CN115794584A - Transformer-based chip whole-chip temperature distribution prediction method, system and medium - Google Patents
Transformer-based chip whole-chip temperature distribution prediction method, system and medium Download PDFInfo
- Publication number
- CN115794584A CN115794584A CN202211487635.XA CN202211487635A CN115794584A CN 115794584 A CN115794584 A CN 115794584A CN 202211487635 A CN202211487635 A CN 202211487635A CN 115794584 A CN115794584 A CN 115794584A
- Authority
- CN
- China
- Prior art keywords
- temperature
- chip
- transformer
- module
- temperature distribution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 claims abstract description 37
- 238000012795 verification Methods 0.000 claims abstract description 13
- 238000004088 simulation Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000013508 migration Methods 0.000 claims description 5
- 230000005012 migration Effects 0.000 claims description 5
- 238000007477 logistic regression Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 abstract description 3
- 230000007246 mechanism Effects 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000008092 positive effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the field of chip temperature prediction, and discloses a method, a system and a medium for predicting the whole chip temperature distribution based on a Transformer, wherein the system comprises a time sequence information coding module, a Transformer neural network training module and a model accuracy verification module; the time sequence information coding module is used for acquiring the performance count and the temperature data of the temperature sensor and coding the time sequence data and the position information of the time sequence data and the temperature data; the Transformer neural network training module is used for predicting the future chip temperature according to the input sequence; and the model accuracy verification module is used for comparing the prediction result with the simulation result of the temperature simulator, outputting a temperature distribution prediction model if the prediction result is in accordance with expectation, and adjusting the training parameters to retrain if the prediction result is not in accordance with the expectation. The invention can predict the temperature of the whole chip and provide the temperature distribution of the chip at the future appointed time, and can assist the dynamic temperature management to actively control the temperature when the chip does not reach the temperature threshold value, thereby reducing the performance loss of forced reduction of the working frequency due to overtemperature.
Description
Technical Field
The invention belongs to the field of chip temperature prediction, and particularly relates to a method, a system and a medium for predicting the whole chip temperature distribution based on a Transformer.
Background
At present, in order to avoid the chip over-temperature, in the processor design, a temperature sensor or a program counter is generally arranged to sense the temperature of the running processor, and whenever the temperature of the processor reaches a temperature alarm value, the operating voltage and the operating frequency of the processor are passively reduced, so as to achieve the purpose of reducing the temperature of the processor. In order to dynamically adjust the power of a multi-core system, task migration and Dynamic Voltage Frequency Scaling (DVFS) are proposed. Based on these basic management actions, researchers have then proposed a number of Dynamic Thermal Management (DTM) methods that can control the temperature of the core while the system is running.
To control the chip temperature, processors typically control the core temperature by using hardware thermal management mechanisms such as Dynamic Voltage and Frequency Scaling (DVFS), clock gating, air cooling mechanisms, and modulating the duty cycle of the processor clock, thereby protecting the system from catastrophic failures. For example, intel processors maintain safe operating temperatures using Dynamic Thermal Management (DTM) solutions, such as thermal monitor 1 (TM 1), thermal monitor 2 (TM 2), on-demand clock modulation (ODCM), or enhanced intel SpeedStep technology. However, these mechanisms are reactive, controlling temperature with either a scaled frequency/voltage or a modulated duty cycle, so the DTM mechanism has a significant impact on the performance of the application.
To improve the effect of DTM, accurate on-line chip thermal distribution prediction is imperative for DTM. However, some research based on the prior art focuses on chip thermal profile estimation, and most thermal estimation methods require accurate power information. In addition, some existing temperature prediction methods can only realize temperature prediction of local hot spots, and cannot realize chip thermal distribution prediction. As for the chip thermal distribution prediction method which does not require power information, there is almost no. Recently, many thermal modeling and estimation methods have been applied to architecture-level thermal estimation. These methods are typically performed off-line to obtain the thermal profile of the chip, although relatively accurate, the computational overhead is unacceptable if used on-line. Therefore, under non-ideal conditions, i.e., less accurate thermal models and power estimation, accurate and efficient estimation and prediction of on-chip thermal profiles is very important for dynamic thermal management.
