Parametric Yield Maximization using Gate Sizing based on Efficient Statistical Power and Delay Gradient Computation

Size: px
Start display at page:

Download "Parametric Yield Maximization using Gate Sizing based on Efficient Statistical Power and Delay Gradient Computation"

Transcription

1 Prmetrc Yeld Mxmzto usg Gte Szg sed o Effcet Sttstcl Power d Dely Grdet Comutto Kvrj Chor, Suml Shh, Ashsh Srvstv, Dvd Bluw, Des Sylvester Uversty of Mchg, EECS Dertmet, A Aror, MI 4809 {kvrj, suml, srvs, Bluw, des}@eecs.umch.edu Astrct Wth the cresed sgfcce of ge ower d erformce vrlty, the yeld of desg s ecomg costred oth y ower d erformce lmts, therey sgfctly comlctg crcut otmzto. I ths er, we roose ew otmzto method for yeld otmzto uder smulteous ge ower d erformce lmts. The otmzto roch uses ovel ge ower d erformce lyss tht s sttstcl ture d cosders the correlto etwee ge ower d erformce to ele ccurte comutto of crcut yeld uder ower d dely lmts. We the roose ew heurstc roch to cremetlly comute the grdet of yeld wth resect to gte szes the crcut wth hgh effcecy d ccurcy. We the show how ths grdet formto c e effectvely used y o-ler otmzer to erform yeld otmzto. We cosder oth ter-de d tr-de vrtos wth correlted d rdom comoets. The roosed roch s mlemeted d tested d we demostrte u to 40% yeld mrovemet comred to determstclly otmzed crcut.. Itroducto d Overvew of Aroch Cotued rocess sclg hs resulted lrge crese rocess vrlty tht leds to lrge fluctutos rocess rmeters from ther oml vlues. These vrtos hve grow due rt to ggressve lthogrhc techques tht re used to tter dmesos smller th the wvelegth of lght. I ddto, smller devce dmesos d smller umer of dog toms crese the fluece of heome such s le edge roughess d rdom dot effects. These rocess vrtos trslte to wde rges erformce metrcs of curret desgs. I rtculr, ge curret whch s extremely sestve to umer of key rocess rmeters hs show huge fluctutos wth [] showg 0X vrto ge ower for 30% vrto erformce cross 000 smles of desg mufctured 80 m techology. I ddto, s the cotruto of ge ower hs grow the fluctuto ower dssto s ow domted y ge ower. Ths results egtve correlto etwee ower dssto d dely of desg. Thus hgh erformce smles of desg re lso exected to hve hgher ower dssto d vce-vers. Ths leds to two-sded costrt o the fesle rego of rmetrc yeld defed y dely d ower lmts [], d cuses sgfct yeld loss uder vrtos rocess rmeters. A umer of lyss techques to cosder the mct of vrlty o tmg [3-7] d ower [8-9] hve ee develoed. However, these roches do ot estmte the true rmetrc yeld of desg cosderg oth ower d erformce correlto. Referece [0] ws the frst to cosder the mct of vrlty o crcut otmzto. The uthors descred the formto of tmg wll due to determstc ower otmzto whch creses the suscetlty of the desg to rocess vrtos, d roosed heurstc roch to revet the uld-u of lrge umer of ths er the crtcl dely of the crcut. Ths ws cheved y ddg elty fucto whch hd egtve mct o the ojectve fucto vlue wheever th hd dely whch ws er crtcl. However, the roch ws determstc ture d dd ot use y sttstcl formto durg otmzto. Recetly, severl sttstcl tmg [-3] d ower [4-6] otmzto roches hve lso ee roosed. However, ll these roches eglect the correlto etwee ower d erformce. Therefore, tmg yeld mrovemets evtly result degrdto ower yeld d vce vers. Moreover, most of these roches suffer from lrge comuttol comlexty d rutmes. Thus, there s crtcl eed to develo roches tht erform true d effcet rmetrc yeld otmzto, where yeld s defed usg oth ower d tmg costrts. Recetly, [7] roosed roch to erform gte-level yeld lyss comuttolly effcet mer whle cosderg the correlto etwee ower d erformce. The roch ws sed o rcl-comoet sed rocess vrto modelg techque to erform tmg d ower lyss usg the sme set of uderlyg rdom vrles (RVs), llowg the correlto ower d erformce to e ctured. Addtolly the roch cosders ll comoets of rocess vrtos: ter-de d tr-de (stlly correlted d rdom) vrlty d c therefore serve s frmework to ele true rmetrc yeld otmzto. I ths work, we roose ovel roch to erform yeld otmzto usg gte szg. The yeld otmzto s formulted s ucostred otmzto rolem, where the ojectve s to mxmze the rmetrc yeld of desg. The otmzto s erformed usg grdet-sed o-ler otmzer. A rute-force grdet comutto roch sed o tertve yeld lyss, however, leds to lrge comuttol overheds. Therefore, we roose effcet heurstc techque to erform the comutto of the yeld grdet. Ths s cheved y erturg the sze of gte the crcut d heurstclly reclcultg the dely d ower rolty dstruto fuctos (dfs) of the ertured crcut. The tmg df of the ertured crcut s clculted sed o ovel cutset roch tht lyzes oly suset of the odes the crcut to estmte the comlete dely df. The ower df of the ertured crcut s clculted wth cremetl ower lyss. Ths volves sutrctg the ower dssto of the ertured gte from the df of totl ower for the comlete crcut d the ddg the ower dssto of the ertured gte whle ccoutg for the gte sze chge. These ertured dfs re the used to comute the yeld of the ertured crcut y tegrtg the ertured vrte Guss dstruto over the rego defed y the ge ower d tmg costrt. Ths grdet comutto techque s the tegrted wth LANCELOT [8], lrge-scle o-ler otmzer, to mrove the rmetrc yeld of the desg, d s foud to rovde 8X mrovemet rutme wth verge error of 0.%. The remder of the er s orgzed s follows. Secto refly revews the rcl comoet sed roch to erform yeld lyss. Secto 3 resets the cremetl tmg d ower lyss techques, whch re the used to comute the grdet of yeld. I Secto 4 we rovde detls regrdg the mlemetto of our yeld otmzto X/05/$ IEEE. 00

2 6 JPDF vlue 4 Fgure. Exmle rttos of crcut usg grd to model the correlted comoet of vrtos roch d reset results cludg comrso of our roch to determstc otmzto. We rovde coclusos Secto 5.. Yeld lyss I ths secto we refly dscuss our modelg ssumtos d yeld lyss roch. We lso defe the yeld otmzto rolem d descre rute-force roch to erform yeld otmzto. The comuttol comlexty of ths rute-force roch motvtes the eed for more effcet grdet comutto techques. Our yeld lyss frmework s sed o the roch [7], d we exress the dely d ge of gte s Dely = d om = Lekge = ex V α om ( P ) = β ( P ) where d om d ex(v om ) re the oml gte dely d ge, resectvely, d α d β ctures the deedece of gte dely d the log of gte ge o the rocess rmeters of terest. The RVs P the ove equto refers to the vrtos these rocess rmeters tht re ssumed to e Guss. The vrto rocess rmeter s the rttoed to correlted d rdom comoet. The correlted comoet s hdled y rttog the desg s show Fgure. For ech rmeter, ech squre the grd s ssged to Guss RV whch ctures the correlted vrto tht rocess rmeter, whch s defed usg correlto mtrx. Ths gves totl of N g RVs for ech rmeter, whch re ssumed to hve jot multorml dstruto. Usg rcl comoets lyss [9] the correlted comoet s exressed s ler comto of N g deedet Guss RVs (z ), d the rdom vrto ll the rocess rmeters s lumed to sgle RV - η for dely d d η for ge, l d the coeffcet of the rdom comoet s clculted y mtchg the vrce of the rdom cotruto. Ths roch gves us cocl exressos for gte dely d ge ower whch re exressed s: Dely = d om α Lekge = ex Vom = = γ z η d R β z R γ ηl = = Tmg lyss s the erformed the srt of [3][7], d the dely s rogted through the crcut, whle mtg the ode delys the sme cocl form wth dfferet coeffcets. The sum oerto s erformed y smly ddg the coeffcets for ech of the RVs, other th the rdom comoet whose coeffcet s oted s the squre root of the sum of the squred coeffcets of the rdom comoets () () Dely(s) log(lekge) Fgure. Jot rolty dstruto fucto for the vrte Guss dstruto for c3540. of the summed dfs. The mx oerto s erformed y mtchg the me, vrce d the correlto of the mx of two RV (whch re oted usg [0]) d the cocl exresso of the mx. Lekge ower lyss s sed o summg logorml RVs usg Wlkso s method [] s roosed [8]. The ge of ech gte s tertvely dded to the sum whch s mted cocl form. I ech ddto the coeffcets of the cocl exresso for the sum re clculted y mtchg the me, vrce d correlto wth the rcl comoets of the sum (oted usg [8]) d the cocl exresso for the sum. At the ed of tmg d ower lyss (tht rovdes the dely d ge ower cocl form) the correlto etwee dely d ge ower s estmted usg: Cov( Dely, Lekge) = α β (3) These fve rmeters (me d vrce of dely d ge ower d ther correlto) re used to defe vrte Guss dstruto for dely d log of ge ower s show Fgure. The rmetrc yeld whch s defed s: = ( D D P ) Y = P 0, P 0 (4) where D s dely of the crcut costred to e less th D 0 d P s the ower of the desg costred to e less th P 0. I ths work, we ssume tht vrtos ower re domted y vrtos ge ower d the dymc ower dssto s ssumed to e fxed umer d sutrcted out of the totl ower udget of the desg to defe the ge ower costrt. Bsed o ths ssumto, we c rewrte (4) s ( D D log P log( P P )) Y = P (5) 0, L 0 D where P L d P D re the ge d dymc ower of the desg. The ove yeld exresso s ow equvlet to the tegrl of vrte Guss RV over rectgulr rego, d c e evluted usg exresso develoed []. Both the tmg d ower comutto stes requre O(N g ) stes, where s the umer of gtes the desg d N g s the umer of regos to whch the desg s rttoed to cture the correlto structure of correlted rocess vrtos. However, the fl yeld comutto ste (5) tself rus costt tme sce the yeld comutto s lwys erformed usg the set of fve rmeters, deedet of the sze of the rolem. Bsed o ths yeld lyss ege, rute-force roch to erform yeld otmzto usg gte szg c e develoed. Ths volves comutg the grdet of yeld to the sze of ech gte, whch c e estmted y reszg ech gte d 0

