Chip dynamic temperature regulation method based on PID control algorithm

Abstract

The application discloses a chip dynamic temperature adjusting method based on a PID control algorithm, and comprises the following steps: S1, collecting and processing multi-point temperature data to obtain a unified temperature vector; S2, extracting a thermal mode, calculating a deviation from a target temperature to generate an error signal; S3, constructing a segmented multi-state PID controller, and calculating a proportional term through mode switching; S4, generating an integral term for each thermal mode, and generating a differential term through a change rate; S5, combining the proportional term, the integral term and the differential term, and distributing an actuator instruction by a distributor; and S6, issuing the actuator instruction, executing amplitude limiting and slope constraint to complete adjustment. Through the segmented multi-state PID control and the mode sensing distributor, the application realizes fast response and stable adjustment of the chip temperature.

Description

Technical Field

[0001] This invention relates to the field of chip temperature management and control technology, and in particular to a chip dynamic temperature regulation method based on a PID control algorithm. Background Technology

[0002] Currently, with the continuous increase in chip integration and operating frequency, the heat generated by chips in high-performance computing and complex application scenarios has increased significantly. Excessive temperature not only affects chip stability and computational accuracy but may also shorten chip lifespan or even lead to functional failure. To ensure reliable chip operation, existing technologies commonly employ dynamic temperature control methods. Among these, the PID control algorithm is widely used in chip temperature control systems due to its simplicity and stable adjustment. A common approach involves collecting temperature data from the chip surface or core area using temperature sensors, comparing the real-time temperature with a preset target temperature, and then using a PID controller to output a control quantity that acts on a fan, dynamic voltage and frequency regulation unit, or thermoelectric cooling unit to achieve stable chip temperature control.

[0003] However, existing PID control methods generally suffer from the following shortcomings. Firstly, traditional PID algorithms often use a single sensor or simple multi-point average temperature as input, making it difficult to reflect the complex thermal distribution characteristics within the chip. This results in lag in control response and poor adjustment performance under conditions of multi-region thermal coupling and rapid hotspot migration. Secondly, conventional PID control parameters are fixed. When the chip's operating state changes from static stability to sudden or sustained high load, the controller parameters cannot adaptively switch, easily causing temperature overshoot or slow response. Furthermore, the integral term is prone to saturation, and the derivative term can amplify interference in the presence of measurement noise, further reducing control accuracy.

[0004] Existing technologies struggle to simultaneously achieve both rapid response and steady-state accuracy, and lack refined control mechanisms for different operating modes. This results in insufficient robustness of chip temperature regulation under complex operating conditions, low energy efficiency, and even problems such as frequent actuator operation, increased energy consumption, and noise.

[0005] Therefore, how to provide a dynamic temperature regulation method for chips based on PID control algorithms is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a dynamic temperature regulation method for chips based on a PID control algorithm. This invention introduces a piecewise multi-state PID controller, an independent modal integrator, fractional derivative processing, and a modal sensing distributor to extract thermal modes from multi-point temperature vectors and switch control parameters in combination with different operating states. It describes in detail the dynamic temperature regulation process under various operating modes such as static stability, sudden high load, and continuous high load, and has the advantages of fast response speed, high steady-state accuracy, smooth actuator action, and optimized energy consumption and noise.

[0007] A chip dynamic temperature regulation method based on a PID control algorithm according to an embodiment of the present invention includes the following steps:

[0008] S1. Collect real-time temperature data from multiple temperature sensors on the chip, process the real-time temperature data, and obtain a multi-point temperature vector.

[0009] S2. Extract thermal modes from the multi-point temperature vector to form several thermal mode variables. Set a target temperature value and calculate the deviation of the thermal mode variables relative to the target temperature value to obtain the mode error signal.

[0010] S3. Construct a segmented multi-state PID controller, which switches between static steady mode, sudden high load mode and continuous high load mode according to the chip's operating state. Call the proportional parameter corresponding to the current mode, perform proportional calculation on the modal error signal, and obtain the modal proportional term.

[0011] S4. In the piecewise multi-mode PID controller, call the integral and derivative parameters corresponding to the current mode, configure an independent integrator for each thermal mode to generate a modal integral term, and perform fractional derivative processing on the rate of change of the modal error signal during the derivative operation to generate a modal derivative term;

[0012] S5. Combine the modal proportional term, modal integral term, and modal differential term to generate a modal control output signal, and input the modal control output signal to the modal sensing distributor to generate corresponding actuator control commands based on the actuator's sensitivity to different thermal modes and constraint conditions.

[0013] S6. The actuator control command is sent to the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit to implement amplitude limiting and slope constraint on the output process, thereby completing the dynamic adjustment of chip temperature.

[0014] Optionally, the processing of the real-time temperature data specifically includes:

[0015] The real-time temperature data collected from the temperature sensor is denoised to remove abnormal fluctuations.

[0016] The denoised real-time temperature data is filtered to suppress high-frequency interference signals;

[0017] The filtered real-time temperature data is fused to form a unified multi-point temperature vector.

[0018] Optionally, S2 specifically includes:

[0019] S21. Read the multi-point temperature vector and extract the thermal modes of the multi-point temperature vector according to the pre-established and saved thermal mode basis vector set to obtain a number of thermal mode variables arranged in order.

[0020] S22. The target temperature value is converted into a modal target value that corresponds one-to-one with the thermal modal variable according to the preset modal mapping rule, so as to obtain a modal target sequence arranged with the same number.

[0021] S23. For each thermal modal variable, first calculate the initial difference between the thermal modal variable and the corresponding modal target value, and then perform the following processing in sequence:

[0022] Within the hysteresis interval formed by the first deviation threshold and the second deviation threshold, the deviation is set to zero. When the deviation exceeds the hysteresis interval, asymmetric deduction is performed according to different thresholds for positive and negative directions.

