Optimal Path Routing in Single and Multiple Clock Domain Systems


 Kathlyn Stafford
 2 years ago
 Views:
Transcription
1 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 1 Opimal Pah Rouing in Single and Muliple Clock Domain Syem Soha Haoun, Senior Member, IEEE, Charle J. Alper, Senior Member, IEEE ) Abrac Shrinking proce geomerie and he increaing ue of IP componen in SoC deign give rie o new problem in rouing and buffer inerion. A paricular concern i ha crochip rouing will require muliple clock cycle. Anoher i he inegraion of independenly clocked componen. Thi paper explore imulaneou rouing and buffer inerion in he conex of ingle and muliple clock domain. We preen wo opimal and efficien polynomial algorihm ha build upon he dynamic programming Fa Pah framework propoed in [17]. The fir algorihm olve he problem of finding he minimum laency pah for a ingle clockdomain yem. The econd conider rouing beween wo componen ha are locally ynchronou ye globally aynchronou o each oher. Boh algorihm can be ued o eimae communicaion overhead for inerconnec and reource planning. Experimenal reul verify he correcne and pracicaliy of our approach. I. INTRODUCTION Three diinc rend will poe new rouing challenge in fuure SoC deign. Fir, SoC will uilize everal IP (Inellecual Propery) componen, boh of and hard, like embedded proceor and memorie. Thi apec will allow IP reue and i will reduce ime o marke. A hore pah for a ignal beween wo chip componen may hu be obruced by IP block. Second, he drive for higher performance will coninue o puh he clocking frequencie. Third, hrinking proce geomerie and improvemen in proce echnologie allow building bigger die. Muliple clock cycle will be required o cro a chip. Furhermore, if he IP are clocked a differen frequencie, a i he cae wih a hard IP ha ofen ha a fixed clock period, hen he ignal roue mu cro from one ime domain (of he ender) o anoher (ha of he receiver) lached hrough he proper circuiry. In conra o a yem wih a ingle clock, or a ingleclock domain yem, a yem wih muliple clock i ofen referred o a a mulipleclock domain yem. The clocking cheme i referred o a Globally Aynchronou Locally Synchronou, or GALS [3], [10]. When muliple clock cycle are needed o roue a ignal acro a chip clocked by he ame clock or i derivaive, hree oluion are poible. The fir oluion i combinaional where he delay of he buffered ignal from ender o receiver i more han one clock cycle. The receiver hen uilize Manucrip received Augu 2, 2002; revied May 19, Thi work wa uppored in par by NSF CAREER gran , and by a Tuf Univeriy Mellon gran. Soha Haoun i wih he Compuer Science Deparmen a Tuf Univeriy, Medford, MA, 02155, U.S.A. Charle J. Alper i wih IBM Auing Reearch Laboraorie, Auin, TX, 78758, U.S.A. circuiry o coun a predeignaed number of cycle before laching new daa. The diadvanage of hi echnique i ha conecuive end canno be overlapped and he hroughpu of he channel i eriouly degraded. A econd oluion i pipelining he roued ignal hrough ynchronizer (edgeriggered regier or leveleniive lache). Buffer are inered whenever needed o boo he elecric ignal and opimize he delay in beween he ynchronizer. The clock ignal i roued o each of he ynchronizer. The hird oluion i wave pipelining. I eliminae one or more regier along he roue, hu allowing he imulaneou exience of everal wavefron along a wire. The key in making wave pipelining a ucce i enuring ha ucceive waveform do no inerfere. While reducing he clock load and eaing clock diribuion, wave pipelining i very eniive o delay, proce, and emperaure variaion effec ha are even more pronounced for long roue. When rouing a ignal beween wo ynchronou domain clocked by differen and unrelaed clock, he criical iue ha mu be addreed i meaabiliy: he clock of he laching regier and he daa may wich imulaneouly. The regier oupu hen ele ino an undefined region neiher a logical high nor a logical low. Several oluion have been propoed o alleviae hi problem [4], [15], [14], [11]. The qualiy of he oluion depend on he mean imeofailure which ofen require an increae in laency or addiional complicaion in he ynchronizaion cheme. Thi paper addree wo problem relaed o rouing and buffering in fuure SoC deign. The fir problem eek an opimal bufferedrouing pah wih ynchronizer and buffer inerion wihin a ingleclock domain yem. The objecive i o minimize he cycle laency, or equivalenly, he oal number of regier along he roue. Conrain are impoed o enure ha when a ignal i lached by a regier, he reuling regieroregier delay are le han he clock period. We wih o avoid phyical obacle, o roue hrough block uch a daapah, and o buffer he ignal a needed o minimize he laency. The econd problem eek o opimize he rouing wihin a muliclock domain yem. Here, we adop he mulidomain communicaion circuiry propoed by Chelcea and Nowick [4] which buffer he ignal from one clock domain o anoher via a pecial circui rucure, a MuliClock FIFO. Thi MCFIFO boh minimize poin of meaabiliy in he deign and allow for buffering (emporarily holding) daa. If he MCFIFO i placed more han one ender clock cycle away from he ender, or more han one receiver clock cycle away from he receiver, hen ynchronizaion of he roued ne o he appropriae
2 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 2 clock i needed. Relay Saion, fir propoed by Carloni e al. [2], are uch ynchronizaion elemen. They provide boh ynchronizaion and emporary orage o allow he correc operaion of he MCFIFO. While our algorihm pecifically ue he Relay Saion and he MCFIFO, i can be eaily adoped o uilize imilar ynchronizaion elemen. Thu, our econd problem conider imulaneou rouing, buffer inerion, relay aion inerion, and MCFIFO inerion o achieve he minimum ime laency. Thi correpond o minimizing he laency of ending daa along he ne when he MCFIFO i empy. Our propoed algorihm o olve hee wo problem are baed on he he Fa Pah framework propoed by Zhou, Wong, Liu, and Aziz [17]. The acual Fa Pah algorihm find a minimum delay pah for a ne while imulaneouly exploring all buffering and rouing oluion. We exend he algorihm o handle addiional ynchronizaion elemen uch a regier, relay aion, and MCFIFO while impoing he iming conrain required by he phyical diance and he communicaion proocol. Recen work in rouing acro muliple clock cycle wihin a ingle domain adop differen approache. Lu e al. decribe a regier/repeaer block planning mehod during archiecural floorplanning/inerconnec planning age[13]. Their mehod i baed on idenifying feaible region in which flipflop and repeaer can be arbirarily inered o aify boh delay and cycle ime conrain. Cocchini exend van Ginneken dynamic programming framework[16] o opimally place regier and repeaer when given a ree rouing opology[5]. Haoun decribe an adapaion o he Fa Pah algorihm o conruc a clocked buffered roue uing leveleniive lache[9]. The algorihm preened here can be uilized eiher for inerconnec planning purpoe or for realizing he final rouing implemenaion. During he deign planning proce, rouing eimae can be achieved during archiecural exploraion o ae communicaion overhead once an iniial floorplan i conruced. Wire widh and layer aignmen are aumed. The early deecion of communicaion overhead allow archiec o explore microarchiecure radeoff ha hide communicaion laencie. Once a roue eimae i obained, he RTLlevel deign decripion i updaed o reflec he added laency aociaed wih mulicycle rouing, and o enure a cycleaccurae deign decripion ha can be properly verified. The algorihm preened here could alo be ued during backend deign o ynheize phyical roue. In hi cae, he roue laency ha been eimaed, and he algorihm deermine he opimal buffer and ynchronizer placemen. The addiion of he regier may complicae he deign flow, verificaion, and DFT. We however believe ha verificaion and eing mehodologie will evolve o accommodae he neceiy of pipelined roue. The remainder of he paper i a follow. Secion 2 preen a deailed overview of he Fa Pah algorihm. Secion 3 inroduce he ingledomain rouing and ynchronized buffering problem and how how o adap he Fa Pah framework o olve i. Secion 4 overview MCFIFO and relay aion communicaion cheme, hereby leading ino a dicuion of he problem of opimal pah conrucion in deign realized uing muliple clock domain. Finally, Secion 5 preen experimenal reul and we conclude in Secion 6. II. BACKGROUND: THE FAST PATH ALGORITHM In rouing in ingle and muliple clock domain, we wih o explore all rouing and ynchronizer inerion opion wihin a given rouing area. Many apec of he Fa Pah algorihm[17] can be exploied o achieve hi goal. The Fa Pah algorihm find he minimum delay ourceoink buffered rouing pah, while conidering boh phyical obacle (e.g., due o IP, memorie, and oher macro block) and wiring blockage (e.g., daa pah). To model phyical and wiring blockage, one may conruc a grid graph G V E (a in [17], [1], [7]) over he poenial rouing area, whereby each node correpond o a poenial inerion poin for a buffer or ynchronizaion elemen, and each edge correpond o par of a poenial roue. Edge in he grid graph which overlap wring blockage are deleed, and node ha overlap phyical obacle are labeled blocked. More preciely, we define a label funcion p : V 0 1 where p v 0 if v V overlap a phyical obacle and p v 1 oherwie. For each edge u v E, le R u v and C u v denoe he capaciance and reiance of a wire connecing u o v. We ue uniform capaciance and reiance for a given lengh auming a fixed widh and layer aignmen. Le R g, K(g), and C g repecively denoe he reiance, inrinic delay, and inpu capaciance of a given buffer or ynchronizaion elemen g. Thee value are deermined for each buffer or ynchronizaion elemen in he rouing library. We ue he reiancecapaciance πmodel o repreen he wire, a wichlevel model o repreen he gae, and he Elmore model o compue pah delay. A pah from node o in he grid graph G i a equence of node v 1 v 2 v k. An opimized pah from o i a pah plu an addiional labeling m of node in he pah. We have m g, m g, and m v i I 0, where I i he e of buffer or ynchronizaion elemen which may be inered on a node in he pah beween ource and ink. Here, g i he driving gae locaed a, g i he receiving gae locaed a, and each inernal node v may be aigned a gae from he e I or no have a gae (correponding o m v 0). A pah i feaible if and only if p v 1 whenever m v I. We aume ha m i iniialized o m g, m g, and m v 0 for v V. Le B be a buffer library coniing of noninvering buffer. The minimumdelay buffered pah problem, or he Fa Pah, can be expreed a follow: Given a rouing graph G V E, he e I B, and wo node V, find a feaible opimized pah from o uch ha he delay from o i minimized. Thi problem can be opimally olved by he Fa Pah algorihm [17] and alo by he hore pah formulaion propoed by Lai and Wong [12]. The laer formulaion can alo be exended o wire izing. We chooe o exend he Fa Pah algorihm o handle he nex wo formulaion ince i doe no require any lookup able compuaion and i likely more efficien when here i no wire izing.