3 erformg yeld lyss d settg the gte ck to ts orgl sze. After comutg the grdet, we use lrge scle o-ler otmzer to mrove the yeld of the crcut. We ow cosder the comuttol comlexty of oe terto of ths roch. Ech grdet comutto requres yeld lyss rus d thus hs overll comlexty of O( N g ). Note tht the sce the sze of the rttos s fxed, the umer of rttos N g c lso e exected to crese wth the sze of the desg. Thus, the overll comuttol requremets soo ecome utele for lrge desgs. Also, ote tht the rute-force roch seds most of the tme reclcultg the sme formto for most of the crcut d motvtes the eed for effcet grdet comutto roch. 3. Grdet Comutto I ths secto we wll dscuss our ew grdet comutto roch tht clcultes the udted tmg d dely dfs sed o chge gte sze. Both the tmg d ower erturto lyss techques udte the coeffcets of the dely d ge df exresso sed o the chge gte sze. These udted dely d ge ower dfs re the used to comute the yeld of the ertured desg. The chge yeld s used to estmte the grdet of yeld to the sze of ech gte the desg. 3. Tmg Perturto Comutto We wll exl our tmg erturto comutto roch sed o cutsets usg the followg grh reresetto for our crcuts. Defto : A tmg grh s drected cyclc grh hvg exctly oe source d oe sk: G={N,E,s,f}, where N={,,.., k } s set of odes, E={e,e,,el} s set of edges, s N s the source ode d f N s the sk ode d ech edge s ordered r of odes e=(, j ) d ech ode s ssocted wth dely for ech f edge, whch deeds o the chrcterstcs of the fout odes. The odes the tmg grh corresod to gtes d the edges corresod to ets crcut. A rolstc tmg grh s defed s tmg grh where ech ode s ssocted wth RV for the dely for ech f edge. Fgure 3 shows exmle tmg grh wth te odes, eght of whch rereset ctul gtes d odes d 0 rereset the source d sk odes, resectvely. The ltest rrvl tme (AT) d requred rrvl tme (RAT) rolty dstruto fuctos (df) for ech ode the tmg grh re ow defed s: Defto : The ltest rrvl tme (AT) t edge e the rolstc tmg grh s RV whose CDF A e (t) gves the rolty tht determstc smle of ths tmg grh hs rrvl tme less th t. Defto 3: The erlest requred rrvl tme (RAT) t edge e the rolstc tmg grh s RV whose CDF R e (t) gves the rolty tht the determstc smle meets the tmg costrt T crt f the determstc rrvl tme t the ode s less th t. Note tht the sum of the AT d RAT t ode reresets the rtl df of dely sce t does ot tke to ccout the fluece of the edges tht re ot reset ether the f or the fout coe of the ode o the df of crcut dely. To exress the deedece of crcut dely o the dely of oe of the odes let us defe the followg terms. Defto 4: A ler toologcl orderg (LTO) of the odes tmg grh s totl order sed o the reltosh tht the order of y ode x tht les the fout coe of ode s strctly lrger th the order of ode, d tht o two odes tmg grh hve the sme order. A LTO of tmg grh c e esly determed y erformg redth-frst trversl of the tmg grh. Though gve tmg grh c hve my LTOs, fdg the otml LTO s ot the focus of ths er. Fgure 3 llustrtes tmg grh wth odes leled ccordg to LTO of the tmg grh. Note tht swg odes 8 d 9 wll stll mt vld LTO of the odes. Defto 5: A cutset of tmg grh wth gve LTO of ode s defed to e the set of edges (, j ) of the tmg grh whch stsfy LTO( ) LTO() d LTO( j )>LTO(). Defto 6: A ode x of the tmg grh elogs to the cutset-source of ode f there exsts edge (x,*) whch elogs to the cutset of ode. Defto 7: The f-set of ode of tmg grh s the set of mmedte redecessor odes of ode. Defto 8: The rrvl tme set or ATSet of ode s the uo of the f-set of ode d the odes the fout coe of the f-set of ode tht hve order less th or equl to the order of ode. Defto 9: The covoluto-set or CovSet of ode s the tersecto of the ATSet d cutset-source of ode. Ay cutset of the tmg grh dvdes the tmg grh to two dscoected comoets d the sttstcl mxmum of the sum of the AT d RAT of ll edges the cutset gves the comlete df of crcut dely. Now, f we ertur the dely chrcterstcs of ode (y gte szg) we lso chge the cctve lodg of the f gtes, ffectg ther dely chrcterstcs s well. To comute the ew crcut dely we ote tht the RAT of the edges the cutset does ot chge, sce ll the gtes ther fout coe cludes gtes tht hve order strctly greter th the order of ode, d hve uchged dely chrcterstcs. Fgure 3. A tmg grh showg ler toologcl orderg for the odes d cutsets for odes 8 d Fgure 4. A tmg grh showg the ATSet (odes wth the shded ellse) d cutset-source set (odes wth the dshed she) for ode 8 Ths eglects the mct of the chge the sloe of the crcut. However, the roch for tmg lyss usg ckwrd rogto [4] c e used to cosder ther mct wth the sme frmework

4 However, the AT for ll edges tht re the fout coe of the f-set of ode chges. However we re oly terested AT chges for edges tht re drve y odes tht hve order less th the order of gte, sce we eed to comute the AT for the edges the cutset oly. Ths s exctly the set of odes defed y the ATSet of ode. If the AT of edge the cutset chges we eed to recomute the covoluto of the AT d RAT t tht edge. These edges re drve y the odes the tersecto of the ATSet d the cutset whch s defed s the CovSet of ode. Let us revst the exmle tmg grh Fgure 3 d cosder ode 8. The cutset for ths ode s the set of edges (6,9), (8,0) d (5,0) s show y the dshed le. The ATSet for the ode c e detfed s the set of odes 6, 7 d 8 s show Fgure 4. The tersecto of the cutset-source d ATSet defes the CovSet d s the set of odes 6 d 8. Note tht the CovSet detfes tht the AT d RAT hs ot chged o the edge (5,0) d we do ot eed to recomute the covoluto of the AT d RAT for ths edge. However, f we cosder ode 9 the cutset s defed y the edges from odes 5, 8 d 9 to ode 0, s show s the dotted le Fgure 3. The seudo-code to clculte the dely df of the ertured crcut s show elow, where we refer to the edge y the me of the drvg ode. The seudo-code volves the comutto of the AT for ll odes the ATSet, covoluto of the AT d RAT for ll odes the CovSet d the sttstcl mxmum of the covoluto for ll edges the cutset. Note tht ll the comuttos CutSetSst re erformed usg the sme cocl exresso for the dely df. Thus, the fl dely df of the ertured crcut s lso exressed the sme form. Although, the roch s descred seems exct, t s heurstc. Ths results from the fct tht the comutto of the mx fucto of dely dfs s ot exct d forwrd d ckwrd trversls of the grh result tmg delys tht re ot exctly sme. However, ths error s very smll s wll e show lter the results secto. 3. Power Perturto Comutto The sttstcl ower comutto s erformed y summg the ower dssto of ech gte crcut to comute the comlete df of ge ower. To erform ower lyss of crcut wth erturtos the sze of gte, we frst erform sttstcl ower lyss of the uertured crcut s descred Secto. Now the ge ower fter the sze of gte hs ee ertured s exressed s P ert crc = P = P uert crc uert crc\ P P uert gte, ert gte, P ert gte, where P ert d P uert refer to the ertured d uertured ower, resectvely d the suscrt dctes whether the ower refers to the crcut or to the gte. Sce the ge ower s exressed logorml (exoetl of Guss) RV, we c roxmte ther sum usg other logorml. I geerl, f we sum P d c P to ot P, whch s CUTSETSSTA () for ech ode (x ATSET ()) Comute AT(x); for ech ode (x CONVSET ()) CT(x) covoluto (AT(x), RT(x)) for ech edge(x CUTSET ()) T mxmum (T, CT(x)) retur T Fgure 5. Pseudo-code for the comutto of erturtos tmg df of crcut (6) mthemtclly exressed s, P = = ex 0 ex 0 z (7) = = z ex c0 = c z c = P P c the coeffcets the exresso for P c e oted y mtchg the me, vrce d the correlto coeffcet wth the exoetl of the rcl comoets (z s). Ths gves us set of equtos vrles whch c e lytclly solved to ot the followg exresso for the coeffcets ssocted wth the rcl comoets [7] = log E z E( Pe ) z ( P ) E( e ) = log z c z E( P ) ( ) e E Pe c z ( ( ) ( ) ( ) E P E P E e Usg the exressos develoed [3], the remg two coeffcets the exresso for P c e exressed s = 0 log 4 ( E( P ) ( ) E P ( E( P ) ( ) ( ) ( ) ( ) E P Vr P Vr Pc Cov P Pc 0.5 ( ) ( ) ( ) Vr P Vr Pc Cov P Pc ( E( P ) E( P ) = = log (8) (9) (0) Note tht to comute the exresso (6) we eed to use the ove exressos twce to clculte the fl ertured ge. However, whe oe of the logormls s sutrcted the sgs ssocted wth ts exected vlue d covrce terms the ove exressos re reversed. 3.3 Yeld Grdet To ths ot we hve develoed effcet roches to erform sttstcl tmg d ower erturto comutto. Now, we wll use these techques to erform the comutto of the grdet of yeld effcet mer. The comutto of yeld grdet volves the comutto of the erturto yeld for smll chges the sze of gtes the desg. The seudo-code for the comutto of yeld s cluded Fgure 6. After ech reszg move, the o-ler FASTYIELDGRADIENT (CIRCUIT, SIZE ) for ech gte (g CIRCUIT) udte lod c d sze of (g) usg SIZE; for ech gte (g CIRCUIT) comute ew gte dely & ower of (g) T FORWARDSSTA (CIRCUIT) P STATPOWERANALYSIS (CIRCUIT) Y YIELD (P, T) do REVERSESSTA (CIRCUIT) for ech gte(g CIRCUIT) sve curret stte of CIRCUIT s SIZE(g) SIZE(g) comute ew gte dely & ower of g for ech gte( FANIN(g)) udte lod c d dely of T CUTSETSSTA (CIRCUIT) P INCREMSPA (CIRCUIT, P) Y YIELD (P, T ) Y(g) (Y -Y)/ SIZE(g) restore the orgl stte of the CIRCUIT retur Y Fgure 6. Pseudo-code for the comutto of the yeld grdet 03

