Search

CN-121209666-B - Chip dynamic temperature adjustment method based on PID control algorithm

CN121209666BCN 121209666 BCN121209666 BCN 121209666BCN-121209666-B

Abstract

The invention discloses a chip dynamic temperature regulation method based on a PID control algorithm, which comprises the following steps of S1, collecting and processing multipoint temperature data to obtain a unified temperature vector, S2, carrying out hot mode extraction, calculating deviation with a target temperature to generate an error signal, S3, constructing a segmented polymorphic PI D controller, carrying out mode switching to calculate a proportional term, S4, generating an integral term for each hot mode, carrying out change rate processing to generate a differential term, S5, combining the proportional, integral and differential terms, outputting an executor instruction by a distributor, S6, issuing the executor instruction, and executing limiting and slope constraint to finish regulation. The invention realizes the rapid response and stable adjustment of the chip temperature through the sectional polymorphic PI D control and the modal sensing distributor.

Inventors

  • QIN KE

Assignees

  • 天津智邦新材料科技有限公司

Dates

Publication Date
20260508
Application Date
20250915

Claims (7)

  1. 1. The chip dynamic temperature regulation method based on the PID control algorithm is characterized by comprising the following steps of: S1, acquiring real-time temperature data of a plurality of temperature sensors on a chip, and processing the real-time temperature data to obtain a multipoint temperature vector; s2, carrying out thermal model extraction on the multipoint temperature vectors to form a plurality of thermal model variables, setting a target temperature value, and calculating deviation of the thermal model variables relative to the target temperature value to obtain a modal error signal; S3, constructing a segmented multi-state PID controller, switching between a steady mode, a burst high-load mode and a continuous high-load mode according to the running state of the chip, calling a proportion parameter corresponding to the current mode, and executing proportion operation on a mode error signal to obtain a mode proportion term; s4, calling integral parameters and differential parameters corresponding to the current mode in the segmented polymorphic PID controller, configuring an independent integrator for each thermal mode to generate a mode integral term, and carrying out fractional differential processing on the change rate of a mode error signal in the differential operation process to generate a mode differential term; S5, combining the mode proportion term, the mode integral term and the mode differential term to generate a mode control output signal, inputting the mode control output signal to a mode sensing distributor, and generating corresponding actuator control instructions according to the sensitivity and constraint conditions of the actuator to different thermal modes; s6, issuing the actuator control instruction to a fan control unit, a dynamic voltage frequency adjusting unit and a thermoelectric cooling unit, and implementing amplitude limiting and slope constraint on the output process to complete dynamic adjustment of the chip temperature.
  2. 2. The method for dynamically adjusting the temperature of a chip based on a PID control algorithm according to claim 1, wherein the processing the real-time temperature data specifically includes: denoising the acquired real-time temperature data of the temperature sensor, and removing abnormal fluctuation data; Filtering the denoised real-time temperature data to inhibit high-frequency interference signals; and carrying out fusion operation on the filtered real-time temperature data to form a unified multipoint temperature vector.
  3. 3. The method for dynamically adjusting the temperature of a chip based on a PID control algorithm according to claim 1, wherein S2 specifically includes: s21, reading a multipoint temperature vector, and carrying out thermal model extraction on the multipoint temperature vector according to a thermal model base vector set which is established and stored in advance to obtain a plurality of thermal model variables which are arranged in sequence; S22, converting the target temperature value into a mode target value corresponding to the thermal mode variable one by one according to a preset mode mapping rule to obtain a mode target sequence arranged according to the same sequence number; S23, for each thermal mode variable, calculating an initial difference value between the thermal mode variable and a corresponding mode target value, and then sequentially executing the following processes: setting the deviation to zero in a hysteresis interval formed by the first deviation threshold and the second deviation threshold, and carrying out asymmetric deduction according to different thresholds of positive direction and negative direction when the deviation exceeds the hysteresis interval; When an ascending trend appears in a plurality of continuous sampling periods and the trend intensity exceeds a preset trend threshold, adding a prospective compensation amount into the deviation, and when a descending trend appears and the trend intensity exceeds the preset trend threshold, adding a moderation compensation amount into the deviation; scaling the deviation according to the sensing consistency evaluation coefficient corresponding to the thermal model variable; amplitude constraint is carried out on the scaled deviation according to preset upper and lower limits, and a mode deviation signal of a thermal mode variable is obtained; S24, combining the mode deviation signals of all the thermal modes according to the sequence numbers of the thermal mode variables to form a mode error signal sequence, and outputting the mode error signal sequence.