3 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 3 Fig. 1. Fa Pah G B m Inpu: G V E Rouing grid graph B Buffer library ource node ink node m iniial labeling Var: Q prioriy queue of candidae α c d m v Candidae a v Oupu: m Complee labeling of  pah 1. Q C m 0 m. 2. while ( Q /0 ) do 3. c m b u exrac min Q 4. if c 0 hen reurn labeling m. 5. if u hen d d R m c K m puh 0 d m u ono Q and prune. coninue 6. for each u v E do c c C u v d d R u v c C u v 2 puh c d m v ono Q and prune 7. if p u 1 and m u 0 hen 8. for each b B do c C b d d R b c K b m u b puh c d m v ono Q and prune The Fa Pah algorihm. The main idea behind he Fa Pah algorihm i o exend Dijkra hore pah algorihm o do a general labeling baed on Elmore delay. Le he quadruple α c d m v repreen a parial oluion a node v where c i he curren inpu capaciance een a v, d i he delay from v o, and m i a labeling for he buffered pah from v o. The oluion α 1 c 1 d 1 m 1 v i aid o be inferior o α 2 c 2 d 2 m 2 v if c1 c2 and d1 d 2. For any pah from o hrough v, a buffer aignmen of m 1 from v o i guaraneed o no be beer han a buffer aignmen of m 2 from v o. Thu, α 1 can be afely deleed (or pruned) wihou acrificing opimaliy. Peudocode of he Fa Pah algorihm [17] i given in Figure 1. The core daa rucure ued by Fa Pah i a prioriy queue of candidae ha key off of he candidae delay value. The algorihm begin by iniializing Q o he e conaining a ingle ink candidae. Then, candidae are ieraively deleed from he Q and expanded eiher o add an edge (Sep 6) or a buffer from he library (Sep 7 and 8). If he ource i reached, i i puhed ono he Q in Sep 5, and when i i evenually popped from he queue, i i reurned a he opimum oluion (Sep 4). Wih each addiion o he queue, candidae for he curren node are checked for inferioriy and hen pruned accordingly. If we aume ha G ha n verice, E 4n (which i rue for a grid graph), and B k, he complexiy of Fa Pah i O n 2 k 2 lognk. III. SINGLE CLOCK DOMAIN ROUTING We now explore he problem of finding a buffered rouing pah from o when muliple clock cycle are required. Rouing over large diance in increaingly aggreive echnologie will require everal clock cycle o cro he die. Hence, one mu periodically clock he ignal by inering ynchronizaion elemen (uch a regier) along he ignal pah. In hi cae, one canno imply rea a regier like a buffer and add he regier delay o he exiing pah delay in he Fa Pah algorihm. The realizable delay beween conecuive regier on a pah will alway be deermined by he clock period, regardle of he acual ignal propagaion ime. Regieroregier ubpah wih delay larger han he permiible clock cycle are no permied. T T T T φ φ 4T φ Fig. 2. Example of he ingle clock domain rouing. The laency i deermined by he number of regier. Le r denoe he regier o be ued for inerion, T φ he clock period, and Seup r o be he eup ime for r. We exend he definiion of node labeling o permi regier aignmen, i.e., m v r for any node v V. We alo aume ha he ource and ink are ynchronizaion elemen, o ha g g r. We define he clock period conrain a follow: a bufferregier pah i feaible if and only if p v 1 whenever m v I and he buffered pah delay beween conecuive regier i le han or equal o T φ Seup r. Since a regier releae i ignal wih each clock wich, he  pah laency i given by T φ! p " 1, where p i he number of regier on he  pah. For example, Figure 2 how an  pah wih hree regier beween and, which mean i ake four clock cycle o ravere from o. Noe ha in he figure he conecuive regier have differen pacing, bu he delay i alway meaured a T φ beween regier. Figure 3 how an example of a bufferedregier pah on a grid graph wih boh circui and wire blockage. The phyical area occupied by a regier can be differen from ha occupied by a buffer. The regier area i a funcion of he underlying circui echnique. In addiion, rouing he clock o clock he regier migh caue added congeion. Our algorihm can be eaily modified o allow regier blockage ha preven inering regier a undeirable grid poin. The problem of finding he minimum bufferregier pah from o can now be aed a: Problem 1: Given a rouing graph G V E, he e I B r, and wo node V, find a feaible bufferregier pah from o uch ha he laency from o i minimized. The objecive i alo equivalen o minimizing v m v r #. To olve Problem 1, one migh iniially ry applying he Fa Pah algorihm and rea he regier like a member of he buffer library wih he following cavea: candidae which violae he regier o regier delay conrain are immediaely pruned. However, he Fa Pah pruning cheme will no behave correcly. Conider he wo parial oluion from v o in Figure 4. Here d 1 and d 2 are he delay from v o he fir regier φ φ
4 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 4 Fig. 3. circui blockage wire blockage An example of rouing wihin a ingleclock domain. pah 1 pah 2 d 2 d 1 v Fig. 4. An example of parial rouing oluion ha canno be compared for pruning. in he op and boom pah, repecively. The op pah ha delay 2T φ " d 1, while he boom pah ha delay T φ " d 2. Since feaibiliy require boh d 1 and d 2 mu be no greaer han T φ Seup r, he boom pah delay i le han he op pah delay. In addiion, ince here i a buffer on he boom pah cloe o v, v ee le downream capaciance on he boom pah han on he op pah. Since he op pah i inferior o he boom pah in erm of capaciance and delay, he fa pah algorihm would prune he candidae correponding o he op pah. However, conider coninuing he roue o node u on he oher ide of he circui blockage from v. I i cerainly poible ha he delay from u o v plu d 2 exceed he delay feaibiliy conrain, while he op pah delay from u o v plu d 1 doe no. In hi cae, only he op pah can uccefully be roued from v o u while ill meeing feaibiliy requiremen. Conequenly, he op pah canno be pruned. The key obervaion i ha one hould only prune ubpah by comparing o oher ubpah wih he ame number of regier. In he previou example, comparing a oneregier pah o a woregier pah lead o an unreolvable inconiency. Had he boom pah had wo regier, hen i could no have had maller delay han he op pah. Thu, one can ill ue he Fa Pah algorihmic framework a long a candidae propagaion proceed in wave of parial u oluion wherein each wave correpond o a differen number of regier. Figure 5 how how one can adap he Fa Pah framework o accomplih hi in he RegieredBuffered Pah (RBP) algorihm. RBP Algorihm G B m r T φ Inpu: G V E Rouing grid graph B Buffer library ource node ink node m iniial labeling wih m m r r regier for clocking ignal T φ required clock period. Var: Q prioriy queue of candidae Q queue holding nex candidae wave α c d m v Candidae a v A Marking of regiered node Oupu: m Labeling of complee  pah 1. Q C r Seup r m. Q /0, A v 0, v V 2. while ( Q /0 ) and ( Q /0 ) do if ( Q /0 ) hen Q Q, Q /0. 3. c d m u exrac min Q 4. if u hen d d R m c K m if d T φ hen reurn labeling m. 5. for each u v E do c c C u v d d R u v c C u v 2 if d T φ K r min R B r c hen puh c d m v ono Q and prune 6. if p u 1 and m u 0 hen 7. for each b B do c C b, m u b d d R b c + K(b) if d T φ K r hen puh c d m u ono Q and prune 8. if A u 0 and d R r c K r T φ hen m u r, A u 1 puh C r Seup r m u ono Q Fig. 5. The RegieredBuffered Pah (RBP) Algorihm. The primary difference beween RBP and he Fa Pah algorihm are a follow: RBP ue a econd queue Q o ore candidae oluion for he ubequen propagaion wave. When a regier i added o a candidae ha i popped from Q i i added o Q and proceed only afer he curren wave i compleed. Thi puhing ono Q i accomplihed in Sep 8, whereby candidae are added only if he inerion of he regier aifie he clock feaibiliy conrain. RBP combine Sep 4 and 5 from Figure 1 ino a ingle Sep 4. RBP ha he luxury of knowing ha a oon a i reached, a minimum laency oluion i guaraneed, hence i can immediaely reurn he oluion, a oppoed o puhing i back ono he queue like Fa Pah. When inering a regier for a candidae a node v, i i guaraneed o be he minimum laency candidae from v o uch ha m v r. Hence, here i no need o conider oher candidae wih m v R. We ue he array A o mark wheher a oluion wih m v r ha been generaed
