CN-121979080-A - Intelligent switching method and system based on chip low-power mode

CN121979080ACN 121979080 ACN121979080 ACN 121979080ACN-121979080-A

Abstract

The invention relates to the technical field of integrated circuit low-power consumption management, in particular to an intelligent switching method and system based on a chip low-power consumption mode, comprising the steps of obtaining an original electric signal and an environment signal when the chip runs; the method comprises the steps of stripping kernel steady-state components and peripheral transient components from an electric signal, constructing a composite electric signal map, carrying out feature decomposition on an environment signal, extracting trend components and disturbance components, carrying out fusion alignment on the electric signal map and the environment components in time to generate a chip-environment unified state model, analyzing the model, identifying key state nodes for predicting power consumption mode conversion, and generating a pre-execution control sequence for voltage and frequency adjustment based on node attributes and relations. According to the invention, by constructing the internal and external state coupling model, more accurate prediction and active control of the low-power-consumption mode switching time of the chip are realized.

Inventors

HONG WENLANG
ZHANG QU
WEI JUNHAI

Assignees

深圳市晨兴达电子有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. The intelligent switching method based on the low-power consumption mode of the chip is characterized by comprising the following steps of: Acquiring an original electric signal and an original environment signal when the chip operates; Stripping a steady-state electric signal component generated by the operation of a chip core and a transient electric signal component generated by the activity of a peripheral circuit from the original electric signal, and constructing a composite electric signal map based on the respective dynamic ranges of the steady-state electric signal component and the transient electric signal component; Carrying out signal characteristic decomposition on the original environment signal, and extracting a trend component which slowly changes along with time and a disturbance component which rapidly fluctuates along with time; The composite electric signal map, the trend component and the disturbance component are fused and aligned in the time dimension to generate a unified state model reflecting the coupling relation between the chip running state and the environment state; analyzing the unified state model, and identifying key state nodes in the model, which correspond to the regions to be accessed or the regions to be released from the low power consumption of the chip; Based on the attributes of the critical state nodes and the relationships between the nodes, a set of pre-execution control sequences including voltage adjustment instructions and frequency adjustment instructions is generated.
2. The chip-based low power mode intelligent switching method according to claim 1, further comprising: Issuing the pre-execution control sequence to a power management interface of a chip, and synchronously acquiring an actual power consumption track fed back by the chip in the execution process; comparing the actual power consumption track with an expected power consumption track in the pre-execution control sequence, and calculating an execution error of power consumption adjustment; Fine tuning parameters of subsequent instructions in the pre-execution control sequence according to the execution error to generate a corrected control sequence; And packaging the corrected control sequence, the actual power consumption track and the execution error into a state report, and transmitting the state report to an external collaborative management node through a short-distance wireless communication link.
3. The intelligent switching method based on the chip low power consumption mode according to claim 1, wherein the stripping of the steady-state electrical signal component generated by the chip core operation and the transient electrical signal component generated by the peripheral circuit activity from the original electrical signal, and constructing a composite electrical signal map based on the respective dynamic ranges of the steady-state electrical signal component and the transient electrical signal component, specifically comprises: Intercepting the original electric signal in a segmented way to obtain signal segments with continuous time; carrying out spectrum analysis on each signal segment, and separating a low-frequency energy concentrated band near a fundamental frequency and discrete energy spectral lines distributed in a high-frequency area; reconstructing the low-frequency energy concentrated band into a steady-state electric signal component representing the continuous operation load of the chip core; Superposing time domain signals corresponding to the discrete energy spectral lines and reconstructing the time domain signals into transient electric signal components representing the peripheral circuit sudden access operation; Respectively extracting characteristic parameters of the steady-state electric signal component and the transient electric signal component in a preset observation window; mapping elements in the steady-state electrical signal component and the transient electrical signal component into feature points on a two-dimensional plane; Connecting corresponding steady-state characteristic points and transient characteristic points under the same time stamp to form a plurality of characteristic line segments; And arranging the characteristic line segments according to a time sequence, and marking the proportional relation between steady state and transient state energy to form the composite electric signal map.
4. The intelligent switching method based on the chip low power consumption mode according to claim 1, wherein the signal characteristic decomposition is performed on the original environmental signal, and a trend component which slowly changes with time and a disturbance component which rapidly fluctuates with time are extracted, specifically: Receiving the raw environmental signals from a plurality of physical position sensors, the raw environmental signals including temperature signals and barometric pressure signals; performing time stamp alignment and signal amplitude normalization processing on temperature signals and air pressure signals from the same physical position sensor to obtain a synchronous environment signal pair; trend fitting is carried out on the temperature signals in each synchronous environment signal pair, and a polynomial curve reflecting the slowly rising and falling change of the environment background temperature is extracted to be used as a temperature trend component; Subtracting the corresponding temperature trend component from the temperature signal in each synchronous environment signal pair to obtain a temperature disturbance component mainly caused by local heat source or air flow; Performing moving average filtering on the air pressure signals in each synchronous environment signal pair to obtain an air pressure baseline reflecting macroscopic change of atmospheric pressure as an air pressure trend component; the original air pressure signal in each synchronous environment signal pair is differenced with the corresponding air pressure base line, and an air pressure disturbance component caused by the opening and closing of a fan start-stop or a door window of the equipment is obtained; And respectively carrying out spatial interpolation on the temperature trend component, the temperature disturbance component, the air pressure trend component and the air pressure disturbance component from all the physical position sensors to form a corresponding temperature trend surface, a corresponding temperature disturbance field, a corresponding air pressure trend surface and a corresponding air pressure disturbance field.
