CN109902807A - A kind of hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network - Google Patents
A kind of hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network Download PDFInfo
- Publication number
- CN109902807A CN109902807A CN201910148729.6A CN201910148729A CN109902807A CN 109902807 A CN109902807 A CN 109902807A CN 201910148729 A CN201910148729 A CN 201910148729A CN 109902807 A CN109902807 A CN 109902807A
- Authority
- CN
- China
- Prior art keywords
- core
- many
- neural network
- recognition
- recurrent neural
- 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.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 36
- 230000000306 recurrent effect Effects 0.000 title claims abstract description 32
- 238000000034 method Methods 0.000 title claims abstract description 16
- 238000009826 distribution Methods 0.000 title claims abstract description 15
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000004088 simulation Methods 0.000 claims 1
- 238000003062 neural network model Methods 0.000 abstract description 4
- 238000004364 calculation method Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 4
- 238000010200 validation analysis Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000009022 nonlinear effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Abstract
The invention belongs to field of electron design automation, disclose a kind of hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network.Dynamic Thermal management can effectively manage the temperature of many-core chip, and a good many-core chip heat modeling can assist Dynamic Thermal management well.However traditional many-core chip lump type heat modeling, with the increase of chip core calculation, its computing cost is exponentially increased.In order to solve the problems, such as that lump type thermal model computing cost is excessive, the hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network that the invention proposes a kind of, it is using each core of chip as individual computing unit, Recognition with Recurrent Neural Network model is established, carries out limited data exchange between core and core.The present invention can simulate the temperature characterisitic of many-core chip with quickish speed and very high precision.
Description
Technical field
The invention belongs to field of electron design automation, are related to depth learning technology field, in particular to a kind of to be based on following
The hot modeling method of many-core chip distribution formula of ring neural network.
Background technique
With the progress of semiconductor technology, chip feature sizes persistently reduce, and by 2018, the commercial chip of 7nm was
Start volume production.Into after nanometer scale, due to the influence of leakage current, the dominant frequency of chip is difficult to improve again, therefore the high property of chip
Energy developing direction is replaced improving dominant frequency by increase chip core calculation, and obtains remarkable effect.
The performance of this many-core chip is greatly improved because of the increase of core number, but also brings simultaneously serious
Chip thermal reliability problem, occur the main reason for this problem be power density greatly caused by chip temperature it is excessively high.
In order to solve the problems, such as that many-core chip thermal reliability, Dynamic Thermal manage this effective and lower-cost scheme and mentioned
Out.This scheme is based on Theory of Automatic Control, by the accurate estimation and adjustment in real time to power consumption, to obtain ideal temperature
Degree distribution.Especially when core number is less, Dynamic Thermal administrative skill can ensure that temperature obtains with lower performance cost
Effective management.However, core piece core number is excessive in the presence of all, lump type thermal model is oversized, and computing cost is with chip
The increase of core number is exponentially increased, and which results in heat management bring processor performance expense itself is excessive.
In view of the above problems, proposing that a kind of many-core chip distribution formula heat modeling is current one of problem anxious to be resolved.
Summary of the invention
In order to solve the problems in above-mentioned technology, the present invention provides a kind of, and the many-core chip based on Recognition with Recurrent Neural Network divides
The hot modeling method of cloth.Many-core chip thermal model is decomposed into several mini Mods, more extreme example by this modeling method
A thermal model exactly is established for each core of many-core chip, limited information exchange is carried out between core and core.This method is first
Recognition with Recurrent Neural Network model is first built, network is then trained by offline temperature and power data.The circulation that training is completed
Neural network can predict the temperature of each core on chip.
The present invention is solved the above problems using following technical scheme:
Step 1 extracts many-core chip thermal model parameters, thermal capacitance and thermal resistance on mainly entire chip from Hotspot
Parameter establishes many-core chip thermal model.
Step 2, obtaining multi-group data using these thermal models, (multiple timing nodes, each node have each core
Power and temperature information), this multi-group data is then made into training set and verifying collects, wherein training set will be used to train circulation
Neural network, verifying collection only carry out verifying trained neural network, and verifying collection data are not used in training.
Step 3, training set is sent into also untrained Recognition with Recurrent Neural Network model, and (weight matrix of the inside is initial at random
Change), the output of Recognition with Recurrent Neural Network can be obtained.Because Recognition with Recurrent Neural Network is not trained to also, Recognition with Recurrent Neural Network
There are also biggish gaps for output and true output.An accurate Recognition with Recurrent Neural Network model in order to obtain, can pass through tune
Whole weight matrix makes its temperature output as close possible to the output of training set.So target reforms into, by adjusting following
The weight matrix of ring neural network minimizes loss function, and loss function is smaller, the output of Recognition with Recurrent Neural Network and true
Output is just closer.
