GaN-HEMT Dynamic ON-state Resistance characterisation and Modelling

Transcription

1 GaN-HEMT Dynamic ON-sae Resisance characerisaion and Modelling Ke Li, Paul Evans, Mark Johnson Power Elecronics, Machine and Conrol group Universiy of Noingham, UK Absrac GaN-HEMTs suffer from rapping effecs which migh increase device ON-sae resisance ( ) values. Thus, dynamic of a commercial GaN-HEMT is characerized a differen bias volages in he paper by a proposed measuremen circui. Based on he measuremen resuls, a behavioural model is proposed o represen device dynamic values, in which rapping and derapping ime consan is represened by a series of RC nework. The model is simulaed in PSPICE, of which he simulaion resuls of values are compared and validaed wih he measuremen when device swiches in a power converer wih differen duy cycles and swiching volages. The resuls show ha values of his device would increase due o rapping effecs. Keywords GaN-HEMT; Dynamic ON-sae resisance; Power semiconducor device characerisaion; Power semiconducor device modelling; Behavioural model I. INTRODUCTION Applicaion of GaN power semiconducor devices is becoming more and more popular in power elecronics sysems design, because GaN devices can operae a high emperaure, swich fas and produce low power loss. In order o beer use GaN devices in power elecronics sysems, i is necessary o undersand heir characerisics. High-Elecron- Mobiliy Transisors (HEMTs) are he mos widely used GaN power elecronic devices, bu hey suffer from elecron rapping effecs ha reduce heir curren carrying capabiliy [1], [2]. Illusraed in Fig. 1, his rapping effec is relaed mainly wih wo effecs when device in OFF-sae, one is he bias V DS volage value and anoher is he bias ime (rapping ime), which would give rise o he increase of GaN device ON-sae resisance ( ) value. According o he resuls presened by auhors in [3], GaN device values would increase by a maximal facor of 4 afer 1ms bias ime depending on bias V DS volage value. The resuls presened by auhors in [4] demonsrae ha device values would furher increase 5% if bias ime is doubled. In he on-sae (low bias), derapping occurs and he values decrease o he seady sae value a a rae characerised by he derapping ime. As shown in [3], GaN device dynamic values would decrease 3% afer 1µs derapping ime. When employing GaN ransisors in power elecronics circuis, GaN device migh swich wih differen periods and duy V DS 1s bias volage 2nd bias volage Trapping ime 1s dynamic 2nd dynamic Derapping ime Fig. 1: GaN-HEMT dynamic values due o rapping effecs cycles leading o a combinaion of rapping and derapping effecs and consequenly uncerainy in he acual value of. This will lead o uncerainy in device power loss, making predicions of converer efficiency and cooling sysem design challenging. The abiliy o characerise and model GaN- HEMT values is hus an imporan design consideraion. There are mainly wo differen mehods o measure GaN- HEMT dynamic values: one mehod is by using direcly a measuremen equipmen [4], and anoher one is by using an elecrical circui, where differen circui opologies are proposed by auhors in [3], [5] [8]. In his paper, a new characerisaion circui is presened o measure GaN- HEMT dynamic values, which could be easily implemened. Compared o he above differen circuis, his new measuremen circui needs fewer componens and offers an alernaive mehod o characerise he device and o compare he resuls. Based on he dynamic measuremen resuls, a behavioural model is hus proposed o presen device rapping and derapping effec, which could be used in a circui simulaor. The paper has he following srucure. Firs he measuremen circui o characerize GaN-HEMT dynamic values is presened. Aferwards, measuremen resuls of differen rapping and derapping ime on values are shown. Based on he measuremen resuls, a behavioural model is proposed o represen device dynamic values. The model is furher validaed by comparing wih he measuremen /16/$31. c 216 IEEE

