基于優(yōu)化算法的半導(dǎo)體器件參數(shù)提取和S參數(shù)的自動化測試

基于優(yōu)化算法的半導(dǎo)體器件參數(shù)提取和S參數(shù)的自動化測試SemiconductorDeviceModelParameterExtractionBasedonOptimizationAlgorithmsandS-parameterAutomatedMeasurement摘要半導(dǎo)體器件模型是影響電路設(shè)計(jì)精度的最主要因素，隨著電路規(guī)模的擴(kuò)大和頻段不斷提高，對器件模型的要求也越來越高，準(zhǔn)確的半導(dǎo)體器件模型對于提高集成電路設(shè)計(jì)的成功率、縮短電路研制周期是非常重要的。在確定半導(dǎo)體器件模型之后，還要有合適的參數(shù)提取方法。參數(shù)提取方法可以分為兩類：一類是直接提取法；另一類是優(yōu)化方法。直接提取法雖然步驟簡單，取值唯一，但提取的參數(shù)值通常還需要進(jìn)一步優(yōu)化。傳統(tǒng)的基于梯度或是微分的優(yōu)化方法有諸多缺點(diǎn)：只有合適的初始值才能求出部分參數(shù)；目標(biāo)函數(shù)含有無法解出的極點(diǎn)和冗余的參量；容易陷入局部最小值。而智能優(yōu)化算法的優(yōu)點(diǎn)是簡單、平穩(wěn)、易于實(shí)現(xiàn)，更重要的是對初值不敏感。優(yōu)化的目的是找到最佳的參數(shù)組合，使之符合器件的特性。為了能夠找到這樣一組優(yōu)質(zhì)解，需要一種提取算法，其全局誤差最小。近幾十年來，應(yīng)用算法提取半導(dǎo)體器件模型的參數(shù)一直是研究的熱點(diǎn)和重點(diǎn)。本文以高電子遷移率晶體管（HEMT）器件為例，首先對遺傳算法，粒子群算法，模擬退火算法和差分進(jìn)化算法進(jìn)行了研究，并對差分進(jìn)化算法進(jìn)行了改進(jìn)。然后基于上述算法進(jìn)行模型參數(shù)提取，主要研究工作包括以下四個(gè)部分：1）用仿真數(shù)據(jù)與優(yōu)化的數(shù)據(jù)相比較，結(jié)果顯示誤差都在可允許的范圍內(nèi)；2）用優(yōu)化算法提取了HEMT器件的小信號等效電路模型參數(shù)，模擬的S參數(shù)和測試的S參數(shù)可以很好的吻合；精度可以達(dá)到40GHz；3）改進(jìn)了差分進(jìn)化算法，其收斂速度比標(biāo)準(zhǔn)的差分進(jìn)化算法快。4）編寫了S參數(shù)的自動化測量軟件，通過控制軟件，可以直接將測試數(shù)據(jù)保存在PC機(jī)內(nèi)，加快了測試進(jìn)度。關(guān)鍵字：HEMT小信號建模優(yōu)化算法自動化測量AbstractModelsofsemiconductordevicesarethemainfactortoaffecttheaccuracyofthecircuitdesign.Withtheexpandingofcircuitscaleandincreasingoffrequencyband,highperformanceofthedevicemodelwasrequired.Itiscriticaloftheaccuratedevicemodelstoimprovethedesignofintegratedcircuitandshortenthedevelopmentcycle.Afterthesemiconductordevicemodelisconfirmed,anefficientmethodforparameterextractionwasinneed.Parameterextractionmethodscanbedividedintotwocategories:oneisthedirectextractionmethod;theotheristheoptimizationmethod.Althoughthedirectextractionmethodissimpleandthesingleresultisgotten,buttheextractedparametervaluestypicallyrequiresfurtheroptimization.Gradientorderivative-basedoptimizationmethodshavemanyshortcomings:onlysuitabletotheinitialvalueofsomeparametersisobtained;objectivefunctioncontainspolesandredundantparametersthatcannotbesolved;easytofallintolocalminimum.Theadvantagesoftheintelligentoptimizationalgorithmsaresimple,stable,easytoimplementand,moreimportantly,isnotsensitivetotheinitialvalue.Thepurposeofoptimizationistofindthebestcombinationofparameters,soastomeetthedevicecharacteristics.Inordertofindsuchasetofhigh-qualitysolution,anextractionalgorithmthatcanmaketheglobalminimumerrorisrequired.Inrecentdecades,theapplicationofalgorithmstoextractthemodelparametersofsemiconductordeviceshasbeenahotresearchandfocus.BasedonHEMTdevices,geneticalgorithm,particleswarmoptimization,simulatedannealingalgorithmanddifferentialalgorithmweredescribed，thenanovelimproveddifferentialalgorithmsisproposed.Onthebasisoffiveoptimizationalgorithmstheparameterswereextracted,thefollowingworkwascompleted:1)Comparedwithoptimizeddataandthesimulationdata,theerroroftheoptimizeddataiswithintheallowablerange;2)Byusingtheoptimizationalgorithm,thesmall-signalequivalentcircuitmodelparametersofHEMTdeviceareextracted,thefrequencyreaches40GHz,theresultshowthattheoptimizedS-parametershavegoodagreementwiththemeasuredS-parameters;3)Improveddifferentialevolutionperformsbetteronconvergencespeedthandifferentialevolution;4)TheautomatedsoftwareformeasuringS-parameterswaswritten.ThemeasureddataisstoreddirectlyinthePCbycontrollingthesoftware;theprogressofthetestistospeedup.Keywords:HEMT,SmallSignalmodeling,Optimizationalgorithms,Automatedmeasurement目錄摘要 圖4-6所示為運(yùn)行程序的操作一般順序。圖4-SEQ圖4-\*ARABIC6 程序操作順序框圖VC++6.0下的程序?qū)崿F(xiàn)儀器的通信是通過GPIB總線來傳輸?shù)?，每臺GPIB儀器設(shè)備都有一個(gè)設(shè)備通信地址。在PC機(jī)中安裝AgilentIOLibrariesSuite，該安裝包中含有AgilentVISA(VirtualInstrumentSoftwareArchitecture)程序庫，根據(jù)AgilentIOLibraries中說明文檔的要求，需要在VC的庫文件（LibraryFiles）目錄中加入VISA32.LIB，在包含文件（IncludeFiles）目錄中也要加入指定的目錄。AgilentVISA為儀器控制的軟件編程提供了很多的功能函數(shù)，其中有六個(gè)常用的函數(shù)：viOpenDefaultRM，viOpen，viPrintf，viScanf，viQueryf，viClose。viOpenDefaultRM函數(shù)用來建立一個(gè)會話。這個(gè)函數(shù)的第一次調(diào)用用來初始化默認(rèn)的資源管理器并返回該資源管理器會話的會話。viOpen函數(shù)對指定的資源打開一個(gè)會話。它返回的會話標(biāo)示符將被隨后使用到該儀器的函數(shù)使用，用以指示該儀器。viPrintf以一定的字符串格式轉(zhuǎn)換、格式化、發(fā)送命令參數(shù)給儀器會話。viScanf從儀器會話的輸出隊(duì)列中讀取數(shù)據(jù)，并以一定格式的字符串輸出數(shù)據(jù)。viQueryf函數(shù)先向儀器會話發(fā)送字符串命令，然后在從儀器的輸出隊(duì)列中讀出數(shù)據(jù)。viClose函數(shù)將結(jié)束一個(gè)會話。并釋放掉分配給該會話的數(shù)據(jù)資源。下面對設(shè)計(jì)程序的關(guān)鍵步驟說明如下：1）連接儀器在InstrumentAddress文本框中輸入GPIB地址，矢量網(wǎng)絡(luò)分析儀的GPIB地址為16，連接關(guān)鍵程序?yàn)椋篿ntaddress=GetDlgItemInt(IDC_EDIT_ADDRESS);//獲取儀器的地址stringstr_address="GPIB0::"+ToString(address)+"::INSTR";viOpenDefaultRM(&defaultRM); //用于第一次打開會話viOpen(defaultRM,&str_address[0],VI_NULL,VI_NULL,&vi);//用于打開詳細(xì)儀器if(viOpen(defaultRM,&str_address[0],VI_NULL,VI_NULL,&vi)==VI_SUCCESS){GetDlgItem(IDC_EDIT_STATUS)->SetWindowText("Connectsuccess!");}else{GetDlgItem(IDC_EDIT_STATUS)->SetWindowText("Connectfailing!");}//打開指定儀器的會話，并返回一個(gè)可以用于調(diào)用其它儀器函數(shù)的標(biāo)識。