5. The intelligent switching method based on the chip low power consumption mode according to claim 4, wherein the fusion alignment of the composite electric signal spectrum, the trend component and the disturbance component in the time dimension is performed to generate a unified state model reflecting the coupling relation between the chip operation state and the environment state, specifically: establishing a unified time reference line, and mapping the time axes of the composite electric signal map, the trend component and the disturbance component onto the unified time reference line; setting sampling points at fixed time intervals on the unified time reference line; Reading steady description vector values and transient description vector values at corresponding moments from the composite electric signal map at each sampling point, reading temperature trend values corresponding to the positions of the chips from the temperature trend surface, reading corresponding temperature disturbance values from the temperature disturbance field, reading air pressure trend values corresponding to the positions of the chips from the air pressure trend surface, and reading corresponding air pressure disturbance values from the air pressure disturbance field; combining the steady state description vector value, the transient description vector value, the temperature trend value, the temperature disturbance value, the air pressure trend value and the air pressure disturbance value read from each sampling point into a multidimensional state feature vector according to a preset sequence; Arranging the multidimensional state feature vectors corresponding to all sampling points according to a time sequence to form a state feature vector sequence evolving along with time; and inputting the state characteristic vector sequence into a network structure, wherein the network structure outputs a dynamic model which can represent the running state and the environment state of the chip at the current moment and can deduce the short-time state change in the future as the unified state model.
6. The method for intelligently switching between low power modes based on a chip according to claim 5, wherein the analyzing the unified state model identifies key state nodes in the model corresponding to the high power area or the low power area to be separated from the chip, specifically comprises: In a state space defined by the unified state model, a low-power-consumption state area and a high-power-consumption state area formed based on historical data clustering are defined; the track of the current state point output by the unified state model in a state space is monitored in real time; When the track shows that the current state point moves from the inside of the low-power-consumption state area to the boundary, calculating the estimated time of the current state point reaching the boundary of the low-power-consumption state area; If the estimated time is smaller than a preset early warning threshold value, marking the current state point as a candidate key state node which is about to be separated from a low-power consumption area; when the track shows that the current state point moves from the outside of the high-power-consumption state area to the boundary, calculating the estimated time of the current state point reaching the boundary of the high-power-consumption state area; If the estimated time is greater than a preset early warning threshold value, marking the current state point as a candidate key state node which is about to enter a high power consumption area; Calculating the curvature and direction consistency of the state evolution path for each candidate key state node; if the curvature is below a threshold and the direction consistency is above a threshold, determining that the state evolution of the candidate critical state node has a deterministic trend, and finally determining the critical state node as the critical state node.
7. The method for intelligent switching between low power consumption modes of a chip according to claim 6, wherein the generating a set of pre-execution control sequences including voltage adjustment instructions and frequency adjustment instructions based on the attributes of the key state nodes and the relationships between the nodes is specifically as follows: According to the type of the key state node, a corresponding voltage frequency adjustment mode is retrieved from a strategy library; acquiring an initial voltage value and an initial frequency value adopted in a state historically according to specific coordinates of the key state nodes in a state space; acquiring the performance rating of a radiator of a current chip and the maximum instantaneous current supply capacity of a power supply unit; Calculating a series of feasible voltage frequency adjustment paths under electrical and thermal safety constraints by taking the initial voltage value and the initial frequency value as starting points and combining the radiator performance rating and the maximum instantaneous current supply capability; selecting a path with the smallest energy consumption increment from the feasible voltage frequency adjustment paths as an optimal adjustment path; Discretizing the optimal conditioning path into a series of chronologically arranged voltage set points and frequency set points; Packaging each voltage set point as an independent voltage regulation command, and packaging each frequency set point as an independent frequency regulation command; And according to the time sequence of the optimal regulation path, the voltage regulation instructions and the frequency regulation instructions are staggered to form the pre-execution control sequence with the execution time stamp.
8. The intelligent switching method based on the low power consumption mode of the chip according to claim 2, wherein the issuing of the pre-execution control sequence to the power management interface of the chip and the synchronous acquisition of the actual power consumption track fed back by the chip during the execution process are specifically as follows: establishing a command channel of a power management interface of a chip, and sequentially sending each instruction in the pre-execution control sequence through the command channel; Recording the sending time and the instruction content of each instruction when the instruction is sent; establishing a data channel with a chip internal power consumption monitoring unit, and continuously collecting a real-time current value and a real-time voltage value fed back by the chip in the process of executing the instruction at a speed higher than the instruction sending frequency through the data channel; Multiplying the real-time current value acquired at each sampling moment by the real-time voltage value to obtain an instantaneous power consumption value at the sampling moment; all the instant power consumption values from the start of the instruction sending to the current moment are connected together to form a power consumption curve according to the time sequence, and the position point corresponding to each instruction sending moment is marked and used as the actual power consumption track.
9. The method for intelligently switching between low-power modes based on a chip according to claim 2, wherein the comparing the actual power consumption track with the expected power consumption track in the pre-execution control sequence calculates an execution error of power consumption adjustment, specifically: according to the voltage value and the frequency value expected to be reached by each instruction in the pre-execution control sequence, calculating an expected steady-state power consumption value by combining a nominal power consumption model of the chip under the voltage frequency combination; Connecting the calculated steady-state power consumption values to form an expected power consumption track; for each instruction in the pre-execution control sequence, finding a stable time period after the instruction sending time on the actual power consumption track; Calculating an average power consumption value of the actual power consumption track in the stable time period as an actual steady-state power consumption value corresponding to the instruction; The actual steady-state power consumption value corresponding to the instruction is differenced with the expected steady-state power consumption value, and a single-step execution error of the instruction is obtained; And counting the average value and standard deviation of single step execution errors of all instructions in the pre-execution control sequence, and respectively taking the average value and standard deviation as the average execution error and the execution dispersion of overall adjustment.
10. A chip-based low power mode intelligent switching system comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that the processor, when executing the computer program, implements the steps of a chip-based low power mode intelligent switching method according to any one of the preceding claims 1 to 9.