5 otmzer clls the yeld comutto route FstYeldGrdet. The frst ste s to tlze the crcut so tht ll odes re ssged the correct lod cctce sed o the szes of the gtes ts mmedte fout, d the correct ge ower sed o ts ow sze. Bsed o the lod cctce of the ode ech ut of ode s ssged to dely df, whch reresets the dely of the tmg rc from tht rtculr ut to the outut of the gte. After the tlzto ste, the ext ste volves the rogto of the AT from the source ode to the sk ode the tmg grh. Ths s rereseted s ForwrdSSTA the seudo-code. The ext ste s to erform sttstcl ower lyss d geerte the ge curret df usg SttPowerAlyss. The Yeld fucto s the used to comute the yeld sed o the tmg d ge ower dfs gve ge ower costrt P d dely costrt D, s outled Secto. To erform the comutto of yeld grdet, we frst rogte the RAT from the sk ode to the source ode usg ReverseSSTA. The we go through ech ode the crcut tertvely d ertur the sze of ech gte y smll mout. The lod cctce of the odes the f-set of the ode d the dely df ssged to ech tmg rc of ths ode d the odes the f-set re udted. The usg the sttstcl tmg d ower erturto comutto techques dscussed Sectos we comute the dely d ge ower dfs of ths ertured crcut. The yeld corresodg to the ertured crcut s the clculted d the chge yeld s used to defe the rtculr comoet of the yeld grdet. Let us cosder the comuttol comlexty of our roosed roch d comre t to the rute-force roch, where ech terto hd comlexty of O( N g ). Ech terto, our roosed roch, volves sgle ru of the comlete yeld lyss roch, s dscussed ove, whch hs comlexty of O(N g ). The tmg d ower erturto comutto s reeted O() tmes. The comlexty of the cremetl ower lyss s O(N g ) sce we requre two sum oertos of the logorml RVs. For the sttstcl tmg erturto comutto most of the mx comuttos the cutset c e reused y storg the formto s he. Thus, the tmg erturto comutto hs comlexty of O(N g log()). Thus the overll comlexty of the roch s O(N g log()), whch s lrge mrovemet comred to the rute-force roch. 4. Results d Imlemetto Detls We mlemeted the roosed roch C d comred our yeld mrovemets to determstc crcut otmzto techque. Our roosed roch for the comutto of the yeld grdet s wrtte s suroute whch the otmzer uses to clculte the grdet of the ojectve fucto. The yeld lyss ege serves s the suroute to clculte the ojectve fucto tself. Followg, we reset the ccurcy d rutme results for the roosed roch. If we ssume tht reverse d forwrd SSTA gve exct tmg dstrutos t ech ode, the the rocedure would e exct s well. However, s oted efore, due to the Guss roxmto cosdered whle comutg the mxmum troduces smll ccurcy whle erformg forwrd d reverse SSTA. Now, sce ths ccurcy s fucto of crcut toology d recovergece structure the sestvty of yeld comuted usg oly FORWARDSSTA sed rute-force s eglgle. Tle shows the rutme comrso d ccurcy results of the roosed grdet comutto rocedure s comred to the ïve rute force roch. The crcut sze terms of the umer of gtes d the verge cut-wdth over ll odes the crcut s lso reorted the secod d the thrd colums, resectvely. Rutme er grdet vector comutto Tle.Comrso of Yeld grdet comutto usg FASTYIELDGRADIENT d rute-force roch. Bech. Averge Brute- Crcut #gtes Cut-Sze Force Grdet Seed-u (%) (%) c E-03 c E-03 c E-0 c E-0 c E-03 c E-03 c E-03 c E-03 c E E E E E E E E-03 Fst Mx. Error Avg. Error usg the rute roch d the roosed rocedure re gve Colums 4 d 5, resectvely. The seed u of the roosed method over the rute-force roch s gve Colum 6, d rges etwee 3X to 0X d s foud to e lrger for gger crcuts. The mxmum error, over ll gtes, foud usg grdet comutto ormlzed wth resect to the rute-force method s gve Colum 7, d s foud to e smll most cses wth verge of.4%. The error verged error over ll gtes the crcut s gve the lst colums of Tle d s foud to e extremely smll. 4. Yeld Otmzto The gtes our stdrd cell lrry re chrcterzed for set of szes the rge from mmum sze to mxmum sze d the dely d ge ower for termedte gte szes s oted usg ler terolto. All desgs re the determstclly otmzed for ower uder dely costrts usg ether desg comler or LANCELOT [8]. We use our sttstcl yeld mxmzto roch to mrove the yeld of ths otmzed desg for set of dfferet ower d tmg costrts. Our results dcte tht erformg sttstcl otmzto c sgfctly mrove the tmg yeld of the desg. We comre our results sed o the ISCAS85 [3] echmrks whch were sytheszed 30 m techology. The yeld otmzto results re gve Tle. The frst su-secto cludg colums, 3, 4 d 5 reort the tl tmg d ower sttstcs resultg from determstclly otmzed crcut. The determstc otmzto ws erformed usg oml dely d ower models. We reset yeld otmzto results for two sets of costrts. The frst set cludes yeld otmzto for ggressve oml vlue costrts. As determstc otmzer s uwre of the vrto ower d tmg d ther correlto, the tl yeld t oml costrts s extremely smll. However, the roosed vrlty wre yeld otmzto sgfctly mroves the yeld. For exmle, the yeld for echmrk crcuts c43, c499 d c880 drmtclly mroves from close to 0% to u to 40%. Colum 6, 7, 8 d 9 reort the ost otmzto tmg d ower sttstcs of the crcut. The tl yeld suject to oml vlue costrts d the yeld fter erformg otmzto re gve Colums 0 d, resectvely. The secod set of results reort the erformce of yeld otmzto uder essmstc costrts. For ths cse we use the oml vlue offset y oe stdrd devto s the costrt for oth ower d tmg whle defg the ojectve fucto for otmzto. Ag colums, 3, 4 d 5 lst the ost otmzto sttstcs of the crcut wheres colums 6 d 7 reort the results cheved fter erformg the roosed yeld otmzto. As the costrts re relxed ths cse the tl yeld of the crcut mroves d for the sme reso the mrovemets cheved re reltvely smller s comred to the revous cse. The mxmum mrovemet ths cse s foud to e greter th 0% for the echmrk crcut c