5 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 5 for node r o preven muliple candidae wih regier inered a v. Sep 8 only add a regier o an exiing candidae if none already exi. The clock feaibiliy conrain i checked before puhing new candidae ono Q in Sep 7 and 8. Thi preven oluion ha can never lead o feaible oluion from furher exploring he grid graph. RBP proceed by expanding all bufferedpah oluion, ju like Fa Pah, unil any furher exploraion violae he clock period conrain. A hi poin Q conain everal newly generaed candidae all of which are ready for wavefron expanion from a node wih an inered regier. Sep 2 dump hee candidae ino Q and reiniialize Q o he empy e. Thee ingleregier candidae are hen expanded, generaing doubleregier candidae ha are ored in Q, ec. When here are no blockage, he wavefron expanion look like Figure 6. Of coure, wih blockage, he wave fron are no nearly a regular Fig. 6. Wavefron expanion (a in [6]). Le N be he number of node ha can be reached from a given node in one clock cycle. When he clock period i ufficienly hor, N n, he complexiy of he RBP algorihm i O nnk 2 lognk. The analyi i imilar o ha of he Fa Pah algorihm [17]. Since he number of elemen in he queue can be a mo O Nk, he ime required o iner an elemen ino he queue O log Nk. Becaue he number of inerion operaion of eiher buffer or lache i bounded by nnk 2, he complexiy i O nnk 2 log Nk. Thi i a leer ime complexiy han he Fa Pah algorihm. The compuaional aving occur becaue we do no have o wae reource exploring he many pah ha violae he clock period conrain. Thi peedup can be een in pracice in he experimenal reul when oberving he number of configuraion ha are examined a well a he run ime. While our algorihm opimize for roue laency, i can be eaily modified o find he minimumlaency pah ha maximize he um of he lack a he ource and a he ink. Each candidae oluion will keep rack of he he ink lack value. When he algorihm reache he ource (i.e. in Sep 4 in Figure 5), all minimumlaency oluion are explored o find he one ha maximize he um of he ink and ource lack. I i alo inereing o noe ha he RBP algorihm could be implemened in an alernaive manner. Inead of uing wo queue, one could ue an array of queue indexed by he number of regier. A new candidae i hen inered ino he queue indexed by he number of regier in he oluion repreened by ha candidae. The ime complexiy of he algorihm would no change, hough hi implemenaion ue more memory. A. Background IV. MULTIPLE CLOCK DOMAIN ROUTING A menioned in Secion I, we adop he MCFIFO propoed by Chelcea and Nowick [4] o roue a ignal beween wo differen clock domain. The MCFIFO i he baic eniy ha eablihe daa communicaion beween wo module operaing a differen frequencie. Like all FIFO, he MCFIFO ha a pu inerface o he ender and a ge inerface o he receiver. Each inerface i clocked by he communicaing domain clock a illuraed in Figure 7. If he full ignal i no aered, hen he ender can reque a pu (Pu Reque ignal) and daa i placed on he Pu Daa wire. The daa i lached ino he FIFO a he nex edge of he ender clock. If he empy ignal i no aered, hen he receiver can reque daa (Ge Reque ignal). The daa i hen made available a he receiver nex clocking edge. The Ge I Valid ignal deermine if he daa i valid. Sender Clock Domain Full Pu Reque Pu Daa Sender Clock Receiver Clock Domain Ge Reque Ge i Valid Empy Ge Daa Receiver Clock Fig. 7. The Mixed Clock FIFO ha inerface wo differen clock domain[4]. Becaue i may ake muliple ender clock cycle o roue a ne from i ource in he rouing grid o he MCFIFO, and muliple receiver clock cycle o roue he ne from he MCFIFO o he ink, ignal mu be ynchronized o he clock of each domain. Chelcea and Nowick exend he concep of a ingledomain relay aion [2] o do o. Thee aion eenially allow breaking long wire ino egmen ha correpond o clock cycle, and hen a chain of relay aion ac like a diribued FIFO. The ingledomain relay aion i hown in Figure 8. I conain a main regier and an auxiliary one. Iniially boh are empy and he conrol elec he main regier for oring
6 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 6 and reading he packe. When he SopIn ignal i aered, he nex incoming packe i ored in he auxiliary regier. SopOu i alo aered on he nex clock cycle o indicae ha he relay aion i full and canno furher accep new daa. packein wich main regier mux packeou T T T 2T + 2T Fig. 10. An example of an MCFIFOregier rouing. The MCFIFO in break he period ino wo clock domain where he period i T on he ource ide of he MCFIFO and T on he ink ide. The oal laency i 2T 2T. T auxiliary regier conrol CLK differen period, he oal laency i given by 2T " 2T. Thi correpond o he ignal flow auming an empy MCFIFO and ignore he wor cae ynchronizaion delay wihin he MCFIFO ha i common o all rouing oluion. Fig. 8. SopOu A relay aion[2]. SR SopIn To adap he ingledomain relay aion o inerface properly wih he MCFIFO, he relay aion bundle he Pu Reque and Pu Daa a he incoming packe, and he Ge I Valid and Ge Daa ignal a he ougoing packe. The full ignal in he MCFIFO i ued o op he incoming flow of packe. An MCFIFO and wo adjacen relay aion are hown in Figure 9. Clocked wih T Clocked wih T FIFO circui blockage wire blockage Fig. 9. Sender Clk Relay Saion Sender Clock domain packein Pu Reque Pu Daa SopOu Mixed Clock FIFO Receiver Clock domain Ge i Valid Ge Daa SopIn packeou Relay Saion The Mixed Clock FIFO and relay aion[4]. B. The GALS Algorihm Receiver CLK Alhough for he MuliClock domain problem we are uing he MCFIFO and relayaion ha require bidirecional ignal flow, we abrac he communicaion a being ingle direcional. We view relay aion a a regier r becaue boh have imilar delay properie. Given any buffered pah beween relay aion r 1 and r 2, if one aume a ingle buffer ype wih he ame delay characeriic a he regier, hen he Elmore delay from r 1 o r 2 i acually idenical o he Elmore delay from r 2 o r 1. Inering a buffer in our Muli Clock domain problem formulaion acually mean requiring he inerion of wo buffer, one for each direcion of ignal flow. Le f denoe he MCFIFO elemen ha mu be inered on he rouing pah, T o be clock period before f and T o be he clock period afer f. Figure 10 how an example where here are wo clock period beween and he MCFIFO and wo clock period afer he MCFIFO. Since he clock have Fig. 11. yem. An example illuraing rouing wihin a muliple domain B r f. For a muliclock domain ourceoink pah (an MCFIFO pah), we ue he following condiion for feaibiliy: an MCFIFO pah i feaible if and only if p v 1 whenever m v I, m v f for exacly one v V, he buffered pah delay beween conecuive regier beween and f i le han or equal o T Seup r, and he buffered pah delay beween conecuive regier beween f and i le han or equal o T Seup r. For example, Figure 11 how a oluion on he rouing graph wih a ingle MCFIFO wih laency T " 2T. Thu, he muliple clock domain, buffered rouing problem Our e of inerable elemen i now I become: Problem 2: Given a rouing graph G V E, he e I B r f, and wo node V, find a feaible MCFIFO pah from o uch ha he laency from o i minimized. One can adop a imilar framework a in RBP algorihm poenially inering an MCFIFO elemen for every candidae, wherever a regier i inered. We call hi new algorihm GALS for Globally Aynchronou, Locally Synchronou. There are everal modificaion o RBP ha mu be conidered o obain he GALS algorihm: A GALS candidae mu know if he MCFIFO ha been inered, o now we ue he ixuple α c d b v z l