[0023] When an upward trend appears within several consecutive sampling periods and the trend intensity exceeds a preset trend threshold, a forward compensation amount is added to the deviation; when a downward trend appears and the trend intensity exceeds a preset trend threshold, a mitigation compensation amount is added to the deviation.

[0024] The deviation is scaled based on the sensor consistency evaluation coefficient corresponding to the thermal modal variable;

[0025] The amplitude of the scaled deviation is constrained according to preset upper and lower limits to obtain the modal deviation signal of the thermal modal variable;

[0026] S24. Combine the modal deviation signals of all thermal modes in the order of the thermal modal variables to form a modal error signal sequence, and output the modal error signal sequence.

[0027] Optionally, S3 specifically includes:

[0028] S31. Obtain the modal error signal sequence, establish a one-to-one mapping relationship with the corresponding thermal modal variables according to their serial numbers, and use it as the input of the piecewise multi-mode PID controller;

[0029] S32. Construct the structure of a segmented multi-state PID controller, the structure including a state manager, a parameter library, a parameter interpolator, a mode resident manager, a small-amplitude jitter suppressor, an abnormal switching protector, and a proportional calculator. The state manager determines the static steady-state mode, the sudden high-load mode, and the continuous high-load mode. The parameter library presets the proportional parameter base value for each thermal modal variable in the three modes.

[0030] S33. The state manager calculates two state determination indices based on the modal error signal sequence. The error amplitude index is the largest absolute amplitude of all modal deviation signals at the same sampling time, and the error change rate index is the largest change amplitude of all modal deviation signals between adjacent sampling times.

[0031] S34. Perform pattern determination in the following order:

[0032] When the error change rate index reaches the error change rate threshold, it enters the sudden high load mode.

[0033] When the error amplitude index continuously reaches the high threshold of error amplitude and the duration reaches the minimum dwell count, it enters the continuous high load mode.

[0034] When the error amplitude index is lower than the low threshold of error amplitude and the error change rate index is lower than the threshold of error change rate, the static steady mode is entered.

[0035] If none of the above conditions are met, the mode of the previous sampling time is maintained, and an upper and lower limit distinction is formed by setting a hysteresis threshold to limit frequent switching;

[0036] S35. After the mode is determined, perform the proportional parameter generation process, specifically:

[0037] The parameter library provides the proportional parameter base value corresponding to the current mode for each thermal modal variable, and the mode dwell manager records the sampling count since the most recent mode switch;

[0038] The parameter interpolator performs time-based smooth interpolation between the scale parameter base value of the previous mode and the scale parameter base value of the current mode;

[0039] When the modal deviation signal is in the preset dead zone, the small-amplitude jitter suppressor sets the proportional parameter adjustment coefficient to zero, and corrects the interpolated proportional parameter according to the sensor consistency evaluation coefficient to obtain the final proportional parameter of each thermal modal variable.

[0040] S36. The proportional operator calculates the modal proportional term for each thermal modal variable according to the linear relationship between the final proportional parameter and the corresponding modal deviation signal, and combines them in the order of the thermal modal variables to form a sequence of modal proportional terms.

[0041] Optionally, S4 specifically includes:

[0042] S41. Obtain the modal error signal sequence, the modal proportional term sequence, and the integral and differential parameters corresponding to the current mode;

[0043] S42. For each thermal modal variable, establish an independent integrator, set the initial integration value, upper integration limit, lower integration limit, integration step size and integration dead zone, and perform integration operation on the modal error signal of the thermal modal variable according to the sampling order to obtain the modal integral term;

[0044] S43. During the integration process, perform integration anti-saturation processing on each thermal modal variable. When the accumulated modal integral term exceeds the integration upper limit, set the modal integral term to the integration upper limit. When the accumulated modal integral term is lower than the integration lower limit, set the modal integral term to the integration lower limit. Record the integration freeze flag. Within the sampling interval where the integration freeze flag is recorded, pause the integration accumulation of the thermal modal variable.

[0045] S44. Calculate the rate of change of the modal error signal for each thermal modal variable. Normalize the rate of change according to the sampling period and set the rate of change dead zone and the rate of change hysteresis threshold. When the absolute value of the rate of change is within the rate of change dead zone, the rate of change is recorded as zero. When the absolute value of the rate of change crosses the rate of change hysteresis threshold, update the rate of change valid flag.

[0046] S45. Select the preset fractional order based on the current mode, perform fractional differentiation on the rate of change sequence with effective rate of change indicators, and obtain the modal differential term of each thermal modal variable by combining the differential parameters. Record the modal differential term as zero for sampling points without effective rate of change indicators.

[0047] S46. Combine the modal integral terms of all thermal modal variables in the order of their serial numbers to form a sequence of modal integral terms, and combine the modal differential terms of all thermal modal variables in the order of their serial numbers to form a sequence of modal differential terms.

[0048] Optionally, S5 specifically includes:

[0049] S51. Obtain the sequence of modal proportional terms, modal integral terms, and modal differential terms arranged by the sequence number of the thermal modal variables. Based on the pre-stored weighting coefficients and fixed combination order, perform weighted superposition on the three terms corresponding to each thermal modal variable to obtain the modal control output value. Combine the modal control output values ​​according to the sequence number to form the modal control output signal sequence.

[0050] S52. Construct a modal sensing allocator, which includes a static parameter set and an operational parameter set. The static parameter set includes an actuator set, a table of sensitivity coefficients of modes and actuators, a table of cost coefficients of actuators, a table of upper and lower limits of actuator amplitude, and a table of maximum rate of change of actuators. The operational parameter set includes the actuator control quantity at the previous sampling time and the modal control output signal sequence at the current sampling time.

