A Novel High-Level Power Model Using Neural Network.pptVIP

下载本文档

1
0
约 23页
2016-11-27 发布于河南
举报
版权申诉

A Novel High-Level Power Model Using Neural Network.ppt

1、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

A Novel High-Level Power Model Using Neural Network

A Novel High-Level Power Model Using Neural Network 劉建男 Chien-Nan Liu Outline Traditional Power Modeling Our Basic Idea Neural Network Concept Power Modeling with Neural Network Experimental Results Conclusions Future Works Power Modeling Approach Traditional Power Modeling Look-up table (LUT) is the most popular Easy to use Number of possible input transitions for an n-input combinational circuit is 22n Impractical to stored the whole input transitions Solution: Use probability measurements liked P(xi), D(xi), etc. … Usage of LUT Power Model LUT-Based Approaches Accuracy depends on how many dimensions and table entries are used The more it use, the more accurate it is The modeling complexity may exponentially increase to reach the desired accuracy For a 4D table: 10x10x10x10=10000 table entries There are several solutions that focus on reducing the table size They might need the gate level (or lower) description of circuit Not suitable for IP protection Basic Idea Replace the look-up table by a neural network “Learn” the relationship between the power consumption and input characteristic values Advantage No need to store all power consumption Unlike LUT, we don’t have to increase the recording points to decrease the error Neural Network (NN) Mathematical model Simulate the biological neural network Weights (W) and bias (b) matrices Ability Learning recognition Application Classification, modeling, data processing … Biological Neural Network Change the response strength at synapse to “learn” something The Operation of NN (1/2) The Operation of NN (2/2) Our Input Data Format Structure of Neural Power Model Training Process Training with a lot of different cases Error Function Mean square error Training Algorithm Levenberg-Marquardt (L-M) algorithm Power Estimation Using NN For a test sequence with m pattern-pair Experiment Setup Circuits ISCAS’85 benchmark (combinational circuits) 0.35um cell library Simulation for real power PowerMill Neural network implemen