2 when device swiches in a power converer. Conclusions are given a las. II. GAN-HEMT DYNAMIC ON-STATE RESISTANCE A. circui MEASUREMENT GaN device values can be obained by measuring device ON-sae volage V DS(on) across i and curren I D hrough i in an elecrical circui. As he measured bias volage when device is OFF (V DS(off) ) could achieve more han several hundred imes higher han device V DS(on), a volage clamping circui is necessary o reduce he measured V DS(off) in order o increase measuremen accuracy. For his reason, he measuremen circui shown in Fig. 2 is consiued by a bias circui o conrol device rapping ime when i is OFF and a volage clamping circui o measure device V DS(on) value when i is ON. In he bias circui, a ransisor T1 is used o conrol DUT rapping ime. A resisive load R load is used o se he curren level when DUT is in ON-sae. Because of he parasiic inducance L para of he R load, wo diodes D1, D2 offer a free wheeling pah of he curren when eiher T1 or DUT is swiched from ON o OFF. The volage clamping circui is consiued by a depleion mode (D-mode) Si-MOSFET and a zener diode. DUT measuremen volage V DS(m.) is measured across he zener diode. The principle of he volage clamping circui is ha when DUT is ON, D-mode Si-MOSFET is in ON-sae (V gms m is superior o MOSFET hreshold volage V h ), so poins s m and d m are almos in he same poenial and DUT V DS(on) can hus be measured direcly (V DS(m.) = V DS(on) ). When DUT is OFF, zener diode juncion capaciance is charged a firs, so V DS(m.) increases o he zener diode clamping volage V clamp, and hen V gms m is inferior o V h. Aferwards, D-mode MOSFET is pinched OFF and is iner-elecrode capaciance C dms m is charged o wihsands almos he whole bias V DS volage (V DS V DS(m.) ). I is o be noed ha as here is a leakage curren balance beween D-mode MOSFET and Zener diode, V DS(m.) is inferior o V clamp in seady-sae. Insead of measuring volage range beween V DS(on) and V DS, a much smaller volage range beween V DS(on) and V clamp is measured, hus he measuremen sensiiviy is increased. Compared o he similar ype volage clamping circuis ha are analyzed by auhors in [6], less componens and no exernal power supply are used in his clamping circui. Device seady-sae could be measured by applying he conrol signal shown in Fig. 2b, where DUT is kep always in ON-sae and T1 is conrolled by a single pulse. Device dynamic could be measured by applying he conrol signal shown in Fig. 2c, where DUT is iniially kep in ON-sae and T1 blocks all he bias volage. Then a 1, DUT is swiched OFF and a 2, T1 is swiched ON, hus Volage bias circui L para L para 56µF D1 1 2 (b) Conrol signal o measure seadysae T1 L para D2 R load (a) configuraion DUT I D measuremen D d m g m S Volage Clamping circui s m V DS measuremen (c) Conrol signal o measure dynamic Fig. 2: GaN-HEMT dynamic ON-sae resisance measuremen circui and conrol signals Load T1 Volage clamping circui D2 D1 DUT Fig. 3: Realizaion of he measuremen circui all he bias volage is across DUT. Aferwards, a 3, DUT is swiched ON again, so curren I D flows hrough he DUT. Finally a 4, T1 is swiched OFF. Thus, DUT rapping ime is conrolled by 2-3 while derapping ime is conrolled by 3-4, so values of differen rapping and derapping ime could be measured. The realizaion of he measuremen circui is shown in Fig. 3. In he measuremen, R load = 1Ω, T1 is a commercial GaN-HEMT (EPC212C, 2V/5A) while D1 and D2 are he

3 V DS(m.) (V) (Ω) I D (A) # 1-4 Dynamic (12V) Dynamic (8V) V GS (V) Time(µs) Fig. 4: Measured waveforms when device is bias a 8V for 1µs Fig. 5: Comparison of seady-sae a 25 C and dynamic same Schoky diode (MBRS421T3G, 2V/4A). Dynamic values of a DUT, which is he same as T1, is measured by he above circui, of which he resuls are presened in he nex secion. B. resuls Several major parameers of he measuremen equipmens and clamping circui devices are summarized in TABLE. I. In he measuremen, he maximal measured V DS volage is 3.3V, which could achieve a measuremen accuracy of a leas =.13V by using a 8-bi resoluion oscilloscope. GaN- HEMT dynamic values are measured when i is biased a 8V and 12V. When GaN-HEMT is biased a 8V and rapping ime is 1µs, he measured volage V DS(m.), curren I D and volage V GS are shown in Fig. 4. Because of he volage clamping circui, V DS(m.) is abou 1.7V when DUT is OFF, which is much smaller han he bias volage (8V). I is o be noed ha maximal V DS(m.) almos does no change when he bias volage increases o 12V, hus he measuremen accuracy is improved in comparison o a direc measuremen. DUT dynamic values are calculaed afer insan in Fig. 4, which is 1µs afer OFF-ON ransiion when each elecrical parameer sabilizes. The obained values based on he above waveforms is shown in Fig. 5, in which values are compared beween seady-sae and dynamic-sae when device is biased a 8V and 12V a 25 C. As shown in he resuls, values increase wih he rapping ime and i decreases wih derapping ime, and i increases more wih a higher bias volage. For his device, i is observed ha when device biased by a cerain rapping ime, i needs a longer derapping ime o reduce is dynamic values o he seady-sae values, which shows ha effecive values are likely o be higher han heoreical values in applicaion. A behavioural model is proposed based on he measuremen resuls, which will be presened in he nex secion B Vcomp V GS (V) Fig. 6: values of differen V GS volages a 25 C III. GAN-HEMT DYNAMIC ON-STATE RESISTANCE A. Behavioural model MODELLING When in seady-sae, he measured device values of differen V GS volages a 25 C is shown in Fig. 6, which presens ha values increase when V GS volage decreases. According o his -V GS relaion, he obained dynamic values could be represened by saic values a he equivalen gae volage shown in Fig. 6, where poin A corresponds o he value when i is obained a seady-sae, and poin B corresponds o he value afer cerain rapping ime. The V GS volage difference beween poin A and poin B, which is defined as p, could be applied o represen change during he rapping and derapping process. Afer adding p in gae circui, which is shown in Fig. 7, device V GS volage (V GS = V G p ) afer rapping ime could be adjused, hus a dynamic value could be obained. I is shown ha p increases wih he rapping ime and decreases wih he derapping ime, in which he phenomena could be modelled in he form of an RC circui, which is presened in Fig. 7. In each RC uni, pn increases when A

4 V G p G D V DS Trapping ime: 1ms 1ms 1ms Model V GS S.1 1µs 1µs 1µs R 1d R 1 C 1 V 1 = k1 V DS p1 R nd R n C n V n = kn V DS p = p1 + + pn pn Fig. 7: Dynamic values represenaion by behavioural model capacior C n is charged by a conrolled volage source V n hrough resisor R n and i decreases when C n is discharged hrough resisor R nd. Thus, GaN-HEMT rapping and derapping effec could be represened by a series of he RC unis, and p value is he sum of he volage of he capacior in each uni. Based on he measuremen resuls shown in Fig. 5 when bias volage is 8V, seven RC unis are used o represen device dynamic values, in which parameers V 1 -V 7, C 1 -C 7, R 1d -R 7d and R 1 -R 7 are obained by fiing mehods. The values of he above parameers are given in TABLE. II. The comparison beween he model and he measuremen is shown in Fig. 8, where he model represens generally well dynamic values of differen rapping and derapping ime. Afer obaining he above parameers, he model illusraed in Fig. 7 could be easily implemened in a circui simulaor. In his paper, he model is implemened in PSPICE, where p and V 1 V 7 are represened by volage conrolled volage source and coefficiens k 1 k 7 could be calculaed once V 1 V 7 are known Fig. 8: Comparison beween he measuremen and model on dynamic values as a funcion of derapping ime for differen rapping imes when V DS = 8V.45 Trapping ime: 1ms.35 1ms ms 1µs 1µs 1µs Model Fig. 9: Comparison beween he measuremen and model on dynamic values when V DS = 12V B. Model exension for differen bias volage Following by ha, he same model is used o represen he measuremen resuls when he device is biased a V DS = 12V. In order o represen device dynamic values of differen bias volages by he model and easily implemen i, only V 1...V 7 are fied again when V DS = 12V, while he oher parameers are he same as obained when V DS = 8V. In such a case, bigger p values migh be obained, so TABLE I: Major parameers of he measuremen equipmens and clamping circui devices Oscilloscope Curren probe Volage probe D-mode MOSFET Zener diode DPO414B (1GHz, 8-bi) TCP312 (1MHz, 3A) TPP1 (1GHz, 3V) BSP149 (2V, V h 1.4V) BZT52C3V3 (3.3V) TABLE II: Parameers using o represen GaN-HEMT rapping effec when device is bias a 8V V 1 V 2 V 3 V 4 V 5 V 6 V 7.51V.6V.28V.25V.66V.1V.97V R 1d R 2d R 3d R 4d R 5d R 6d R 7d Ω Ω Ω Ω Ω Ω Ω R 1 R 2 R 3 R 4 R 5 R 6 R Ω 5.6Ω 31.5Ω Ω Ω Ω.5Ω C 1 C 2 C 3 C 4 C 5 C 6 C F F F F F F F

5 .25.2 Simulaion 1 2 Fig. 1: Conrol signal when device swiches coninuously in a power converer bigger dynamic values would be represened when device is biased a he same rapping ime as V DS = 8V. I corresponds o he measuremen resuls shown in Fig. 5 ha bigger dynamic values are obained when device is bias a 12V han a 8V. Obviously, all he above 28 parameers could be fied again o represen he measuremen resuls when V DS = 12V. However, i migh be difficul o find he rend of each parameer wih he bias volage, especially ha of R nd, R n and C n. In conrary, he rend of V n wih he bias volage is easier o be found and implemened in he model. The obained parameers of V 1 -V 7 in his siuaion are given in TABLE III. The comparison beween he model and he measuremen when device biased a 12V is shown in Fig. 9, where i is shown ha he model could represen he measuremen in a reasonable way. As GaN-HEMT suffered from rapping effec, is values migh increase when i swiches coninuously in a power converer. For his reason, values esimaed by he above model is compared wih he measuremen, and he resuls will be presened in he nex secion. A. Swiching volage a 8V IV. MODEL VALIDATION The same elecrical circui shown in Fig. 2a wih he conrol signal shown in Fig. 1 is used o measure device dynamic values when i swiches coninuously. In order o avoid device self heaing, i swiches a 1kHz wih differen duy cycles and a duraion of.1s. When swiching volage is 8V and duy cycle (D) is 5% and 9%, he comparison of he values is shown in Fig. 11a and in Fig. 11b separaely. As explained previously, device derapping ime o reach is seady-sae value is longer han rapping ime o increase is dynamic values, which explains ha he measured values increase o a facor of wo higher han is seady-sae value afer.1s in he measuremen. The rapping ime in he above measuremens varies from 5µs o 1µs and derapping ime Simulaion (a) Duy cycle is 5% (b) Duy cycle is 9% Fig. 11: Comparison beween he measuremen and simulaion on values when device swiches a 8V varies from 5µs o 9 µs. According he measuremen resuls shown in Fig. 8, dynamic values are no influenced by derapping ime obviously in his derapping ime range, which migh explain ha a change of duy cycle does no change device values obviously. The rapping effecs can increase values very quickly when rapping ime inferior o 1µs, hus i migh sill increase device values when device using in high frequency converer (more han 1kHz). However, if device swiches in high frequency, he swiching loss migh increase device juncion emperaure T j, which would also increase device values. I is also shown in Fig. 11 ha he rend of he values increase in he measuremen is represened generally well by he simulaion, where he difference beween he measuremen and he simulaion in boh siuaions is wihin 17%. The difference beween he simulaion and he measuremen migh be due o he difference beween he model and he measuremen shown in Fig. 8 on dynamic values. TABLE III: Parameers using o represen GaN-HEMT rapping effec when device is bias a 12V V 1 V 2 V 3 V 4 V 5 V 6 V 7.143V.1V.32V.1V.74V.1V 1.59V

6 Simulaion Simulaion (a) Device swiches a 12V (D=5%) (b) Device swiches a 1V (D=5%) Fig. 12: Comparison beween he measuremen and simulaion on values when device swiches a differen volages B. Swiching volage a 12V and 1V The device is hen swiched a 12V, of which he comparison beween he measuremen and simulaion resuls is shown in Fig. 12a. Despie some difference beween he measuremen and simulaion, he increase rend of he values is represened well by he model. The bigger values ripper in he simulaion han in measuremens is mainly relied on difference illusraed in Fig. 9 beween he model and he measuremen on change influenced by derapping ime, where he model represens a bigger value variaion han he measuremen. The device is hen swiched a 1V, of which he comparison beween he measuremen and simulaion resuls is shown in Fig. 12b. In his condiion, parameers R nd, R n and C n are he same as ha shown in TABLE II, and parameers V n is inerpolaed linearly based on he resuls obained a TABLE II and TABLE III. I is shown in Fig. 12b ha he simulaion represens well he measuremen, and obained dynamic values in his condiion is beween he values when device biased a 12V and 8V. Even wih some difference, all he above resuls could validae he proposed model, which could be used o esimae device conducion loss including rapping effec in power converers a differen swiching volages. V. CONCLUSION In his paper, dynamic ON-sae resisance ( ) values of a commercial GaN-HEMT is measured a differen bias volages by a proposed elecrical circui, which is consiued by a volage bias circui o conrol DUT rapping ime and a volage clamping circui o measure DUT on-sae volage. Compared o oher elecrical characerisaion circuis, his circui has less elecrical componens and could be easily implemened. I is shown in he characerisaion resuls ha boh rapping and derapping ime could influence device dynamic values and a higher bias volage would give rise o a higher dynamic value. Based on he characerisaion resuls, a behavioural model, which is consiued by a series of RC nework, is proposed o represen device dynamic values of differen bias volages. The model is hen implemened in PSPICE and validaed wih he measuremen when device swiches in a power converer wih differen duy cycles and swiching volages. Boh he measuremen and simulaion resuls show ha device values would increase due o he rapping effecs. The rapping effecs of he device migh sill increase is values in more han 1kHz high frequency power converers, because device derapping ime o reach is seadysae value is longer han rapping ime o increase is dynamic values. The proposed model could be hus used o esimae device conducion loss in a power converer when using GaN-HEMT of differen swiching volages. ACKNOWLEDGMENT The auhors would like o hank UK Engineering and Physical Sciences Research Council (EPSRC) for he financial suppor under EP/K14471/1 and all he parners of he projec Silicon Compaible GaN Power Elecronics for echnical discussions. REFERENCES [1] T. Mizuani, Y. Ohno, M. Akia, S. Kishimoo, and K. Maezawa, A sudy on curren collapse in AlGaN/GaN HEMTs induced by bias sress, Elecron Devices, IEEE Transacions on, vol. 5, no. 1, pp , 23. [2] M. Meneghini, D. Bisi, D. Marcon, S. Soffels, M. Van Hove, T.-L. Wu, S. Decouere, G. Meneghesso, and E. Zanoni, Trapping and reliabiliy assessmen in d-mode gan-based mis-hems for power applicaions, Power Elecronics, IEEE Transacions on, vol. 29, no. 5, pp , 214. [3] B. Lu, T. Palacios, D. Risbud, S. Bahl, and D. Anderson, Exracion of Dynamic On-Resisance in GaN Transisors: Under Sof- and Hard- Swiching Condiions, in Compound Semiconducor Inegraed Circui Symposium (CSICS), 211 IEEE, pp. 1 4, Oc 211. [4] D. Jin and J. Del Alamo, Mehodology for he Sudy of Dynamic ON-Resisance in High-Volage GaN Field-Effec Transisors, Elecron Devices, IEEE Transacions on, vol. 6, pp , Oc 213. [5] G. Cao, A. Ansari, and H.-J. Kim, A New Circui o Evaluae Curren Collapse Effec of GaN HEMTs Under Pracical Condiions, Insrumenaion and, IEEE Transacions on, vol. 64, pp , July 215.

7 [6] R. Gelagaev, P. Jacqmaer, and J. Driesen, A Fas Volage Clamp Circui for he Accurae of he Dynamic ON-Resisance of Power Transisors, Indusrial Elecronics, IEEE Transacions on, vol. 62, pp , Feb 215. [7] R. Chu, A. Corrion, M. Chen, R. Li, D. Wong, D. Zehnder, B. Hughes, and K. Bouros, 12-V Normally Off GaN-on-Si Field-Effec Transisors Wih Low Dynamic on -Resisance, Elecron Device Leers, IEEE, vol. 32, pp , may 211. [8] J. Bocker, H. Jus, O. Hil, N. Badawi, J. Wurfl, and S. Dieckerhoff, Experimenal analysis and modeling of GaN normally-off HFETs wih rapping effecs, in Power Elecronics and Applicaions (EPE 15 ECCE- Europe), h European Conference on, pp. 1 1, Sep 215.