Through the above analysis, the problems and defects of the prior art are as follows:
(1) The conventional temperature prediction method can only realize the temperature prediction of local hot spots and cannot realize the chip heat distribution prediction. There are few chip thermal distribution prediction methods that do not require power information.
(2) The computational overhead of the existing thermal modeling and estimation methods is very large for online use.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method, a system and a medium for predicting the whole chip temperature distribution based on a Transformer.
The invention is realized in such a way that a Transformer-based chip whole temperature distribution prediction method comprises the following steps:
the method comprises the steps of obtaining performance counts and temperature sensor temperature data from a chip, conducting temperature sequence coding, then training by using a Transformer neural network, predicting the whole chip temperature distribution of the chip by the Transformer neural network according to input information, and supplying the predicted information to a dynamic thermal management system (DTM) to make an active decision.
Further, the method for predicting the temperature distribution of the whole chip based on the Transformer comprises the following steps: the system comprises a time sequence information coding module, a Transformer neural network training module and a model accuracy verification module;
the time sequence information coding module is used for acquiring the performance count and the temperature data of the temperature sensor and coding the time sequence data and the position information of the time sequence data and the temperature data;
the Transformer neural network training module is used for predicting the future chip temperature according to an input sequence;
and the model accuracy verification module is used for comparing the prediction result with the simulation result of the temperature simulator, outputting the temperature distribution prediction model if the prediction result is in accordance with the expectation, and otherwise, adjusting the training parameters for retraining.
Further, the time sequence information coding module acquires performance counting frequency data of each functional module of the current chip at each time point and temperature data of a temperature sensor in a specified time period, codes information according to time sequence information, codes positions according to position information, adds the information codes and the position codes and sends the information codes and the position codes as an input sequence of a model to the transform neural network training module;
the position code is used for generating a vector containing absolute position information and relative position information of the temperature sensor and the functional module according to the position of the temperature sensor and the position of the functional module.
The transform neural network training module comprises an encoder and a decoder, after the vector representation matrix of the input sequence is obtained, the vector representation matrix of the input sequence is transmitted into the encoder, the encoded information matrix is obtained after the vector representation matrix passes through six encoder units, the encoded information matrix finally output by the encoder is transmitted into the decoder, and the prediction result of the temperature of the whole chip lattice point of the chip in a certain period of time in the future is obtained through a linear layer and a Softmax logistic regression layer after the encoded information matrix passes through six decoding units;
the encoder unit includes: multi-head-orientation, residual join and layer normalization (Add & norm), feed forward layer, and residual join and layer normalization (Add & norm);
the decoder unit includes: mask Multi-head-orientation, residual join and layer normalization (Add & norm), feed forward layer, and residual join and layer normalization (Add & norm).
Further, the model accuracy verification module compares temperature simulation data calculated by a temperature simulator according to a temperature prediction result of the Transformer neural network training module, and judges whether a loss function value exceeds a pre-threshold value or not; if so, optimizing model parameters in a time sequence information coding module to be trained and a Transformer neural network training module, returning to the step of the time sequence information coding module, coding time sequence information and coding position again, and sending the time sequence information and the position to the Transformer neural network training module for retraining; if not, obtaining the trained temperature prediction model.
Further, the Transformer neural network training process comprises:
firstly, acquiring performance counting data of each functional module of a current chip at each time point and temperature data of each temperature sensor according to a set time period so as to generate an input vector, generating position codes according to different positions of the temperature sensors and the functional modules, and adding the position codes to obtain a vector representation matrix of an input sequence;
after the input of the full connection layer is adjusted to be the dimensionality required by the Transformer, the Transformer predicts the temperature of a certain period of time in the future according to the input information;
and judging whether the predicted value is in accordance with the expectation or not by calculating a loss function of the predicted value, outputting a temperature distribution prediction model if the predicted value is in accordance with the expectation, otherwise, adjusting the training parameters and re-training.