7 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 7 where z 0 if α doe no conain an MCFIFO and z 1 oherwie. Le T 0 T and T 1 T be a funcion which reurn he required clock period for a given z value. The laency l i dicued below. GALS pruning ake place only wih candidae wih he ame value of z. Two candidae wih oppoing value of z are no direcly compared for pruning. Inead of oring a ingle li of candidae for each grid node, now we ore wo li: one for z 0 and one for z 1. If we have no ye inered an MCFIFO ono he pah, pruning i done wih repec o he z 0 li, oherwie, i i done wih repec o he z 1 li. T, one canno find idenical elemen for wave fron expanion a eaily a in he ingleclock domain cae. In RBP, he number of regier obviouly deermined he laency. For GALS, one fourregier pah may have laency 2T " 3T while anoher fourregier Becaue T pah ha laency T " 4T. Whichever pah ha maller laency mu be explored fir. Thu, he candidae value l ore he laency from he mo recenly inered regier or MCFIFO back o he ink. Ju like he RBP algorihm, d ill ore he combinaional delay from he curren node o he mo recenly inered regier or MCFIFO. The elemen in Q are ill ordered by d, bu he elemen in Q are ordered by l. We define he operaion Q ExracAllMin Q o pull all elemen off of Q wih he ame laency and load hem ino Q. Thi operaion exrac he nex wave fron of elemen wih equal laency from Q. In RBP, he fir regier inered a a grid node v preclude he need o iner regier a v for any oher pah. RBP ue A v 0 1 o repreen wheher a regier had been een in a pah a v. In GALS, we need o eparae he cae of inering a regier before f and afer f. Le A 0 v 0 1 repreen wheher a regier wa inered beween f and a v and A 1 v 0 1 repreen wheher a regier wa inered beween and f a v. Alo, le F v 0 1 denoe wheher an MCFIFO wa inered a v. Figure 12 give a emplae of he GALS algorihm. The main difference beween GALS and RBP i he addiion of Sep 9 for inering MCFIFO elemen. Ju like regier in RBP, GALS conider inering an MCFIFO a each poible inernal node a he wavefron expanion proceed. Oher difference include uing T z o look up he curren clock period conrain, reurning a oluion in Sep 4 only if i ha an MCFIFO, and he wavefron queue mechanim of Sep 2. If N i he number of node ha can be reached in max T T, he ime complexiy of GALS i O nnk 2 lognk which i ame a he RBP algorihm. V. EXPERIMENTAL RESULTS We obained code for he FaPah algorihm from he auhor of [17], hen implemened RBP and GALS uing hi framework. The code i wrien in C and wa run on a Sun Solari Enerprie 250. To perform he experimen, we ue GALS Algorihm G B m r f T T Inpu: G V E Rouing grid graph B Buffer library ource node ink node m iniial labeling wih m, m r regier for clocking ignal f MCFIFO elemen T required clock period from f o. T required clock period from o f. Var: Q prioriy queue keyed by d Q prioriy queue keyed by l α c d m v z l Candidae a v A 0 A 1 Marking of regiered node Oupu: m Labeling of complee  pah 1. Q C r Seup r m 0 0. Q /0, A 0 v A 1 v 0, v V 2. while ( Q /0 ) and ( Q /0 ) do if ( Q /0 ) hen Q ExracAllMin Q coninue. 3. c m b u z l exrac min Q 4. if u hen d d R m c K m if z 1 and d T hen reurn labeling m. 5. for each u v E do c c C u v d d R u v c C u v 2 if d T z hen puh c d m v z l ono Q and prune 6. if p u 1 and m u 0 hen 7. for each b B do c C b, m u b d d R b c + K(b) if d T z hen puh c d m u z l ono Q and prune 8. if A z u 0 and d R r c K r T z hen m u r, A z u 1 puh C r Seup r m u z l T z ono Q 9. if z 0, F u 0 and d R f c K f T z hen m u f, F u 1 puh C f Seup f m u 1 l T ono Q Fig. 12. The GALS Algorihm. eimaed parameer for a 0 07µ echnology a repored by Cong and Pan [8]. We ue a ingle buffer ize of 100 ime minimum gae widh, riple wide wire, and aume delay characeriic for he regier and MCFIFO o be idenical o ha of he buffer. A in [6], we ue a 25 by 25 mm chip and place he ource and ink 40 mm apar. Thee choice enure ha a ignifican number of clock cycle will be required o propagae a ignal from o. Our fir wo experimen focu on he RBP algorihm while he hird experimen focue on he GALS algorihm. A. Single Clock Domain wih Varying Period Our fir experimen inveigae he behavior of he RBP algorihm a a funcion of he clock period. Given a grid eparaion of 0 125mm and a grid ize of 200 x 200, we varied he number of regier ha can be placed along he pah ha i eparaed by 159 grid edge.
8 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 8 TABLE I FASTREGPATH STATISTICS AS A FUNCTION OF T φ. MAXREGSEP/MINREGSEP REFERS TO THE MAX/MIN NUMBER OF GRID POINTS BETWEEN SUCCESSIVE REGISTERS. MAX R/B SEP REFERS TO THE MAXIMUM GRID SEPARATION BETWEEN A REGISTER OR A BUFFER AND THE FOLLOWING REGISTER OR BUFFER. ON THIS GRID, THE NUMBER OF GRID EDGES ALONG A SHORTEST PATH BETWEEN SINK AND SOURCE WAS 159. T φ Laency Regier Buffer MaxRegSep MinRegSep Max R/B Sep Min R/B Sep Config MaxQSize ime (p) (p) (ec) The reul are ummarized in Table I. The fir daa row in he able (wih T φ ) preen he reul of running he Fa Pah algorihm, where he repored laency i acually he minimumbuffered pah delay. The oher row are he reul of running he RBP algorihm wih he indicaed clock period. 1 The clock period and laencie are repored in picoecond. The maximum and minimum regier eparaion i given, a well a he eparaion beween any inered elemen (i.e. a regier or a buffer followed by a conecuive regier or buffer). We make he following obervaion: A he clock period decreae, he number of regier along he pah increae while he number of buffer and he maximum and minimum regier eparaion all decreae. However, he maximum and minimum eparaion beween conecuive regier and buffer do no conienly decreae unle he clock period i o mall ha regier are inered every one or every oher grid poin. The number of configuraion inveigaed (i.e., candidae popped off he queue Q in Figure 5) decreae wih decreaing clock period. Thi empirically confirm ha he run ime complexiy of RBP become more efficien a he clock period decreae, becaue he pace of feaible wavefron expanion in a ingle cycle i reduced. Becaue RBP ha addiional overhead for regier inerion, only when he clock period drop below a cerain hrehold do we ee a runime improvemen over Fa Pah, e.g., in hi cae i i for T phi 463. B. Single Clock Domain wih Varying Grid Size Nex, we inveigae he behavior RBP a a funcion of he clock period and he grid eparaion. We experimened wih hree grid eparaion: 0.5mm,.25mm, and.125mm. We 1 The eemingly odd choice for he clock period are acually he fae clock period required o achieve he given number of regier (rounded o he neare picoecond). For example, a T φ of 686 i he fae clock period ha achieve a hree regier oluion. ummarize he reul in Table II. The fir daa row for each grid ize correpond o he reul of he Fa Pah algorihm, while he oher repreen he reul of running our RBP algorihm. We oberve he following: A he grid become more refined, Fa Pah laency improve lighly, from 2741 o 2739 p. The improvemen may be more ignifican when blockage along he pah are preen. I i poible o achieve a maller laency wih a more refined grid. For example, a a clock period of 925 and a 50! 50 grid, he laency i 3700, bu i i 2775 when we ue a 100! 100 grid. In ome cae, no improvemen were poible uch a for clock period 67 and 62. Wih a coare grid and a very mall clock period, i i impoible o find a rouing oluion a he grid eparaion demand placing regier le han one grid edge apar. The finer grid allow placing he regier cloer. No oluion for example a fond A clock period 53 and 49 for grid ize 50! 50, and for clock period 49 for grid ize 100! 100. Wih larger clock period, i i poible o achieve a laency cloe o he opimal bufferedpah delay. For example, uing a 200! 200 grid, a all clock period hown above 84p i i poible o be wihin one clock period from he opimal pah delay of C. GALS for Muliple Clock Domain Our final experimen explore he behavior of he GALS algorihm for differen period of he clock domain. Given our new problem aemen, comparing o Fa Pah i no poible. We imply illurae reul of he echnique. We ran GALS on he ame e cae in he previou experimen uing a grid eparaion of mm. Table III repor he number of buffer inered, he number of regier on he ink ide of he MCFIFO (Reg), he number of regier on he ource ide of he MCFIFO (Reg) and he laency. The relaive value of Reg and Reg indicae
9 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 9 TABLE II RBP PERFORMANCE AS A FUNCTION OF CLOCK PERIOD AND GRID SIZE. MAX. (MIN.) SEPARATION REFERS TO THE MAXIMUM (MINIMUM) BUFFER SEPARATION WHEN THE CLOCK PERIOD IS, AND TO THE CORRESPONDING REGISTER SEPARATION OTHERWISE. Grid Separaion: 0 5mm:50 50grid. Period(p) Regier Buffer Laency Max. Sep Min. Sep ime() Grid Separaion: 0 25mm: grid. Period Regier Buffer Laency Max. Sep Min. Sep ime() Grid Separaion: 0 125mm: grid. Period Regier Buffer Laency Max. Sep Min. Sep ime() wheher he MCFIFO wa placed cloe o he ource or o he ink. For example, when T T 300, he algorihm place he MCFIFO cloe o he ource, bu when T 200 i place i cloer o he ink. Thu, we canno generalize he behavior on he opimal locaion of he MCFIFO, i depend on he blockage map, clock period T and T and he echnology parameer. For all cae, we oberve ha he oal laency i no ignificanly higher han he minimum ourceink delay of 2739 p (from Table I). TABLE III GALS STATISTICS AS A FUNCTION OF T AND T WITH A GRID SEPARATION OF MM. T T Buffer Reg Reg laency VI. CONCLUDING REMARKS Auomaed buffered rouing i a neceiy in modern VLSI deign. The conribuion of hi paper are wo new problem formulaion for buffered rouing for ingle and muliple clock domain. Boh of hee formulaion addre problem ha will become more prominen in fuure deign. Any CAD ool currenly performing buffer inerion will evenually have o deal wih ynchronizer inerion. Furhermore, any SoC rouing CAD ool will have o handle rouing acro muliple clock domain due o he increaing ue of IP. We olve boh problem opimally in polynomial ime via he RBP and GALS algorihm ha build upon he Fa Pah algorihm of [17]. Experimenal reul validae he correcne and pracicaliy of he wo algorihm for an aggreive echnology. ACKNOWLEDGMENTS The auhor are graeful o Hai Zhou for upplying Fa Pah code, and alo o Meera Thiagarajan for help wih he figure and reearching he background maerial on he MCFIFO. REFERENCES [1] C. J. Alper, G. Gandham, J. Hu, S. T. Quay J. L. Neve, and S. S. Sapnekar. STeiner Tree Opimizaion for Buffer and Blockage and Bay. IEEE Tranacion on CompuerAided Deign, 20(4): , April [2] L Carloni, K. McMillan, A. Saldanha, and A. SangiovanniVincenelli. A Mehodology for CorrecbyConrucion Laency Ineniive Deign. In Proc. of he IEEE Inernaional Conference on CompuerAided Deign (ICCAD), [3] D. Chapiro. Globally Aynchronou Locally Aynchronou Syem. PhD hei, Sanford Univeriy, [4] T. Chelcea and S. Nowick. Robu Inerface for MixedTiming Syem wih Applicaion o LaencyIneniive Proocol. In Proc. of he ACM/IEEE Deign Auomaion Conference (DAC), page 21 6, [5] P. Cocchini. Concurren FlipFlop and Repeaer Inerion for High Performance Inegraed Circui. In Proc. of he IEEE Inernaional Conference on CompuerAided Deign (ICCAD), page , [6] J. Cong. Timing Cloure Baed on Phyical Hierarchy. In Proceeding of he Inernaional Sympoium on Phyical Deign, page , [7] J. Cong, J. Fang, and K.Y. Khoo. An Implici Connecion Graph Maze Rouing Algorihm for ECO Rouing. In Proc. of he IEEE Inernaional Conference on CompuerAided Deign (ICCAD), page , [8] J. Cong and Z. Pan. Inerconnec Performance Eimaion Model for Deign Planning. IEEE Tranacion on CompuerAided Deign, 20(6): , June [9] S. Haoun. Opimal Ue of 2Phae Tranparen Lache in Buffered Maze Rouing. In Proc. of he IEEE Inernaional Sympoium on Circui and Syem, [10] A. Hemani, T. Meincke, S. Kumar, A. Poula, T. Olon, P. Nilon, J. Ober, P. Ellervee, and D. Lundqvi. Lowering Power Conumpion in Clock by Uing Globally Aynchronou Locally Synchronou Deign Syle. In Proc. of he ACM/IEEE Deign Auomaion Conference (DAC), 1999.
10 IEEE TRANSACTIONS ON COMPUTERAIDED DESIGN, TO APPEAR. 10 [11] J. Muerbach, T. Villiger, H. Kaelin, N. Felber, and W. Fichner. Globallyaynchronou locallyynchronou archiecure o implify he deign of onchip yem. In Twelfh Annual IEEE Inernaional ASIC/SOC Conference, [12] M. Lai and D. F. Wong. Maze Rouing wih Buffer Inerion and Wireizing. In Proc. of he ACM/IEEE Deign Auomaion Conference (DAC), page , [13] R. Lu, G. Zhong, C. Koh,, and K. Chao. FlipFlop and Repeaer Inerion for Early Inerconnec Planning. In Proc. of Deign Auomaion and Te in Europe Conference (DATE), page 690 5, [14] J. Rabaey. Digial Inegraed Circui. Prenice Hall, [15] JN. Seizovic. Pipeline Synchronizaion. In IEEE ASYNC, [16] L.P.P.P. van Ginneken. Nework for Minimal Elmore Delay. In Proc. of he IEEE Inernaional Sympoium on Circui and Syem, page 865 8, [17] H. Zhou, D. F. Wong, I.M. Liu, and A. Aziz. Simulaneou Rouing and Buffer Inerion wih Rericion on Buffer Locaion. IEEE Tranacion on CompuerAided Deign, 19(7): , July 2000.
2.4 Network flows. Many direct and indirect applications telecommunication transportation (public, freight, railway, air, ) logistics
.4 Nework flow Problem involving he diribuion of a given produc (e.g., waer, ga, daa, ) from a e of producion locaion o a e of uer o a o opimize a given objecive funcion (e.g., amoun of produc, co,...).
More informationPhysical Topology Discovery for Large MultiSubnet Networks
Phyical Topology Dicovery for Large MuliSubne Nework Yigal Bejerano, Yuri Breibar, Mino Garofalaki, Rajeev Raogi Bell Lab, Lucen Technologie 600 Mounain Ave., Murray Hill, NJ 07974. {bej,mino,raogi}@reearch.belllab.com
More informationChapter 13. Network Flow III Applications. 13.1 Edge disjoint paths. 13.1.1 Edgedisjoint paths in a directed graphs
Chaper 13 Nework Flow III Applicaion CS 573: Algorihm, Fall 014 Ocober 9, 014 13.1 Edge dijoin pah 13.1.1 Edgedijoin pah in a direced graph 13.1.1.1 Edge dijoin pah queiong: graph (dir/undir)., : verice.
More informationOn the Connection Between MultipleUnicast Network Coding and SingleSource SingleSink Network Error Correction
On he Connecion Beween MulipleUnica ework Coding and SingleSource SingleSink ework Error Correcion Jörg Kliewer JIT Join work wih Wenao Huang and Michael Langberg ework Error Correcion Problem: Adverary
More informationFortified financial forecasting models: nonlinear searching approaches
0 Inernaional Conference on Economic and inance Reearch IPEDR vol.4 (0 (0 IACSIT Pre, Singapore orified financial forecaing model: nonlinear earching approache Mohammad R. Hamidizadeh, Ph.D. Profeor,
More informationHow Much Can Taxes Help Selfish Routing?