6 Bech. Crcuts Tle : Yeld Otmzto results for dfferet ower d tmg costrts Itl soluto D < Dm, P < Pm D < Dm Ds, P < Pm Ps Dely(s) Power (µw) Dely(s) Power (µw) Yeld(%) Dely(s) Power (µw) Yeld(%) Dm Ds Pm Ps Dm Ds Pm Ps It. Ot. Dm Ds Pm Ps It. Ot. c ~ c ~ c ~ c c c c ~ c c ~ Coclusos To the est of our kowledge, we hve reseted the frst roch to erform gte-level rmetrc yeld otmzto cosderg costrts o ower d erformce, log wth ther correlto. The roch for yeld comutto s show to e comuttolly effcet d s show to rovde 8X mrovemet rutme, s comred to rute-force grdet comutto roch. The yeld grdet s used to gude lrge-scle o ler otmzer to mrove the yeld of desg tht hs ee otmzed determstclly, uder vryg ower d dely costrts. The results show tht we c cheve mrovemets yeld whch re s lrge s 40%. Ackowledgemets Ths work ws suorted rt y NSF, SRC d MARCO/DARPA. Refereces [] T. Krk, S. Borkr, d V. De, Su-90 m techologes chlleges d oortutes for CAD, ACM/IEEE ICCAD, , 00. [] R. Ro et l., Prmetrc yeld lyss d costred-sed suly voltge otmzto, ACM/IEEE ISQED, , 005. [3] H. Chg d S. S. Stekr, Sttstcl tmg lyss cosderg stl correltos usg sgle PERT-lke trversl, ACM/IEEE ICCAD,. 6-65, 003. [4] C. Vsweswerh et l., Frst-order cremetl lock-sed sttstcl tmg lyss, ACM/IEEE DAC, , 004. [5] A. Agrwl et l., Sttstcl tmg lyss usg ouds d selectve eumerto, IEEE Trs. o CAD, , , Set [6] A. Devg d C. Kshy, Block-Bsed sttstcl tmg lyss wth ucertty, ACM/IEEE ICCAD, , 003. [7] M. Orshsky d A. Bdodhyy, Fst sttstcl tmg lyss hdlg rtrry dely correltos, ACM/IEEE DAC, , 004. [8] R.R. Ro, et l., Sttstcl lyss of suthreshold ge curret for VLSI crcuts, IEEE Trs. VLSI Systems,.3-39, Fe [9] S. Nredr et l., Full-ch su-threshold ge ower redcto model for su-0.8µm CMOS, ACM/IEEE ISLPED,. 9-3, 00. [0] X. B, et l., Ucertty wre crcut otmzto, ACM/IEEE DAC,.58-63, 00. [] S. Rj, S. Vrudhul, d J. Wg, A methodology to mrove tmg yeld the resece of rocess vrtos, ACM/IEEE DAC, , 004. [] S. Cho, B. Pul d K. Roy, Novel szg lgorthm for yeld mrovemet uder rocess vrto ometer techology, IEEE/ACM DAC, , 004. [3] A. Agrwl et l., Sttstcl tmg sed otmzto usg gte szg, ACM/IEEE DATE, , 005. [4] A. Srvstv, D. Sylvester, d D. Bluw, Sttstcl otmzto of ge ower cosderg rocess vrtos usg dul-vth d szg, ACM/IEEE DAC, , 004. [5] A. Dvood, V. Khdelwl, d A. Srvstv, Vrlty sred mlemetto selecto rolem, ACM/IEEE ICCAD, , 004. [6] M. M d M. Orshsky, A ew sttstcl lgorthm for gte szg, ACM/IEEE ICCD,. 7-77, 004. [7] A. Srvstv et l., Accurte d effcet gte level rmetrc yeld estmto cosderg correlted vrtos ge ower d erformce, ACM/IEEE DAC, , [8] A. R. Co, N. I. M. Gould, d Ph. L. Tot. LANCELOT: A Fortr ckge for lrge-scle o-ler otmzer (Relese A). Srger, Verlg, 99. [9] D. F. Morrso, Multvrte sttstcl methods, McGrw-Hll Book Comy, 967. [0] C. Clrk, The gretest of fte set of rdom vrles, Oertos Reserch, vol. 9,. 85-9, 96. [] S.C. Schwrtz d Y.S. Yeh, O the dstruto fucto d momets of ower sums wth logorml comoets, Bell Systems Techcl Jourl, vol.6,.44-46, Se. 98. [] J. H. Cdwell, The vrte orml tegrl, Bometrk,. 3-35, Dec. 95. [3] F. Brglez d H. Fujwr, A eutrl etlst of 0 comtol echmrk crcuts d trget trsltor Fortr, Proc. ISCAS, , My 989. [4] D. Lee, V. Zolotov d D. Bluw, Sttc tmg lyss usg ckwrd rogto, ACM/IEEE DAC, ,

Soo King Lim Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Figure 7: Figure 8: Figure 9: Figure 10: Figure 11:

Soo King Lim Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Figure 7: Figure 8: Figure 9: Figure 10: Figure 11: Soo Kg Lm 1.0 Nested Fctorl Desg... 1.1 Two-Fctor Nested Desg... 1.1.1 Alss of Vrce... Exmple 1... 5 1.1. Stggered Nested Desg for Equlzg Degree of Freedom... 7 1.1. Three-Fctor Nested Desg... 8 1.1..1

More information

Density estimation II

Density estimation II CS 750 Mche Lerg Lecture 6 esty estmto II Mlos Husrecht mlos@tt.edu 539 Seott Squre t: esty estmto {.. } vector of ttrute vlues Ojectve: estmte the model of the uderlyg rolty dstruto over vrles X X usg

More information

Solutions Manual for Polymer Science and Technology Third Edition

Solutions Manual for Polymer Science and Technology Third Edition Solutos ul for Polymer Scece d Techology Thrd Edto Joel R. Fred Uer Sddle Rver, NJ Bosto Idols S Frcsco New York Toroto otrel Lodo uch Prs drd Cetow Sydey Tokyo Sgore exco Cty Ths text s ssocted wth Fred/Polymer

More information

ITERATIVE METHODS FOR SOLVING SYSTEMS OF LINEAR ALGEBRAIC EQUATIONS

ITERATIVE METHODS FOR SOLVING SYSTEMS OF LINEAR ALGEBRAIC EQUATIONS Numercl Alyss for Egeers Germ Jord Uversty ITERATIVE METHODS FOR SOLVING SYSTEMS OF LINEAR ALGEBRAIC EQUATIONS Numercl soluto of lrge systems of ler lgerc equtos usg drect methods such s Mtr Iverse, Guss

More information

MTH 146 Class 7 Notes

MTH 146 Class 7 Notes 7.7- Approxmte Itegrto Motvto: MTH 46 Clss 7 Notes I secto 7.5 we lered tht some defte tegrls, lke x e dx, cot e wrtte terms of elemetry fuctos. So, good questo to sk would e: How c oe clculte somethg

More information

Area and the Definite Integral. Area under Curve. The Partition. y f (x) We want to find the area under f (x) on [ a, b ]

Area and the Definite Integral. Area under Curve. The Partition. y f (x) We want to find the area under f (x) on [ a, b ] Are d the Defte Itegrl 1 Are uder Curve We wt to fd the re uder f (x) o [, ] y f (x) x The Prtto We eg y prttog the tervl [, ] to smller su-tervls x 0 x 1 x x - x -1 x 1 The Bsc Ide We the crete rectgles

More information

6. Chemical Potential and the Grand Partition Function

6. Chemical Potential and the Grand Partition Function 6. Chemcl Potetl d the Grd Prtto Fucto ome Mth Fcts (see ppedx E for detls) If F() s lytc fucto of stte vrles d such tht df d pd the t follows: F F p lso sce F p F we c coclude: p I other words cross dervtves

More information

Accurate and Efficient Gate-Level Parametric Yield Estimation Considering Correlated Variations in Leakage Power and Performance

Accurate and Efficient Gate-Level Parametric Yield Estimation Considering Correlated Variations in Leakage Power and Performance Aurte d Effet Gte-Level rmetr Yeld Estmto Cosderg Correlted Vrtos Lekge ower d erforme Ashsh Srvstv es Sylvester Suml Shh vd Bluw Kk Agrwl Stehe retor Uversty of Mhg, EECS ertmet, A Aror, MI 489 {srvs,

More information

MATRIX AND VECTOR NORMS

MATRIX AND VECTOR NORMS Numercl lyss for Egeers Germ Jord Uversty MTRIX ND VECTOR NORMS vector orm s mesure of the mgtude of vector. Smlrly, mtr orm s mesure of the mgtude of mtr. For sgle comoet etty such s ordry umers, the

More information

Introduction to mathematical Statistics

Introduction to mathematical Statistics Itroducto to mthemtcl ttstcs Fl oluto. A grou of bbes ll of whom weghed romtely the sme t brth re rdomly dvded to two grous. The bbes smle were fed formul A; those smle were fed formul B. The weght gs

More information

Patterns of Continued Fractions with a Positive Integer as a Gap

Patterns of Continued Fractions with a Positive Integer as a Gap IOSR Jourl of Mthemtcs (IOSR-JM) e-issn: 78-578, -ISSN: 39-765X Volume, Issue 3 Ver III (My - Ju 6), PP -5 wwwosrjourlsorg Ptters of Cotued Frctos wth Postve Iteger s G A Gm, S Krth (Mthemtcs, Govermet

More information

Chapter 2 Intro to Math Techniques for Quantum Mechanics

Chapter 2 Intro to Math Techniques for Quantum Mechanics Wter 3 Chem 356: Itroductory Qutum Mechcs Chpter Itro to Mth Techques for Qutum Mechcs... Itro to dfferetl equtos... Boudry Codtos... 5 Prtl dfferetl equtos d seprto of vrbles... 5 Itroducto to Sttstcs...