Further, the specific process of the DTM for making the active decision is as follows:
after the temperature distribution prediction model is obtained, the performance count and the sensor temperature data in a past period of time are input into the temperature distribution prediction model to obtain a temperature prediction result in a future period of time, and a dynamic thermal management system (DTM) is supplied to make an active decision, perform task migration on cores with overhigh temperature, and transfer the cores to cores with lower temperature.
Another object of the present invention is to provide a transform-based system for predicting temperature distribution of an entire chip by implementing the transform-based method for predicting temperature distribution of an entire chip, the transform-based system for predicting temperature distribution of an entire chip comprising:
the time sequence information coding unit is used for acquiring the performance count and the temperature data of the temperature sensor and coding the time sequence data and the position information of the time sequence data and the temperature data;
the Transformer neural network training unit is used for predicting the future chip temperature according to the input sequence;
and the model accuracy verification unit is used for comparing the prediction result with the simulation result of the temperature simulator, outputting the temperature distribution prediction model if the prediction result is in accordance with the expectation, and otherwise, adjusting the training parameters for retraining.
Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the steps of the transform-based chip whole temperature distribution prediction method.
Another object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the processor is enabled to execute the steps of the transform-based chip whole temperature distribution prediction method.
Another object of the present invention is to provide an information data processing terminal, where the information data processing terminal is configured to implement the transform-based system for predicting temperature distribution of a whole chip.
By combining the technical scheme and the technical problem to be solved, the technical scheme to be protected by the invention has the advantages and positive effects that:
first, the existing temperature prediction technology has the problems that only local hot spots can be predicted and the online calculation cost is large, so that the difficulty in online prediction of the whole chip temperature is large. The specific description is as follows:
the invention provides a whole chip temperature distribution prediction method based on a Transformer. According to the method, performance counts of all functional modules of a chip and temperature data of a temperature sensor are obtained, time sequence and position information of the performance counts and the temperature data are coded and then sent to a transform neural network for training, so that the future whole temperature of the chip is predicted, and then the predicted information is supplied to a dynamic thermal management system (DTM) to make an active decision, so that the running reliability of a multi-core system is ensured.
Secondly, compared with the prior art, the method solves the problem of predicting the whole chip temperature, has the advantages of low calculation cost, settable prediction time and capability of reducing the performance loss in the dynamic temperature management of the chip, and is specifically described as follows:
the invention can predict the temperature of the whole chip and give the temperature distribution of the chip at the future designated time. Meanwhile, the dynamic temperature management can be assisted to actively control the temperature when the chip does not reach the temperature threshold, and the performance loss of forced reduction of the working frequency due to overtemperature is reduced.
Third, as an inventive supplementary proof of the claims of the present invention, there are also presented several important aspects:
(1) The expected income and commercial value after the technical scheme of the invention is converted are as follows:
compared with a temperature control method for passively reducing the working voltage and the working frequency of the processor to reduce the temperature of the processor, the temperature control method can actively make a temperature management decision according to the possible future over-temperature condition, reduce the performance loss of the chip in operation, improve the working efficiency of the chip, and has high commercial value and wide application scenes.
(2) The technical scheme of the invention solves the technical problem that people are eagerly to solve but can not be successfully solved all the time:
in order to provide a thermal prediction algorithm with high prediction precision, the invention obtains the performance count of each functional module and the temperature data of the temperature sensor from the chip, and then trains through a Transformer neural network to predict the whole temperature distribution of the chip in a certain period of time in the future, thereby improving the temperature prediction precision, helping DTM to better perform dynamic thermal management and solving the technical problem existing in the prior method.
Drawings
FIG. 1 is a process diagram of a transform-based method for predicting temperature distribution of a whole chip according to an embodiment of the present invention;
FIG. 2 is a diagram of a transform neural network according to an embodiment of the present invention;
FIG. 3 is a flowchart of the training of the Transformer neural network according to the embodiment of the present invention;
fig. 4 is a schematic diagram of a DTM decision process assisted by a temperature prediction result according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.