How Much Can Taxe Help Selfih Rouing? Tim Roughgarden (Cornell) Join wih Richard Cole (NYU) and Yevgeniy Dodi (NYU) Selfih Rouing a direced graph G = (V,E) a ource and a deinaion one uni of raffic from
More informationTSGRAN Working Group 1 (Radio Layer 1) meeting #3 Nynashamn, Sweden 22 nd 26 th March 1999
TSGRAN Working Group 1 (Radio Layer 1) meeing #3 Nynashamn, Sweden 22 nd 26 h March 1999 RAN TSGW1#3(99)196 Agenda Iem: 9.1 Source: Tile: Documen for: Moorola Macrodiversiy for he PRACH Discussion/Decision
More informationChapter 7. Response of FirstOrder RL and RC Circuits
Chaper 7. esponse of FirsOrder L and C Circuis 7.1. The Naural esponse of an L Circui 7.2. The Naural esponse of an C Circui 7.3. The ep esponse of L and C Circuis 7.4. A General oluion for ep and Naural
More informationMultiprocessor SystemsonChips
Par of: Muliprocessor SysemsonChips Edied by: Ahmed Amine Jerraya and Wayne Wolf Morgan Kaufmann Publishers, 2005 2 Modeling Shared Resources Conex swiching implies overhead. On a processing elemen,
More informationHeat demand forecasting for concrete district heating system
Hea demand forecaing for concree diric heaing yem Bronilav Chramcov Abrac Thi paper preen he reul of an inveigaion of a model for horerm hea demand forecaing. Foreca of hi hea demand coure i ignifican
More informationA Comparative Study of Linear and Nonlinear Models for Aggregate Retail Sales Forecasting
A Comparaive Sudy of Linear and Nonlinear Model for Aggregae Reail Sale Forecaing G. Peer Zhang Deparmen of Managemen Georgia Sae Univeriy Alana GA 30066 (404) 6514065 Abrac: The purpoe of hi paper i
More informationEmpirical heuristics for improving Intermittent Demand Forecasting
Empirical heuriic for improving Inermien Demand Forecaing Foio Peropoulo 1,*, Konanino Nikolopoulo 2, Georgio P. Spihouraki 1, Vailio Aimakopoulo 1 1 Forecaing & Sraegy Uni, School of Elecrical and Compuer
More informationAutomatic measurement and detection of GSM interferences
Auomaic measuremen and deecion of GSM inerferences Poor speech qualiy and dropped calls in GSM neworks may be caused by inerferences as a resul of high raffic load. The radio nework analyzers from Rohde
More informationNanocubes for RealTime Exploration of Spatiotemporal Datasets
Nanocube for RealTime Exploraion of Spaioemporal Daae Lauro Lin, Jame T Kloowki, and arlo Scheidegger Fig 1 Example viualizaion of 210 million public geolocaed Twier po over he coure of a year The daa
More informationFormulating CyberSecurity as Convex Optimization Problems
Formulaing CyberSecuriy a Convex Opimizaion Problem Kyriako G. Vamvoudaki, João P. Hepanha, Richard A. Kemmerer, and Giovanni Vigna Univeriy of California, Sana Barbara Abrac. Miioncenric cyberecuriy
More informationOverview. AdHoc DFT Methods. Definition. AdHoc DFT Methods. AdHoc DFT Methods. VLSI Design Verification and Testing
VLSI esign Verificaion and Tesing esign for Tesabiliy (FT)  1 Mohammad Tehranipoor Elecrical and Compuer Engineering Universiy of Connecicu Overview efiniion Adhoc mehods Scan design esign rules Scan
More informationCHARGE AND DISCHARGE OF A CAPACITOR
REFERENCES RC Circuis: Elecrical Insrumens: Mos Inroducory Physics exs (e.g. A. Halliday and Resnick, Physics ; M. Sernheim and J. Kane, General Physics.) This Laboraory Manual: Commonly Used Insrumens:
More informationThree Dimensional Grounding Grid Design
Three Dimenional Grounding Grid Deign Fikri Bari Uzunlar 1, Özcan Kalenderli 2 1 Schneider Elecric Turkey, Ianbul, Turkey bari.uzunlar@r.chneiderelecric.com 2 Ianbul Technical Univeriy, ElecricalElecronic
More informationThe Application of Multi Shifts and Break Windows in Employees Scheduling
The Applicaion of Muli Shifs and Brea Windows in Employees Scheduling Evy Herowai Indusrial Engineering Deparmen, Universiy of Surabaya, Indonesia Absrac. One mehod for increasing company s performance
More informationProcess Modeling for Object Oriented Analysis using BORM Object Behavioral Analysis.
Proce Modeling for Objec Oriened Analyi uing BORM Objec Behavioral Analyi. Roger P. Kno Ph.D., Compuer Science Dep, Loughborough Univeriy, U.K. r.p.kno@lboro.ac.uk 9RMW FKMerunka Ph.D., Dep. of Informaion
More informationHow has globalisation affected inflation dynamics in the United Kingdom?
292 Quarerly Bullein 2008 Q3 How ha globaliaion affeced inflaion dynamic in he Unied Kingdom? By Jennifer Greenlade and Sephen Millard of he Bank Srucural Economic Analyi Diviion and Chri Peacock of he
More informationRobust Bandwidth Allocation Strategies
Robu Bandwidh Allocaion Sraegie Oliver Heckmann, Jen Schmi, Ralf Seinmez Mulimedia Communicaion Lab (KOM), Darmad Univeriy of Technology Merckr. 25 D64283 Darmad Germany {Heckmann, Schmi, Seinmez}@kom.udarmad.de
More informationFormulating CyberSecurity as Convex Optimization Problems Æ
Formulaing CyberSecuriy a Convex Opimizaion Problem Æ Kyriako G. Vamvoudaki,João P. Hepanha, Richard A. Kemmerer 2, and Giovanni Vigna 2 Cener for Conrol, Dynamicalyem and Compuaion (CCDC), Univeriy
More informationChabot College Physics Lab RC Circuits Scott Hildreth
Chabo College Physics Lab Circuis Sco Hildreh Goals: Coninue o advance your undersanding of circuis, measuring resisances, currens, and volages across muliple componens. Exend your skills in making breadboard
More informationCHAPTER 11 NONPARAMETRIC REGRESSION WITH COMPLEX SURVEY DATA. R. L. Chambers Department of Social Statistics University of Southampton
CHAPTER 11 NONPARAMETRIC REGRESSION WITH COMPLEX SURVEY DATA R. L. Chamber Deparmen of Social Saiic Univeriy of Souhampon A.H. Dorfman Office of Survey Mehod Reearch Bureau of Labor Saiic M.Yu. Sverchkov
More informationCrosssectional and longitudinal weighting in a rotational household panel: applications to EUSILC. Vijay Verma, Gianni Betti, Giulio Ghellini
Croecional and longiudinal eighing in a roaional houehold panel: applicaion o EUSILC Viay Verma, Gianni Bei, Giulio Ghellini Working Paper n. 67, December 006 CROSSSECTIONAL AND LONGITUDINAL WEIGHTING
More informationTask is a schedulable entity, i.e., a thread
RealTime Scheduling Sysem Model Task is a schedulable eniy, i.e., a hread Time consrains of periodic ask T:  s: saring poin  e: processing ime of T  d: deadline of T  p: period of T Periodic ask T
More informationPart 1: White Noise and Moving Average Models
Chaper 3: Forecasing From Time Series Models Par 1: Whie Noise and Moving Average Models Saionariy In his chaper, we sudy models for saionary ime series. A ime series is saionary if is underlying saisical
More informationRC, RL and RLC circuits
Name Dae Time o Complee h m Parner Course/ Secion / Grade RC, RL and RLC circuis Inroducion In his experimen we will invesigae he behavior of circuis conaining combinaions of resisors, capaciors, and inducors.
More informationModule 4. Singlephase AC circuits. Version 2 EE IIT, Kharagpur
Module 4 Singlephase A circuis ersion EE T, Kharagpur esson 5 Soluion of urren in A Series and Parallel ircuis ersion EE T, Kharagpur n he las lesson, wo poins were described:. How o solve for he impedance,
More informationAppendix A: Area. 1 Find the radius of a circle that has circumference 12 inches.
Appendi A: Area workedou s o OddNumbered Eercises Do no read hese workedou s before aemping o do he eercises ourself. Oherwise ou ma mimic he echniques shown here wihou undersanding he ideas. Bes wa
More information9. Capacitor and Resistor Circuits
ElecronicsLab9.nb 1 9. Capacior and Resisor Circuis Inroducion hus far we have consider resisors in various combinaions wih a power supply or baery which provide a consan volage source or direc curren
More informationCapacitors and inductors
Capaciors and inducors We coninue wih our analysis of linear circuis by inroducing wo new passive and linear elemens: he capacior and he inducor. All he mehods developed so far for he analysis of linear
More informationIssues Using OLS with Time Series Data. Time series data NOT randomly sampled in same way as cross sectional each obs not i.i.d
These noes largely concern auocorrelaion Issues Using OLS wih Time Series Daa Recall main poins from Chaper 10: Time series daa NOT randomly sampled in same way as cross secional each obs no i.i.d Why?
