionless conductivity correction factor relative to the reference temperature state. The product of the frequency domain coupling coefficient and the conductivity correction factor is determined as the inter-module transfer coefficient update multiplier. The inter-module transfer coefficient update multiplier is used to correct the corresponding initial inter-module electromagnetic interference transfer coefficient to obtain the dynamic electromagnetic interference transfer matrix.

[0032] As a preferred embodiment of the present invention, the low-power optimization strategy further includes:

[0033] The dynamic electromagnetic interference transfer coefficient between the digital module and the radio frequency module is obtained in the dynamic electromagnetic interference transfer matrix. When the dynamic electromagnetic interference transfer coefficient is greater than the coupling isolation threshold, the transient switching time window of the digital module and the radio frequency activity time window of the radio frequency module are extracted. The radio frequency activity time window includes at least one of the radio frequency transmission time window or the radio frequency reception sensitive time window.

[0034] The equivalent relative permittivity at the current temperature is obtained by correcting the dielectric properties of the electromagnetic interference propagation path. The propagation delay time of the electromagnetic interference is determined according to the relationship that the propagation delay is proportional to the physical length of the propagation path, inversely proportional to the speed of light in vacuum, and proportional to the square root of the equivalent relative permittivity.

[0035] By utilizing the electromagnetic interference propagation delay time, the transient switching time window of the digital module is mapped to the physical time axis of the location of the RF module, and a unified physical time axis is established with the RF activity time window to obtain the overlapping interval of the two time windows.

[0036] The transient coupling power sequence at the location of the RF module is formed based on the transient power sequence of the digital module within the overlapping interval and the dynamic electromagnetic interference transfer coefficient from the digital module to the RF module, and the maximum value in the transient coupling power sequence is determined as the transient coupling peak power.

[0037] When the transient coupling peak power is greater than the power safety threshold, search for the time offset within the timing adjustment range allowed by the digital module so that the transient switching time window and the RF activity time window do not overlap, and select the time offset with the smallest absolute value as the minimum time offset.

[0038] Generate timing misalignment control instructions containing the minimum time offset, incorporate the timing misalignment control instructions into the working mode switching strategy, and change the clock trigger edge of the digital module through the clock delay adjustment module connected to the clock control terminal of the digital module, so that the transient switching time window of the digital module and the RF activity time window are separated on a unified physical time axis.

[0039] The present invention also discloses a microsystem SiP power consumption analysis and low power optimization verification system for performing the microsystem SiP power consumption analysis and low power optimization verification method as described above, including a parameter acquisition interface, a model solving module, a power consumption analysis and strategy generation module, a verification and calibration module, a processor, and a memory;

[0040] The parameter acquisition interface is connected to the working status monitoring node of the microsystem system-level package to obtain the working status parameters of the digital module, analog module and radio frequency module, and output the working status parameters to the model solving module;

[0041] The model solving module is implemented by the processor executing model solving instructions in the memory. It is used to establish the electromagnetic interference transmission relationship and heat conduction network based on the working state parameters, obtain the module operating temperature based on the module power consumption estimate, generate dielectric property correction and conductivity correction based on the module operating temperature and update the electromagnetic interference transmission relationship, obtain candidate power consumption estimates based on the updated electromagnetic interference transmission relationship, determine the iterative update weights based on the change state of the candidate power consumption estimates, and output power distribution data that meets the convergence condition in a loop.

[0042] The power consumption analysis and strategy generation module receives the power distribution data output by the model solving module, combines the spatial location and functional status of the heterogeneous functional modules to form high power consumption regions, power consumption abnormal modules and low power consumption optimization strategies, and sends the low power consumption optimization strategies to the verification and calibration module.

[0043] The verification and calibration module establishes data connections with the packaging performance simulation environment and the test data acquisition interface, respectively. It receives low-power optimization strategies for simulation verification and test verification, corrects the electromagnetic interference transmission relationship and heat conduction network parameters based on the test power consumption data, and returns the corrected model parameters to the model solving module.

[0044] When the dynamic electromagnetic interference propagation coefficient meets the timing misalignment control conditions, the power consumption analysis and strategy generation module is also used to generate timing misalignment control instructions based on the transient switching time window of the digital module, the radio frequency activity time window of the radio frequency module, and the electromagnetic interference propagation delay time, and send the timing misalignment control instructions to the clock delay adjustment module of the digital module through the clock control interface.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] This invention acquires the operating state parameters of digital, analog, and RF modules, establishes electromagnetic interference (EMI) propagation relationships and heat conduction networks, and forms a multi-dimensional power consumption coupling model. This model ensures that module power consumption, module operating temperature, and inter-module electromagnetic coupling states form a continuous data dependency within the same model. Compared to existing technologies where power consumption analysis, thermal analysis, and electromagnetic analysis are separated and only unidirectional data is transferred, this invention uses the module operating temperature output from the heat conduction network to generate dielectric and conductivity correction values. These correction values ​​are then used to update the EMI propagation relationships, which are then re-involved in power consumption calculations. This establishes a closed-loop feedback between temperature, electromagnetic coupling, and power consumption, improving the accuracy of power consumption distribution calculations under complex thermo-electromagnetic coupling conditions.

[0047] Furthermore, this invention dynamically adjusts the iterative update weights based on the gradient of the change between the candidate power consumption estimate and the power consumption estimate of the previous iteration cycle. When the power consumption changes significantly, the single update magnitude is reduced, and when the power consumption change slows down, the candidate power consumption absorption ratio is increased. This avoids the repeated fluctuations of the power consumption estimate under strong thermal feedback conditions and improves the convergence stability and iteration efficiency of the multidimensional power consumption coupling model.

[0048] This invention generates a signed resonant frequency offset based on dielectric property correction and updates the inter-module transfer coefficient by integrating the spectrum overlap between the offset interference source spectrum and the disturbance sensitivity weighting function. This enables the dynamic electromagnetic interference transfer matrix to reflect the frequency domain drift effect of spectral energy moving towards the sensitive frequency band of the target module, improving the resolution capability of electromagnetic interference coupling calculation in microwave and millimeter-wave bands. Furthermore, when the dynamic electromagnetic interference transfer coefficient exceeds the coupling isolation threshold, this invention maps the transient switching time window of the digital module and the activity time window of the RF module to a unified physical time axis through the electromagnetic interference propagation delay time. When the transient coupling peak power exceeds the power safety threshold, a timing misalignment control command is generated. This reduces the transient coupling risk without changing the computing power of the digital module or the operating state of the RF module, enhancing the electromagnetic compatibility and low-power control flexibility of the system-in-package. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the method of the present invention.