Step 4 takes gradient optimization algorithm to optimize loss function, sets learning rate and calculates separately loss letter
Then the local derviation of the weight matrix of several pairs of Recognition with Recurrent Neural Network is iterated update to them.By successive ignition, until loss
Function no longer reduces or reaches the maximum number of iterations of our settings, means that trained completion, records loss function value at this time,
That is training error.Meanwhile also verifying collection can be sent into trained Recognition with Recurrent Neural Network, on verifying collection, iteration is not updated, only
Record its loss function value, i.e. validation error.Then change the number of plies of hidden layer and the number of hidden layer neuron, again
Training new model, records its training error and validation error.Finally, from these models, select validation error it is the smallest that,
Thermal model as chip core.For each core on many-core chip, a thermal model will be trained for it, combined,
It is many-core chip distribution formula thermal model.
Compared with prior art, beneficial outcomes of the invention are: Recognition with Recurrent Neural Network can effectively fit non-linear letter
Number enables the many-core chip distribution formula thermal model established using it with very high accuracy rate and cracking response speed come mould
The temperature characterisitic of quasi- many-core chip.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 be 16 core pieces layout, number and the 6th core the location drawing.
Fig. 2 is the thermal model structure of the 6th core and the location diagram of core adjacent with other.
Fig. 3 is Recognition with Recurrent Neural Network structure chart, its loop structure is from output layer to hidden layer.
Fig. 4 is the comparison figure of the 6th nuclear heat model prediction temperature value and true temperature value.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with the attached drawing in present example,
Technical solution in present example is clearly and completely described, it is clear that described example is that a part of the invention is real
Example, rather than whole examples.Based on the example in the present invention, those of ordinary skill in the art are not making creative work
Under the premise of every other example obtained, shall fall within the protection scope of the present invention.
Fig. 1 is the layout no of 16 core pieces and the location drawing of the 6th core.
In present example, the many-core chip of 16 cores is provided, number is as shown in Figure 1, entire many-core chip
Distributed heat model will build a model to each core, then combine, be exactly the distributed hot-die of entire many-core chip
Type.Here the 6th core color is prominent, will explain the process for establishing model using the 6th core as example.
Fig. 2 is the thermal model structure of the 6th core and the location diagram of core adjacent with other.
The thermal model structure of the 6th core and the positional relationship of core adjacent with other are depicted, here it and the 2nd, 5,7,10
Nuclear phase is adjacent, is connected with thermal resistance；There are also ground connection thermal capacitances；There are also externally input power.
Fig. 3 is Recognition with Recurrent Neural Network structure chart, it is contemplated that nonlinear effect, the present invention are built with neural network to each core
Vertical thermal model, but because temperature has a series of value in time, sequence vector is constituted, and Recognition with Recurrent Neural Network is one special
It is engaged in the neural network of Series Modeling, it can handle the data of sequence vector form well, so final use is followed here
Ring neural network to each core establishes thermal model.Here, it inputs as Pi(k) and Ti_near(k), the i-th of k moment is respectively indicated
The power of core and temperature with the core of the i-th nuclear phase neighbour, state Hi(k) hidden layer, table are also designated as in Recognition with Recurrent Neural Network
Show the state of the i-th core of k moment；Export Ti(k) temperature of the i-th core of k moment is indicated.WihIt is weight square of the input layer to hidden layer
Battle array, WhoIt is weight matrix of the hidden layer to output layer, WohIt is weight matrix of the output layer to hidden layer.In order to make circulation nerve
Network preferably fitting function, hidden layer can have multilayer, show the circulation nerve containing only a hidden layer here
Network.
Fig. 4 is the comparison figure of the 6th nuclear heat model prediction temperature value and true temperature value.
The thermal model based on Recognition with Recurrent Neural Network for training the 6th core come of the invention is used to predict the temperature of the 6th core
Degree, it can be seen that the temperature of model prediction can be fitted true temperature well.
The invention discloses a kind of the many-core chip distribution formula thermal model method based on Recognition with Recurrent Neural Network, above example pair
The present invention is described in detail, but be not limited to that this, it later still can be to technical side documented by example before
Case is modified, this can't make the spirit and scope of each case technology scheme of the essence disengaging present invention of corresponding technical solution.