This section is an explanatory embodiment expanding on the claims so as to fully understand how the present invention is embodied by those skilled in the art.
As shown in fig. 1, the method for predicting temperature distribution of a whole chip based on a transform according to an embodiment of the present invention includes: the system comprises a time sequence information coding module, a Transformer neural network training module and a model accuracy verification module;
the time sequence information coding module is used for acquiring performance counting of the functional module and temperature data of the sensor and coding the time sequence data and the position information of the functional module; the transform neural network training module is used for predicting the future chip temperature according to the input sequence; and the model accuracy verification module is used for comparing the prediction result with the simulation result of the temperature simulator, outputting the temperature model if the prediction result is in line with the simulation result of the temperature simulator, and otherwise, adjusting the training parameters for retraining.
Further, the time sequence information coding module is used for acquiring performance counting frequency data of each functional module of the current chip at each time point and temperature data of each temperature sensor in a specified time period, coding the data according to the time sequence information, coding the position according to the position information, adding the data and the position information as the input of the model, and sending the input into the subsequent transform neural network training module.
Further, after the vector representation matrix of the input sequence is obtained by the transform neural network training module, the vector representation matrix of the input sequence is transmitted into an encoder, and an encoding information matrix can be obtained after the vector representation matrix passes through six encoder units. And transmitting the encoding information matrix finally output by the encoder to a decoder, and obtaining the prediction result of the temperature of the whole chip lattice point of the chip in a certain period of time in the future through a linear layer and a Softmax logistic regression layer after passing through six decoding units. The structure of the Transformer neural network is shown in fig. 2.
Further, the model accuracy verification module compares temperature simulation data calculated by a temperature simulator according to a temperature prediction result of the Transformer neural network training module, and judges whether a loss function value exceeds a pre-threshold value or not; if so, optimizing model parameters in a time sequence information coding module to be trained and a Transformer neural network training module, returning to the step of the time sequence information coding module, coding time sequence information again, and sending the time sequence information to the Transformer neural network training module for retraining; if not, obtaining the trained temperature prediction model.
FIG. 3 is a transform neural network training process. The method comprises the steps of firstly, collecting performance counting data of each function module of a current chip at each time point and temperature data of each temperature sensor according to a set time period so as to generate an input vector, generating position codes according to different positions of the temperature sensors and the function modules, and adding to obtain a vector representation matrix of an input sequence. After the fully-connected layer adjusts the input to the dimensions required by the Transformer, the Transformer will predict the temperature for some period of time in the future based on the input information. And (4) judging whether the predicted value accords with the expectation or not by calculating a loss function of the predicted value, outputting the model if the predicted value accords with the expectation, otherwise, adjusting the training parameters and re-training.
Fig. 4 is a temperature prediction result assisted DTM decision process. After a Transformer temperature distribution prediction model is obtained, the performance count and sensor temperature information in a past period of time are input into the temperature distribution prediction model to obtain a temperature prediction result in a future period of time, and the temperature prediction result is supplied to a dynamic thermal management system (DTM) to make an active decision, and the steps are repeated in such a way, so that accurate on-line chip heat distribution prediction is provided for the DTM decision, task migration is performed on a core with overhigh temperature, and the core with lower temperature is transferred to the core with lower temperature, and therefore performance loss is avoided.
In order to prove the creativity and the technical value of the technical scheme of the invention, the part is the application example of the technical scheme of the claims on specific products or related technologies.
The invention can be applied to the chip temperature prediction and control under a multi-core system, such as a server, a personal computer, a tablet computer, a mobile phone and other mobile equipment. By utilizing the invention, the application equipment can improve the working efficiency, stabilize the working temperature and reduce the online temperature prediction cost.
It should be noted that embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portions may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or specially designed hardware. Those skilled in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such code being provided on a carrier medium such as a disk, CD-or DVD-ROM, programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier, for example. The apparatus and its modules of the present invention may be implemented by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of hardware circuits and software, e.g., firmware.