  4. 4. The method for dynamically adjusting the temperature of a chip based on a PID control algorithm according to claim 1, wherein S3 specifically includes: s31, acquiring a modal error signal sequence, establishing a one-to-one mapping relation corresponding to the modal variable according to the serial number of the thermal modal variable, and taking the modal error signal sequence as the input of the segmented polymorphic PID controller; S32, constructing a structure of a segmented polymorphic PID controller, wherein the structure comprises a state manager, a parameter library, a parameter interpolator, a mode residence manager, a small-amplitude jitter suppressor, an abnormal switching protector and a proportion arithmetic unit, the state manager determines a steady mode, a burst high-load mode and a continuous high-load mode, and the parameter library presets proportion parameter basic values for each thermal mode variable in three modes respectively; s33, calculating two state judgment indexes based on the modal error signal sequence by a state manager, wherein the error amplitude index is the largest of absolute amplitudes of all modal deviation signals at the same sampling moment, and the error change rate index is the largest of the change amplitudes of all modal deviation signals between adjacent sampling moments; S34, performing mode judgment according to the following sequence: When the error change rate index reaches an error change rate threshold value, entering a burst high-load mode; when the error amplitude index continuously reaches the error amplitude high threshold value and the duration reaches the minimum resident count, entering a continuous high load mode; When the error amplitude index is lower than the error amplitude low threshold value and the error change rate index is lower than the error change rate threshold value, entering a stationary mode; when any condition is not met, the mode of the last sampling time is kept, and an upper limit and a lower limit are formed through hysteresis threshold setting so as to limit frequent switching; s35, after the mode is determined, performing proportional parameter generation processing, specifically: Providing a proportional parameter base value corresponding to the current mode for each thermal mode variable by a parameter library, and recording a sampling count from the last mode switching by a mode residence manager; Performing time-based smooth interpolation between the proportional parameter base value of the previous mode and the proportional parameter base value of the current mode by a parameter interpolator; Setting the scale parameter adjustment coefficient to zero when the modal deviation signal is in a preset dead zone by a small-amplitude jitter suppressor, and correcting the interpolated scale parameter according to the sensing consistency evaluation coefficient to obtain the final scale parameter of each thermal modal variable; S36, calculating a modal proportion term for each thermal modal variable by the proportion arithmetic unit according to the linear relation between the final proportion parameter and the corresponding modal deviation signal, and combining the modal proportion term sequences according to the sequence number sequence of the thermal modal variable.
  5. 5. The method for dynamically adjusting the temperature of a chip based on a PID control algorithm according to claim 1, wherein S4 specifically includes: s41, acquiring an integral parameter and a differential parameter corresponding to a mode error signal sequence, a mode proportion term sequence and a current mode; S42, establishing an independent integrator for each thermal model variable, setting an integration initial value, an integration upper limit, an integration lower limit, an integration step length and an integration dead zone, and carrying out integration operation on a mode error signal of the thermal model variable according to a sampling sequence to obtain a mode integration term; s43, executing integral anti-saturation processing on each thermal mode variable in the integral operation process, setting a mode integral term as an integral upper limit when the accumulated mode integral term exceeds the integral upper limit, setting the mode integral term as an integral lower limit when the accumulated mode integral term is lower than the integral lower limit, recording an integral freezing mark, and suspending integral accumulation of the thermal mode variable in a sampling interval in which the integral freezing mark is recorded; s44, calculating the change rate of the modal error signal of each thermal model variable, normalizing the change rate according to a sampling period, setting a change rate dead zone and a change rate hysteresis threshold, recording the change rate as zero when the absolute value of the change rate is in the change rate dead zone, and updating a change rate effective mark when the absolute value of the change rate crosses the change rate hysteresis threshold; S45, selecting a preset fractional order according to a current mode, performing fractional differential processing on a change rate sequence with a change rate effective mark, combining differential parameters to obtain a modal differential term of each thermal mode variable, and marking the modal differential term as zero for sampling points without the change rate effective mark; S46, combining the modal integral terms of all the thermal model variables according to the sequence numbers of the thermal model variables to form a modal integral term sequence, and combining the modal differential terms of all the thermal model variables according to the sequence numbers of the thermal model variables to form a modal differential term sequence.