More information

Lecture 3-4 Solutions of System of Linear Equations

Lecture 3-4 Solutions of System of Linear Equations Lecture - Solutos of System of Ler Equtos Numerc Ler Alger Revew of vectorsd mtrces System of Ler Equtos Guss Elmto (drect solver) LU Decomposto Guss-Sedel method (tertve solver) VECTORS,,, colum vector

More information

Level-2 BLAS. Matrix-Vector operations with O(n 2 ) operations (sequentially) BLAS-Notation: S --- single precision G E general matrix M V --- vector

Level-2 BLAS. Matrix-Vector operations with O(n 2 ) operations (sequentially) BLAS-Notation: S --- single precision G E general matrix M V --- vector evel-2 BS trx-vector opertos wth 2 opertos sequetlly BS-Notto: S --- sgle precso G E geerl mtrx V --- vector defes SGEV, mtrx-vector product: r y r α x β r y ther evel-2 BS: Solvg trgulr system x wth trgulr

More information

Optimality of Strategies for Collapsing Expanded Random Variables In a Simple Random Sample Ed Stanek

Optimality of Strategies for Collapsing Expanded Random Variables In a Simple Random Sample Ed Stanek Optmlt of Strteges for Collpsg Expe Rom Vrles Smple Rom Smple E Stek troucto We revew the propertes of prectors of ler comtos of rom vrles se o rom vrles su-spce of the orgl rom vrles prtculr, we ttempt

More information

CURVE FITTING LEAST SQUARES METHOD

CURVE FITTING LEAST SQUARES METHOD Nuercl Alss for Egeers Ger Jord Uverst CURVE FITTING Although, the for of fucto represetg phscl sste s kow, the fucto tself ot be kow. Therefore, t s frequetl desred to ft curve to set of dt pots the ssued

More information

Vapor-Liquid Equilibria for HFCs + Propane

Vapor-Liquid Equilibria for HFCs + Propane Vor-Lqud Equlbr for HFCs + Proe Altertves Reserch Ceter Hyu Sg J, Byug Gwo Lee, De Ryook Yg, Jog Sug Lm * CFC Altertve Reserch Ceter, Evrometl d Process, KIST Chemcl Egeerg, Kore Uversty Chemcl Egeerg

More information

β (cf Khan, 2006). In this model, p independent

β (cf Khan, 2006). In this model, p independent Proc. ICCS-3, Bogor, Idoes December 8-4 Vol. Testg the Equlty of the Two Itercets for the Prllel Regresso Model Bud Prtko d Shhjh Kh Dertmet of Mthemtcs d Nturl Scece Jederl Soedrm Uversty, Purwokerto,

More information

In Calculus I you learned an approximation method using a Riemann sum. Recall that the Riemann sum is

In Calculus I you learned an approximation method using a Riemann sum. Recall that the Riemann sum is Mth Sprg 08 L Approxmtg Dete Itegrls I Itroducto We hve studed severl methods tht llow us to d the exct vlues o dete tegrls However, there re some cses whch t s ot possle to evlute dete tegrl exctly I

More information

Stats & Summary

Stats & Summary Stts 443.3 & 85.3 Summr The Woodbur Theorem BCD B C D B D where the verses C C D B, d est. Block Mtrces Let the m mtr m q q m be rttoed to sub-mtrces,,,, Smlrl rtto the m k mtr B B B mk m B B l kl Product

More information

A Technique for Constructing Odd-order Magic Squares Using Basic Latin Squares

A Technique for Constructing Odd-order Magic Squares Using Basic Latin Squares Itertol Jourl of Scetfc d Reserch Publctos, Volume, Issue, My 0 ISSN 0- A Techque for Costructg Odd-order Mgc Squres Usg Bsc Lt Squres Tomb I. Deprtmet of Mthemtcs, Mpur Uversty, Imphl, Mpur (INDIA) tombrom@gml.com

More information

Some Unbiased Classes of Estimators of Finite Population Mean

Some Unbiased Classes of Estimators of Finite Population Mean Itertol Jourl O Mtemtcs Ad ttstcs Iveto (IJMI) E-IN: 3 4767 P-IN: 3-4759 Www.Ijms.Org Volume Issue 09 etember. 04 PP-3-37 ome Ubsed lsses o Estmtors o Fte Poulto Me Prvee Kumr Msr d s Bus. Dertmet o ttstcs,

More information

Problem Set 4 Solutions

Problem Set 4 Solutions 4 Eoom Altos of Gme Theory TA: Youg wg /08/0 - Ato se: A A { B, } S Prolem Set 4 Solutos - Tye Se: T { α }, T { β, β} Se Plyer hs o rte formto, we model ths so tht her tye tke oly oe lue Plyer kows tht

More information

On Several Inequalities Deduced Using a Power Series Approach

On Several Inequalities Deduced Using a Power Series Approach It J Cotemp Mth Sceces, Vol 8, 203, o 8, 855-864 HIKARI Ltd, wwwm-hrcom http://dxdoorg/02988/jcms2033896 O Severl Iequltes Deduced Usg Power Seres Approch Lored Curdru Deprtmet of Mthemtcs Poltehc Uversty

More information

PubH 7405: REGRESSION ANALYSIS REGRESSION IN MATRIX TERMS

PubH 7405: REGRESSION ANALYSIS REGRESSION IN MATRIX TERMS PubH 745: REGRESSION ANALSIS REGRESSION IN MATRIX TERMS A mtr s dspl of umbers or umercl quttes ld out rectgulr rr of rows d colums. The rr, or two-w tble of umbers, could be rectgulr or squre could be

More information

Section 7.2 Two-way ANOVA with random effect(s)

Section 7.2 Two-way ANOVA with random effect(s) Secto 7. Two-wy ANOVA wth rdom effect(s) 1 1. Model wth Two Rdom ffects The fctors hgher-wy ANOVAs c g e cosdered fxed or rdom depedg o the cotext of the study. or ech fctor: Are the levels of tht fctor

More information

An Extended Mixture Inverse Gaussian Distribution

An Extended Mixture Inverse Gaussian Distribution Avlble ole t htt://wwwssstjscssructh Su Sudh Scece d Techology Jourl 016 Fculty o Scece d Techology, Su Sudh Rjbht Uversty A Eteded Mture Iverse Guss Dstrbuto Chookt Pudrommrt * Fculty o Scece d Techology,

More information

14.2 Line Integrals. determines a partition P of the curve by points Pi ( xi, y

14.2 Line Integrals. determines a partition P of the curve by points Pi ( xi, y 4. Le Itegrls I ths secto we defe tegrl tht s smlr to sgle tegrl except tht sted of tegrtg over tervl [ ] we tegrte over curve. Such tegrls re clled le tegrls lthough curve tegrls would e etter termology.

More information

Dopant Compensation. Lecture 2. Carrier Drift. Types of Charge in a Semiconductor

Dopant Compensation. Lecture 2. Carrier Drift. Types of Charge in a Semiconductor Lecture OUTLIE Bc Semcoductor Phycs (cot d) rrer d uo P ucto odes Electrosttcs ctce ot omesto tye semcoductor c be coverted to P tye mterl by couter dog t wth ccetors such tht >. comested semcoductor mterl

More information

The z-transform. LTI System description. Prof. Siripong Potisuk

The z-transform. LTI System description. Prof. Siripong Potisuk The -Trsform Prof. Srpog Potsuk LTI System descrpto Prevous bss fucto: ut smple or DT mpulse The put sequece s represeted s ler combto of shfted DT mpulses. The respose s gve by covoluto sum of the put

More information

this is the indefinite integral Since integration is the reverse of differentiation we can check the previous by [ ]

this is the indefinite integral Since integration is the reverse of differentiation we can check the previous by [ ] Atervtves The Itegrl Atervtves Ojectve: Use efte tegrl otto for tervtves. Use sc tegrto rules to f tervtves. Aother mportt questo clculus s gve ervtve f the fucto tht t cme from. Ths s the process kow

More information

Mathematically, integration is just finding the area under a curve from one point to another. It is b

Mathematically, integration is just finding the area under a curve from one point to another. It is b Numerl Metods or Eg [ENGR 9] [Lyes KADEM 7] CHAPTER VI Numerl Itegrto Tops - Rem sums - Trpezodl rule - Smpso s rule - Rrdso s etrpolto - Guss qudrture rule Mtemtlly, tegrto s just dg te re uder urve rom

More information

Cooper and McGillem Chapter 4: Moments Linear Regression

Cooper and McGillem Chapter 4: Moments Linear Regression Cooper d McGllem Chpter 4: Momets Ler Regresso Chpter 4: lemets of Sttstcs 4-6 Curve Fttg d Ler Regresso 4-7 Correlto Betwee Two Sets of Dt Cocepts How close re the smple vlues to the uderlg pdf vlues?

More information

Chapter Gauss-Seidel Method

Chapter Gauss-Seidel Method Chpter 04.08 Guss-Sedel Method After redg ths hpter, you should be ble to:. solve set of equtos usg the Guss-Sedel method,. reogze the dvtges d ptflls of the Guss-Sedel method, d. determe uder wht odtos

More information

Math 2414 Activity 16 (Due by end of class August 13) 1. Let f be a positive, continuous, decreasing function for x 1, and suppose that

Math 2414 Activity 16 (Due by end of class August 13) 1. Let f be a positive, continuous, decreasing function for x 1, and suppose that Mth Actvty 6 (Due y ed of clss August ). Let f e ostve, cotuous, decresg fucto for x, d suose tht f. If the seres coverges to s, d we cll the th rtl sum of the seres the the remder doule equlty r 0 s,

More information

Chapter 4: Distributions

Chapter 4: Distributions Chpter 4: Dstrbutos Prerequste: Chpter 4. The Algebr of Expecttos d Vrces I ths secto we wll mke use of the followg symbols: s rdom vrble b s rdom vrble c s costt vector md s costt mtrx, d F m s costt

More information

Chapter 7. Bounds for weighted sums of Random Variables

Chapter 7. Bounds for weighted sums of Random Variables Chpter 7. Bouds for weghted sums of Rdom Vrbles 7. Itroducto Let d 2 be two depedet rdom vrbles hvg commo dstrbuto fucto. Htczeko (998 d Hu d L (2000 vestgted the Rylegh dstrbuto d obted some results bout

More information

Chapter 2 Intro to Math Techniques for Quantum Mechanics

Chapter 2 Intro to Math Techniques for Quantum Mechanics Fll 4 Chem 356: Itroductory Qutum Mechcs Chpter Itro to Mth Techques for Qutum Mechcs... Itro to dfferetl equtos... Boudry Codtos... 5 Prtl dfferetl equtos d seprto of vrbles... 5 Itroducto to Sttstcs...