The embodiment of the invention has some positive effects in the process of research and development or use, and indeed has great advantages compared with the prior art, and the following contents are described by combining data, charts and the like in the test process.
The experiment employed a 16-core system, all connected via one network on chip. DTM is performed in one of the cores, and temperature information collection and transmission of the calculated recommended power for all cores are operated by the NoC. The chips were divided into a 20x20 grid, i.e., each chip had 25 grid points. At the very beginning of the control, the ambient temperature was set at 20 ℃ and the target temperature at 70 ℃. The temperature range for the rise was set from 20 ℃ to 70 ℃. And setting the temperature of the whole chip within 1s in the future. All experiments were performed on a PC equipped with an Intel i7-8750H CPU and 16GB memory.
The existing DTM works at a high power when the temperature is low until the temperature reaches a temperature threshold, a system is triggered to reduce the frequency so as to prevent the chip from overtemperature, and the working frequency fluctuation is relatively large in a set temperature threshold range. The method for predicting the whole chip temperature distribution based on the Transformer helps the DTM to adjust the power frequency or perform task migration in advance when the temperature is about to reach the temperature threshold in the future 1s, and further the working frequency fluctuation is relatively small in the range of the set temperature threshold. Compared with the two methods, the traditional DTM forced frequency reduction method can effectively improve the energy efficiency ratio of the chip by comparing the two methods, and the temperature of the chip can be more stably in the range of the set temperature threshold.
The above description is only for the purpose of illustrating the embodiments of the present invention, and the scope of the present invention should not be limited thereto, and any modifications, equivalents and improvements made by those skilled in the art within the technical scope of the present invention as disclosed in the present invention should be covered by the scope of the present invention.
Claims (10)
1. A whole chip temperature distribution prediction method based on a Transformer is characterized by comprising the following steps:
acquiring performance counts and temperature data of a temperature sensor from a chip, carrying out temperature sequence coding, and then training by using a Transformer neural network, wherein the Transformer neural network predicts the temperature distribution of the whole chip according to input information and supplies the predicted information to a dynamic thermal management system (DTM) to make an active decision.
2. The Transformer-based chip whole temperature distribution prediction method is characterized by comprising a time sequence information coding module, a Transformer neural network training module and a model accuracy verification module;
the time sequence information coding module is used for acquiring the performance count and the temperature data of the temperature sensor and coding the time sequence data and the position information of the time sequence data and the temperature data;
the transform neural network training module is used for predicting the future chip temperature according to the input sequence;
and the model accuracy verification module is used for comparing the prediction result with the simulation result of the temperature simulator, outputting the temperature distribution prediction model if the prediction result is in accordance with the expectation, and adjusting the training parameters to retrain if the prediction result is in accordance with the expectation.
3. The Transformer-based chip whole temperature distribution prediction method as claimed in claim 1, wherein the timing information coding module obtains performance counting frequency data of each functional module of a current chip at each time point and temperature data of a temperature sensor in a specified time period, performs information coding according to time sequence information, performs position coding according to position information, and feeds the sum of the information coding and the position coding into the Transformer neural network training module as an input sequence of a model;
the position code is a vector which is generated according to the position of the temperature sensor and the position of the functional module and contains absolute position information and relative position information of the temperature sensor and the functional module.
4. The transform-based chip whole-chip temperature distribution prediction method is characterized in that the transform neural network training module comprises an encoder and a decoder, after a vector representation matrix of the input sequence is obtained, the vector representation matrix of the input sequence is transmitted into the encoder, after six encoder units are passed, an encoding information matrix is obtained, the encoding information matrix finally output by the encoder is transmitted into the decoder, after six decoding units are passed, a linear layer and a Softmax logistic regression layer are further used for obtaining a prediction result of the chip whole-chip lattice point temperature in a certain period in the future.