[0051] Figure 2 This is a flowchart of the iterative update weight determination and convergence judgment sub-process of the present invention.

[0052] Figure 3 This is a flowchart of the dynamic electromagnetic interference transmission matrix update sub-process of the present invention.

[0053] Figure 4 This is a flowchart of the timing misalignment control sub-process of the present invention. Detailed Implementation

[0054] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0055] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0056] The microsystem SiP power consumption analysis and low-power optimization verification method operates on an analysis cycle basis. A complete analysis cycle begins with the acquisition of operating state parameters and ends with the writing of model calibration results into the subsequent analysis cycle. Model parameters can be formed based on design data, package structure data, material data, electromagnetic calibration data, thermal calibration data, and test data.

[0057] Each heterogeneous functional module has a unique module identifier. Operating status data includes the module identifier, analysis cycle identifier, parameter name, parameter value, and parameter unit. Digital module status data forms a digital module status vector, analog module status data forms an analog module status vector, and RF module status data forms an RF module status vector. These different status vectors are combined according to the analysis cycle to form the system module status vector.

[0058] Parameter preprocessing checks whether the input parameters are within the allowable range of the corresponding interface and standardizes the units of power, temperature, frequency, voltage, current, and other physical quantities. Activity status can use normalized indices within the range of 0 to 1. Continuously sampled data is used to generate average, peak, or activity statistics required for the power consumption model within one analysis cycle. The parameter preprocessing output retains the original module identifier and analysis cycle identifier.

[0059] The base power consumption of a module can be obtained from transistor-level, gate-level, behavioral-level, or module-level power consumption models. The base power consumption of a digital module can be determined based on the power supply voltage, operating frequency, and active state; the base power consumption of an analog module can be determined based on the power supply voltage, bias current, and mode state; and the base power consumption of an RF module can be determined based on its transmit, receive, or standby operating state. The base power consumption is further incorporated into a multi-dimensional power coupling model and is corrected for thermal conditions and electromagnetic coupling between modules.

[0060] Each valid inter-module transfer path from the source module to the target module has a path identifier, which corresponds to the source module identifier and the target module identifier. The path record includes the reference temperature, the initial electromagnetic interference transfer coefficient between modules, the applicable frequency band, and the calibration source; when medium temperature correction is required, the path record saves the medium-related calibration data; when conductivity state temperature correction is required, the path record saves the conductivity-related calibration data.

[0061] Each thermal node in the heat conduction network has a thermal node identifier, and a mapping is established between module thermal nodes and module identifiers. Substrate thermal nodes and heat dissipation boundary nodes are established according to the packaging structure, and the thermal transfer connections record the equivalent thermal parameters between two thermal nodes. Transient thermal analysis can also configure equivalent thermal capacity at thermal nodes, and each set of thermal network parameters is associated with the corresponding packaging structure version.

[0062] The electromagnetic interference (EMI) transmission matrix has a reference temperature state, which is selected when the electromagnetic model is calibrated. The initial EMI transmission coefficient between modules corresponds to the reference temperature state, and the temperature difference between the current module's operating temperature and the reference temperature is incorporated into the process of forming dielectric and conductivity correction values.

[0063] When using a lookup table to generate dielectric property correction values, the input to the lookup table is temperature, and the output is the correction of inter-module transfer relationships relative to a reference temperature state. Interpolation can be performed when the current temperature is between two calibration temperature points. When using a multi-temperature-point electromagnetic calibration method, the package geometry, source module excitation, and target module observation position are kept consistent. The material and conductivity parameters at the corresponding temperature states are changed, and temperature correction data is established based on the inter-module coupling response at different operating temperature points.

[0064] When using a lookup table to generate the conductivity correction, the input to the lookup table is the equivalent temperature corresponding to the propagation path between modules, and the output is the correction of the inter-module transmission relationship relative to the reference temperature state. The equivalent temperature of the propagation path can be determined based on the operating temperature of the source module, the operating temperature of the target module, and the corresponding hot node temperature. When using a distributed thermal network, the equivalent temperature of the propagation path can be determined based on the temperatures of multiple hot nodes.

[0065] The model solving module stores the current iteration cycle index, the module power consumption estimate from the previous iteration cycle, the power consumption estimate from the most recently stable module, the current power consumption change gradient, the current iteration update weights, and the consecutive convergence and divergence counts. Different state variables are used to control power consumption iteration, convergence termination, and divergence backoff.

[0066] The convergence threshold is determined based on the required accuracy of power consumption analysis and the power consumption magnitude of the module. When normalized power consumption change is used as the convergence criterion, the convergence threshold does not depend on a specific power unit. The minimum iterative update weight is used to limit the proportion of candidate power consumption in a single cycle entering the next state under strong feedback conditions, and the maximum iterative update weight is used to limit the maximum update proportion when approaching a stable state. The boundary of the iterative update weight can be determined based on the numerical stable state under different high-power operating conditions.

[0067] The criteria for judging power consumption anomalies can be formed based on the module's design power consumption budget or historical stable operating power consumption range. Different operating modes can each have a corresponding reference power consumption range set, thus associating power consumption anomaly judgments with the current module's operating mode. When a high-power module has no controllable resources, the high-power analysis results continue to be used for thermal management or design adjustments; when controllable resources are available, the power consumption analysis and strategy generation module generates a low-power strategy based on the module's functional state and control resource mapping.

[0068] Clock gating operates on digital functional areas with independent clock control and meeting idle conditions; clock recovery conditions are associated with the corresponding function trigger state. Power domain management forms a set of legal power states based on the power supply boundaries of the target system. When there are modules that cannot be powered off or whose power supply state cannot be changed, the corresponding power domain will not enter a control state that does not meet the functional conditions. Operating mode switching is performed between the legal operating modes already defined in the target system, with mode switching trigger conditions formed based on abnormal power consumption states and system functional load.