如果打開成功，則在文本框顯示連接成功，否則顯示連接失敗。2）頻率設(shè)置，設(shè)置要測量的頻率范圍。viPrintf(vi,":SENSe:FREQuency:STARt%@3lf\n",m_startf); //設(shè)置起始頻率viPrintf(vi,":SENSe:FREQuency:STOP%@3lf\n",m_stopf); //設(shè)置終止頻率viPrintf(vi,":SENSe:SWEep:POINts%@1d\n",m_points); //設(shè)置掃描點(diǎn)數(shù)3）儀器的校準(zhǔn)，儀器在使用前一定要校準(zhǔn)。先選擇連接到矢量網(wǎng)絡(luò)分析儀端口和校準(zhǔn)件的類型viPrintf(vi,":SENSe:CORRection:COLLect:GUIDed:CONNector:PORT1:SELect\"%s\"\n",conn1); //定義端口1的連接頭類型然后根據(jù)提示連接相應(yīng)的校準(zhǔn)件，直到校準(zhǔn)完成。viPrintf(vi,":SENSe:CORRection:COLLect:GUIDed:DESCription?%@1d\n",step);viPrintf(vi,":SENSe:CORRection:COLLect:GUIDed:ACQuireSTAN%s\n",step);4）測量并將文件保存在PC機(jī)中。viPrintf(vi,":MMEMory:STORe:FILE\"%s\"\n",ext);viPrintf(vi,":MMEMory:TRANsfer?\"%s\"\n",ext);下REF_Ref353106969\h圖4-7為使用軟件獲得的數(shù)據(jù)文件。圖4-SEQ圖4-\*ARABIC7 保存的數(shù)據(jù)文件本章小結(jié)本文設(shè)計(jì)的測試系統(tǒng)實(shí)現(xiàn)了對參數(shù)的自動化測量，無論是軟件平臺或是硬件平臺，都可以根據(jù)需要繼續(xù)設(shè)計(jì)。如果想增添設(shè)備，只需將儀器通過接口接入系統(tǒng)；也可以添加新的代碼實(shí)現(xiàn)其他的測試項(xiàng)目。如今還出現(xiàn)了LAN接口的儀器，可以直接用網(wǎng)線連接，傳輸速率變得更高，各種儀器組合方便，對儀器進(jìn)行遠(yuǎn)程控制，實(shí)現(xiàn)儀器的遠(yuǎn)距離共享?？偨Y(jié)與展望工作總結(jié)隨著集成電路技術(shù)的發(fā)展，器件的集成度越來越高，器件的模型也越來越復(fù)雜，如何快速而準(zhǔn)確的提取器件模型參數(shù)也成為研究的重點(diǎn)?，F(xiàn)如今，一些先進(jìn)的優(yōu)化算法被應(yīng)用到半導(dǎo)體器件模型參數(shù)的提取中。本文主要是針對模型參數(shù)提取過程而展開，具體內(nèi)容包括：（1）介紹并在MATLAB中實(shí)現(xiàn)了四種優(yōu)化算法（遺傳算法、差分進(jìn)化算法、模擬退火算法和粒子群算法），對差分進(jìn)化算法進(jìn)行了改進(jìn)，實(shí)現(xiàn)了在不影響算法精度的同時(shí)，加快了算法的收斂速度。（2）在ADS軟件中建立HMET小信號模型，給定小信號模型的16個(gè)參數(shù)值，在100MHz~40GHz頻率范圍內(nèi)進(jìn)行小信號仿真，將得到的參數(shù)數(shù)據(jù)作為擬合目標(biāo)，分別使用上述四種優(yōu)化算法，提取出HEMT小信號模型的參數(shù)值。并將優(yōu)化得到的參數(shù)值與所設(shè)定的參數(shù)值相比較，發(fā)現(xiàn)偏差都在可接受的范圍內(nèi)，從而驗(yàn)證了用優(yōu)化算法提取半導(dǎo)體器件模型參數(shù)的可行性。（3）應(yīng)用測試的GaAsHEMT器件參數(shù)，分別使用四種優(yōu)化算法對HEMT小信號等效電路模型進(jìn)行參數(shù)提取，取擬合參數(shù)與測試參數(shù)的差值作為優(yōu)化目標(biāo)，其值越小，提取的模型參數(shù)值也就越準(zhǔn)，結(jié)果發(fā)現(xiàn)改進(jìn)的算法大大縮短了尋找最優(yōu)值的時(shí)間。（4）應(yīng)用微波射頻器件測量時(shí)常用的矢量網(wǎng)絡(luò)分析儀和半導(dǎo)體分析儀，編寫了一個(gè)軟件，用于自動化測量參數(shù)，可以用來和模型參數(shù)的提取過程相結(jié)合，在測量器件參數(shù)時(shí)，獲得器件模型的參數(shù)。工作展望本文使用四種優(yōu)化算法對HEMT器件的小信號等效電路模型參數(shù)進(jìn)行了提取，并提出了基于差分進(jìn)化算法的改進(jìn)算法，雖然取得了一定的成果，但由于時(shí)間緊迫，還可以在以下幾方面做更深入的研究：1）將設(shè)計(jì)的自動化測試軟件與優(yōu)化算法相結(jié)合，在獲得測試數(shù)據(jù)的同時(shí)，可以直接應(yīng)用優(yōu)化方法來提取器件模型的參數(shù)。2）本文使用優(yōu)化算法對HMET器件的小信號等效電路模型進(jìn)行了提取，對FET器件的非線性等效電路模型進(jìn)行參數(shù)提取是下一步的工作。