Claims (5)
1. a kind of hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network, it is characterised in that: every to many-core chip
A core carries out hot modeling；The thermal model of each core is established using Recognition with Recurrent Neural Network；To each core of many-core chip, around
The selection mode of temperature is fixed to reduce error.
2. the many-core chip distribution formula hot modeling method according to claim 1 based on Recognition with Recurrent Neural Network, feature exist
In: hot modeling is carried out to each core of many-core chip, the position of chip core is different, and heat modeling is also different；Given chip core
Position, its temperature can come out according to the power of itself and the temperature computation of surrounding core.
3. the many-core chip distribution formula hot modeling method according to claim 1 based on Recognition with Recurrent Neural Network, feature exist
In: it can use the fine simulation nonlinear function of Recognition with Recurrent Neural Network；It can use Recognition with Recurrent Neural Network processing sequence vector
Data；The input layer of Recognition with Recurrent Neural Network has carried out weight dismantling, to solve the problems, such as that the existing power of input has temperature again.
4. the many-core chip distribution formula hot modeling method according to claim 1 based on Recognition with Recurrent Neural Network, feature exist
In: to each core of many-core chip, the selection mode of ambient temperature is fixed to reduce error, not in accordance with core number from small
Longer spread is arrived, but surface is first, is then rotated clockwise, the temperature until taking all adjacent cores exists in this way
When handling outermost core, just do not allow error-prone, because they are adjacent with external environment, and there are also areas for adjacent position
Not.
5. the many-core chip distribution formula hot modeling method according to claim 1 based on Recognition with Recurrent Neural Network, feature exist
In: the input layer of Recognition with Recurrent Neural Network has carried out weight dismantling, to solve the temperature that the existing power of input has temperature again, defeated
Enter weight matrix and split into two parts, a part gives power, and another part is to temperature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910148729.6A CN109902807B (en) | 2019-02-27 | 2019-02-27 | Many-core chip distributed thermal modeling method based on recurrent neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910148729.6A CN109902807B (en) | 2019-02-27 | 2019-02-27 | Many-core chip distributed thermal modeling method based on recurrent neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109902807A true CN109902807A (en) | 2019-06-18 |
| CN109902807B CN109902807B (en) | 2022-07-05 |
Family
ID=66945767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910148729.6A Expired - Fee Related CN109902807B (en) | 2019-02-27 | 2019-02-27 | Many-core chip distributed thermal modeling method based on recurrent neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109902807B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113467590A (en) * | 2021-09-06 | 2021-10-01 | 南京大学 | Many-core chip temperature reconstruction method based on correlation and artificial neural network |
| CN109902807B (en) * | 2019-02-27 | 2022-07-05 | 电子科技大学 | Many-core chip distributed thermal modeling method based on recurrent neural network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110172984A1 (en) * | 2010-01-08 | 2011-07-14 | International Business Machines Corporation | Efficiency of static core turn-off in a system-on-a-chip with variation |
| US20150106306A1 (en) * | 2013-10-16 | 2015-04-16 | University Of Tennessee Research Foundation | Method and apparatus for constructing a neuroscience-inspired artificial neural network with visualization of neural pathways |
| US9015093B1 (en) * | 2010-10-26 | 2015-04-21 | Michael Lamport Commons | Intelligent control with hierarchical stacked neural networks |
| CN106681453A (en) * | 2016-11-24 | 2017-05-17 | 电子科技大学 | Dynamic heat treatment method of high-performance multi-core microprocessor |
| GB201810314D0 (en) * | 2018-06-22 | 2018-08-08 | Moixa Energy Holdings Ltd | Systems for machine learning, optimising and managing local multi-asset flexibility of distributed energy storage resources |
| CN108563949A (en) * | 2018-04-16 | 2018-09-21 | 电子科技大学 | For the duty mapping method of multi-core processor information security |
| CN109685207A (en) * | 2018-11-27 | 2019-04-26 | 电子科技大学 | A kind of multi core chip thermal management algorithm based on Recognition with Recurrent Neural Network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902807B (en) * | 2019-02-27 | 2022-07-05 | 电子科技大学 | Many-core chip distributed thermal modeling method based on recurrent neural network |
-
2019
- 2019-02-27 CN CN201910148729.6A patent/CN109902807B/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110172984A1 (en) * | 2010-01-08 | 2011-07-14 | International Business Machines Corporation | Efficiency of static core turn-off in a system-on-a-chip with variation |