More informationInfrastructure and Evolution in Division of Labour
Infrarucure and Evoluion in Diviion of Labour Mei Wen Monah Univery (Thi paper ha been publihed in RDE. (), 906) April 997 Abrac Thi paper udie he relaionhip beween infrarucure ependure and endogenou
More informationThe Transport Equation
The Transpor Equaion Consider a fluid, flowing wih velociy, V, in a hin sraigh ube whose cross secion will be denoed by A. Suppose he fluid conains a conaminan whose concenraion a posiion a ime will be
More informationChapter 2 Problems. s = d t up. = 40km / hr d t down. 60km / hr. d t total. + t down. = t up. = 40km / hr + d. 60km / hr + 40km / hr
Chaper 2 Problems 2.2 A car ravels up a hill a a consan speed of 40km/h and reurns down he hill a a consan speed of 60 km/h. Calculae he average speed for he rip. This problem is a bi more suble han i
More informationUsing RCtime to Measure Resistance
Basic Express Applicaion Noe Using RCime o Measure Resisance Inroducion One common use for I/O pins is o measure he analog value of a variable resisance. Alhough a builin ADC (Analog o Digial Converer)
More informationDifferential Equations and Linear Superposition
Differenial Equaions and Linear Superposiion Basic Idea: Provide soluion in closed form Like Inegraion, no general soluions in closed form Order of equaion: highes derivaive in equaion e.g. dy d dy 2 y
More informationPolicies & Procedures. I.D. Number: 1071
Policie & Procedure Tile: Licened Pracical Nure (LPN ) ADDED SKILLS (Aigned Funcion) Auhorizaion: [x] SHR Nuring Pracice Commiee I.D. Number: 1071 Source: Nuring Dae Revied: Sepember 2004 Dae Effecive:
More informationAP Calculus BC 2010 Scoring Guidelines
AP Calculus BC Scoring Guidelines The College Board The College Board is a noforprofi membership associaion whose mission is o connec sudens o college success and opporuniy. Founded in, he College Board
More informationANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS
ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS R. Caballero, E. Cerdá, M. M. Muñoz and L. Rey () Deparmen of Applied Economics (Mahemaics), Universiy of Málaga,
More informationSELFEVALUATION FOR VIDEO TRACKING SYSTEMS
SELFEVALUATION FOR VIDEO TRACKING SYSTEMS Hao Wu and Qinfen Zheng Cenre for Auomaion Research Dep. of Elecrical and Compuer Engineering Universiy of Maryland, College Park, MD20742 {wh2003, qinfen}@cfar.umd.edu
More informationCalculation of variable annuity market sensitivities using a pathwise methodology
cuing edge Variable annuiie Calculaion of variable annuiy marke eniiviie uing a pahwie mehodology Under radiional finie difference mehod, he calculaion of variable annuiy eniiviie can involve muliple Mone
More informationIndividual Health Insurance April 30, 2008 Pages 167170
Individual Healh Insurance April 30, 2008 Pages 167170 We have received feedback ha his secion of he e is confusing because some of he defined noaion is inconsisen wih comparable life insurance reserve
More informationLecture 15 Isolated DCDC converters
ELEC440/940 Lecure 15 olae CC converer Ofen, he oupu C volage fro a CC converer u be iolae fro he inpu AC upply. C power upplie for appliance an equipen are goo exaple. i avanageou o have he iolaion
More informationAn approach for designing a surface pencil through a given geodesic curve
An approach for deigning a urface pencil hrough a given geodeic curve Gülnur SAFFAK ATALAY, Fama GÜLER, Ergin BAYRAM *, Emin KASAP Ondokuz Mayı Univeriy, Faculy of Ar and Science, Mahemaic Deparmen gulnur.affak@omu.edu.r,
More informationRC Circuit and Time Constant
ircui and Time onsan 8M Objec: Apparaus: To invesigae he volages across he resisor and capacior in a resisorcapacior circui ( circui) as he capacior charges and discharges. We also wish o deermine he
More informationOPTIMAL BATCH QUANTITY MODELS FOR A LEAN PRODUCTION SYSTEM WITH REWORK AND SCRAP. A Thesis
OTIMAL BATH UANTITY MOELS FOR A LEAN ROUTION SYSTEM WITH REWORK AN SRA A Thei Submied o he Graduae Faculy of he Louiiana Sae Univeriy and Agriculural and Mechanical ollege in parial fulfillmen of he requiremen
More informationStrategic Optimization of a Transportation Distribution Network
Sraegic Opimizaion of a Transporaion Disribuion Nework K. John Sophabmixay, Sco J. Mason, Manuel D. Rossei Deparmen of Indusrial Engineering Universiy of Arkansas 4207 Bell Engineering Cener Fayeeville,
More informationMorningstar Investor Return
Morningsar Invesor Reurn Morningsar Mehodology Paper Augus 31, 2010 2010 Morningsar, Inc. All righs reserved. The informaion in his documen is he propery of Morningsar, Inc. Reproducion or ranscripion
More informationEquity Valuation Using Multiples. Jing Liu. Anderson Graduate School of Management. University of California at Los Angeles (310) 2065861
Equiy Valuaion Uing Muliple Jing Liu Anderon Graduae School of Managemen Univeriy of California a Lo Angele (310) 2065861 jing.liu@anderon.ucla.edu Doron Niim Columbia Univeriy Graduae School of Buine
More informationDistributed Echo Cancellation in Multimedia Conferencing System
Disribued Echo Cancellaion in Mulimedia Conferencing Sysem Balan Sinniah 1, Sureswaran Ramadass 2 1 KDU College Sdn.Bhd, A Paramoun Corporaion Company, 32, Jalan Anson, 10400 Penang, Malaysia. sbalan@kdupg.edu.my
More informationChapter 8: Regression with Lagged Explanatory Variables
Chaper 8: Regression wih Lagged Explanaory Variables Time series daa: Y for =1,..,T End goal: Regression model relaing a dependen variable o explanaory variables. Wih ime series new issues arise: 1. One
More informationConstant Data Length Retrieval for Video Servers with Variable Bit Rate Streams
IEEE Inernaional Conference on Mulimedia Compuing & Sysems, June 173, 1996, in Hiroshima, Japan, p. 151155 Consan Lengh Rerieval for Video Servers wih Variable Bi Rae Sreams Erns Biersack, Frédéric Thiesse,
More informationCircuit Types. () i( t) ( )
Circui Types DC Circuis Idenifying feaures: o Consan inpus: he volages of independen volage sources and currens of independen curren sources are all consan. o The circui does no conain any swiches. All
More informationsdomain Circuit Analysis
Domain ircui Analyi Operae direcly in he domain wih capacior, inducor and reior Key feaure lineariy i preered c decribed by ODE and heir Order equal number of plu number of Elemenbyelemen and ource ranformaion
More informationChapter 2: Principles of steadystate converter analysis
Chaper 2 Principles of SeadySae Converer Analysis 2.1. Inroducion 2.2. Inducor volsecond balance, capacior charge balance, and he small ripple approximaion 2.3. Boos converer example 2.4. Cuk converer
More informationExplore the Application of Financial Engineering in the Management of Exchange Rate Risk
SHS Web o Conerence 17, 01006 (015) DOI: 10.1051/ hcon/01517 01006 C Owned by he auhor, publihed by EDP Science, 015 Explore he Applicaion o Financial Engineering in he Managemen o Exchange Rae Rik Liu
More informationThe naive method discussed in Lecture 1 uses the most recent observations to forecast future values. That is, Y ˆ t + 1
Business Condiions & Forecasing Exponenial Smoohing LECTURE 2 MOVING AVERAGES AND EXPONENTIAL SMOOTHING OVERVIEW This lecure inroduces imeseries smoohing forecasing mehods. Various models are discussed,
More informationRC (ResistorCapacitor) Circuits. AP Physics C
(ResisorCapacior Circuis AP Physics C Circui Iniial Condiions An circui is one where you have a capacior and resisor in he same circui. Suppose we have he following circui: Iniially, he capacior is UNCHARGED
More information4. International Parity Conditions
4. Inernaional ariy ondiions 4.1 urchasing ower ariy he urchasing ower ariy ( heory is one of he early heories of exchange rae deerminaion. his heory is based on he concep ha he demand for a counry's currency
More informationSinglemachine Scheduling with Periodic Maintenance and both Preemptive and. Nonpreemptive jobs in Remanufacturing System 1
Absrac number: 050407 Singlemachine Scheduling wih Periodic Mainenance and boh Preempive and Nonpreempive jobs in Remanufacuring Sysem Liu Biyu hen Weida (School of Economics and Managemen Souheas Universiy
More informationMonotonic, Inrush Current Limited StartUp for Linear Regulators
Applicaion epor SLA156 March 2004 Monoonic, Inrush urren Limied SarUp for Linear egulaors Jeff Falin PMP Porable Producs ABSA he oupu volage of a linear regulaor ends o rise quickly afer i is enabled.
More informationOPERATION MANUAL. Indoor unit for air to water heat pump system and options EKHBRD011ABV1 EKHBRD014ABV1 EKHBRD016ABV1
OPERAION MANUAL Indoor uni for air o waer hea pump sysem and opions EKHBRD011ABV1 EKHBRD014ABV1 EKHBRD016ABV1 EKHBRD011ABY1 EKHBRD014ABY1 EKHBRD016ABY1 EKHBRD011ACV1 EKHBRD014ACV1 EKHBRD016ACV1 EKHBRD011ACY1
More informationGraduate Macro Theory II: Notes on Neoclassical Growth Model
Graduae Macro Theory II: Noes on Neoclassical Growh Model Eric Sims Universiy of Nore Dame Spring 2011 1 Basic Neoclassical Growh Model The economy is populaed by a large number of infiniely lived agens.
More informationInductance and Transient Circuits
Chaper H Inducance and Transien Circuis Blinn College  Physics 2426  Terry Honan As a consequence of Faraday's law a changing curren hrough one coil induces an EMF in anoher coil; his is known as muual
More informationAP Calculus AB 2010 Scoring Guidelines
AP Calculus AB 1 Scoring Guidelines The College Board The College Board is a noforprofi membership associaion whose mission is o connec sudens o college success and opporuniy. Founded in 1, he College
More informationSc i e n c e a n d t e a c h i n g:
Dikuionpapierreihe Working Paper Serie Sc i e n c e a n d e a c h i n g: Tw o d i m e n i o n a l i g n a l l i n g in he academic job marke Andrea Schneider Nr./ No. 95 Augu 2009 Deparmen of Economic
More informationRealtime Particle Filters
Realime Paricle Filers Cody Kwok Dieer Fox Marina Meilă Dep. of Compuer Science & Engineering, Dep. of Saisics Universiy of Washingon Seale, WA 9895 ckwok,fox @cs.washingon.edu, mmp@sa.washingon.edu Absrac
More informationSubsistence Consumption and Rising Saving Rate
Subience Conumpion and Riing Saving Rae Kenneh S. Lin a, HiuYun Lee b * a Deparmen of Economic, Naional Taiwan Univeriy, Taipei, 00, Taiwan. b Deparmen of Economic, Naional Chung Cheng Univeriy, ChiaYi,
More informationDifferential Equations. Solving for Impulse Response. Linear systems are often described using differential equations.