附錄下面是用MATLAB實(shí)現(xiàn)半導(dǎo)體參數(shù)提取的主要程序。%主程序functiontest clc; clearall; formatlong; formatcompact;time=1;%總的運(yùn)行次數(shù)totalTime=20;whiletime<=totalTimeticrand('seed',sum(100*clock));testdomy(@pso);toc %計(jì)算每次循環(huán)所用時(shí)間time=time+1;endend%%具體應(yīng)用算法functiontestdomy(optimizer)[bestX,bestFitness,bestFitnesss,evaluations]=feval(optimizer,@fitnesshemt,hemtproblemdata);end%以下是對問題的初始設(shè)置functiondata=hemtproblemdata(maxEvaluations)data=struct(...'Dim',16,...%要求參數(shù)個(gè)數(shù)'AcceptableFitness',0.000001,...%允許適應(yīng)值的最大值.'MaxEvaluations',100000,...%迭代的最大次數(shù)'LowerInit',pa/100,...%初始值參數(shù)的下邊界'UpperInit',pa*100,...%初始值參數(shù)的上邊界'LowerBound',pa/100,...%參數(shù)的下邊界'UpperBound',pa*100);%參數(shù)的上邊界end%以下是計(jì)算所用函數(shù)的適應(yīng)值functionvalue=fitnesshemt(pop,fname,datas2p) npop=size(pop,1); value=zeros(npop,1); pointsnum=size(datas2p,1); %計(jì)算每個(gè)個(gè)體的適應(yīng)值 fori=1:npop pa=pop(i,:); forpointnum=1:pointsnum f=datas2p(pointnum,1); Ypad=myYpad(pa(1),pa(2),pa(3),f); Zrl=myZrl(pa(4),pa(5),pa(6),pa(7),pa(8),pa(9),f); Yint=myYint(pa(10),pa(11),pa(12),pa(13),pa(14),pa(15),pa(16),f); [S_11,S_12,S_21,S_22]=chig(Ypad,Zrl,Yint); datnew(pointnum,:)=[f,abs(S_11),angle(S_11)*180/pi,abs(S_21),... angle(S_21)*180/pi,abs(S_12),angle(S_12)*180/pi,abs(S_22),angle(S_22)*180/pi]; end %每個(gè)個(gè)體適應(yīng)值 tmp=(((datas2p(:,2:end)-datnew(:,2:end))./datas2p(:,2:end))).^2; value(i)=sum(tmp(:)); end cleardatas2pdatanewend%HEMT器件的小信號等效電路模型的S參數(shù)矩陣function[S_11,S_12,S_21,S_22]=chig(Ypad,Zrl,Yint)Y=Ypad+inv(Zrl+inv(Yint));s_params=y2smy(Y);S_11=s_params(1,1);S_12=s_params(1,2);S_21=s_params(2,1);S_22=s_params(2,2);End%PAD電容導(dǎo)納矩陣functionYpad=myYpad(Cpg,Cpgd,Cpd,f)Ypad=[1i*2*pi*f*(Cpg+Cpgd)-1i*2*pi*f*Cpgd -1i*2*pi*f*Cpgd1i*2*pi*f*(Cpgd+Cpd)];End%寄生電感和寄生電阻的Z參數(shù)矩陣functionZrl=myZrl(Rg,Lg,Rs,Ls,Rd,Ld,f)Zrl=[Rg+Rs+1i*2*pi*f*(Ls+Lg)Rs+1i*2*pi*f*Ls Rs+1i*2*pi*f*LsRs+Rd+1i*2*pi*f*(Ls+Ld)];End%本征部分導(dǎo)納矩陣functionYint=myYint(Cgd,Cgs,Cds,Yint=[i*2*pi*f*Cgs/(1+1i*2*pi*f*Ri*Cgs)+1i*2*pi*f*Cgd-1i*2*pi*f*Cgdgm*exp(-1i*2*pi*f*tau)/(1+1i*2*pi*f*Cgs*Ri)-1i*2*pi*f*Cgdgds+1i*2*pi*f*(Cds+Cgd)];end%Y參數(shù)矩陣轉(zhuǎn)換成S參數(shù)矩陣functionSout=y2smy(Y)Zo=50;I=eye(2);Sout=(I+Zo*Y)\(I-Zo*Y);end攻讀碩士期間發(fā)表論文[1]用進(jìn)化算法提取HEMT小信號模型的參數(shù)電子器件已錄用[2]漸變槽線天線研究2012年上海市研究生學(xué)術(shù)論壇(射頻·微電子)ADDINNE.Bib參考文獻(xiàn)[1]陳星弼，張慶中.晶體管原理與設(shè)計(jì)第2版[M].