  6. 6. The method for dynamically adjusting the temperature of a chip based on a PID control algorithm according to claim 1, wherein S5 specifically includes: S51, acquiring a modal proportion term sequence, a modal integral term sequence and a modal differential term sequence which are arranged according to a thermal modal variable sequence, carrying out weighted superposition on three terms corresponding to each thermal modal variable according to a pre-stored weighting coefficient and a fixed combination sequence to obtain a modal control output value, and combining the modal control output values according to the sequence to form a modal control output signal sequence; S52, constructing a modal sensing distributor, wherein the modal sensing distributor comprises a static parameter set and an operation parameter set, the static parameter set comprises an actuator set, a modal and sensitivity coefficient table of the actuator, a cost coefficient table of the actuator, an upper limit table and a lower limit table of the amplitude of the actuator and a maximum change rate table of the actuator, and the operation parameter set comprises an actuator control quantity at the last sampling moment and a modal control output signal sequence at the current sampling moment; S53, calculating an intermediate control quantity set which minimizes the sum of modal fitting error and actuator cost weighting by a modal sensing distributor based on a modal control output signal sequence, a sensitivity coefficient table and a cost coefficient table at each sampling moment, and outputting the intermediate control quantity set in a one-to-one correspondence manner according to an actuator channel; s54, sequentially executing change rate constraint and amplitude constraint on the intermediate control quantity set: In the change rate constraint, comparing the control quantity difference value of each actuator at the current sampling time and the last sampling time, and limiting the control quantity difference value to be within the maximum change rate range allowed by the actuators; in the amplitude constraint, taking the upper limit and the lower limit of the amplitude of each actuator as a boundary, projecting the control quantity, when any actuator channel touches the boundary in any constraint, carrying out equal scaling on the intermediate control quantity of all actuator channels according to a preset rule, and recording a boundary touching mark and a scaling factor; s55, converting the control quantity set into an executable control instruction set, wherein the control instruction of the fan control unit is a target rotating speed or duty ratio, the control instruction of the dynamic voltage frequency adjusting unit is a target running state or duty ratio, and the control instruction of the thermoelectric cooling unit is a target driving power.
  7. 7. The method for dynamically adjusting the temperature of a chip based on a PID control algorithm according to claim 1, wherein S6 specifically includes: S61, reading a control instruction set and an actuator control quantity set at the last sampling moment, and loading a boundary touch mark and a scaling factor; s62, setting a minimum residence count and a maximum switching frequency for a fan control unit, a dynamic voltage frequency adjusting unit and a thermoelectric cooling unit, and converting a control instruction set into a target control amount format, wherein the fan control unit adopts a duty ratio format, the dynamic voltage frequency adjusting unit adopts a target running state, and the thermoelectric cooling unit adopts a target driving power format; S63, executing slope constraint, comparing the difference value between the current target control quantity of each actuator and the control quantity of the last sampling moment, and limiting the difference value within the maximum change rate range allowed by the actuator to obtain a control quantity set after slope constraint; S64, amplitude limiting is carried out, the control quantity of each actuator is limited in the range of the corresponding upper limit and the lower limit of the amplitude, when any actuator touches the boundary, the control quantity of all actuators is scaled in equal proportion according to a preset rule according to the boundary touching mark and the scaling factor, and an executable control quantity set is formed; S65, issuing an executable control quantity set to the fan control unit, the dynamic voltage frequency adjusting unit and the thermoelectric cooling unit, executing minimum residence time check on the dynamic voltage frequency adjusting unit, executing soft start and soft stop on the fan control unit and the thermoelectric cooling unit, collecting execution feedback, updating an actuator control quantity set and a mode residence count at the last sampling moment, detecting a fault juxtaposition fault sign, and outputting a degradation instruction set.