[0051] S53. At each sampling time, the modal sensing distributor calculates the set of intermediate control quantities that minimizes the weighted sum of modal fitting error and actuator cost based on the modal control output signal sequence, sensitivity coefficient table and cost coefficient table, and outputs the set of intermediate control quantities one by one according to the actuator channel.

[0052] S54. Apply rate of change constraints and magnitude constraints sequentially to the intermediate control variable set:

[0053] In the rate of change constraint, the difference between the control quantity of each actuator at the current sampling time and the previous sampling time is compared and limited to the maximum allowable rate of change of the actuator.

[0054] In amplitude constraints, the control quantity is projected using the upper and lower limits of the amplitude of each actuator as boundaries. When any actuator channel touches the boundary in any constraint, the intermediate control quantity of all actuator channels is scaled proportionally according to preset rules, and the boundary touch mark and scaling factor are recorded.

[0055] S55. Convert the set of control quantities into a set of executable control instructions, wherein the control instructions for the fan control unit are the target speed or duty cycle, the control instructions for the dynamic voltage and frequency adjustment unit are the target operating state or duty cycle, and the control instructions for the thermoelectric cooling unit are the target drive power.

[0056] Optionally, S6 specifically includes:

[0057] S61. Read the set of control instructions and the set of actuator control quantities at the previous sampling time, and load the boundary touch flag and scaling factor;

[0058] S62. Set the minimum dwell count and maximum switching frequency for the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit, and convert the control command set into the target control quantity format, wherein the fan control unit adopts the duty cycle format, the dynamic voltage and frequency adjustment unit adopts the target operating state, and the thermoelectric cooling unit adopts the target drive power format.

[0059] S63. Execute slope constraints, compare the difference between the current target control quantity of each actuator and the control quantity at the previous sampling time, limit the difference to the maximum allowable rate of change of the actuator, and obtain the set of control quantities after slope constraints.

[0060] S64. Execute amplitude limiting, which limits the control quantity of each actuator to the corresponding upper and lower amplitude limits. When any actuator touches the boundary, the control quantities of all actuators are scaled proportionally according to the boundary touch flag and scaling factor according to the preset rules to form a set of executable control quantities.

[0061] S65. Send the set of executable control quantities to the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit. Perform minimum dwell time check on the dynamic voltage and frequency adjustment unit. Perform soft start and soft stop on the fan control unit and thermoelectric cooling unit. Collect execution feedback and update the set of actuator control quantities and mode dwell count at the previous sampling time. Detect faults and set fault flags. Output a set of degradation instructions.

[0062] The beneficial effects of this invention are:

[0063] This invention solves the problems of lag in multi-point temperature control and inability to adapt to different operating modes in existing technologies by introducing a dynamic temperature regulation method for chips based on PID control algorithms. By introducing piecewise multi-state control and modal sensing allocation mechanisms into the PID controller, this invention can flexibly adjust control parameters for different chip operating states, ensuring high efficiency and accuracy in the temperature regulation process. In particular, when the chip load switches from a static steady state to a sudden high load or a sustained high load, the controller can automatically switch parameters, avoiding the problems of temperature overshoot or slow response in traditional PID control when the load changes significantly.

[0064] This invention introduces an anti-saturation mechanism in the integral term to prevent system instability caused by excessive accumulation of the integral term in traditional PID control. Fractional-order derivative processing reduces the impact of high-frequency noise in the temperature signal on control accuracy, further improving the stability of temperature regulation. The modal sensing distributor intelligently allocates control inputs based on the relationship between thermal modes and actuator sensitivity, resulting in more precise actuator control, avoiding actuator over-working or frequent switching, and significantly improving the system's energy efficiency ratio.

[0065] This invention enables precise control of chip temperature, while improving system robustness and adaptability, reducing excessive energy consumption and noise, and enhancing chip operational reliability and lifespan. Attached Figure Description

[0066] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0067] Figure 1 This is a flowchart of a chip dynamic temperature regulation method based on PID control algorithm proposed in this invention;

[0068] Figure 2 This is a schematic diagram of the segmented multi-state PID controller for a chip dynamic temperature regulation method based on PID control algorithm proposed in this invention. Detailed Implementation

[0069] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0070] refer to Figure 1 and Figure 2 A method for dynamic temperature regulation of a chip based on a PID control algorithm includes the following steps:

[0071] S1. Collect real-time temperature data from multiple temperature sensors on the chip, process the real-time temperature data, and obtain a multi-point temperature vector.

[0072] S2. Extract thermal modes from the multi-point temperature vector to form several thermal mode variables. Set a target temperature value and calculate the deviation of the thermal mode variables relative to the target temperature value to obtain the mode error signal.

[0073] S3. Construct a segmented multi-state PID controller, which switches between static steady mode, sudden high load mode and continuous high load mode according to the chip's operating state. Call the proportional parameter corresponding to the current mode, perform proportional calculation on the modal error signal, and obtain the modal proportional term.

[0074] S4. In the piecewise multi-mode PID controller, call the integral and derivative parameters corresponding to the current mode, configure an independent integrator for each thermal mode to generate a modal integral term, and perform fractional derivative processing on the rate of change of the modal error signal during the derivative operation to generate a modal derivative term;

[0075] S5. Combine the modal proportional term, modal integral term, and modal differential term to generate a modal control output signal, and input the modal control output signal to the modal sensing distributor to generate corresponding actuator control commands based on the actuator's sensitivity to different thermal modes and constraint conditions.

[0076] S6. The actuator control command is sent to the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit to implement amplitude limiting and slope constraint on the output process, thereby completing the dynamic adjustment of chip temperature.

[0077] In this embodiment, the processing of real-time temperature data specifically includes:

[0078] The real-time temperature data collected from the temperature sensor is denoised to remove abnormal fluctuations.