5. The Transformer-based chip whole-chip temperature distribution prediction method of claim 1, wherein the model accuracy verification module compares temperature simulation data calculated by a temperature simulator according to a temperature prediction result of the Transformer neural network training module to judge whether a loss function value exceeds a pre-threshold value; if so, optimizing model parameters in a time sequence information coding module to be trained and a Transformer neural network training module, returning to the step of the time sequence information coding module, coding time sequence information and coding position again, and sending the time sequence information and the position to the Transformer neural network training module for retraining; if not, obtaining the trained temperature prediction model.
6. The Transformer-based chip whole-chip temperature distribution prediction method of claim 1, wherein the Transformer neural network training process comprises:
firstly, acquiring performance counting data of each functional module of a current chip at each time point and temperature data of each temperature sensor according to a set time period so as to generate an input vector, generating position codes according to different positions of the temperature sensors and the functional modules, and adding to obtain a vector representation matrix of an input sequence;
after the input of the full connection layer is adjusted to the dimension required by the Transformer, the Transformer predicts the temperature of a certain period of time in the future according to the input information;
and judging whether the predicted value is in accordance with the expectation or not by calculating a loss function of the predicted value, outputting a temperature distribution prediction model if the predicted value is in accordance with the expectation, otherwise, adjusting the training parameters and re-training.
7. The method for predicting the whole chip temperature distribution based on the Transformer as claimed in claim 1, wherein the specific process of the DTM for making the active decision is as follows:
after the temperature distribution prediction model is obtained, the performance count and the sensor temperature data in a past period of time are input into the temperature distribution prediction model to obtain a temperature prediction result in a future period of time, and a dynamic thermal management system (DTM) is supplied to make an active decision, and task migration is performed on a core with an overhigh temperature, and the core with a lower temperature is transferred to the core with the overhigh temperature.
8. A Transformer-based system for predicting temperature distribution of a chip on board according to any one of claims 1-7, wherein the Transformer-based system for predicting temperature distribution of a chip on board comprises:
the time sequence information coding unit is used for acquiring the performance count and the temperature data of the temperature sensor and coding the time sequence data and the position information of the time sequence data and the temperature data;
the Transformer neural network training unit is used for predicting the future chip temperature according to the input sequence;
and the model accuracy verification unit is used for comparing the prediction result with the simulation result of the temperature simulator, outputting the temperature distribution prediction model if the prediction result is in accordance with the expectation, and otherwise, adjusting the training parameters for retraining.
9. A computer device, characterized in that the computer device comprises a memory and a processor, the memory storing a computer program, the computer program, when executed by the processor, causing the processor to perform the steps of the Transformer-based chip whole temperature distribution prediction method according to any one of claims 1-7.
10. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the transform-based chip full-scale temperature distribution prediction method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211487635.XA CN115794584A (en) | 2022-11-24 | 2022-11-24 | Transformer-based chip whole-chip temperature distribution prediction method, system and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211487635.XA CN115794584A (en) | 2022-11-24 | 2022-11-24 | Transformer-based chip whole-chip temperature distribution prediction method, system and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115794584A true CN115794584A (en) | 2023-03-14 |
Family
ID=85441338
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211487635.XA Pending CN115794584A (en) | 2022-11-24 | 2022-11-24 | Transformer-based chip whole-chip temperature distribution prediction method, system and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115794584A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116820896A (en) * | 2023-08-30 | 2023-09-29 | 南京南瑞信息通信科技有限公司 | Physical signal-based non-invasive industrial control terminal abnormality detection method |