| US9015093B1 (en) * | 2010-10-26 | 2015-04-21 | Michael Lamport Commons | Intelligent control with hierarchical stacked neural networks |
| US20150106306A1 (en) * | 2013-10-16 | 2015-04-16 | University Of Tennessee Research Foundation | Method and apparatus for constructing a neuroscience-inspired artificial neural network with visualization of neural pathways |
| CN106681453A (en) * | 2016-11-24 | 2017-05-17 | 电子科技大学 | Dynamic heat treatment method of high-performance multi-core microprocessor |
| CN108563949A (en) * | 2018-04-16 | 2018-09-21 | 电子科技大学 | For the duty mapping method of multi-core processor information security |
| GB201810314D0 (en) * | 2018-06-22 | 2018-08-08 | Moixa Energy Holdings Ltd | Systems for machine learning, optimising and managing local multi-asset flexibility of distributed energy storage resources |
| CN109685207A (en) * | 2018-11-27 | 2019-04-26 | 电子科技大学 | A kind of multi core chip thermal management algorithm based on Recognition with Recurrent Neural Network |
Non-Patent Citations (7)
| Title |
|---|
| HAI WANG: "A Fast Leakage-Aware Full-Chip Transient Thermal Estimation Method", 《IEEE TRANSACTIONS ON COMPUTERS》 * |
| LEI XU等: "Distributed model predictive control for dynamic thermal management of multi-core systems", 《2016 13TH IEEE INTERNATIONAL CONFERENCE ON SOLID-STATE AND INTEGRATED CIRCUIT TECHNOLOGY (ICSICT)》 * |
| 张明: "众核芯片热建模与功耗管理技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 肖涛: "基于深度学习的3DIC分布式热管理技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 胥雷: "众核芯片分布式热建模与热管理技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
| 郭星星: "基于循环神经网络的多核芯片热管理技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
| 黄柯衡等: "基于混合优化的多核处理器动态热管理方法", 《电子工艺技术》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902807B (en) * | 2019-02-27 | 2022-07-05 | 电子科技大学 | Many-core chip distributed thermal modeling method based on recurrent neural network |
| CN113467590A (en) * | 2021-09-06 | 2021-10-01 | 南京大学 | Many-core chip temperature reconstruction method based on correlation and artificial neural network |
| CN113467590B (en) * | 2021-09-06 | 2021-12-17 | 南京大学 | Many-core chip temperature reconstruction method based on correlation and artificial neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109902807B (en) | 2022-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107169527B (en) | Medical image classification method based on collaborative deep learning | |
| Wu et al. | Recent advances in optimization and game theoretic control for networked systems | |
| WO2021109644A1 (en) | Hybrid vehicle working condition prediction method based on meta-learning | |
| CN109063291B (en) | Intelligent topological optimization design method for cooling channel structure of electromechanical equipment | |
| CN106875002A (en) | Complex value neural network training method based on gradient descent method Yu generalized inverse | |
| CN103905246B (en) | Link prediction method based on grouping genetic algorithm | |
| CN109902807A (en) | A kind of hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network | |
| CN109725537A (en) | A kind of improved linear optimal semi-active control method | |
| CN103971160A (en) | Particle swarm optimization method based on complex network | |
| CN104539601B (en) | Dynamic network attack process analysis method for reliability and system | |
| CN110866640A (en) | Power load prediction method based on deep neural network | |
| CN102170516B (en) | Color space transition method based on fuzzy logic and neural network | |
| CN104050505A (en) | Multilayer-perceptron training method based on bee colony algorithm with learning factor | |
| CN114818483A (en) | Electromechanical disturbance positioning and propagation prediction method based on graph neural network | |
| CN113807040A (en) | Optimal design method for microwave circuit | |
| CN109685207A (en) | A kind of multi core chip thermal management algorithm based on Recognition with Recurrent Neural Network | |
| CN109145503A (en) | A kind of Dynamic High-accuracy modeling method of photovoltaic plant cluster | |
| CN114707613B (en) | Layered depth strategy gradient network-based power grid regulation and control method | |
| CN107918920B (en) | Output correlation analysis method for multiple photovoltaic power stations | |
| CN115600669A (en) | High-efficiency deep pulse neural network learning method based on local classifier | |
| CN116894504A (en) | Wind power cluster power ultra-short-term prediction model establishment method | |
| CN115797732A (en) | Image retrieval model training method and system used in open category scene | |
| CN111103799B (en) | Energy efficiency optimization modeling method based on teacher-student network online exploration learning | |
| CN115169235A (en) | Super surface unit structure inverse design method based on improved generation of countermeasure network | |
| CN103383743A (en) | Chrominance space transformation method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220705 |