More information

Mathematical models for computer systems behaviour

Mathematical models for computer systems behaviour Mthemtcl models for comuter systems ehvour Gols : redct comuter system ehvours - erformces mesuremets, - comrso of systems, - dmesog, Methodology : - modellg evromet (stochstc rocess) - modellg system

More information

A METHOD FOR THE RAPID NUMERICAL CALCULATION OF PARTIAL SUMS OF GENERALIZED HARMONICAL SERIES WITH PRESCRIBED ACCURACY

A METHOD FOR THE RAPID NUMERICAL CALCULATION OF PARTIAL SUMS OF GENERALIZED HARMONICAL SERIES WITH PRESCRIBED ACCURACY UPB c Bull, eres D, Vol 8, No, 00 A METHOD FOR THE RAPD NUMERAL ALULATON OF PARTAL UM OF GENERALZED HARMONAL ERE WTH PRERBED AURAY BERBENTE e roue o etodă ouă etru clculul rd l suelor rţle le serlor roce

More information

Chapter 3 Supplemental Text Material

Chapter 3 Supplemental Text Material S3-. The Defto of Fctor Effects Chpter 3 Supplemetl Text Mterl As oted Sectos 3- d 3-3, there re two wys to wrte the model for sglefctor expermet, the mes model d the effects model. We wll geerlly use

More information

Roberto s Notes on Integral Calculus Chapter 4: Definite integrals and the FTC Section 2. Riemann sums

Roberto s Notes on Integral Calculus Chapter 4: Definite integrals and the FTC Section 2. Riemann sums Roerto s Notes o Itegrl Clculus Chpter 4: Defte tegrls d the FTC Secto 2 Rem sums Wht you eed to kow lredy: The defto of re for rectgle. Rememer tht our curret prolem s how to compute the re of ple rego

More information

Advanced Algorithmic Problem Solving Le 3 Arithmetic. Fredrik Heintz Dept of Computer and Information Science Linköping University

Advanced Algorithmic Problem Solving Le 3 Arithmetic. Fredrik Heintz Dept of Computer and Information Science Linköping University Advced Algorthmc Prolem Solvg Le Arthmetc Fredrk Hetz Dept of Computer d Iformto Scece Lköpg Uversty Overvew Arthmetc Iteger multplcto Krtsu s lgorthm Multplcto of polyomls Fst Fourer Trsform Systems of

More information

Systems of second order ordinary differential equations

Systems of second order ordinary differential equations Ffth order dgolly mplct Ruge-Kutt Nystrom geerl method solvg secod Order IVPs Fudzh Isml Astrct A dgolly mplct Ruge-Kutt-Nystróm Geerl (SDIRKNG) method of ffth order wth explct frst stge for the tegrto

More information

under the curve in the first quadrant.

under the curve in the first quadrant. NOTES 5: INTEGRALS Nme: Dte: Perod: LESSON 5. AREAS AND DISTANCES Are uder the curve Are uder f( ), ove the -s, o the dom., Prctce Prolems:. f ( ). Fd the re uder the fucto, ove the - s, etwee,.. f ( )

More information

PATTERNS IN CONTINUED FRACTION EXPANSIONS

PATTERNS IN CONTINUED FRACTION EXPANSIONS PATTERNS IN CONTINUED FRACTION EXPANSIONS A THESIS SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MINNESOTA BY SAMUEL WAYNE JUDNICK IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR

More information

University of California at Berkeley College of Engineering Dept. of Electrical Engineering and Computer Sciences.

University of California at Berkeley College of Engineering Dept. of Electrical Engineering and Computer Sciences. Uversty of Clfor t Berkeley College of Egeerg et. of Electrcl Egeerg Comuter Sceces EE 5 Mterm I Srg 6 Prof. Mg C. u Feb. 3, 6 Gueles Close book otes. Oe-ge formto sheet llowe. There re some useful formuls

More information

DATA FITTING. Intensive Computation 2013/2014. Annalisa Massini

DATA FITTING. Intensive Computation 2013/2014. Annalisa Massini DATA FITTING Itesve Computto 3/4 Als Mss Dt fttg Dt fttg cocers the problem of fttg dscrete dt to obt termedte estmtes. There re two geerl pproches two curve fttg: Iterpolto Dt s ver precse. The strteg

More information

Rendering Equation. Linear equation Spatial homogeneous Both ray tracing and radiosity can be considered special case of this general eq.

Rendering Equation. Linear equation Spatial homogeneous Both ray tracing and radiosity can be considered special case of this general eq. Rederg quto Ler equto Sptl homogeeous oth ry trcg d rdosty c be cosdered specl cse of ths geerl eq. Relty ctul photogrph Rdosty Mus Rdosty Rederg quls the dfferece or error mge http://www.grphcs.corell.edu/ole/box/compre.html

More information

Proceedings of the ASME 2011 International Mechanical Engineering Congress & Exposition IMECE2011 November 11-17, 2011, Denver, Colorado, USA

Proceedings of the ASME 2011 International Mechanical Engineering Congress & Exposition IMECE2011 November 11-17, 2011, Denver, Colorado, USA Proceedgs of the ASME Itertol Mechcl Egeerg ogress & Exosto IMEE ovember -7 Dever olordo USA IMEE-6869 ALORITHM OF LAUHED VEHILES MASS ALULATIO AT THE EARLY STAE OF DESII IMEE-6869 A. AYUTDIOVA Dertmet

More information

Random variables and sampling theory

Random variables and sampling theory Revew Rdom vrbles d smplg theory [Note: Beg your study of ths chpter by redg the Overvew secto below. The red the correspodg chpter the textbook, vew the correspodg sldeshows o the webste, d do the strred

More information

The Computation of Common Infinity-norm Lyapunov Functions for Linear Switched Systems

The Computation of Common Infinity-norm Lyapunov Functions for Linear Switched Systems ISS 746-7659 Egd UK Jour of Iformto d Comutg Scece Vo. 6 o. 4. 6-68 The Comutto of Commo Ifty-orm yuov Fuctos for er Swtched Systems Zheg Che Y Go Busess Schoo Uversty of Shgh for Scece d Techoogy Shgh

More information

ME 501A Seminar in Engineering Analysis Page 1

ME 501A Seminar in Engineering Analysis Page 1 Mtr Trsformtos usg Egevectors September 8, Mtr Trsformtos Usg Egevectors Lrry Cretto Mechcl Egeerg A Semr Egeerg Alyss September 8, Outle Revew lst lecture Trsformtos wth mtr of egevectors: = - A ermt

More information

St John s College. UPPER V Mathematics: Paper 1 Learning Outcome 1 and 2. Examiner: GE Marks: 150 Moderator: BT / SLS INSTRUCTIONS AND INFORMATION

St John s College. UPPER V Mathematics: Paper 1 Learning Outcome 1 and 2. Examiner: GE Marks: 150 Moderator: BT / SLS INSTRUCTIONS AND INFORMATION St Joh s College UPPER V Mthemtcs: Pper Lerg Outcome d ugust 00 Tme: 3 hours Emer: GE Mrks: 50 Modertor: BT / SLS INSTRUCTIONS ND INFORMTION Red the followg structos crefull. Ths questo pper cossts of

More information

Department of Statistics, Dibrugarh University, Dibrugarh, Assam, India. Department of Statistics, G. C. College, Silchar, Assam, India.

Department of Statistics, Dibrugarh University, Dibrugarh, Assam, India. Department of Statistics, G. C. College, Silchar, Assam, India. A Dscrete Power Dstruto Surt Chkrort * d Dhrujot Chkrvrt Dertet of Sttstcs Drugrh Uverst Drugrh Ass Id. Dertet of Sttstcs G. C. College Slchr Ass Id. *el: surt_r@hoo.co. Astrct A ew dscrete dstruto hs

More information

Implementation of Nested Dissection Method Using Block Elimination

Implementation of Nested Dissection Method Using Block Elimination Itertol Jourl of Aled cece d echology Vol. o. Jury 0 Imlemetto of ested Dssecto ethod sg Bloc lmto Hshm er * Dertmet of themtcs College of cece d themtcs versty of orth Georg A Chrles Bry Dertmet of themtcs

More information

å 1 13 Practice Final Examination Solutions - = CS109 Dec 5, 2018

å 1 13 Practice Final Examination Solutions - = CS109 Dec 5, 2018 Chrs Pech Fal Practce CS09 Dec 5, 08 Practce Fal Examato Solutos. Aswer: 4/5 8/7. There are multle ways to obta ths aswer; here are two: The frst commo method s to sum over all ossbltes for the rak of

More information

Chapter Unary Matrix Operations

Chapter Unary Matrix Operations Chpter 04.04 Ury trx Opertos After redg ths chpter, you should be ble to:. kow wht ury opertos mes,. fd the trspose of squre mtrx d t s reltoshp to symmetrc mtrces,. fd the trce of mtrx, d 4. fd the ermt

More information

Math 1313 Final Exam Review

Math 1313 Final Exam Review Mth 33 Fl m Revew. The e Compy stlled ew mhe oe of ts ftores t ost of $0,000. The mhe s depreted lerly over 0 yers wth srp vlue of $,000. Fd the vlue of the mhe fter 5 yers.. mufturer hs mothly fed ost

More information

The definite Riemann integral

The definite Riemann integral Roberto s Notes o Itegrl Clculus Chpter 4: Defte tegrls d the FTC Secto 4 The defte Rem tegrl Wht you eed to kow lredy: How to ppromte the re uder curve by usg Rem sums. Wht you c ler here: How to use

More information

Preliminary Examinations: Upper V Mathematics Paper 1

Preliminary Examinations: Upper V Mathematics Paper 1 relmr Emtos: Upper V Mthemtcs per Jul 03 Emer: G Evs Tme: 3 hrs Modertor: D Grgortos Mrks: 50 INSTRUCTIONS ND INFORMTION Ths questo pper sts of 0 pges, cludg swer Sheet pge 8 d Iformto Sheet pges 9 d 0

More information

2SLS Estimates ECON In this case, begin with the assumption that E[ i

2SLS Estimates ECON In this case, begin with the assumption that E[ i SLS Estmates ECON 3033 Bll Evas Fall 05 Two-Stage Least Squares (SLS Cosder a stadard lear bvarate regresso model y 0 x. I ths case, beg wth the assumto that E[ x] 0 whch meas that OLS estmates of wll

More information

PTAS for Bin-Packing

PTAS for Bin-Packing CS 663: Patter Matchg Algorthms Scrbe: Che Jag /9/00. Itroducto PTAS for B-Packg The B-Packg problem s NP-hard. If we use approxmato algorthms, the B-Packg problem could be solved polyomal tme. For example,

More information

Numerical Differentiation and Integration

Numerical Differentiation and Integration Numerl Deretto d Itegrto Overvew Numerl Deretto Newto-Cotes Itegrto Formuls Trpezodl rule Smpso s Rules Guss Qudrture Cheyshev s ormul Numerl Deretto Forwrd te dvded deree Bkwrd te dvded deree Ceter te

More information

The Schur-Cohn Algorithm

The Schur-Cohn Algorithm Modelng, Estmton nd Otml Flterng n Sgnl Processng Mohmed Njm Coyrght 8, ISTE Ltd. Aendx F The Schur-Cohn Algorthm In ths endx, our m s to resent the Schur-Cohn lgorthm [] whch s often used s crteron for

More information

Analytical Approach for the Solution of Thermodynamic Identities with Relativistic General Equation of State in a Mixture of Gases

Analytical Approach for the Solution of Thermodynamic Identities with Relativistic General Equation of State in a Mixture of Gases Itertol Jourl of Advced Reserch Physcl Scece (IJARPS) Volume, Issue 5, September 204, PP 6-0 ISSN 2349-7874 (Prt) & ISSN 2349-7882 (Ole) www.rcourls.org Alytcl Approch for the Soluto of Thermodymc Idettes

More information

MATLAB PROGRAM FOR THE NUMERICAL SOLUTION OF DUHAMEL CONVOLUTION INTEGRAL

MATLAB PROGRAM FOR THE NUMERICAL SOLUTION OF DUHAMEL CONVOLUTION INTEGRAL Bullet of te Trslv Uversty of Brşov Vol 5 (54-1 Seres 1: Specl Issue No 1 MATLAB PROGRAM FOR THE NUMERICAL SOLUTION OF DUHAMEL CONVOLUTION INTEGRAL M BOTIŞ 1 Astrct: I te ler lyss of structures troug modl

More information

Computations with large numbers

Computations with large numbers Comutatos wth large umbers Wehu Hog, Det. of Math, Clayto State Uversty, 2 Clayto State lvd, Morrow, G 326, Mgshe Wu, Det. of Mathematcs, Statstcs, ad Comuter Scece, Uversty of Wscos-Stout, Meomoe, WI

More information

DISCRETE TIME MODELS OF FORWARD CONTRACTS INSURANCE

DISCRETE TIME MODELS OF FORWARD CONTRACTS INSURANCE G Tstsshvl DSCRETE TME MODELS OF FORWARD CONTRACTS NSURANCE (Vol) 008 September DSCRETE TME MODELS OF FORWARD CONTRACTS NSURANCE GSh Tstsshvl e-ml: gurm@mdvoru 69004 Vldvosto Rdo str 7 sttute for Appled

More information

Channel Models with Memory. Channel Models with Memory. Channel Models with Memory. Channel Models with Memory

Channel Models with Memory. Channel Models with Memory. Channel Models with Memory. Channel Models with Memory Chael Models wth Memory Chael Models wth Memory Hayder radha Electrcal ad Comuter Egeerg Mchga State Uversty I may ractcal etworkg scearos (cludg the Iteret ad wreless etworks), the uderlyg chaels are

More information

6.6 Moments and Centers of Mass

6.6 Moments and Centers of Mass th 8 www.tetodre.co 6.6 oets d Ceters of ss Our ojectve here s to fd the pot P o whch th plte of gve shpe lces horzotll. Ths pot s clled the ceter of ss ( or ceter of grvt ) of the plte.. We frst cosder

More information

Differential Entropy 吳家麟教授

Differential Entropy 吳家麟教授 Deretl Etropy 吳家麟教授 Deto Let be rdom vrble wt cumultve dstrbuto ucto I F s cotuous te r.v. s sd to be cotuous. Let = F we te dervtve s deed. I te s clled te pd or. Te set were > 0 s clled te support set

More information

Physics 220: Worksheet5 Name

Physics 220: Worksheet5 Name ocepts: pctce, delectrc costt, resstce, seres/prllel comtos () coxl cle cossts of sultor of er rdus wth chrge/legth +λ d outer sultg cylder of rdus wth chrge/legth -λ. () Fd the electrc feld everywhere

More information

European Journal of Mathematics and Computer Science Vol. 3 No. 1, 2016 ISSN ISSN

European Journal of Mathematics and Computer Science Vol. 3 No. 1, 2016 ISSN ISSN Euroe Jour of Mthemtcs d omuter Scece Vo. No. 6 ISSN 59-995 ISSN 59-995 ON AN INVESTIGATION O THE MATRIX O THE SEOND PARTIA DERIVATIVE IN ONE EONOMI DYNAMIS MODE S. I. Hmdov Bu Stte Uverst ABSTRAT The

More information

D KL (P Q) := p i ln p i q i

D KL (P Q) := p i ln p i q i Cheroff-Bouds 1 The Geeral Boud Let P 1,, m ) ad Q q 1,, q m ) be two dstrbutos o m elemets, e,, q 0, for 1,, m, ad m 1 m 1 q 1 The Kullback-Lebler dvergece or relatve etroy of P ad Q s defed as m D KL

More information

2. Independence and Bernoulli Trials

2. Independence and Bernoulli Trials . Ideedece ad Beroull Trals Ideedece: Evets ad B are deedet f B B. - It s easy to show that, B deedet mles, B;, B are all deedet ars. For examle, ad so that B or B B B B B φ,.e., ad B are deedet evets.,

More information

Functions of Random Variables

Functions of Random Variables Fuctos of Radom Varables Chapter Fve Fuctos of Radom Varables 5. Itroducto A geeral egeerg aalyss model s show Fg. 5.. The model output (respose) cotas the performaces of a system or product, such as weght,

More information

An Alternative Method to Find the Solution of Zero One Integer Linear Fractional Programming Problem with the Help of -Matrix

An Alternative Method to Find the Solution of Zero One Integer Linear Fractional Programming Problem with the Help of -Matrix Itertol Jourl of Scetfc d Reserch Pulctos, Volue 3, Issue 6, Jue 3 ISSN 5-353 A Altertve Method to Fd the Soluto of Zero Oe Iteger Ler Frctol Progrg Prole wth the Help of -Mtr VSeeregsy *, DrKJeyr ** *

More information

5. Lighting & Shading

5. Lighting & Shading 3 4 Rel-Worl vs. eg Rel worl comlex comuttos see otcs textoos, hotorelstc reerg eg smlfe moel met, ffuse seculr lght sources reflectos esy to tue fst to comute ght sources ght reflecto ght source: Reflecto

More information

EVALUATING COMPARISON BETWEEN CONSISTENCY IMPROVING METHOD AND RESURVEY IN AHP

EVALUATING COMPARISON BETWEEN CONSISTENCY IMPROVING METHOD AND RESURVEY IN AHP ISAHP 00, Bere, Stzerld, August -4, 00 EVALUATING COMPARISON BETWEEN CONSISTENCY IMPROVING METHOD AND RESURVEY IN AHP J Rhro, S Hlm d Set Wto Petr Chrst Uversty, Surby, Idoes @peter.petr.c.d sh@peter.petr.c.d

More information

UNIVERSITI KEBANGSAAN MALAYSIA PEPERIKSAAN AKHIR SEMESTER I SESI AKADEMIK 2007/2008 IJAZAH SARJANAMUDA DENGAN KEPUJIAN NOVEMBER 2007 MASA : 3 JAM

UNIVERSITI KEBANGSAAN MALAYSIA PEPERIKSAAN AKHIR SEMESTER I SESI AKADEMIK 2007/2008 IJAZAH SARJANAMUDA DENGAN KEPUJIAN NOVEMBER 2007 MASA : 3 JAM UNIVERSITI KEBANGSAAN MALAYSIA PEPERIKSAAN AKHIR SEMESTER I SESI AKADEMIK 7/8 IJAZAH SARJANAMUDA DENGAN KEPUJIAN NOVEMBER 7 MASA : 3 JAM KOD KURSUS : KKKQ33/KKKF33 TAJUK : PENGIRAAN BERANGKA ARAHAN :.