| CN116933608A (en) * | 2023-09-15 | 2023-10-24 | 深圳市正和兴电子有限公司 | Storage chip management method, system and storage medium |
| CN117055658A (en) * | 2023-10-11 | 2023-11-14 | 江苏徕德生物医药有限公司 | Self-adaptive temperature control system and method for tiarelvone crystallization production process |
| US12026550B1 (en) * | 2023-07-27 | 2024-07-02 | Auradine, Inc. | Cryptomining heat management |
-
2022
- 2022-11-24 CN CN202211487635.XA patent/CN115794584A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12026550B1 (en) * | 2023-07-27 | 2024-07-02 | Auradine, Inc. | Cryptomining heat management |
| CN116820896A (en) * | 2023-08-30 | 2023-09-29 | 南京南瑞信息通信科技有限公司 | Physical signal-based non-invasive industrial control terminal abnormality detection method |
| CN116820896B (en) * | 2023-08-30 | 2023-11-24 | 南京南瑞信息通信科技有限公司 | Physical signal-based non-invasive industrial control terminal abnormality detection method |
| CN116933608A (en) * | 2023-09-15 | 2023-10-24 | 深圳市正和兴电子有限公司 | Storage chip management method, system and storage medium |
| CN116933608B (en) * | 2023-09-15 | 2023-12-22 | 深圳市正和兴电子有限公司 | Storage chip management method, system and storage medium |
| CN117055658A (en) * | 2023-10-11 | 2023-11-14 | 江苏徕德生物医药有限公司 | Self-adaptive temperature control system and method for tiarelvone crystallization production process |
| CN117055658B (en) * | 2023-10-11 | 2024-01-02 | 江苏徕德生物医药有限公司 | Self-adaptive temperature control system and method for tiarelvone crystallization production process |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115794584A (en) | Transformer-based chip whole-chip temperature distribution prediction method, system and medium | |
| Yuan et al. | Parameter identification of PEMFC based on Convolutional neural network optimized by balanced deer hunting optimization algorithm | |
| CN115759658B (en) | Enterprise energy consumption data management system suitable for smart city | |
| Hu et al. | Improved multistep ahead photovoltaic power prediction model based on LSTM and self-attention with weather forecast data | |
| Bal et al. | Spikingbert: Distilling bert to train spiking language models using implicit differentiation | |
| Zha et al. | A wind speed vector-wind power curve modeling method based on data denoising algorithm and the improved transformer | |
| CN116187548A (en) | Photovoltaic power generation power prediction method and device, storage medium and electronic device | |
| CN114548571A (en) | Photovoltaic power generation prediction method and device based on Bayesian ridge regression | |
| CN117612053A (en) | Asphalt waterproof coiled material processing and preparing system and method thereof | |
| He et al. | Resilient operation of DC microgrid against FDI attack: A GRU based framework | |
| Xu et al. | Cement rotary kiln temperature prediction based on time-delay calculation and residual network and bidirectional novel gated recurrent unit multi-model fusion | |
| Zhai et al. | A degradation fault prognostic method of radar transmitter combining multivariate long short-term memory network and multivariate Gaussian distribution | |
| Zhou et al. | An accurate ensemble forecasting approach for highly dynamic cloud workload with vmd and r-transformer | |
| CN117744855A (en) | Load prediction system and method based on machine learning | |
| CN116522785A (en) | Electromagnetic transient simulation equation set solving method and device based on quantum computation | |
| Wei et al. | A short‐term electricity consumption forecasting approach based on feature processing and hybrid modelling | |
| Shang et al. | Effinformer: A deep-learning-based data-driven modeling of DC–DC bidirectional converters | |
| Wang et al. | Convolutional preprocessing Transformer-based fault diagnosis for rectifier-filter circuits in nuclear power plants | |
| Li et al. | Medium‐term load forecasting of power system based on BiLSTM and parallel feature extraction network | |
| Yang et al. | A novel ultra short-term wind power prediction model based on double model coordination switching mechanism | |
| CN115270383A (en) | Temperature prediction method and device, storage medium and computer equipment | |
| Ma et al. | Fmamba: Mamba based on fast-attention for multivariate time-series forecasting | |
| Xuan et al. | An Enhanced Bidirectional Transformer Model with Temporal-Aware Self-Attention for Short-Term Load Forecasting | |
| Fu et al. | A non-stationary transformer-based remaining useful life prediction method for proton exchange membrane fuel cells | |
| Gao et al. | Distillation enhanced time series forecasting network with momentum contrastive learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |