Power Loss and Thermal Characterization of IGBT Modules in the Alternate Arm Converter

Transcription

1 Power oss and Thermal haracerizaion of Modules in he Alernae Arm onverer P. D. Judge, M. M.. Merlin, P. D. Micheson, T.. Green onrol and Power Group, Deparmen of Elecrical and Elecronic Engineering, Imperial ollege ondon Absrac Power losses in high power HVD converers are dominaed by hose ha occur wihin he power elecronic devices. This power loss is dissipaed as hea a he juncion of semiconducor devices. The cooling sysem ensures ha he generaed hea is evacuaed ouside he converer saion bu emperaure managemen remains criical for he lifeime of he semiconducor devices. This paper presens he resuls of a sudy on he emperaure profile of he differen swiches inside a mulilevel converer. The seady sae juncion emperaures are observed hrough he simulaion of a 2 MW Alernae Arm onverer using 1.2kA 3.3 kv modules. A comparison of he Alernae Arm onverer is made agains he case of boh he halfbridge and fullbridge Modular Mulilevel onverer opologies. Furhermore, he concep of varying he duycycle of he wo alernaive zerovolage saes of he Hbridge modules is inroduced. Simulaion resuls demonsrae ha i can change he balance of elecrical and hermal sress beween he wo op swiches and he wo boom swiches of a fullbridge cell. I. THERMA DESIGN ONSIDERATIONS AND IMPORTANE Manufacurers ypically require ha he maximum juncion emperaure of modules is limied o below 1, or 125, and ha he operaing emperaure under normal condiions be limied o 8% ha of he maximum value [1], [2]. Devices wih higher curren raings are ypically limied o he lower value of 125. Operaing he device above his level can lead o a breakdown in reverse volage blocking capabiliy and hermal runaway, leading o device desrucion. The sysem mus be designed so ha he juncion emperaure ( ) never exceeds his value. The level of power and hermal cycling ha a module undergoes during is operaion can have a large impac on he expeced lifeime of he module. Power cycling, which occurs a ime frames of around 2 seconds causes sress in he soldering of he elecrical connecions wihin he module due o he differen coefficiens of linear expansion. If he power cycling causes he soldering o deeriorae sufficienly hen he device will experience elecrical failure. A a longer ime frame and wih a larger change in emperaure ( ), hermal cycling of modules causes unequal expansion of he insulaion subsrae and he copper baseplae and faigue o he solder layer beween hem [1]. This can lead o device failure as i causes he hermal resisance of he module o increase, The auhors graefully acknowledge he financial suppor of he Research ouncils UK hrough he HubNe Projec (gran number: EP/I13636/1) leading o hermal runaway. Manufacurers usually supply power cycle and hermal cycle capabiliy curves for modules which allow he esimaion of he lifeime of he modules. The degradaion of lifeime is dependen on absolue juncion emperaure as well as he magniude of he power or hermal cycle [3]. ooking beyond lifeime issues for high power applicaions, i is desirable o keep he juncion emperaure as low as reasonably possible. This is because he level of hermally generaed elecron hole pairs increases wih emperaure, hus decreasing he devices conduciviy and ulimaely leading o an increase in power loss. II. POWER OSS MODEING A Malab/Simulink block was creaed o esimae he power losses wihin each module wihin a converer following he echnique presened in [4]. This mehod uses he curren flowing hrough each module as well as he gae signal o deec swiching evens o deermine he expeced losses. The waveforms in he simulaed converer are assumed o no be impaired by he onsae resisances of swiching elemens wihin Malab/Simulink as he overall losses are small. This Simulink block oupus he power losses occurring a he juncion and he diode juncion wihin a module, which can hen be applied o a hermal model of he device o deermine he expeced juncion emperaures. This mehod is suiable for use wihin fixedsep discree realime simulaions and calculaes he power loss a each ime sep wihin he simulaion. This mehod has been verified agains he resuls of a posprocessing scrip mehod used in [5]. The loss curves needed for his mehod were exraced from he manufacurers daashee and where possible scaled for an operaing emperaure of 75. III. THERMA MODEING The hermal model of he used wihin his sudy was developed using he Finie Elemen Mehod (FEM) elecronics hermal simulaion package ANSYS IcePak. The Device used is a 1.2kA 3.3kV 5SNA 12E331 module from ABB Semiconducors [6]. The module modeled wihin ANSYS IcePak is shown in Figure 1. The hermal modeling of a similar hea sink mouned module and validaion agains experimenal resuls has been exensively covered in [7]. The hermal model wihin his sudy

2 1 1 Z h(j a) Z h(j ) Figure 1: Heasink mouned modeled wihin ANSYS IcePak Z h(j c) [K/W], has been compared agains he resuls presened in [7], and agrees wih close approximaion. The inernal consrucion and hermal design of he device is deailed in [2]. Inernally he device consiss of 24 dies and 12 diode dies conneced by inernal bus bars and bond wires. The diode dies are locaed closes o he cener of he device. In he model i is assumed ha all hea flow is from he acive juncions o he cooling liquid of he heaplae and ha any hea flow from he juncion o ambien air is negligible, for his reason he plasic case, bond wires and inernal bus bars of he module are no included in he model. The power losses are modeled as 2D power sources, dissipaing a fixed amoun of power, placed upon he op surface of each die. The dies wihin he module are assumed o equally share he overall power loss wihin he module due o he posiive emperaure coefficien of he onsae volages of each die. The juncion emperaures are measured a he geomeric cener of each silicon die. Due o he inernal consrucion of he device here is a varying amoun of cross coupling beween he dies which leads o an imbalance beween juncion emperaures of individual dies wihin he module. The juncion emperaure measuremens presened in his sudy are an average of he juncion emperaures measured a each or diode die wihin he module. The hermal model has also been verified agains he ransien hermal impedance curves provided by he manufacurer in he daashee by using a 2D emperaure source o hold he exernal case emperaure o a fixed value. In real applicaions, holding he case emperaure a a consan emperaure is no pracical so i is necessary o model he hea sink on which he module is mouned. The hea sink also acs as a hea spreader, increasing he level of cross coupling beween he dies wihin he module [7]. Z h () = γ i.(1 exp( τ i )) (1) The ransien hermal impedance response curves of he module o boh ses of dies being heaed are shown in Figure 2. To derive hermal models suiable for use wihin he Malab/Simulink model he ransien hermal impedance curves are fied o a finie series of exponenial erms in he form of (1). The aplace ransform is hen applied o his finie series o give a ransfer funcion suiable for use in a Malab/Simulink simulaion of he converer. Z h(j c) [K/W], [s] (a) dies heaed Z h(j a) Z h(j ) [s] (b) dies heaed Figure 2: Transien hermal impedance response IV. MODUAR MUTIEVE ONVERTERS AND THE ATERNATE ARM ONVERTER The primary focus of his paper is he hermal characerizaion of he modules wihin he Alernae Arm onverer (AA), which was inroduced in [5] and is illusraed in Figure 3. However for comparison purposes an analysis of halfbridge Modular Mulilevel onverer (MM), illusraed in Figure 4, and he fullbridge MM is also presened. Funcionally he fullbridge MM and sandard halfbridge MM have similar operaion hough he fullbridge MM adds he poenial o block D side fauls due o is abiliy o conrol he converers inducor currens during a faul. This abiliy comes a a cos of needing wice as many semiconducor swiches and a doubling of he conducion losses wihin he converer. The AA is a hybrid opology beween he 2level converer and he MM firs inroduced in [8]. As wih he MM, he AA uilizes sacks of Hbridge cells (in heir 4 swich version, shown in Figure 3) in is arms bu wih direcor swiches o alernae he working of he arms leading o several key feaures disinc from he MM. Firsly, his mode of operaion recifies he A curren direcly insead of requiring an addiional D curren running coninuously hrough he

3 I D D Terminal MM AA Uniy pf A Grid Transformer I D arm arm arm I Ā D D I arm arm arm I V D eading pf agging pf D D D Figure 5: AA and MM op arm curren and volage V Ā 1 3 I Sack I D D Terminal Figure 3: Alernae Arm onverer V 2 4 I Sack I D D Terminal (a) Fullbridge cell I Sack 1 A Grid Transformer I I arm arm arm arm arm arm I Ā I V D V 2 I Sack (b) Halfbridge cell Figure 6: Fullbridge and halfbridge cell configuraions V Ā a negaive D volage which allows an overmodulaed A volage, greaer han he D volage. In comparison each cell wihin an MM conducs curren for he enire elecrical cycle and is unable o generae a negaive volage. Figure 4: Halfbridge Modular Mulievel onverer I D D Terminal arms. Secondly, he number of cells in he sacks is much reduced because each arm is used o produce only half of he converer A volage waveform. Thirdly, he AA is able o block Dside faul currens passing hrough is arms because fullbridge cells can produce negaive volage o oppose he A grid volage, and hence keep conrol of he sack. Finally, he direcor swiches can be operaed in sofswiching mode, i.e. swiching off when heir curren reaches zero, hanks o he arm currens being kep under igh conrol by he sacks. A comparison of he arm currens and volages wihin he AA and he MM is shown in Figure 5. The primary difference beween he wo is ha each arm of he AA conducs curren for half of each elecrical cycle and can generae V. FUBRIDGE AND HAFBRIDGE ONFIGURATIONS Boh he AA and he fullbridge MM make use of a fullbridge cells raher han a halfbridge cells as used in he sandard MM opology. Boh cell arrangemens are shown in Figure 6, wih volage and curren definiions given using a source convenion. A fullbridge has effecively four saes wih hree differen volage oupus. The AA uilizes all hree possible volage oupus whils he fullbridge MM is limied o using he posiive volage oupu and he wo zero volage oupu saes during normal operaion. The pah ha he arm curren akes hrough he cell depends on curren direcion and cell volage oupu (or sae). For a fullbridge cell his pah is always hrough wo separae modules. The pah ha he curren akes is shown in Table I. A halfbridge has wo saes wih wo possible volage oupus. The curren pahs for hese saes are shown in Table II.

4 Table I: urren pah hrough a fullbridge cell depending on sack curren direcion and cell volage oupu (a) I sack Posiive Posiion V ou V cell V cell (b) I sack Negaive Posiion V ou V cell V cell Table II: urren pah hrough a halfbridge cell depending on sack curren direcion and cell volage oupu (a) I sack Posiive Posiion V ou 1 2 V cell (b) I sack Negaive Posiion V ou 1 2 V cell For converer wih fullbridge cells, he zerovolage sae duy cycle (δ zv ) of each cell has been defined as he raio of how ofen he combinaion of s 2 and 4 is seleced, as opposed o he combinaion of s 1 and 3, o how ofen eiher zerovolage sae is used. A δ zv of % corresponds o only s 1 and 3 being seleced for zero volage generaion and a δ zv of 1% corresponds o only s 2 and 4 being seleced for zero volage generaion. Unless specified oherwise, a δ zv of % has been used in he simulaions, meaning ha a random disribuion beween he wo possible saes is used. VI. POWER OSS AND THERMA HARATERIZATION The power losses wihin each module of a single cell were deermined by simulaion and are ploed for differen operaing poins of he hree converer opologies under examinaion. Each poin is an average power loss over 3 s of seadysae operaion so as o avoid any shorerm effecs arising from he conrol sysem. The power flow convenion used is ha negaive power is A power being recified o D and posiive power is D power being invered o A. The cooling waer emperaure for all resuls shown is assumed o be 4. A. The Alernae Arm onverer The power losses and juncion emperaures as a funcion of real power for each module of a single cell wihin an AA in addiion o one forming par of he direcor swich in he same arm are shown in Figure 7. As is shown in Figure 7, he power losses wihin each module occur almos exclusively wihin eiher he or he diode, depending on wheher he converer is in recifying or invering mode and varying wih real power flow. This is because, a uniy power facor, he curren flowing hrough each arm of he AA is eiher nominally coninually posiive or coninually negaive. The direcor swich exhibis higher power losses, leading o a higher juncion emperaure han any of he swiches in he cells in spie of having close o zero swiching losses. This can be explained by he fac ha he direcor swiches are in conducion for he whole working period of heir arm while he swiches in he cells are periodically swiched ou of he conducion pah, in addiion conducion losses dominae swiching losses in MM derivaive converers. The resuls indicae ha he losses wihin each cell should be approximaely equal regardless of he power flow direcion. However he power losses and juncion emperaures reached wihin he direcor swich modules will be higher during invering operaion due o he higher on sae volage drop of he in comparison o he diode. The resuls indicae ha he AA will operae wih a maximum juncion emperaure ha is significanly below he maximum permissible value of 125 (given he 4 ambien coolan emperaure). In he direcor swiches, here is 63 of headroom o 125 for he s and 59 for he diode. In he cells, here is 69 of headroom o 125 for he s and 68 for he diodes. Furhermore, he resuls show ha here is a good disribuion of elecrical and hermal sress beween each of he modules wihin he cell when δ zv is se o %. The B 1 lifeime curves, provided by he manufacurer in [3], indicae ha ha hermal and power cycling impacs on lifeime should no be significan, even when exreme cases such as regular power flow reversals are considered. B. The HalfBridge Modular Mulievel onverer The power losses and juncion emperaures wihin each module of a single cell wihin a halfbridge MM are shown in Figure 8. As shown in Figure 8, here is a large imbalance beween he power losses generaed wihin he wo modules, wih mos of he losses occurring in. Addiionally In he power losses are spli almos equally beween he and he diode whereas in he power losses occur almos exclusively in eiher he or diode, depending on power flow direcion. This can be explained by examining he volage waveforms shown in Figure 5. When he sacks are generaing a low volage, i.e. mos of he cells wihin an arm are bypassed using wihin each cell, he curren flowing hough he arm approaches is highes level resuling in high losses in eiher he diode or wihin his module. During he res of he cycle when he sack is generaing an inermediae or high volage, i.e. when he cells are more likely o be o be generaing an oupu of E, he curren is lower in magniude and for par of he cycle reversed in direcion, changing he nominal curren pah hrough he module which

5 (a) Power osses 1 4 Direcor Swich Direcor Swich ( o ) ( o ) ( o ) ( o ) ( o ) (b) Juncion Temperaures 4 4 Figure 7: Power oss and juncion emperaures of modules wihin he AA a uniy power facor ( o ) (a) Power osses ( o ) (b) Juncion Temperaures 4 Figure 8: Power oss and juncion emperaures of modules wihin he halfbridge MM a uniy power facor resuls in he spli of losses beween and diode observed in module 1. Similar o he AA direcor swich, he resuls indicae ha he losses wihin each cell will be higher during invering operaion. The resuls also indicae ha he maximum juncion emperaures reached wihin he halfbridge MM are similar o hose reached by he direcor swich module wihin he AA, bu lower han he emperaures reached wihin he cells of he AA.. The FullBridge Modular Mulievel onverer The power losses and juncion emperaures wihin each module of a single cell wihin an fullbridge MM are shown in Figure9. Similar o he halfbridge MM, he power losses are spli beween he and he diode wihin each module. s 1 and 4 experience larger losses han s 2 and 3. This can be explained by examining he curren pah hrough each cell shown in Table I. s 1 and 4 each form par of he curren carrying pah for wo possible saes, during he posiive volage oupu sae, and each during one of he wo possible zero volage oupu saes. As he negaive volage oupu sae of he cell is unused in he fullbridge MM opology, S 2 and 3 each only form par of he curren carrying pah during one of he wo possible zero volage oupu saes. As s 2 and 3 are uilized less han s 1 and 4 hey experience lower losses. In conras he halfbridge MM cell and he direcor swich of he AA he power losses wihin appear o be greaer during recifying operaion. Given he maximum juncion emperaure reached, here is 62 of headroom o 125. VII. IMPAT OF DUTYYE OF ZEROVOTAGE STATES As discussed in he inroducion, by varying he duycycle of he zerovolage saes (δ zv ), he elecrical and hermal sress wihin each cell can be redisribued. This can be used o more evenly spread he hermal and elecrical sress beween swiches. The exen o which his mehod is effecive for he AA operaing a raed power is illusraed in Figure 1. In boh recifying and invering operaion he opimal duy cycle for balancing he losses beween he modules wihin a cell appears o be close %. The resuls indicae ha by varying he value of δ zv, he elecrical load upon individual swiches can be varied by up o approximaely 8 W. This represens a percenage change in power losses of approximaely 57% above or below he nominal power losses when δ zv is se o %. orrespondingly he juncion emperaure of individual devices can be varied over a range of approximaely 2. This mehod has a limiaion in ha he

6 1 1 1 (a) Power osses ( o ) ( o ) ( o ) ( o ) 4 4 (b) Juncion Temperaures 4 4 Figure 9: Power loss and juncion emperaures of he fullbridge MM a uniy power facor Power osses when invering 2 MW Power osses when recifying 2 MW Juncion Temperaure (º) < Zero Volage Paern (%) > Temperaures when invering 2 MW < Zero Volage Paern (%) > 2 4 Juncion Temperaure (º) < Zero Volage Paern (%) > Temperaures when recifying 2 MW < Zero Volage Paern (%) > 2 4 (a) Invering operaion (b) Recifying operaion Figure 1: Power losses and emperaures of modules wihin an AA for differen values of δ zv a raed power power losses wihin s 1 and 3 are coupled and similarly he power losses wihin s 2 and 4 are coupled. VIII. ONINE THERMA MANAGEMENT USING THE ZERO VOTAGE DUTY YE I has been shown ha i is possible o significanly affec which modules wihin each cell of he AA experience he mos power losses by varying he zero volage duy cycle. By using his mehod wih a feedback mechanism, i should be possible o move some of he hermal and elecrical sress being placed upon a module which has been deeced as deerioraing on o he oher healhier modules wihin he cell. This could be achieved by measuring a emperaure dependen characerisic, such as he V ce I ce relaionship or by aking a measuremen of he modules case emperaure. Using he ANSYS