[0069] After the strategy candidate is simulated, if a certain performance constraint is not met, the corresponding over-limit index and the current strategy control object are recorded. The control state is adjusted and the simulation verification is re-executed. The low-power strategy that passes the simulation verification enters the test verification. Before the test data enters the model calibration, it is time-aligned according to the working state, and the correspondence between temperature measurement points and hot nodes is established.

[0070] Model parameter calibration adjusts the electromagnetic interference transfer coefficient and thermal network parameters between modules within the physical boundaries of the parameters. When the error minimization result exceeds the allowable parameter range, the corresponding parameters are kept within the limits defined by material, structure, or calibration data. After parameter adjustment, the same operating condition analysis is re-executed to gradually reduce the residual between the model output and the test power consumption. Once the model residual meets the calibration conditions, the updated multidimensional power coupling model is saved.

[0071] To facilitate a better understanding of the present invention by those skilled in the art, the present invention will be further illustrated below with reference to specific implementation examples.

[0072] Example 1: This example uses a digital module, an analog module, and an RF module to illustrate the core processing procedure. The digital module's operating frequency is set to 1GHz for example, and its switching activity state is set to 0.5 for example; the analog module's power supply voltage is set to 3.3V for example, and its bias current is set to 10mA for example; the RF module's transmit power is set to 20mW for example, and its operating frequency band is set to 2.4GHz for example.

[0073] The initial power consumption of the digital module is, for example, 500mW, the initial power consumption of the analog module is, for example, 100mW, and the initial power consumption of the RF module is, for example, 300mW. Based on the package space information and electromagnetic calibration data, the initial electromagnetic interference (EMI) transmission coefficient between the digital module and the analog module is, for example, 0.020; the initial EMI transmission coefficient between the digital module and the RF module is, for example, 0.015; and the initial EMI transmission coefficients between the analog module and the digital module, and between the RF module and the digital module, are both, for example, 0.005.

[0074] Based on the current power consumption of the three modules and the initial electromagnetic interference transmission coefficients between the corresponding modules, the electromagnetic coupling power consumption correction caused by the influence of the analog and RF modules on the digital module is 2mW, the electromagnetic coupling power consumption correction caused by the influence of the digital module on the analog module is 10mW, and the electromagnetic coupling power consumption correction caused by the influence of the digital module on the RF module is 7.5mW.

[0075] The power consumption estimates of the three modules are simultaneously fed into the thermal conduction network. For example, the thermal conduction network outputs the operating temperature of the digital module as 318K, the analog module as 305K, and the RF module as 315K. Based on the operating temperatures of the three modules and the calibration data of the corresponding transmission paths between the modules, dielectric and conductivity correction values ​​are generated, and the electromagnetic interference transmission matrix is ​​updated accordingly.

[0076] The dynamic electromagnetic interference transfer matrix and the operating temperatures of the three modules re-enter the multidimensional power consumption coupling model.

[0077] For example, in the first iteration cycle, the candidate power consumption of the digital module is 515mW, the candidate power consumption of the analog module is 114mW, and the candidate power consumption of the RF module is 314mW. After normalization using the corresponding initial power consumption as a reference power consumption, the normalized power consumption change of the digital module is 0.030, the normalized power consumption change of the analog module is 0.140, and the normalized power consumption change of the RF module is approximately 0.0467. When using the maximum normalization difference, the power consumption change gradient in this iteration cycle is 0.140.

[0078] For example, setting the minimum iteration update weight to 0.10, the maximum iteration update weight to 0.80, and the adjustment coefficient to 5, the current iteration update weight is approximately 0.4476 according to the iteration update weight formula. After weighting the candidate power consumption estimate with the power consumption estimate of the previous iteration cycle according to the iteration update weight, the updated power consumption of the digital module is approximately 506.71mW, the updated power consumption of the analog module is approximately 106.27mW, and the updated power consumption of the RF module is approximately 306.27mW.

[0079] The updated power consumption of the three modules is re-entered into the heat conduction network, followed by sequential execution of module operating temperature formation, dielectric property correction amount formation, conductivity property correction amount formation, dynamic electromagnetic interference transfer matrix update, candidate power consumption formation, and iterative update. The convergence threshold is set to, for example, 0.001, and it is required that the normalized power consumption changes of the three heterogeneous functional modules are all lower than the convergence threshold within two consecutive judgment periods.

[0080] After continuous iterations, for example, the power consumption of the digital module was found to be 521.8mW, the analog module power consumption was 116.7mW, and the RF module power consumption was 317.1mW, with a total system power consumption of 955.6mW. The power distribution data continued to enter the high-power region and power-abnormal module identification process. When the digital module was identified as the main high-power object and there were controllable idle digital sub-functional units, corresponding clock gating strategy candidates were formed, and simulation verification and testing were carried out.

[0081] Example 2: This example uses a digital signal processing module, an analog power management module, and a millimeter-wave RF transceiver module to form a high-power-density microsystem-in-package. The digital signal processing module has high-speed switching activity, the millimeter-wave RF transceiver module has high-frequency transmission and reception operation, and the analog power management module provides corresponding power to the digital signal processing module and the millimeter-wave RF transceiver module.

[0082] For example, under a high-temperature operating condition, the heat conduction network output source module operates at 345K and the target module operates at 358K, the reference temperature is set to 298K, the initial electromagnetic interference transmission coefficient between the source module and the target module is set to 0.025, the equivalent dielectric characteristic temperature correction coefficient is set to 0.0008K⁻¹, and the equivalent conductivity characteristic temperature correction coefficient is set to 0.0005K⁻¹.

[0083] According to the formula for the dynamic electromagnetic interference transmission coefficient, the temperature difference between the source module and the reference temperature is 47K, and the temperature difference between the target module and the reference temperature is 60K. The corresponding dielectric property correction term is 0.0376, and the corresponding conductivity property correction term is 0.0300. The transmission coefficient correction factor is 1.0676, and the dynamic electromagnetic interference transmission coefficient is approximately 0.02669.

[0084] After the dynamic electromagnetic interference (EMI) transfer coefficient is written into the EMI transfer matrix, the multidimensional power consumption coupling model regenerates candidate power consumption estimates. These candidate power consumption estimates, along with the power consumption estimate from the previous iteration, form a normalized power consumption change gradient. When the power consumption change gradient is large, the iterative update weights approach the minimum boundary; as the power consumption change gradient decreases, the iterative update weights gradually increase.