Description

Chip dynamic temperature adjustment method based on PID control algorithm Technical Field The invention relates to the technical field of chip temperature management and control, in particular to a chip dynamic temperature adjustment method based on a PID control algorithm. Background At present, with the continuous improvement of the integration level and the running frequency of a chip, the heat generated by the chip in high-performance calculation and complex application scenes is obviously increased, and the stability and the operation precision of the chip can be influenced by the overhigh temperature, so that the service life of the chip can be shortened, and even the function is invalid. In order to ensure reliable operation of the chip, a temperature control method is generally adopted in the prior art for dynamic adjustment, wherein a PID control algorithm is widely applied to a chip temperature control system due to simple implementation and stable adjustment. The common practice is to collect temperature data of the chip surface or core area through a temperature sensor, compare the real-time temperature with a preset target temperature, output control quantity by using a PID controller, and act on a fan, a dynamic voltage frequency adjusting unit or a thermoelectric cooling unit to realize stable control of the chip temperature. However, the PID control method in the prior art has the following disadvantages. On one hand, the traditional PID algorithm mostly adopts a single sensor or simple multipoint average temperature as input, and is difficult to reflect complex heat distribution characteristics inside a chip, so that under the conditions of multi-region thermal coupling and hot spot rapid migration, control response is lagged, and the adjusting effect is poor. On the other hand, the conventional PID control parameters are fixed, and when the running state of the chip is changed from steady state to burst high load or continuous high load, the parameters of the controller cannot be adaptively switched, so that temperature overshoot or response delay is easily caused. In addition, the integration link is easy to generate saturation phenomenon, the differential link is easy to amplify interference when measurement noise exists, and control accuracy is further reduced. In the prior art, both the quick response and the steady-state precision are difficult to be simultaneously considered, and a refinement control mechanism under different working modes is also lacking. The chip temperature regulation has insufficient robustness under the complex working condition, the energy utilization efficiency is low, and even the problems of frequent action of an actuator, energy consumption, noise increase and the like can occur. Therefore, how to provide a chip dynamic temperature adjustment method based on a PID control algorithm is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a chip dynamic temperature regulation method based on a PID control algorithm, which has the advantages of fast response speed, high steady-state precision, smooth actuator action and optimized energy consumption and noise by introducing a segmented polymorphic PID controller, an independent modal integrator, a fractional differential processing and modal sensing distributor, carrying out hot mode extraction on a multipoint temperature vector, and carrying out control parameter switching by combining different running states. According to the embodiment of the invention, the chip dynamic temperature adjustment method based on the PID control algorithm comprises the following steps of: S1, acquiring real-time temperature data of a plurality of temperature sensors on a chip, and processing the real-time temperature data to obtain a multipoint temperature vector; s2, carrying out thermal model extraction on the multipoint temperature vectors to form a plurality of thermal model variables, setting a target temperature value, and calculating deviation of the thermal model variables relative to the target temperature value to obtain a modal error signal; S3, constructing a segmented multi-state PID controller, switching between a steady mode, a burst high-load mode and a continuous high-load mode according to the running state of the chip, calling a proportion parameter corresponding to the current mode, and executing proportion operation on a mode error signal to obtain a mode proportion term; s4, calling integral parameters and differential parameters corresponding to the current mode in the segmented polymorphic PID controller, configuring an independent integrator for each thermal mode to generate a mode integral term, and carrying out fractional differential processing on the change rate of a mode error signal in the differential operation process to generate a mode differential term; S5, combining the mode proportion term, the mode integral term and the mode dif