model of he heasink mouned module he ransien hermal impedance from he and diode juncions o an exernal poin on he baseplae of he module was deermined. This poin was seleced so ha he hermal resisance from IGBuncion and he diode juncion o he measuring poin was approximaely equal. As he hermal sress placed s 1 and 4 and upon s 2 and 3 are coupled he proposed conrol scheme was picked o ry drive he difference beween he sum of he hermal sress on each coupled pair o zero. The conrol scheme is shown schemaically in Figure 11, where (s) represens he conroller. The conroller used was a simple proporionalinegral conroller wih a low gain, given he slow dynamics, o ensure good seady sae regulaion. The resuls of a 2 second

7 ase Temp 1 ase Temp 3 ase Temp 2 ase Temp 4 _ (s) Figure 11: Zero volage duy cycle feedback conrol scheme overview ( o ) Temperaure ( o ) δ zv Device Juncion Temperaures Time (s) Device ase Temperaures ase 1 ase 2 ase 3 ase Time (s) 65 Zero Volage Duy ycle Time (s) Figure 12: Zero volage duy cycle feedback demonsraion long simulaion are shown in Figure 12. A 4 seconds, he degradaion of module 1 is simulaed by increasing he power losses wihin he module by %, resuling in increase in is juncion emperaures and he case emperaure a he measuremen poin. This imbalance causes he compensaor o adjus he value of δ zv, moving some of he hermal sress away from modules 1 and 4 and on o modules 2 and 3. The sysem has a seling ime of approximaely 2 seconds before i reaches a seadysae value. The power flow during his simulaion remained a a consan 2MW in recifying mode. δ zv However, he devices in direcor swiches of he AA suffer relaively high conducion losses and herefore relaively high juncion emperaure. The power loss and hermal analysis of hese converers has also led o he developmen of a convenien mechanism for balancing emperaure rise of he devices wihin a cell. I is called he zerovolage sae duycycle (δ zv ) and i ses how ofen each of he wo possible zerovolage oupu saes of a fullbridge cell is seleced. By varying his duycycle, a significan amoun of elecrical and hermal sresses can be ransferred beween s 1 and 3, which form he curren carrying pah in one sae, and s 2 and 4, which form he curren carrying pah in he oher sae. A simple feedback mechanism for conrolling his duycycle online and hereby balancing he power losses beween modules wihin a cell wihou disruping he operaion of he converer has also been demonsraed by simulaion. REFERENES [1] Misubish Elecric d, Modules Applicaion Noe: The 5h Generaion hip use, December 27. [Online]. Available: hp:// files/manuals/igb_noe_e.pdf [2] ABB, Applicaion Noe: Thermal Design and Temperaure Raings of Modules. [Online]. Available: hp:// com/semiconducors [3] ABB, oadcycling capabiliy of HiPak Modules. [Online]. Available: {hp:// [4] Z. uo, A Thermal Model For Igb Modules And Is Implemenaion In A Real Time Simulaor, Ph.D. disseraion, Universiy of Pisburgh, 22. [5] M. M.. Merlin, T.. Green, P. D. Micheson, D. R. Trainer, D. R. richley, and R. W. rookes, A new hybrid mulilevel VolageSource onverer wih D faul blocking capabiliy, in A and D Power Transmission, 21. AD. 9h IET Inernaional onference on, oc. 21, pp [6] ABB HiPak Module 5SNA 12E331. Daashee available online. [Online]. Available: hp:// [7] U. Drofenik, D. oe, A. Müsing, J. M. Meyer, and J. W. Kolar, Modelling he hermal coupling beween inernal power semiconducor dies of a waercooled 33V/12A HiPak module, in Proceedings of Power onversion and Inelligen Moion onference, 27. [8] A. esnicar and R. Marquard, An innovaive modular mulilevel converer opology suiable for a wide power range, in Power Tech onference Proceedings, 23 IEEE Bologna, vol. 3, june 23, p. 6 pp. Vol.3. IX. ONUSION A power loss and hermal sudy of he semiconducor devices in he AA and he halfbridge and fullbridge varians of he MM has been conduced and he resuls indicae ha he devices in all of hese converers should be expeced o operae well below he maximum allowable juncion emperaure and allow a significan safey margin. omparing he AA o boh he halfbridge MM and fullbridge MM shows ha he maximum juncion emperaure of devices wihin he cells of he AA should be expeced o be lower han eiher of he MM formas because of lower overall cell power loss.