[0085] For example, the minimum iteration update weight is set to 0.10, the maximum iteration update weight is set to 0.80, and the adjustment coefficient is set to 5. When the dimensionless power consumption gradient is 0.30, the iteration update weight is approximately 0.2562; when the dimensionless power consumption gradient decreases to 0.02, the iteration update weight is approximately 0.7334. This reduces the magnitude of each update during periods of significant power consumption change and increases the proportion of candidate power consumption values ​​entering the next state when the model approaches a stable state.

[0086] Under a set of iterative parameter configurations, a fixed update method requires, for example, 20 iterations to reach the set convergence condition, while a method using gradient adjustment of power consumption to adjust iterative update weights requires, for example, 12 iterations to reach the same convergence condition. After convergence, the power consumption distribution enters the high-power region identification stage. When the millimeter-wave RF transceiver module and the adjacent digital signal processing module form a high-power region, candidates for RF operating state adjustment, digital module clock gating, and power domain management can be generated, respectively.

[0087] The strategy candidate input is packaged into a performance simulation environment. The simulation environment outputs the total system power consumption, module operating temperature, and performance parameters corresponding to the millimeter-wave radar's operating state. Strategies that meet the power consumption, thermal, and performance constraints are then tested and verified. The test power consumption is aligned with the model's calculated power consumption to form model residuals. These residuals are used to correct the electromagnetic interference transmission coefficients and thermal network parameters between corresponding modules. The corrected model is then re-analyzed under the same operating state, and the updated model parameters are saved for subsequent analysis cycles.

[0088] Example 3: This example uses a digital control module, an analog power management module, and an RF transceiver module to form a mobile terminal RF front-end system-in-package. The parameter acquisition interface obtains the activity status of the digital control module, the power supply and bias status of the analog power management module, and the transmit power, receive status, and operating frequency band of the RF transceiver module.

[0089] Different communication operating states correspond to different analysis periods. In data transmission mode, the digital control module and RF transceiver module exhibit high activity levels; in standby mode, some digital control functions enter a low-activity state; in weak signal mode, the RF transceiver module enters a transmit or receive operating state corresponding to the link conditions. Each operating state is input into a multidimensional power coupling model with the same structure, resulting in a corresponding power distribution.

[0090] The power consumption distribution in standby mode can be used to identify digital sub-functional units with clock gating conditions. The power consumption distribution in data transmission mode can be used to identify high-power regions formed by RF transceiver modules and analog power management modules. The power consumption distribution in weak signal mode can be used to analyze the relationship between RF high-power states and local module operating temperatures.

[0091] For example, in a data transmission state, the system power consumption before optimization could be 1000mW. After formulating a low-power optimization strategy based on the power distribution and applying simulation constraints, the optimized system power consumption can be reduced to 880mW. Further testing and verification were conducted to obtain module power consumption, module operating temperature, and communication performance parameters. The test data was used simultaneously for evaluating the optimization effect and correcting the parameters of the multi-dimensional power coupling model.

[0092] Example 4: This example illustrates the frequency domain dynamic drift of electromagnetic interference propagation between modules. The high-speed interconnect structure within a microsystem-in-package (SiP) exhibits distributed parameter characteristics in the microwave or millimeter-wave frequency bands. The package can employ silicon interposer boards, high-density fan-out packages, or other package forms containing multilayer redistribution structures. Taking a propagation path containing a through-silicon via (TSV) array and multilayer metal redistribution as an example, the TSVs, insulating dielectric, redistribution layers, and adjacent ground structures collectively form parasitic capacitance, parasitic inductance, and high-frequency losses.

[0093] The thermal conductivity network outputs the current operating temperatures of the digital module, analog module, and RF module. For the cross-layer coupling path between the digital and RF modules, the multi-dimensional power coupling model reads the corresponding three-dimensional structural data. The three-dimensional structural data may include the diameter of the through-silicon via (TSV), the TSV center-to-center spacing, the insulation layer thickness, the redistribution layer trace width, the redistribution layer trace spacing, and the location of the reference ground structure.

[0094] The model solver module reads the material parameters corresponding to the reference temperature of the propagation path, including the relative permittivity, conductivity, and equivalent parasitic parameters at the reference temperature. The module operating temperature and the thermal node temperatures of the propagation path provided by the thermal network are used to determine the equivalent temperature of the current propagation path.

[0095] The rate of change of equivalent capacitance at the current temperature is obtained based on the dielectric property correction and the equivalent parasitic capacitance at the reference temperature. The rate of change of equivalent capacitance is the relative change of the current equivalent parasitic capacitance with respect to the equivalent parasitic capacitance at the reference temperature.

[0096] Within a temperature analysis range, when the equivalent inductance of the propagation path uses the same reference calibration value, the resonant frequency at the current temperature is calculated according to the inverse relationship between the resonant frequency and the square root of the equivalent capacitance. Subtracting the initial resonant frequency from the current resonant frequency yields the signed resonant frequency offset. When an increase in equivalent capacitance causes a shift in the resonant position to a lower frequency, the signed resonant frequency offset is negative.

[0097] The interference source spectral description is formed from the operating state parameters of the digital module. For example, the time-domain current change sequence corresponding to the logic transition state of the digital module is converted into discrete frequency points and corresponding power spectral densities to obtain the interference source spectral description. The interference source spectral description at least saves the frequency coordinates and the corresponding power spectral density.

[0098] The power spectral density of the interference source is obtained by shifting the frequency coordinates of the interference source spectrum according to the signed resonant frequency offset. When the resonant frequency shifts to lower frequencies, the corresponding spectral energy position shifts to lower frequencies according to the signed frequency offset.

[0099] The disturbance sensitivity description of the RF module is formed by the disturbance response corresponding to different frequency positions within the target frequency band. The disturbance response can be established based on the input interference power of the RF module and the corresponding performance changes. After benchmark normalization, a dimensionless disturbance sensitivity weighting function is formed.