[0079] The denoised real-time temperature data is filtered to suppress high-frequency interference signals;

[0080] The filtered real-time temperature data is fused to form a unified multi-point temperature vector.

[0081] In this embodiment, S2 specifically includes:

[0082] S21. Read the multi-point temperature vector and extract the thermal modes of the multi-point temperature vector according to the pre-established and saved thermal mode basis vector set to obtain a number of thermal mode variables arranged in order.

[0083] S22. The target temperature value is converted into a modal target value that corresponds one-to-one with the thermal modal variable according to the preset modal mapping rule, so as to obtain a modal target sequence arranged with the same number.

[0084] S23. For each thermal modal variable, first calculate the initial difference between the thermal modal variable and the corresponding modal target value, and then perform the following processing in sequence:

[0085] Within the hysteresis interval formed by the first deviation threshold and the second deviation threshold, the deviation is set to zero. When the deviation exceeds the hysteresis interval, asymmetric deduction is performed according to different thresholds for positive and negative directions.

[0086] When an upward trend appears within several consecutive sampling periods and the trend intensity exceeds a preset trend threshold, a forward compensation amount is added to the deviation; when a downward trend appears and the trend intensity exceeds a preset trend threshold, a mitigation compensation amount is added to the deviation.

[0087] The deviation is scaled based on the sensor consistency evaluation coefficient corresponding to the thermal modal variable;

[0088] The amplitude of the scaled deviation is constrained according to preset upper and lower limits to obtain the modal deviation signal of the thermal modal variable;

[0089] S24. Combine the modal deviation signals of all thermal modes in the order of the thermal modal variables to form a modal error signal sequence, and output the modal error signal sequence.

[0090] This invention introduces hysteresis interval control, trend compensation, sensor consistency correction, and amplitude constraint mechanisms into the modal error signal generation process, effectively solving the problems of high noise, slow response, and insufficient accuracy in existing technologies. By extracting thermal modes and mapping them to target values, temperature control is expanded from single-point measurement to multi-point fusion, enabling a more realistic reflection of the internal thermal distribution of the chip. The introduction of hysteresis intervals and asymmetric deduction ensures that small fluctuations do not trigger invalid responses, while guaranteeing reasonable correction of positive and negative deviations at different amplitudes, avoiding malfunctions. The trend compensation mechanism identifies continuous upward or downward trends to correct deviations earlier or later, giving the controller a certain degree of foresight and buffering capability, improving dynamic adaptability. The scaling mechanism of the sensor consistency evaluation coefficient reduces the impact of abnormal sensor data, improving the reliability of modal deviations. Finally, the amplitude constraint ensures that the deviation signal remains within a controllable range. This invention achieves smoother, more stable, and more accurate modal error signal generation, laying the foundation for precise adjustment of the subsequent controller.

[0091] In this embodiment, S3 specifically includes:

[0092] S31. Obtain the modal error signal sequence, establish a one-to-one mapping relationship with the corresponding thermal modal variables according to their serial numbers, and use it as the input of the piecewise multi-mode PID controller;

[0093] S32. Construct the structure of a segmented multi-state PID controller, the structure including a state manager, a parameter library, a parameter interpolator, a mode resident manager, a small-amplitude jitter suppressor, an abnormal switching protector, and a proportional calculator. The state manager determines the static steady-state mode, the sudden high-load mode, and the continuous high-load mode. The parameter library presets the proportional parameter base value for each thermal modal variable in the three modes.

[0094] S33. The state manager calculates two state determination indices based on the modal error signal sequence. The error amplitude index is the largest absolute amplitude of all modal deviation signals at the same sampling time, and the error change rate index is the largest change amplitude of all modal deviation signals between adjacent sampling times.

[0095] S34. Perform pattern determination in the following order:

[0096] When the error change rate index reaches the error change rate threshold, it enters the sudden high load mode.

[0097] When the error amplitude index continuously reaches the high threshold of error amplitude and the duration reaches the minimum dwell count, it enters the continuous high load mode.

[0098] When the error amplitude index is lower than the low threshold of error amplitude and the error change rate index is lower than the threshold of error change rate, the static steady mode is entered.

[0099] If none of the above conditions are met, the mode of the previous sampling time is maintained, and an upper and lower limit distinction is formed by setting a hysteresis threshold to limit frequent switching;

[0100] S35. After the mode is determined, perform the proportional parameter generation process, specifically:

[0101] The parameter library provides the proportional parameter base value corresponding to the current mode for each thermal modal variable, and the mode dwell manager records the sampling count since the most recent mode switch;

[0102] The parameter interpolator performs time-based smooth interpolation between the scale parameter base value of the previous mode and the scale parameter base value of the current mode;

[0103] When the modal deviation signal is in the preset dead zone, the small-amplitude jitter suppressor sets the proportional parameter adjustment coefficient to zero, and corrects the interpolated proportional parameter according to the sensor consistency evaluation coefficient to obtain the final proportional parameter of each thermal modal variable.

[0104] S36. The proportional operator calculates the modal proportional term for each thermal modal variable according to the linear relationship between the final proportional parameter and the corresponding modal deviation signal, and combines them in the order of the thermal modal variables to form a sequence of modal proportional terms.

[0105] This invention introduces modules such as a state manager, parameter library, parameter interpolator, mode dwell manager, small-amplitude jitter suppressor, and abnormal switching protector into a piecewise multi-state PID controller, achieving dynamic adaptive adjustment of temperature control parameters under different chip operating modes. By jointly determining the error amplitude and error change rate, it can accurately identify static steady-state, sudden high-load, and continuous high-load modes, avoiding the lag and instability problems of traditional PID control during state switching. The parameter interpolator combined with the mode dwell manager ensures smooth transition of proportional parameters, suppressing abrupt changes caused by frequent switching. The addition of the small-amplitude jitter suppressor and sensor consistency correction mechanism effectively eliminates noise interference and improves the reliability of multi-sensor fusion. This invention achieves accurate generation of modal proportional parameters and stable output of modal proportional terms, striking a balance between fast response and steady-state accuracy, significantly improving the robustness and energy efficiency of the chip's dynamic temperature regulation.

[0106] In this embodiment, S4 specifically includes:

[0107] S41. Obtain the modal error signal sequence, the modal proportional term sequence, and the integral and differential parameters corresponding to the current mode;

[0108] S42. For each thermal modal variable, establish an independent integrator, set the initial integration value, upper integration limit, lower integration limit, integration step size and integration dead zone, and perform integration operation on the modal error signal of the thermal modal variable according to the sampling order to obtain the modal integral term;

[0109] S43. During the integration process, perform integration anti-saturation processing on each thermal modal variable. When the accumulated modal integral term exceeds the integration upper limit, set the modal integral term to the integration upper limit. When the accumulated modal integral term is lower than the integration lower limit, set the modal integral term to the integration lower limit. Record the integration freeze flag. Within the sampling interval where the integration freeze flag is recorded, pause the integration accumulation of the thermal modal variable.

[0110] S44. Calculate the rate of change of the modal error signal for each thermal modal variable. Normalize the rate of change according to the sampling period and set the rate of change dead zone and the rate of change hysteresis threshold. When the absolute value of the rate of change is within the rate of change dead zone, the rate of change is recorded as zero. When the absolute value of the rate of change crosses the rate of change hysteresis threshold, update the rate of change valid flag.

[0111] S45. Select the preset fractional order based on the current mode, perform fractional differentiation on the rate of change sequence with effective rate of change indicators, and obtain the modal differential term of each thermal modal variable by combining the differential parameters. Record the modal differential term as zero for sampling points without effective rate of change indicators.

[0112] S46. Combine the modal integral terms of all thermal modal variables in the order of their serial numbers to form a sequence of modal integral terms, and combine the modal differential terms of all thermal modal variables in the order of their serial numbers to form a sequence of modal differential terms.

[0113] This invention introduces independent integrators, an integral anti-saturation mechanism, and fractional derivative processing into the design of the integral and derivative stages, overcoming the limitations of traditional PID control in complex thermal environments. By establishing an independent integrator for each thermal modal variable and setting upper and lower limits, step sizes, and dead zones, it not only avoids saturation problems caused by excessive accumulation of integral terms but also maintains system stability during integral freeze periods. Combined with rate-of-change normalization and hysteresis threshold determination, it effectively suppresses misjudgments caused by sensor noise. Furthermore, by using fractional derivatives instead of conventional integer derivatives, the processing of the modal error signal rate of change becomes more flexible, preserving fast response while reducing the risk of high-frequency noise amplification. Through these innovative measures, this invention can output smoother and more reliable modal integral and derivative terms, exhibiting excellent control performance in both rapid thermal changes and long-term stable operation, significantly improving the stability and robustness of chip temperature regulation.

[0114] In this embodiment, S5 specifically includes:

[0115] S51. Obtain the sequence of modal proportional terms, modal integral terms, and modal differential terms arranged by the sequence number of the thermal modal variables. Based on the pre-stored weighting coefficients and fixed combination order, perform weighted superposition on the three terms corresponding to each thermal modal variable to obtain the modal control output value. Combine the modal control output values ​​according to the sequence number to form the modal control output signal sequence.

[0116] S52. Construct a modal sensing allocator, which includes a static parameter set and an operational parameter set. The static parameter set includes an actuator set, a table of sensitivity coefficients of modes and actuators, a table of cost coefficients of actuators, a table of upper and lower limits of actuator amplitude, and a table of maximum rate of change of actuators. The operational parameter set includes the actuator control quantity at the previous sampling time and the modal control output signal sequence at the current sampling time.

[0117] S53. At each sampling time, the modal sensing distributor calculates the set of intermediate control quantities that minimizes the weighted sum of modal fitting error and actuator cost based on the modal control output signal sequence, sensitivity coefficient table and cost coefficient table, and outputs the set of intermediate control quantities one by one according to the actuator channel.

[0118] S54. Apply rate of change constraints and magnitude constraints sequentially to the intermediate control variable set:

[0119] In the rate of change constraint, the difference between the control quantity of each actuator at the current sampling time and the previous sampling time is compared and limited to the maximum allowable rate of change of the actuator.

[0120] In amplitude constraints, the control quantity is projected using the upper and lower limits of the amplitude of each actuator as boundaries. When any actuator channel touches the boundary in any constraint, the intermediate control quantity of all actuator channels is scaled proportionally according to preset rules, and the boundary touch mark and scaling factor are recorded.

[0121] S55. Convert the set of control quantities into a set of executable control instructions, wherein the control instructions for the fan control unit are the target speed or duty cycle, the control instructions for the dynamic voltage and frequency adjustment unit are the target operating state or duty cycle, and the control instructions for the thermoelectric cooling unit are the target drive power.

[0122] This invention introduces a modal sensing allocator and a multi-constraint mechanism into the mapping process from modal control output to actuators, solving the problems of coarse actuator allocation, high energy consumption, and excessive action in existing technologies. By weighting and superimposing proportional, integral, and derivative terms, a complete modal control signal sequence is formed. Combined with the actuator's sensitivity coefficient and cost coefficient, the adjustment behavior of each actuator is matched with its contribution to the thermal mode and its energy consumption weight, thereby avoiding overload or ineffective action of a single actuator. The intermediate control quantity is dynamically corrected under rate of change and amplitude constraints, preventing oscillations and over-limit problems caused by excessively rapid adjustment. At the same time, the introduction of boundary touch flags and scaling factors achieves collaborative protection among actuators. The final generated instruction set can achieve division of labor and complementarity among different actuators, taking into account both fast response and energy consumption optimization. The innovation of this invention lies in the allocation strategy of modal sensing and multi-constraint fusion, which makes the temperature regulation process more stable, energy-saving, and controllable, significantly improving the robustness of the system and the service life of the actuators.

[0123] In this embodiment, S6 specifically includes:

[0124] S61. Read the set of control instructions and the set of actuator control quantities at the previous sampling time, and load the boundary touch flag and scaling factor;

[0125] S62. Set the minimum dwell count and maximum switching frequency for the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit, and convert the control command set into the target control quantity format, wherein the fan control unit adopts the duty cycle format, the dynamic voltage and frequency adjustment unit adopts the target operating state, and the thermoelectric cooling unit adopts the target drive power format.

[0126] S63. Execute slope constraints, compare the difference between the current target control quantity of each actuator and the control quantity at the previous sampling time, limit the difference to the maximum allowable rate of change of the actuator, and obtain the set of control quantities after slope constraints.

[0127] S64. Execute amplitude limiting, which limits the control quantity of each actuator to the corresponding upper and lower amplitude limits. When any actuator touches the boundary, the control quantities of all actuators are scaled proportionally according to the boundary touch flag and scaling factor according to the preset rules to form a set of executable control quantities.

[0128] S65. Send the set of executable control quantities to the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit. Perform minimum dwell time check on the dynamic voltage and frequency adjustment unit. Perform soft start and soft stop on the fan control unit and thermoelectric cooling unit. Collect execution feedback and update the set of actuator control quantities and mode dwell count at the previous sampling time. Detect faults and set fault flags. Output a set of degradation instructions.

[0129] Example 1:

[0130] To verify the feasibility of this invention in practice, it was applied to an information technology laboratory where researchers conducted a temperature management experiment on a 32-core server chip. When performing mixed workload tasks (parallel database queries and AI inference), the chip's peak power consumption approached 240 watts, and the internal temperature rose significantly. Traditional PID methods often suffer from large temperature overshoot, frequent fan switching, and high noise levels when dealing with such rapidly changing loads, severely impacting stability and energy efficiency under experimental conditions.

[0131] To address the aforementioned issues, researchers employed the chip dynamic temperature regulation method based on the PID control algorithm proposed in this invention. Twelve sensors deployed within the chip collect real-time temperature data. After filtering and data fusion, a multi-point temperature vector accurately reflecting local hotspots is obtained. This temperature vector is mapped to multiple thermal modal variables and compared with the target temperature value to generate a modal error signal.

[0132] The system calls a segmented multi-state PID controller, which dynamically switches between static steady-state, burst high-load, and continuous high-load modes based on the chip's operating state. When the temperature rise rate exceeds the threshold, the controller enters burst high-load mode, increasing the proportional and derivative actions for rapid response; when the temperature remains high, it enters continuous high-load mode, and the integrator is limited to prevent overshoot.

[0133] Each thermal mode variable is equipped with an independent integrator, and the integration process has limiting boundaries to avoid integration saturation. At the same time, fractional derivatives are used in the differentiation stage to smooth the trend of error signal changes and reduce the differential amplification effect caused by sensor noise.

[0134] The modal sensing distributor rationally allocates control signals to the fan control unit, dynamic voltage and frequency adjustment unit, and thermoelectric cooling unit based on the relationship between thermal modal variables and actuator sensitivity. Compared with traditional methods, this invention avoids overloading a single actuator and instead achieves coordinated adjustment of multiple actuators. For example, when hot spots appear in the chip edge area, the distributor appropriately increases the fan speed, while when the core area temperature rises rapidly, the dynamic voltage and frequency adjustment unit intervenes first to reduce power consumption.

[0135] The actuator control commands are sent to each actuator after slope constraint and amplitude limiting processing. In actual experiments, the fan speed was limited to between 2800 and 3500 rpm, the dynamic voltage frequency adjustment unit reduced its frequency by approximately 0.2 GHz during high load periods, and the thermoelectric cooling unit's power consumption did not exceed 75% of its rated value. After a 20-minute stress test, the chip temperature stabilized below 83°C with fluctuations less than ±2°C, noise decreased by approximately 5 dB, and system power consumption decreased by approximately 15 watts. The results demonstrate that this invention effectively improves energy efficiency and user experience while maintaining stable chip operation.

[0136] Table 1 Comparative experimental data of the method of the present invention and the traditional PID method

[0137]

[0138]

[0139] As shown in Table 1, under static and stable operating conditions, the average temperature maintained by traditional PID control is 72℃, with a temperature fluctuation range of ±3.0℃, a maximum overshoot of 5.0℃, an average fan speed of approximately 2800 rpm, a system power consumption of 230 watts, and a noise level of 44 dB. However, after adopting the method of this invention, the average temperature is reduced to 70℃, the fluctuation is reduced to ±1.2℃, the maximum overshoot is only 2.0℃, the fan speed decreases to 2500 rpm, the system power consumption decreases to 220 watts, and the noise level also decreases to 39 dB. This indicates that the present invention can provide smoother temperature control and lower energy consumption and noise under static and stable conditions.

[0140] In sudden high-load scenarios, traditional PID control causes the chip temperature to rise rapidly, averaging 91°C with a fluctuation range of ±4.2°C and a maximum overshoot of 8.0°C. This necessitates increasing the average fan speed to 4000 rpm, increasing power consumption to 245 watts, and generating 56 dB of noise. In contrast, the method of this invention controls the average temperature to 83°C, reduces temperature fluctuation to ±1.8°C, lowers the maximum overshoot to 3.2°C, reduces the fan speed to 3200 rpm, reduces system power consumption to 230 watts, and keeps noise levels below 50 dB. This demonstrates that when dealing with rapid load changes, this invention effectively suppresses temperature overshoot and excessive fluctuations, and reduces over-reliance on actuators.