Description

Intelligent switching method and system based on chip low-power mode Technical Field The invention relates to the technical field of low-power consumption management of integrated circuits, in particular to an intelligent switching method and system based on a low-power consumption mode of a chip. Background The existing chip low-power consumption management technology mainly relies on monitoring single aggregate signals such as overall power consumption, current or temperature and the like, and makes mode switching decisions based on a preset fixed threshold. The method can not distinguish the steady-state current generated by the computing core from the transient current caused by the peripheral circuit activity by aliasing the electric activity characteristics of different functional units in the chip. Because of the lack of fine recognition of the source of power consumption and the behavior pattern, the system can only make a lagging and rough response to the overall power consumption changes that have occurred, and it is difficult to achieve precise power consumption control. Existing schemes typically treat environmental factors as parallel judgment conditions or simple linear correction factors independent of the internal state of the chip. The processing mode breaks the internal correlation between the dynamic change of the environment and the electric activity inside the chip, and the differential influence possibly generated by the gradual change trend and the rapid disturbance of the environment parameters on different circuit modules is not considered. Therefore, the prior art cannot capture and utilize the complex space-time coupling relation between the chip running state and the environment state, so that the power consumption mode switching time prediction capability is insufficient, and the optimal energy efficiency point is often missed. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides an intelligent switching method and system based on a low-power-consumption mode of a chip. In order to achieve the purpose, the invention adopts the following technical scheme that the intelligent switching method based on the low-power consumption mode of the chip comprises the following steps: Acquiring an original electric signal and an original environment signal when the chip operates; Stripping a steady-state electric signal component generated by the operation of a chip core and a transient electric signal component generated by the activity of a peripheral circuit from the original electric signal, and constructing a composite electric signal map based on the respective dynamic ranges of the steady-state electric signal component and the transient electric signal component; Carrying out signal characteristic decomposition on the original environment signal, and extracting a trend component which slowly changes along with time and a disturbance component which rapidly fluctuates along with time; The composite electric signal map, the trend component and the disturbance component are fused and aligned in the time dimension to generate a unified state model reflecting the coupling relation between the chip running state and the environment state; analyzing the unified state model, and identifying key state nodes in the model, which correspond to the regions to be accessed or the regions to be released from the low power consumption of the chip; Based on the attributes of the critical state nodes and the relationships between the nodes, a set of pre-execution control sequences including voltage adjustment instructions and frequency adjustment instructions is generated. As a further aspect of the present invention, the method further includes: Issuing the pre-execution control sequence to a power management interface of a chip, and synchronously acquiring an actual power consumption track fed back by the chip in the execution process; comparing the actual power consumption track with an expected power consumption track in the pre-execution control sequence, and calculating an execution error of power consumption adjustment; Fine tuning parameters of subsequent instructions in the pre-execution control sequence according to the execution error to generate a corrected control sequence; And packaging the corrected control sequence, the actual power consumption track and the execution error into a state report, and transmitting the state report to an external collaborative management node through a short-distance wireless communication link. As a further scheme of the present invention, the method includes stripping a steady-state electrical signal component generated by the operation of a chip core and a transient electrical signal component generated by the activity of a peripheral circuit from the original electrical signal, and constructing a composite electrical signal spectrum based on respective dynamic ranges of the steady-state electrical signal compone