[0100] The target frequency band is divided into multiple frequency intervals according to frequency resolution. The model solver reads the power spectral density of the offset interference source and the corresponding disturbance sensitivity weights for each frequency interval, and calculates the spectral overlap integral. Numerical integration can use a trapezoidal integration method; a smaller frequency step size can be used for frequency ranges with faster spectral changes, while a larger frequency step size can be used for frequency ranges with slower spectral changes.

[0101] The reference spectrum overlap integral is calculated using the same target frequency band under the reference temperature condition. The dimensionless frequency domain coupling coefficient is obtained by dividing the current spectrum overlap integral by the reference spectrum overlap integral.

[0102] The model solver module synchronously generates a conductivity correction value based on the current equivalent temperature of the propagation path, and converts the conductivity correction value into a dimensionless conductivity correction factor. The conductivity correction factor for the reference temperature state is 1.

[0103] The frequency domain coupling coefficient is multiplied by the conductance correction factor to form the inter-module transfer coefficient update multiplier. The inter-module transfer coefficient update multiplier is multiplied by the initial inter-module electromagnetic interference transfer coefficient at the reference temperature state to obtain the dynamic electromagnetic interference transfer coefficient at the current temperature state.

[0104] For example, the digital module's operating frequency is set to 2.4 GHz, the reference temperature is set to 298 K, the reference relative permittivity corresponding to the propagation path is set to 3.9, the initial resonant frequency of the electromagnetic interference propagation path is set to 4.8 GHz, and the target frequency band is set to 4.0 GHz to 6.0 GHz. The spectral overlap integral at the reference temperature is set to 2.15 mW, and the initial electromagnetic interference transmission coefficient between modules is set to 0.0150.

[0105] Table 1 shows the frequency domain coupling parameters and dynamic electromagnetic interference transmission coefficients at different module operating temperatures.

[0106] Table 1 Frequency domain coupling parameters and dynamic electromagnetic interference transmission coefficients under different temperature conditions;

[0107] 298 0.000 0.00 2.15 1.0000 1.0000 0.0150 315 0.042 -97.73 3.84 1.7860 0.8772 0.0235 335 0.105 -233.75 8.12 3.7767 0.8438 0.0478 355 0.188 -396.14 14.65 6.8140 0.7944 0.0812 375 0.294 -580.37 25.30 11.7674 0.7620 0.1345

[0108] Taking the 315K state as an example, after normalizing the current spectral overlap integral of 3.84mW and the reference spectral overlap integral of 2.15mW, the frequency domain coupling coefficient is approximately 1.7860. The conductance correction factor is, for example, 0.8772, and the product of the two dimensionless factors is approximately 1.5667. Multiplying the initial inter-module electromagnetic interference transmission coefficient of 0.0150 by 1.5667 yields a dynamic electromagnetic interference transmission coefficient of approximately 0.0235.

[0109] Taking the 355K state as an example, after normalizing the current spectral overlap integral of 14.65mW and the reference spectral overlap integral of 2.15mW, the frequency domain coupling coefficient is approximately 6.8140. The conductance correction factor is, for example, 0.7944, corresponding to an inter-module transfer coefficient update multiplier of approximately 5.4133. Multiplying the initial inter-module electromagnetic interference transfer coefficient of 0.0150 by 5.4133 yields a dynamic electromagnetic interference transfer coefficient of approximately 0.0812.

[0110] The above processing ensures that the dynamic electromagnetic interference transfer coefficient simultaneously reflects changes in the frequency domain position of the propagation path and changes in high-frequency conductivity loss. When temperature changes cause the harmonic energy of the digital module to shift towards the sensitive frequency band of the RF module, the spectral overlap integral increases, thereby improving the frequency domain coupling coefficient; changes in conductivity loss are further corrected for the frequency domain coupling result by a conductivity correction factor.

[0111] After obtaining the updated dynamic electromagnetic interference transfer matrix, the dynamic electromagnetic interference transfer matrix continues to enter the candidate power consumption calculation and subsequent iteration process without changing the power consumption-temperature closed-loop relationship described in Examples 1 and 2.

[0112] Example 5: This example illustrates the time-domain active avoidance process in the low-power optimization strategy based on the frequency-domain dynamic coupling analysis in Example 4.

[0113] After the dynamic electromagnetic interference (EMI) transfer matrix reaches the power consumption iteration convergence condition, the power consumption analysis and strategy generation module reads the EMI transfer coefficients from the digital module to the RF module in the matrix. The coupling isolation threshold is set according to the allowable electromagnetic coupling budget between the digital module and the RF module. When the EMI transfer coefficient is not higher than the coupling isolation threshold, the existing strategy is maintained; when the EMI transfer coefficient is higher than the coupling isolation threshold, time-domain coupling analysis is initiated.

[0114] The transient switching time window of the digital module is formed based on the transient current sequence corresponding to the high-concurrency logic activities within the digital module. The parameter acquisition interface reads the clock trigger information and logic activity information of the digital module, identifies the continuous time interval within an analysis cycle where the transient power exceeds the set activity threshold, and determines the corresponding continuous time interval as the transient switching time window.

[0115] The radio frequency (RF) activity time window of the RF module is formed based on the RF control state. When the RF module is in the transmit state, the RF transmit time window is determined according to the on and off times of the RF transmit control signal; when the RF module is in the sensitive receive state, the RF receive sensitive time window is determined according to the sensitive operating time of the RF receive link.

[0116] The transient switching time window of digital modules and the radio frequency (RF) activity time window may be formed by different control clocks or different protocol scheduling. In order to determine whether the transient interference of digital modules actually overlaps with the RF activity when it arrives at the location of the RF module, it is necessary to compensate for the propagation delay caused by the electromagnetic interference propagation path between modules.

[0117] The model solver reads the physical length of the propagation path from the digital module to the RF module and obtains the current equivalent relative permittivity of the propagation path based on the current dielectric property correction. The propagation delay time is calculated according to the relationship that the propagation delay is proportional to the propagation path length, inversely proportional to the speed of light in vacuum, and proportional to the square root of the equivalent relative permittivity.

[0118] For example, the physical length of the electromagnetic interference propagation path is 0.015m, the current equivalent relative permittivity is 4.2, and the propagation delay time is about 102.55ps according to the above propagation relationship.

[0119] By increasing the start and end times of the digital module transient switching time window by 102.55 ps, the arrival time window of the digital transient interference at the RF module location is obtained. The arrival time window and the RF activity time window are mapped to the same physical time base, and the intersection of the two windows is calculated.

[0120] When the intersection is not empty, the time ranges corresponding to the intersection form an overlapping interval. Within the overlapping interval, the transient power sequence of the digital module is read, and the transient coupled power sequence at the location of the RF module is formed based on the dynamic electromagnetic interference transfer coefficient from the digital module to the RF module. The maximum power value in the transient coupled power sequence is taken as the transient coupled peak power.

[0121] The power safety threshold is set according to the allowable interference power corresponding to the operating state of the RF module. When the transient coupling peak power exceeds the power safety threshold, the minimum time offset is calculated.

[0122] For example, the timing adjustment range corresponding to the system task is 10μs, and the clock delay adjustment module supports a time adjustment step size of 1ns. The model searches for candidate time offsets in both the advance and delay directions, starting from the smallest non-zero time adjustment step size. For each candidate time offset formed, the transient switching time window of the digital module is re-shifted, and the overlap interval between it and the RF activity time window is calculated.

[0123] If the overlapping interval corresponding to the candidate time offset is still not empty, the time offset is increased. If the overlapping interval is empty and the candidate time offset is within the allowed timing adjustment range, the corresponding candidate time offset is added to the available time offset set. The time offset with the smallest absolute value is selected from the available time offset set to obtain the minimum time offset.

[0124] The power consumption analysis and strategy generation module generates timing misalignment control instructions that include the minimum time offset, the target clock object, and the offset direction. These instructions are then sent to the clock delay adjustment module of the digital module via the clock control interface.

[0125] The clock delay adjustment module can consist of a digitally programmable delay chain, a delay phase-locked loop, or a clock management unit with clock phase adjustment functionality. Taking a digitally programmable delay chain as an example, the clock delay adjustment module selects the corresponding delay level according to the timing misalignment control command, thereby causing the clock trigger edge of the target digital arithmetic unit to generate a corresponding time offset.

[0126] In one working state, the propagation delay time and the time domain parameters before and after timing misalignment control are shown in Table 2.

[0127] Table 2 Transient coupling states before and after timing misalignment control;

[0128] Before control 102.55 0 45.0 328.4 12.5 First adjustment state 102.55 25 20.0 145.2 18.2 Second adjustment state 102.55 40 5.0 42.8 24.6 steady state 102.55 45 0.0 3.1 31.4

[0129] Before control, the transient switching time window of the digital module after propagation delay compensation has a 45.0ns overlap with the sensitive time window of the radio frequency receiver, and the transient coupling peak power is 328.4mW.

[0130] After a 25ns delay in the clock trigger edge, the overlap interval shortens to 20.0ns, and the transient coupling peak power decreases to 145.2mW. After the cumulative clock offset increases to 40ns, the overlap interval shortens to 5.0ns, and the transient coupling peak power decreases to 42.8mW.

[0131] After the cumulative clock offset reaches 45ns, the transient switching time window of the digital module and the sensitive time window of the RF receiver no longer overlap. In this state, the peak transient coupling power is 3.1mW, and the RF receiver signal-to-noise ratio is 31.4dB.

[0132] After timing misalignment control is completed, a control rule is established between the 45ns time offset and the corresponding digital module's operating state. When subsequently entering the same digital module's operating state and RF activity state, the clock offset can be set according to the control rule first, and verification can continue using the dynamic electromagnetic interference transfer coefficient and a unified physical time axis.

[0133] If a change in the digital module's task causes a 45ns time offset to exceed the current timing adjustment range, the original timing misalignment control instruction is not executed directly; instead, the minimum time offset is recalculated. This ensures that the timing misalignment strategy is always constrained by the digital module's functional timing.

[0134] Comparative Example 1: This comparative example uses the same digital module, analog module, and RF module as Example 1, with the same operating state parameters and the same initial module power consumption. Only the multi-dimensional power consumption coupling process consisting of cross-module electromagnetic coupling and thermal feedback is eliminated, and the independent power consumption of the modules is directly used for summation.

[0135] The power consumption of the digital module is, for example, 500mW, the power consumption of the analog module is, for example, 100mW, and the power consumption of the RF module is, for example, 300mW, which directly sums up to a system power consumption of 900mW. In Example 1, using the same initial input, after iterating through dynamic electromagnetic interference propagation and thermal feedback, the total system power consumption is, for example, 955.6mW. This allows for a comparison of the power distribution differences between ignoring cross-module coupling processing and using multi-dimensional power coupling processing.

[0136] Comparative Example 2: This comparative example uses the same heterogeneous functional modules, the same operating state parameters, the same initial module power consumption, the same heat conduction network, the same initial electromagnetic interference transfer matrix, and the same power consumption iteration process as Example 1. The only difference is that the module operating temperature no longer generates dielectric property correction and conductivity property correction, so that the electromagnetic interference transfer matrix maintains the initial inter-module transfer coefficient corresponding to the reference temperature state throughout the entire iteration cycle.

[0137] This comparative example still calculates the module operating temperature based on the module power consumption and performs power consumption iteration according to the same convergence conditions. For example, when using a fixed electromagnetic interference transfer matrix, the converged system power consumption can be 940mW; when using the module operating temperature to drive the dynamic electromagnetic interference transfer matrix update, the converged system power consumption can be 955.6mW. Whether changing the module operating temperature between the two processes triggers the electromagnetic interference transfer relationship update is used to compare the impact of temperature-driven dynamic electromagnetic interference transfer relationship on the power consumption solution results.

[0138] Taking a microsystem system-in-package consisting of a digital module D, an analog module A, and an RF module R as an example, the initial power consumption of the digital module D can be 500mW, the initial power consumption of the analog module A can be 100mW, and the initial power consumption of the RF module R can be 300mW. The initial electromagnetic interference transmission coefficient between the digital module D and the analog module A can be 0.020, the initial electromagnetic interference transmission coefficient between the digital module D and the RF module R can be 0.015, and the initial electromagnetic interference transmission coefficients between the analog module A and the digital module D, and between the RF module R and the digital module D, can be 0.005 respectively.