北京市:電子工業(yè)出版社,2007:385.[2]KilbyJ.S.Inventionoftheintegratedcircuit[J].ElectronDevices,IEEETransactionson.1976,23(7):648-654.[3]RabaeyJ.M.，ChandrakasanA.，Nikoli?B.數(shù)字集成電路:設(shè)計(jì)透視[M].清華大學(xué)出版社,2003.[4]吳濤，杜剛，劉曉彥，等.基于遺傳算法的半導(dǎo)體器件模型參數(shù)提取[J].北京大學(xué)學(xué)報(bào):自然科學(xué)版.2007,43(5):687-693.[5]CostaD.,LiuW.U.,HarrisJrJ.S.DirectextractionoftheAlGaAs/GaAsheterojunctionbipolartransistorsmall-signalequivalentcircuit[J].ElectronDevices,IEEETransactionson.1991,38(9):2018-2024.[6]DambrineG.,CappyA.,HeliodoreF.,etal.AnewmethodfordeterminingtheFETsmall-signalequivalentcircuit[J].MicrowaveTheoryandTechniques,IEEETransactionson.1988,36(7):1151-1159.[7]ArnoldE.,GolioM.,MillerM.,etal.DirectextractionofGaAsMESFETintrinsicelementandparasiticinductancevalues[C].IEEE,1990:359-362.[8]Man-YoungJ.,Byung-GyuK.,Young-JinJ.,etal.Atechniqueforextractingsmall-signalequivalent-circuitelementsofHEMTs[J].IEICEtransactionsonelectronics.1999,82(11):1968-1976.[9]ChigaevaE.,WalthesW.,WiegnerD.,etal.Determinationofsmall-signalparametersofGaN-basedHEMTs[C].IEEE,2000:115-122.[10]Zarate-DeLandaA.,Zuniga-JuarezJ.E.,Reynoso-HernandezJ.A.,etal.ANewandBetterMethodforExtractingtheParasiticElementsofOn-WaferGaNTransistors[C].2007:791-794.[11]BradyR.G.,OxleyC.H.,BrazilT.J.Animprovedsmall-signalparameter-extractionalgorithmforGaNHEMTdevices[J].MicrowaveTheoryandTechniques,IEEETransactionson.2008,56(7):1535-1544.[12]胡曉萍，胡建萍，鄭梁，等.一種新的HBT小信號模型參數(shù)優(yōu)化提取法[J].電子學(xué)報(bào).2010,38(003):567-571.[13]ChengY.,HuC.MOSFETmodelingandBSIM3user'sguide[M].Springer,1999.[14]ColeyD.A.AnIntroductiontoGeneticAlgorithmsforScientistsandEngineers[M].WorldScientificPubl.,1999.[15]MenozziR.,PiazziA.,ContiniF.Small-signalmodelingformicrowaveFETlinearcircuitsbasedonageneticalgorithm[J].CircuitsandSystemsI:FundamentalTheoryandApplications,IEEETransactionson.1996,43(10):839-847.[16]MenozziR.,BorgarinoM.,BaeyensY.,etal.OntheeffectsofhotelectronsontheDCandRFcharacteristicsoflattice-matchedInAlAs/InGaAs/InPHEMTs[J].MicrowaveandGuidedWaveLetters,IEEE.1997,7(1):3-5.[17]WattsJ.,BittnerC.,HeaberlinD.,etal.ExtractionofcompactmodelparametersforULSIMOSFETsusingageneticalgorithm[C].1999:176-179.[18]PrasadK.,RanjanR.,SahooN.C.,etal.Optimalreconfigurationofradialdistributionsystemsusingafuzzymutatedgeneticalgorithm[J].PowerDelivery,IEEETransactionson.2005,20(2):1211-1213.[19]謝曉鋒，李釗，阮駿，等.應(yīng)用遺傳算法實(shí)現(xiàn)MOS器件綜合[J].