[0141] Under sustained high load conditions, traditional PID control maintains an average temperature of 86℃ with a fluctuation range of ±3.8℃, a maximum overshoot of 7.0℃, a fan speed of 3700 rpm, system power consumption of 240 watts, and noise level of 54 dB. The method of this invention reduces the average temperature to 82℃, narrows the fluctuation range to ±1.6℃, controls the maximum overshoot at 2.9℃, reduces the fan speed to 3100 rpm, lowers system power consumption to 225 watts, and reduces noise to 48 dB. This demonstrates that the present invention can maintain stable temperature control even under prolonged high load operation, reducing energy consumption and noise accumulation, and extending the lifespan of the chip and actuator.

[0142] The method of this invention is superior to the traditional PID method in terms of average temperature control, temperature fluctuation suppression, overshoot reduction, actuator speed optimization, and energy consumption and noise reduction. It improves the accuracy and stability of chip dynamic temperature regulation and verifies the practicality and superiority of the method.

[0143] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic temperature regulation of a chip based on a PID control algorithm, characterized in that, Includes the following steps: S1. Collect real-time temperature data from multiple temperature sensors on the chip, process the real-time temperature data, and obtain a multi-point temperature vector. S2. Extract thermal modes from the multi-point temperature vector to form several thermal mode variables. Set a target temperature value and calculate the deviation of the thermal mode variables relative to the target temperature value to obtain the mode error signal. S3. Construct a segmented multi-state PID controller, which switches between static steady mode, sudden high load mode and continuous high load mode according to the chip's operating state. Call the proportional parameter corresponding to the current mode, perform proportional calculation on the modal error signal, and obtain the modal proportional term. S4. In the piecewise multi-mode PID controller, call the integral and derivative parameters corresponding to the current mode, configure an independent integrator for each thermal mode to generate a modal integral term, and perform fractional derivative processing on the rate of change of the modal error signal during the derivative operation to generate a modal derivative term; S5. Combine the modal proportional term, modal integral term, and modal differential term to generate a modal control output signal, and input the modal control output signal to the modal sensing distributor to generate corresponding actuator control commands based on the actuator's sensitivity to different thermal modes and constraint conditions. S6. The actuator control command is sent to the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit to implement amplitude limiting and slope constraint on the output process, thereby completing the dynamic adjustment of chip temperature.

2. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, The processing of real-time temperature data specifically includes: The real-time temperature data collected from the temperature sensor is denoised to remove abnormal fluctuations. The denoised real-time temperature data is filtered to suppress high-frequency interference signals; The filtered real-time temperature data is fused to form a unified multi-point temperature vector.

3. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, S2 specifically includes: S21. Read the multi-point temperature vector and extract the thermal modes of the multi-point temperature vector according to the pre-established and saved thermal mode basis vector set to obtain a number of thermal mode variables arranged in order. S22. The target temperature value is converted into a modal target value that corresponds one-to-one with the thermal modal variable according to the preset modal mapping rule, so as to obtain a modal target sequence arranged with the same number. S23. For each thermal modal variable, first calculate the initial difference between the thermal modal variable and the corresponding modal target value, and then perform the following processing in sequence: Within the hysteresis interval formed by the first deviation threshold and the second deviation threshold, the deviation is set to zero. When the deviation exceeds the hysteresis interval, asymmetric deduction is performed according to different thresholds for positive and negative directions. When an upward trend appears within several consecutive sampling periods and the trend intensity exceeds a preset trend threshold, a forward compensation amount is added to the deviation; when a downward trend appears and the trend intensity exceeds a preset trend threshold, a mitigation compensation amount is added to the deviation. The deviation is scaled based on the sensor consistency evaluation coefficient corresponding to the thermal modal variable; The amplitude of the scaled deviation is constrained according to preset upper and lower limits to obtain the modal deviation signal of the thermal modal variable; S24. Combine the modal deviation signals of all thermal modes in the order of the thermal modal variables to form a modal error signal sequence, and output the modal error signal sequence.

4. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, S3 specifically includes: S31. Obtain the modal error signal sequence, establish a one-to-one mapping relationship with the corresponding thermal modal variables according to their serial numbers, and use it as the input of the piecewise multi-mode PID controller; S32. Construct the structure of a segmented multi-state PID controller, the structure including a state manager, a parameter library, a parameter interpolator, a mode resident manager, a small-amplitude jitter suppressor, an abnormal switching protector, and a proportional calculator. The state manager determines the static steady-state mode, the sudden high-load mode, and the continuous high-load mode. The parameter library presets the proportional parameter base value for each thermal modal variable in the three modes. S33. The state manager calculates two state determination indices based on the modal error signal sequence. The error amplitude index is the largest absolute amplitude of all modal deviation signals at the same sampling time, and the error change rate index is the largest change amplitude of all modal deviation signals between adjacent sampling times. S34. Perform pattern determination in the following order: When the error change rate index reaches the error change rate threshold, it enters the sudden high load mode. When the error amplitude index continuously reaches the high threshold of error amplitude and the duration reaches the minimum dwell count, it enters the continuous high load mode. When the error amplitude index is lower than the low threshold of error amplitude and the error change rate index is lower than the threshold of error change rate, the system enters static stability mode. If none of the above conditions are met, the mode of the previous sampling time is maintained, and an upper and lower limit distinction is formed by setting a hysteresis threshold to limit frequent switching; S35. After the mode is determined, perform the scaling parameter generation process, specifically: The parameter library provides the proportional parameter base value corresponding to the current mode for each thermal mode variable, and the mode dwell manager records the sampling count since the most recent mode switch; The parameter interpolator performs time-based smooth interpolation between the scale parameter base value of the previous mode and the scale parameter base value of the current mode; When the modal deviation signal is in the preset dead zone, the small-amplitude jitter suppressor sets the proportional parameter adjustment coefficient to zero, and corrects the interpolated proportional parameter according to the sensor consistency evaluation coefficient to obtain the final proportional parameter of each thermal modal variable. S36. The proportional operator calculates the modal proportional term for each thermal modal variable according to the linear relationship between the final proportional parameter and the corresponding modal deviation signal, and combines them in the order of the thermal modal variables to form a sequence of modal proportional terms.

5. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, S4 specifically includes: S41. Obtain the modal error signal sequence, the modal proportional term sequence, and the integral and differential parameters corresponding to the current mode; S42. For each thermal modal variable, establish an independent integrator, set the initial integration value, upper integration limit, lower integration limit, integration step size and integration dead zone, and perform integration operation on the modal error signal of the thermal modal variable according to the sampling order to obtain the modal integral term; S43. During the integration process, perform integration anti-saturation processing on each thermal modal variable. When the accumulated modal integral term exceeds the integration upper limit, set the modal integral term to the integration upper limit. When the accumulated modal integral term is lower than the integration lower limit, set the modal integral term to the integration lower limit. Record the integration freeze flag. Within the sampling interval where the integration freeze flag is recorded, pause the integration accumulation of the thermal modal variable. S44. Calculate the rate of change of the modal error signal for each thermal modal variable. Normalize the rate of change according to the sampling period and set the rate of change dead zone and the rate of change hysteresis threshold. When the absolute value of the rate of change is within the rate of change dead zone, the rate of change is recorded as zero. When the absolute value of the rate of change crosses the rate of change hysteresis threshold, update the rate of change valid flag. S45. Select the preset fractional order based on the current mode, perform fractional differentiation on the rate of change sequence with effective rate of change indicators, and obtain the modal differential term of each thermal modal variable by combining the differential parameters. Record the modal differential term as zero for sampling points without effective rate of change indicators. S46. Combine the modal integral terms of all thermal modal variables in the order of their serial numbers to form a sequence of modal integral terms, and combine the modal differential terms of all thermal modal variables in the order of their serial numbers to form a sequence of modal differential terms.

6. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, S5 specifically includes: S51. Obtain the sequence of modal proportional terms, modal integral terms, and modal differential terms arranged by the sequence number of the thermal modal variables. Based on the pre-stored weighting coefficients and fixed combination order, perform weighted superposition on the three terms corresponding to each thermal modal variable to obtain the modal control output value. Combine the modal control output values ​​according to the sequence number to form the modal control output signal sequence. S52. Construct a modal sensing allocator, which includes a static parameter set and an operational parameter set. The static parameter set includes an actuator set, a table of sensitivity coefficients of modes and actuators, a table of cost coefficients of actuators, a table of upper and lower limits of actuator amplitude, and a table of maximum rate of change of actuators. The operational parameter set includes the actuator control quantity at the previous sampling time and the modal control output signal sequence at the current sampling time. S53. At each sampling time, the modal sensing distributor calculates the set of intermediate control quantities that minimizes the weighted sum of modal fitting error and actuator cost based on the modal control output signal sequence, sensitivity coefficient table and cost coefficient table, and outputs the set of intermediate control quantities one by one according to the actuator channel. S54. Apply rate of change constraints and magnitude constraints sequentially to the intermediate control variable set: In the rate of change constraint, the difference between the control quantity of each actuator at the current sampling time and the previous sampling time is compared and limited to the maximum rate of change allowed by the actuator. In amplitude constraints, the control quantity is projected using the upper and lower limits of the amplitude of each actuator as boundaries. When any actuator channel touches the boundary in any constraint, the intermediate control quantity of all actuator channels is scaled proportionally according to preset rules, and the boundary touch mark and scaling factor are recorded. S55. Convert the set of control quantities into a set of executable control instructions, wherein the control instructions for the fan control unit are the target speed or duty cycle, the control instructions for the dynamic voltage and frequency adjustment unit are the target operating state or duty cycle, and the control instructions for the thermoelectric cooling unit are the target drive power.

7. The chip dynamic temperature regulation method based on PID control algorithm according to claim 1, characterized in that, S6 specifically includes: S61. Read the set of control instructions and the set of actuator control quantities at the previous sampling time, and load the boundary touch flag and scaling factor; S62. Set the minimum dwell count and maximum switching frequency for the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit, and convert the control command set into the target control quantity format, wherein the fan control unit adopts the duty cycle format, the dynamic voltage and frequency adjustment unit adopts the target operating state, and the thermoelectric cooling unit adopts the target drive power format. S63. Execute slope constraints, compare the difference between the current target control quantity of each actuator and the control quantity at the previous sampling time, limit the difference to the maximum allowable rate of change of the actuator, and obtain the set of control quantities after slope constraints. S64. Execute amplitude limiting, which limits the control quantity of each actuator to the corresponding upper and lower amplitude limits. When any actuator touches the boundary, the control quantities of all actuators are scaled proportionally according to the boundary touch flag and scaling factor according to the preset rules to form a set of executable control quantities. S65. Send the set of executable control quantities to the fan control unit, dynamic voltage and frequency adjustment unit and thermoelectric cooling unit. Perform minimum dwell time check on the dynamic voltage and frequency adjustment unit. Perform soft start and soft stop on the fan control unit and thermoelectric cooling unit. Collect execution feedback and update the set of actuator control quantities and mode dwell count at the previous sampling time. Detect faults and set fault flags. Output a set of degradation instructions.

Images