[0139] Based on the initial electromagnetic interference transmission coefficient between modules, the current power consumption of the source module is weighted, and the electromagnetic coupling power consumption correction caused by the influence of analog module A and RF module R on digital module D is 2mW, the electromagnetic coupling power consumption correction caused by the influence of digital module D on analog module A is 10mW, and the electromagnetic coupling power consumption correction caused by the influence of digital module D on RF module R is 7.5mW.

[0140] After obtaining the power consumption of the three modules through the heat conduction network, for example, the operating temperature of the output digital module D is 318K, the operating temperature of the analog module A is 305K, and the operating temperature of the RF module R is 315K.

[0141] For a high-temperature operating path, for example, setting the reference temperature to 298K, the source module operating temperature to 345K, the target module operating temperature to 358K, the initial electromagnetic interference (EMI) transmission coefficient between modules to 0.025, the equivalent dielectric property temperature correction coefficient to 0.0008K⁻¹, and the equivalent conductivity property temperature correction coefficient to 0.0005K⁻¹, based on the dynamic EMI transmission coefficient relationship, the dynamic transmission coefficient correction factor is obtained to be 1.0676, and the dynamic EMI transmission coefficient is 0.02669.

[0142] The dynamic electromagnetic interference transfer matrix and module operating temperature are further input into the multidimensional power consumption coupling model, resulting in candidate power consumption of 515mW for digital module D, 114mW for analog module A, and 314mW for RF module R. Normalizing these initial power consumptions yields a normalized power consumption change of 0.030 for digital module D, 0.140 for analog module A, and approximately 0.0467 for RF module R. The power consumption change gradient is 0.140 when using the maximum normalization difference.

[0143] For example, setting the minimum iteration update weight to 0.10, the maximum iteration update weight to 0.80, and the adjustment coefficient to 5, the current iteration update weight is approximately 0.4476 according to the iteration update weight relationship. After updating according to the iteration update weight, the power consumption of digital module D is approximately 506.71mW, the power consumption of analog module A is approximately 106.27mW, and the power consumption of RF module R is approximately 306.27mW. The updated power consumption of the three modules re-enters the heat conduction network and continues to form the module operating temperature and dynamic electromagnetic interference transmission matrix for the next iteration cycle.

[0144] The above process is repeated until the normalized power consumption changes of the three heterogeneous functional modules are all below the set convergence threshold and continuously meet the predetermined judgment period. For example, the final power consumption of digital module D can be 521.8mW, analog module A can be 116.7mW, RF module R can be 317.1mW, and the total system power consumption is 955.6mW.

[0145] After power distribution analysis identifies digital module D as the primary high-power component, idle digital sub-functional units within digital module D can form candidate clock gating strategies. For example, when clock gating reduces the power consumption of digital module D by 70mW, the optimized system power consumption is 885.6mW. The low-power optimization strategy continues to be verified under module operating temperature and timing performance constraints, and after simulation verification, it enters the testing verification phase.

[0146] For operating conditions where there is significant high-frequency coupling between digital module D and RF module R, further frequency domain dynamic drift processing can be invoked. The equivalent capacitance change rate is obtained based on the dielectric characteristic correction of the propagation path from digital module D to RF module R. A signed resonant frequency offset is formed based on the initial resonant frequency at the corresponding reference temperature. The interference source spectrum of digital module D is shifted, and the dynamic electromagnetic interference transmission coefficient from digital module D to RF module R is updated using the spectral overlap integral.

[0147] If the updated dynamic electromagnetic interference transmission coefficient from digital module D to RF module R exceeds the coupling isolation threshold, then the transient switching time window of digital module D and the active time window of RF module R are obtained. Based on the propagation path length and the current equivalent relative permittivity, a propagation delay time is generated. After compensating for the propagation delay, the transient switching time window of digital module D is overlapped with the active time window of RF module R on the time axis.

[0148] When the peak power of transient coupling corresponding to the overlapping interval exceeds the power safety threshold, the minimum time offset is searched within the timing adjustment range allowed by digital module D, and a timing misalignment control command is generated. The clock delay adjustment module changes the clock trigger edge of digital module D according to the timing misalignment control command until the transient switching time window of digital module D and the active time window of RF module R are separated on a unified physical time axis.

[0149] The testing and verification process obtains module power consumption, module operating temperature, and corresponding performance data. The test power consumption is aligned with the output of the multidimensional power consumption coupling model according to the operating state and module identification to form model residuals. The model residuals are used to jointly correct the electromagnetic interference transmission coefficient between modules and the thermal network parameters. The corrected thermal network outputs the module operating temperature again. The new module operating temperature continues to form dielectric property corrections and conductivity corrections, allowing the corrected multidimensional power consumption coupling model to enter the next analysis cycle.

[0150] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for power consumption analysis and low-power optimization verification of microsystems (SiP), characterized in that, Includes the following steps: Obtain the operating state parameters of heterogeneous functional modules within the system-level package of a microsystem, establish electromagnetic interference transmission relationships and heat conduction networks based on the operating state parameters, and form a multidimensional power consumption coupling model. The module power consumption estimate is input into the multidimensional power consumption coupling model to obtain the module operating temperature and electromagnetic interference response. Based on the module operating temperature, dielectric property correction and conductivity correction are generated, and the electromagnetic interference transmission relationship is updated accordingly. The updated electromagnetic interference transmission relationship and module operating temperature are fed back into the multidimensional power consumption coupling model to obtain candidate power consumption estimates. The iteration update weights are determined based on the gradient of the change between the candidate power consumption estimates and the power consumption estimates of the previous iteration cycle, and the module power consumption estimates are updated. This process is repeated until the convergence condition is met to obtain power consumption distribution data. Based on the power distribution data, high-power regions and modules with abnormal power consumption are identified. Based on the identification results, low-power optimization strategies including clock gating, power domain management or working mode switching are formed and simulation verification is completed. The low-power optimization strategy verified by simulation is tested and verified. The power consumption data obtained from the test is used to correct the multidimensional power coupling model and output the evaluation results of the optimization effect.

2. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, The heterogeneous functional modules include digital modules, analog modules, and radio frequency (RF) modules. The operating status parameters of the digital modules include operating frequency, logic transition state, and switching activity state. The operating status parameters of the analog modules include supply voltage, bias current, and signal gain. The operating status parameters of the RF modules include transmit power, receive state, and operating frequency band. The operating status parameters are used to form a module state vector according to a unified sampling time marker, and the module state vector is input into a multidimensional power consumption coupling model.

3. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, The establishment of electromagnetic interference transmission relationships includes: forming an interference source spectrum description based on the switching states of digital modules; forming a disturbance sensitivity description by combining the disturbance responses of analog and radio frequency modules in the corresponding frequency bands; forming a propagation path description by combining the physical location relationships, interconnection relationships, and packaging material parameters between heterogeneous functional modules; generating initial electromagnetic interference transmission coefficients between modules according to the interference source spectrum description, disturbance sensitivity description, and propagation path description, and forming an electromagnetic interference transmission matrix from the initial electromagnetic interference transmission coefficients between modules.

4. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, The establishment of the heat conduction network includes: establishing thermal nodes and heat transfer connections between thermal nodes based on heterogeneous functional modules, packaging substrates and packaging heat dissipation boundaries; configuring thermal network parameters for thermal nodes and heat transfer connections based on packaging structure data, material thermal parameters or calibration data; mapping the estimated power consumption of the modules to the heat input of the corresponding thermal nodes, and obtaining the module operating temperature of each heterogeneous functional module based on the heat transfer connections between thermal nodes.

5. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, The dielectric and conductivity corrections are generated based on the module's operating temperature. This includes: for the inter-module transmission paths in the electromagnetic interference transmission matrix, a dielectric correction is generated based on the module's operating temperature and the material parameter data, packaging structure electromagnetic simulation calibration results, or test calibration results corresponding to the inter-module transmission paths; a conductivity correction is generated based on the module's operating temperature and the interconnect conductivity parameter data or test calibration results corresponding to the inter-module transmission paths; and the dielectric and conductivity corrections are applied together to the corresponding initial inter-module electromagnetic interference transmission coefficients to generate a dynamic electromagnetic interference transmission matrix.

6. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, Determining the iterative update weights and updating the module power consumption estimates includes: normalizing the difference between the candidate power consumption estimates and the power consumption estimates of the previous iteration cycle to form a power consumption change gradient reflecting the current iteration state; decreasing the iterative update weights when the power consumption change gradient increases and increasing the iterative update weights when the power consumption change gradient decreases; using the iterative update weights to perform weighted updates on the candidate power consumption estimates and the power consumption estimates of the previous iteration cycle; outputting power consumption distribution data when the normalized power consumption changes of all heterogeneous functional modules meet the convergence threshold; decreasing the iterative update weights when a continuous increase in power consumption change is detected, and re-iterating from the most recently stable module power consumption estimate.

7. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, Identifying high-power regions and modules with abnormal power consumption includes: determining the relative power consumption levels between heterogeneous functional modules based on power consumption distribution data; forming a set of high-power modules by combining the spatial location and relative power consumption levels of the heterogeneous functional modules; forming high-power regions based on the spatial adjacency relationships of the high-power module set; identifying at least one power consumption abnormality state among power consumption surge, sustained high power consumption, or periodic abnormality using power consumption distribution data from continuous analysis periods, and marking the corresponding heterogeneous functional modules as modules with abnormal power consumption.

8. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 7, characterized in that, Before formulating a low-power optimization strategy, the power distribution data of continuous analysis cycles and the current state of the multidimensional power coupling model are combined as prediction inputs. The power changes of subsequent analysis cycles are calculated according to the thermal feedback relationship and dynamic electromagnetic interference transmission relationship of the multidimensional power coupling model. When the calculation results meet the power anomaly judgment conditions, the power warning information of the corresponding heterogeneous functional modules is output, and the power warning information is used as the strategy object input of the low-power optimization strategy.

9. The method for power consumption analysis and low-power optimization verification of a microsystem SiP according to claim 1, characterized in that, The multidimensional power coupling model is corrected using power consumption data obtained from testing, including: aligning the test power consumption data under the same operating condition with the power distribution data output by the multidimensional power coupling model to form model residuals; performing joint parameter fitting on the inter-module transfer coefficients in the electromagnetic interference transfer matrix and the thermal network parameters in the heat conduction network based on the model residuals; updating the multidimensional power coupling model with the fitted inter-module transfer coefficients and thermal network parameters, and using the updated multidimensional power coupling model in subsequent analysis cycles.

10. A microsystem SiP power consumption analysis and low-power optimization verification system, used to execute the microsystem SiP power consumption analysis and low-power optimization verification method according to any one of claims 1-9, characterized in that, It includes a parameter acquisition interface, a model solving module, a power consumption analysis and strategy generation module, a verification and calibration module, a processor, and memory; The parameter acquisition interface is connected to the working status monitoring node of the microsystem system-level package to obtain the working status parameters of the digital module, analog module and radio frequency module, and output the working status parameters to the model solving module; The model solving module is implemented by the processor executing model solving instructions in the memory. It is used to establish the electromagnetic interference transmission relationship and heat conduction network based on the working state parameters, obtain the module operating temperature based on the module power consumption estimate, generate dielectric property correction and conductivity correction based on the module operating temperature and update the electromagnetic interference transmission relationship, obtain candidate power consumption estimates based on the updated electromagnetic interference transmission relationship, determine the iterative update weights based on the change state of the candidate power consumption estimates, and output power distribution data that meets the convergence condition in a loop. The power consumption analysis and strategy generation module receives the power distribution data output by the model solving module, combines the spatial location and functional status of the heterogeneous functional modules to form high power consumption regions, power consumption abnormal modules and low power consumption optimization strategies, and sends the low power consumption optimization strategies to the verification and calibration module. The verification and calibration module establishes data connections with the packaging performance simulation environment and the test data acquisition interface, respectively. It receives low-power optimization strategies for simulation verification and test verification, corrects the electromagnetic interference transmission relationship and heat conduction network parameters based on the test power consumption data, and returns the corrected model parameters to the model solving module. When the dynamic electromagnetic interference propagation coefficient meets the timing misalignment control conditions, the power consumption analysis and strategy generation module is also used to generate timing misalignment control instructions based on the transient switching time window of the digital module, the radio frequency activity time window of the radio frequency module, and the electromagnetic interference propagation delay time, and send the timing misalignment control instructions to the clock delay adjustment module of the digital module through the clock control interface.

Images