半導(dǎo)體學(xué)報(bào).2002,23(1):95-101.[20]石瑞英，劉訓(xùn)春.用改進(jìn)的遺傳算法從S參數(shù)中提取HBT交流小信號等效電路模型參數(shù)[J].半導(dǎo)體學(xué)報(bào).2002,23(9):957-961.[21]CiouS.H.,WangY.Y.,MengB.C.,etal.ANovelParameterExtractionMethodforHEMTModelsbyUsingGenericAlgorithms[C].IEEE,2007:241-245.[22]張有濤，夏冠群，高建峰，等.用改進(jìn)的遺傳算法精確提取GaAsMESFET小信號等效電路參數(shù)[J].2004.[23]宋文斌.并行遺傳算法在器件模型參數(shù)提取中的應(yīng)用[J].軟件工程師.2012(4):45-47.[24]QinA.K.,HuangV.L.,SuganthanP.N.Differentialevolutionalgorithmwithstrategyadaptationforglobalnumericaloptimization[J].EvolutionaryComputation,IEEETransactionson.2009,13(2):398-417.[25]SabatS.L.,UdgataS.K.DifferentialEvolutionalgorithmforMESFETsmallsignalmodelParameterextraction[C].IEEE,2010:203-207.[26]LiuB.,LuJ.,WangY.,etal.Aneffectiveparameterextractionmethodbasedonmemeticdifferentialevolutionalgorithm[J].MicroelectronicsJournal.2008,39(12):1761-1769.[27]LaarhovenP.J.M.,AartsE.H.SimulatedAnnealing:TheoryandApplications[M].Springer,1987.[28]VaiM.K.,PrasadS.,LiN.C.,etal.Modelingofmicrowavesemiconductordevicesusingsimulatedannealingoptimization[J].ElectronDevices,IEEETransactionson.1989,36(4):761-762.[29]VaiM.K.,NgM.Atechnology-independentdevicemodelingprogramusingsimulatedannealingoptimization[C].IEEE,1989:4-9.[30]VaiM.K.,PrasadS.Computer-aideddesignofmonolithicMESFETdistributedamplifiers[J].MicrowaveTheoryandTechniques,IEEETransactionson.1990,38(4):345-349.[31]VaiM.K.,LinJ.S.,PrasadS.Accelerationofsimulatedannealinganditsapplicationtomicrowavedeviceandcircuitoptimization[C].IEEE,1991:1213-1216.[32]MeskoobB.,PrasadS.,VaiM.,etal.Asmall-signalequivalentcircuitforthecollector-upInGaAs/InAlAs/InPheterojunctionbipolartransistor[J].ElectronDevices,IEEETransactionson.1992,39(11):2629-2632.[33]BilbroG.L.,SteerM.B.,TrewR.J.,etal.Extractionoftheparametersofequivalentcircuitsofmicrowavetransistorsusingtreeannealing[J].MicrowaveTheoryandTechniques,IEEETransactionson.1990,38(11):1711-1718.[34]朱磊，尤煥成，金香菊.用改進(jìn)算法提取GaNHEMT小信號等效電路模型[J].微納電子技術(shù).2007,44(3):120-124.[35]ClercM.Particleswarmoptimization[M].Wiley-ISTE,2010.[36]SabatS.L.,UdgataS.K.,MurthyK.SmallsignalparameterextractionofMESFETusingquantumparticleswarmoptimization[J].MicroelectronicsReliability.2010,50(2):199-206.[37]SabatS.L.,UdgataS.K.,AbrahamA.ArtificialbeecolonyalgorithmforsmallsignalmodelparameterextractionofMESFET[J].EngineeringApplicationsofArtificialIntelligence.2010,23(5):689-694.[38]YegnanarayanaB.Artificialneuralnetworks[M].PHILearningPvt.Ltd.,2004.