More information

CS473-Algorithms I. Lecture 3. Solving Recurrences. Cevdet Aykanat - Bilkent University Computer Engineering Department

CS473-Algorithms I. Lecture 3. Solving Recurrences. Cevdet Aykanat - Bilkent University Computer Engineering Department CS473-Algorthms I Lecture 3 Solvg Recurreces Cevdet Aykt - Blket Uversty Computer Egeerg Deprtmet Solvg Recurreces The lyss of merge sort Lecture requred us to solve recurrece. Recurreces re lke solvg

More information

12 Iterative Methods. Linear Systems: Gauss-Seidel Nonlinear Systems Case Study: Chemical Reactions

12 Iterative Methods. Linear Systems: Gauss-Seidel Nonlinear Systems Case Study: Chemical Reactions HK Km Slghtly moded //9 /8/6 Frstly wrtte t Mrch 5 Itertve Methods er Systems: Guss-Sedel Noler Systems Cse Study: Chemcl Rectos Itertve or ppromte methods or systems o equtos cosst o guessg vlue d the

More information

A Brief Introduction to Olympiad Inequalities

A Brief Introduction to Olympiad Inequalities Ev Che Aprl 0, 04 The gol of ths documet s to provde eser troducto to olympd equltes th the stdrd exposto Olympd Iequltes, by Thoms Mldorf I ws motvted to wrte t by feelg gulty for gettg free 7 s o problems

More information

ICS141: Discrete Mathematics for Computer Science I

ICS141: Discrete Mathematics for Computer Science I Uversty o Hw ICS: Dscrete Mthemtcs or Computer Scece I Dept. Iormto & Computer Sc., Uversty o Hw J Stelovsy bsed o sldes by Dr. Be d Dr. Stll Orgls by Dr. M. P. Fr d Dr. J.L. Gross Provded by McGrw-Hll

More information

Sequences and summations

Sequences and summations Lecture 0 Sequeces d summtos Istructor: Kgl Km CSE) E-ml: kkm0@kokuk.c.kr Tel. : 0-0-9 Room : New Mleum Bldg. 0 Lb : New Egeerg Bldg. 0 All sldes re bsed o CS Dscrete Mthemtcs for Computer Scece course

More information

MATH2999 Directed Studies in Mathematics Matrix Theory and Its Applications

MATH2999 Directed Studies in Mathematics Matrix Theory and Its Applications MATH999 Drected Studes Mthemtcs Mtr Theory d Its Applctos Reserch Topc Sttory Probblty Vector of Hgher-order Mrkov Ch By Zhg Sho Supervsors: Prof. L Ch-Kwog d Dr. Ch Jor-Tg Cotets Abstrct. Itroducto: Bckgroud.

More information

Keywords: Heptic non-homogeneous equation, Pyramidal numbers, Pronic numbers, fourth dimensional figurate numbers.

Keywords: Heptic non-homogeneous equation, Pyramidal numbers, Pronic numbers, fourth dimensional figurate numbers. [Gol 5: M 0] ISSN: 77-9655 IJEST INTENTIONL JOUNL OF ENGINEEING SCIENCES & ESECH TECHNOLOGY O the Hetc No-Hoogeeous Euto th Four Ukos z 6 0 M..Gol * G.Suth S.Vdhlksh * Dertet of MthetcsShrt Idr Gdh CollegeTrch

More information

Regression. By Jugal Kalita Based on Chapter 17 of Chapra and Canale, Numerical Methods for Engineers

Regression. By Jugal Kalita Based on Chapter 17 of Chapra and Canale, Numerical Methods for Engineers Regresso By Jugl Klt Bsed o Chpter 7 of Chpr d Cle, Numercl Methods for Egeers Regresso Descrbes techques to ft curves (curve fttg) to dscrete dt to obt termedte estmtes. There re two geerl pproches two

More information

Z = = = = X np n. n n. npq. npq pq

Z = = = = X np n. n n. npq. npq pq Stt 4, secto 4 Goodess of Ft Ctegory Probbltes Specfed otes by Tm Plchowsk Recll bck to Lectures 6c, 84 (83 the 8 th edto d 94 whe we delt wth populto proportos Vocbulry from 6c: The pot estmte for populto

More information

Available online through

Available online through Avlble ole through wwwmfo FIXED POINTS FOR NON-SELF MAPPINGS ON CONEX ECTOR METRIC SPACES Susht Kumr Moht* Deprtmet of Mthemtcs West Begl Stte Uverst Brst 4 PrgsNorth) Kolt 76 West Begl Id E-ml: smwbes@yhoo

More information

TiCC TR November, Gauss Sums, Partitions and Constant-Value Codes. A.J. van Zanten. TiCC, Tilburg University Tilburg, The Netherlands

TiCC TR November, Gauss Sums, Partitions and Constant-Value Codes. A.J. van Zanten. TiCC, Tilburg University Tilburg, The Netherlands Tlburg ceter for Cogto d Coucto P.O. Box 953 Tlburg Uversty 5 LE Tlburg, The Netherlds htt://www.tlburguversty.edu/reserch/sttutes-d-reserch-grous/tcc/cc/techcl-reorts/ El: tcc@uvt.l Coyrght A.J. v Zte,

More information

Review of Linear Algebra

Review of Linear Algebra PGE 30: Forulto d Soluto Geosstes Egeerg Dr. Blhoff Sprg 0 Revew of Ler Alger Chpter 7 of Nuercl Methods wth MATLAB, Gerld Recktewld Vector s ordered set of rel (or cople) uers rrged s row or colu sclr

More information

1 4 6 is symmetric 3 SPECIAL MATRICES 3.1 SYMMETRIC MATRICES. Defn: A matrix A is symmetric if and only if A = A, i.e., a ij =a ji i, j. Example 3.1.

1 4 6 is symmetric 3 SPECIAL MATRICES 3.1 SYMMETRIC MATRICES. Defn: A matrix A is symmetric if and only if A = A, i.e., a ij =a ji i, j. Example 3.1. SPECIAL MATRICES SYMMETRIC MATRICES Def: A mtr A s symmetr f d oly f A A, e,, Emple A s symmetr Def: A mtr A s skew symmetr f d oly f A A, e,, Emple A s skew symmetr Remrks: If A s symmetr or skew symmetr,

More information

Chapter Simpson s 1/3 Rule of Integration. ( x)

Chapter Simpson s 1/3 Rule of Integration. ( x) Cpter 7. Smpso s / Rule o Itegrto Ater redg ts pter, you sould e le to. derve te ormul or Smpso s / rule o tegrto,. use Smpso s / rule t to solve tegrls,. develop te ormul or multple-segmet Smpso s / rule

More information

The linear system. The problem: solve

The linear system. The problem: solve The ler syste The prole: solve Suppose A s vertle, the there ests uue soluto How to effetly opute the soluto uerlly??? A A A evew of dret ethods Guss elto wth pvotg Meory ost: O^ Coputtol ost: O^ C oly

More information

CHAPTER 7 SOLVING FUZZY LINEAR FRACTIONAL PROGRAMMING PROBLEM BY USING TAYLOR'S SERIES METHOD

CHAPTER 7 SOLVING FUZZY LINEAR FRACTIONAL PROGRAMMING PROBLEM BY USING TAYLOR'S SERIES METHOD 67 CHAPTER 7 SOLVING FUZZY LINEAR FRACTIONAL PROGRAMMING PROBLEM BY USING TAYLOR'S SERIES METHOD 7. INTRODUCTION The eso mers the setors le fl ororte lg routo lg mretg me seleto uversty lg stuet mssos

More information

i+1 by A and imposes Ax

i+1 by A and imposes Ax MASSACHUSETTS INSTITUTE OF TECHNOLOGY DEPARTMENT OF MECHANICAL ENGINEERING CAMBRIDGE, MASSACHUSETTS 09.9 NUMERICAL FLUID MECHANICS FALL 009 Mody, October 9, 009 QUIZ : SOLUTIONS Notes: ) Multple solutos

More information

CHAPTER 3 NETWORK ADMITTANCE AND IMPEDANCE MATRICES

CHAPTER 3 NETWORK ADMITTANCE AND IMPEDANCE MATRICES CHAPTER NETWORK ADTTANCE AND PEDANCE ATRCES As we hve see i Chter tht ower system etwor c e coverted ito equivlet imedce digrm. This digrm forms the sis of ower flow (or lod flow) studies d short circuit

More information

ON NILPOTENCY IN NONASSOCIATIVE ALGEBRAS

ON NILPOTENCY IN NONASSOCIATIVE ALGEBRAS Jourl of Algebr Nuber Theory: Advces d Applctos Volue 6 Nuber 6 ges 85- Avlble t http://scetfcdvces.co. DOI: http://dx.do.org/.864/t_779 ON NILOTENCY IN NONASSOCIATIVE ALGERAS C. J. A. ÉRÉ M. F. OUEDRAOGO

More information