Differenial Equaions Linear sysems are ofen described using differenial equaions. For example: d 2 y d 2 + 5dy + 6y f() d where f() is he inpu o he sysem and y() is he oupu. We know how o solve for y given
More informationTwo Compartment Body Model and V d Terms by Jeff Stark
Two Comparmen Body Model and V d Terms by Jeff Sark In a onecomparmen model, we make wo imporan assumpions: (1) Linear pharmacokineics  By his, we mean ha eliminaion is firs order and ha pharmacokineic
More informationMaintenance scheduling and process optimization under uncertainty
Compuers and Chemical Engineering 25 (2001) 217 236 www.elsevier.com/locae/compchemeng ainenance scheduling and process opimizaion under uncerainy C.G. Vassiliadis, E.N. Piikopoulos * Deparmen of Chemical
More informationDISTRIBUTED DATA PARALLEL TECHNIQUES FOR CONTENTMATCHING INTRUSION DETECTION SYSTEMS. G. Chapman J. Cleese E. Idle
DISTRIBUTED DATA PARALLEL TECHNIQUES FOR CONTENTMATCHING INTRUSION DETECTION SYSTEMS G. Chapman J. Cleee E. Idle ABSTRACT Content matching i a neceary component of any ignaturebaed network Intruion Detection
More informationµ r of the ferrite amounts to 1000...4000. It should be noted that the magnetic length of the + δ
Page 9 Design of Inducors and High Frequency Transformers Inducors sore energy, ransformers ransfer energy. This is he prime difference. The magneic cores are significanly differen for inducors and high
More informationMolding. Injection. Design. GE Plastics. GE Engineering Thermoplastics DESIGN GUIDE
apple GE Plasics GE Engineering Thermoplasics DESIGN GUIDE Wall Thickness Paring Lines Ejecion Appearance Pars Ribs/Gusses Bosses Holes Depressions Radii, Filles and Corners Molding Design Injecion s for
More informationChapter 2 Kinematics in One Dimension
Chaper Kinemaics in One Dimension Chaper DESCRIBING MOTION:KINEMATICS IN ONE DIMENSION PREVIEW Kinemaics is he sudy of how hings moe how far (disance and displacemen), how fas (speed and elociy), and how
More informationSKF Documented Solutions
SKF Documened Soluions Real world savings and we can prove i! How much can SKF save you? Le s do he numbers. The SKF Documened Soluions Program SKF is probably no he firs of your supplier parners o alk
More informationSignal Processing and Linear Systems I
Sanford Universiy Summer 214215 Signal Processing and Linear Sysems I Lecure 5: Time Domain Analysis of Coninuous Time Sysems June 3, 215 EE12A:Signal Processing and Linear Sysems I; Summer 1415, Gibbons
More informationPROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE
Profi Tes Modelling in Life Assurance Using Spreadshees PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Erik Alm Peer Millingon 2004 Profi Tes Modelling in Life Assurance Using Spreadshees
More informationCaring for trees and your service
Caring for rees and your service Line clearing helps preven ouages FPL is commied o delivering safe, reliable elecric service o our cusomers. Trees, especially palm rees, can inerfere wih power lines and
More informationAdvanced Traffic Routing as Part of the USA Intelligent Telecommunications Network
Advanced Traffic Rouing as Par of he USA Inelligen Telecommunicaions Nework Edmund P. GOULD Bell Communicaions Research 435 Souh Sree Morrisown, New Jersey 07960 The USA Telecommunicaions Nework is evolving
More informationThe Chase Problem (Part 2) David C. Arney
The Chae Problem Par David C. Arne Inroducion In he previou ecion, eniled The Chae Problem Par, we dicued a dicree model for a chaing cenario where one hing chae anoher. Some of he applicaion of hi kind
More informationPerformance Center Overview. Performance Center Overview 1
Performance Cener Overview Performance Cener Overview 1 ODJFS Performance Cener ce Cener New Performance Cener Model Performance Cener Projec Meeings Performance Cener Execuive Meeings Performance Cener
More informationCLOCK SKEW CAUSES CLOCK SKEW DUE TO THE DRIVER EROSION OF THE CLOCK PERIOD
DESIGNING WITH HIGH SPEED CLOCK DRIERS CONFERENCE PAPER CP19 Inegraed Device Technology, Inc. By Sanley Hronik ABSTRACT Today s high speed sysems are encounering problems wih clocking ha were no consideraions
More informationcooking trajectory boiling water B (t) microwave 0 2 4 6 8 101214161820 time t (mins)
Alligaor egg wih calculus We have a large alligaor egg jus ou of he fridge (1 ) which we need o hea o 9. Now here are wo accepable mehods for heaing alligaor eggs, one is o immerse hem in boiling waer
More informationGloballyOptimal Greedy Algorithms for Tracking a Variable Number of Objects
GloballyOpimal Greedy Algorihm for Tracking a Variable Number of Objec Hamed Piriavah Deva Ramanan Charle C. Fowlke Deparmen of Compuer Science, Univeriy of California, Irvine {hpiriav,dramanan,fowlke}@ic.uci.edu
More informationA Mathematical Description of MOSFET Behavior
10/19/004 A Mahemaical Descripion of MOSFET Behavior.doc 1/8 A Mahemaical Descripion of MOSFET Behavior Q: We ve learned an awful lo abou enhancemen MOSFETs, bu we sill have ye o esablished a mahemaical
More informationMaking Use of Gate Charge Information in MOSFET and IGBT Data Sheets
Making Use of ae Charge Informaion in MOSFET and IBT Daa Shees Ralph McArhur Senior Applicaions Engineer Advanced Power Technology 405 S.W. Columbia Sree Bend, Oregon 97702 Power MOSFETs and IBTs have
More informationMaking a Faster Cryptanalytic TimeMemory TradeOff
Making a Faser Crypanalyic TimeMemory TradeOff Philippe Oechslin Laboraoire de Securié e de Crypographie (LASEC) Ecole Polyechnique Fédérale de Lausanne Faculé I&C, 1015 Lausanne, Swizerland philippe.oechslin@epfl.ch
More informationProduct Operation and Setup Instructions
A9 Please read and save hese insrucions. Read carefully before aemping o assemble, insall, operae, or mainain he produc described. Proec yourself and ohers by observing all safey informaion. Failure o
More informationESIGN Rendering Service
Markeing maerials on demand wihou phoo shoos or seup Wih he ESIGN Rendering Service, we produce new, prinready markeing maerials for you in a cosefficien and imely manner for he design of brochures,
More informationRotational Inertia of a Point Mass
Roaional Ineria of a Poin Mass Saddleback College Physics Deparmen, adaped from PASCO Scienific PURPOSE The purpose of his experimen is o find he roaional ineria of a poin experimenally and o verify ha
More informationAcceleration Lab Teacher s Guide
Acceleraion Lab Teacher s Guide Objecives:. Use graphs of disance vs. ime and velociy vs. ime o find acceleraion of a oy car.. Observe he relaionship beween he angle of an inclined plane and he acceleraion
More informationDividend taxation, share repurchases and the equity trap
Working Paper 2009:7 Deparmen of Economic Dividend axaion, hare repurchae and he equiy rap Tobia Lindhe and Jan Söderen Deparmen of Economic Working paper 2009:7 Uppala Univeriy May 2009 P.O. Box 53 ISSN
More informationEfficient Onetime Signature Schemes for Stream Authentication *
JOURNAL OF INFORMATION SCIENCE AND ENGINEERING, 61164 (006) Efficien Oneime Signaure Schemes for Sream Auhenicaion * YONGSU PARK AND YOOKUN CHO + College of Informaion and Communicaions Hanyang Universiy
More informationPI4ULS5V202 2Bit Bidirectional Level Shifter with Automatic Sensing & Ultra Tiny Package
Feaures can be Less han, Greaer han or Equal o V CCB 1.2V o 5.5V on A Por and 1.2V o 5.5V on B Por High Speed wih 20 Mb/s Daa Rae for pushpull applicaion High Speed wih 2 Mb/s Daa Rae for opendrain applicaion
More informationTEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS
TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS RICHARD J. POVINELLI AND XIN FENG Deparmen of Elecrical and Compuer Engineering Marquee Universiy, P.O.
More informationChapter 2 Problems. 3600s = 25m / s d = s t = 25m / s 0.5s = 12.5m. Δx = x(4) x(0) =12m 0m =12m
Chaper 2 Problems 2.1 During a hard sneeze, your eyes migh shu for 0.5s. If you are driving a car a 90km/h during such a sneeze, how far does he car move during ha ime s = 90km 1000m h 1km 1h 3600s = 25m
More information