[39]ZhangQ.J.,GuptaK.C.,DevabhaktuniV.K.ArtificialneuralnetworksforRFandmicrowavedesign-fromtheorytopractice[J].MicrowaveTheoryandTechniques,IEEETransactionson.2003,51(4):1339-1350.[40]ZhangL.,ZhangQ.SimpleandEffectiveExtrapolationTechniqueforNeural-BasedMicrowaveModeling[J].MicrowaveandWirelessComponentsLetters,IEEE.2010,20(6):301-303.[41]KabirH.,WangY.,YuM.,etal.High-dimensionalneural-networktechniqueandapplicationstomicrowavefiltermodeling[J].MicrowaveTheoryandTechniques,IEEETransactionson.2010,58(1):145-156.[42]LiS.,HanB.,ChengJ.,etal.Neuro‐spacemapping‐basedDCmodelingfor130‐nmMOSFET[J].InternationalJournalofRFandMicrowaveComputer‐AidedEngineering.2010,20(5):587-592.[43]LiS.,ChengJ.,HanB.,etal.Bias‐dependentsmall‐signalmodelingbasedonneuro‐spacemappingforMOSFET[J].InternationalJournalofRFandMicrowaveComputer‐AidedEngineering.2011,21(2):182-189.[44]LiS.,LiX.,GaoJ.Neural‐spacemapping‐basedlarge‐signalmodelingforMOSFET[J].InternationalJournalofRFandMicrowaveComputer‐AidedEngineering.2011,21(3):353-362.[45]TianY.Z.,ZhangQ.J.Knowledge-BasedNeuralNetworksforModelingofRadio-Frequency/MicrowaveComponents[J].電子科技大學(xué)學(xué)報(bào).2011,40(6):816.[46]GaoJ.RFandMicrowaveModelingandMeasurementTechniquesforCompundFieldEffectTransistors[M].SciTechPub.,2010.[47]SabatS.L.,UdgataS.K.,AbrahamA.ArtificialbeecolonyalgorithmforsmallsignalmodelparameterextractionofMESFET[J].EngineeringApplicationsofArtificialIntelligence.2010,23(5):689-694.[48]解蘭，於洪標(biāo).基于遺傳算法的場效應(yīng)管模型參數(shù)的提取[J].微波學(xué)報(bào).2009(S1).[49]HollandJ.OutlineforaLogicalTheoryofAdaptiveSystems[J].JournaloftheACM(JACM).1962,9(3):297-314.[50]BagleyJ.D.thebehaviorofadaptivesystemswhichemploygeneticandcorrelationalgorithms[D].ProQuest,UMIDissertationsPublishing,1967.[51]HollandJ.H.Adaptationinnaturalandartificialsystems,UniversityofMichiganpress[J].AnnArbor,MI.1975,1(97):5.[52]陸金桂，李謙，遺傳學(xué)，等.遺傳算法原理及其工程應(yīng)用[M].中國礦業(yè)大學(xué)出版社,1997.[53]李敏強(qiáng).遺傳算法的基本理論與應(yīng)用[M].科學(xué)出版社,2002.[54]王小平，曹立明.遺傳算法:理論,應(yīng)用及軟件實(shí)現(xiàn)[M].西安交通大學(xué)出版社,2002.[55]KennedyJ.,EberhartR.Particleswarmoptimization[C].IEEE,1995:1942-1948.[56]StornR.,PriceK.Differentialevolution–asimpleandefficientheuristicforglobaloptimizationovercontinuousspaces[J].Journalofglobaloptimization.1997,11(4):341-359.[57]PriceK.V.,StornR.M.,LampinenJ.A.Differentialevolution[M].Springer,1997.[58]PriceK.V.Anintroductiontodi