- Email: [email protected]
Numerical study of avalanche breakdown of 6H-SiC planar p-n junctions
Diamond
and Related Materials 6 ( 1997) : 500- 1503
Numerical study of avalanche breakdown of 6 -Sic planar p-n junctions E . Stefanov a,*, L.. Baijlon b, J. Barbolla b a LAAS - CNRS, 7, Avenue Colonel Rock, 31077 Toulouse, Cedex. France b Department ofElectrici1.v and Electronics - Faculty Sciences*University of Vailadolid, 47071 YuNadufic?,Spain
of
Abstract
The effect of the junction curvature, oxide charge density and n-epitaxial layer doping on GH-SiC planar p-n junction breakdown capabilities is reported. The calculated breakdown voltage of a cylindrical unprotected junction is compared with that of fieldplate- (FP-) and multiple field-limiting-ring- (FLR-) guarded junctions with optimized geometries. 0 1997 Elsevier Science S.A. Kqwortk
Breakdown voltage;
GH-SIC,p-n junction; Junction termination extensions; Floating ring; 2-D numerical analysis
1. Introduction
The design of pawcr devices relies considerably on design of junctions that have the necessary high voltage handling cap;\bilitics. Recent technology dcvclol~~~~e~~ts to fabricate planar devices based on 6H-SK are encouraging [I J, and the opportunity to use some junction termination extension (57X) techniques directly from Si technology bccomcs realistic. Various JTF%for planar junctions [2-41 have been report4 for Sic dcviscs, but only 50% of the i&al plant breakdown voltage was observed. We report a numerical study of the efrect of the junction err1vature on breakdown voltage VU,for planar y-n junctions in 6H-Sic. The efl’ect of the junction radius rJ, background doping N,, and fixed charge density N, on V,, is analyzed for a cylindrical p-n junction with a gaussian profile. Two concepts for planar JTE (rJ= I pm ijnd NB= 1.5 x IO’”cm “-‘) are studied: ( 1) FP-: and (2) multiple-FLR-structures. Ninety percent of the ideal plane parallel junction breakdown voltage was obtained for an optimized FLR system with five rings. The two-dimensional off-state program POWER was used to simulate the breakdown. the
. T
The two-dimensional otY=statesimulator POWER [5] is a user-oriented program for power devices. The simu_m___l_*Corresponding niithor. ~25-9635i%‘/$l7.~ 0 1997 Elsevier ScienceS.A. All rights reserved. HI 50925-9635(97)0005G.3
lation is based on the solution of Poisson’s equation for an arbitrary two-dimensional structure: BEBI(/=-q(NI:
-N,;
i-p-n)-pr,
(1)
where $ is the electrostatic potential; /jr is the fixed charge density; E is the pcrmitivity, q is the elementary Clli\~@, Iz and 11 urc the electron and hole densilies. respectively. and N,{ and IV, arc the ionized donor and acceptor densities. The ionization integrals for lh! carricrs from thr rcsultcd clcctric field arc calculated. The criterion for achieving breakdown is when these integrals tend to unity. A special algorithm included in the program ensures fast and automatic search of breakdown voltage. The generation rates were calculated by using the average ser of the ionization parameters fol GH-SIC given by Ruff et al. [6]. The program has been extended by an efficient algorithm to reduce the number of iterations when adjusting the quasi-Fermi potential of floating guard rings.
wirestructure
and si
Pkmc parallel p --)Ijunctions with depths Xj, and planur cylindrical junctions with curvature radius rj of 0.1. 0.3, I, 3 and IOpm are studied. The doping profile in all p--r1junctions is approximated by a gaussian in both the vertical and lateral directions. a surface concentration ,V,= 1 x 10zocme3. The ratio of lateral to vertical diffusion for the planar junction is assumed to be unity. The background doping Nn is in
E. Srefarlov et nl. / Dianlodand
Related Materials 6 (1997) 1500-1503
the range 1 x 10”-1 x 1018cmm3 and the diodes are considered non reach-through. The fixed charge density Nr is taken into account. Two JTE techniques have been studied for breakdown voltage capabilities. Fig. 1 shows the device structure of multiple FLR protected planar p+n diode with substrate doping ND= 1.5 x lOi cmm3 and junction radius rj= 1 pm. The ring spacing d,(i) is the width of the IZsubstrate at the surface between two adjacent p-rings. The ring width W,(i) is the width of the ith ring at the surface. The main ,junction is reverse biased with applied bias I$,. The second JTE is an FP-guarded planar diode with the same junction radius and substrate doping as for FLR, and is optimized for both the field plate length and oxide thickness. Fig. 2 shows the calculated breakdown voltage VBR dependence on the background doping for plane parallel p-n junctions with gaussian profile. A comparison is given with the ideal V,, of an abrupt junction. Fig. 3 shows the maximum electrir field and depletion width at breakdown. When decreasing the junction depth Xj from 10 to 0.1 urn, the abrupt junction behavior is approached. In this case, the doping gradient in the depletion layer increases due to decrease of the gaussian standard deviation (the surface concentration is kept
1501 1e+02
1
,
ie+l5
,
, ,.,,.,,
1
le+16
,
,
11111,
,
,
le+17
Background doping, (cm -3
Fig. 3. Maximum electric field and depletion tiersus doping of plan; parallel junctions.
,
le-01
,,,,,,
le+l8
) width at breakdown
constant). As a result, one side of the junction becomes considerably more doped than the other, and the maximum electric field at the junction becomes higher when increasing the substrate doping. Fig. 4 shows the maximum electric field dependence on NB in the case of planar cylindrical junction with gaussian profile. The parameter is the curvature radius rj and the solid line is for plane abrupt junction. The critical field behavior differs from the one shown in Fig. 3, owing to the additional etkct of the junction curvature. In all casts. E!,, varies by less than 3 factur of three in going from rj=O. 1 urn to IO urn. I& of t is normalized in respect of I planar junction
l&b04
I@+01 -I-” It?+15
le+16
1@*17
btckground doping, (cm -’ ) Fig. 2. Vbi,,versus doping for a plane parallel junction
lecl8 Background dopiflg, (cm 2~)
Fig.4. Maximumelectric field versus doping
for in pia:r~’ junclwn.
E. Stt$erov
1502
et cd. / Dimondnttd
breakdown voltage in plane parallel gaussian doped junction with the same junction depth rj= Xj. Both & and normalized breakdown voltage versus the substrate doping is given in Fig. 5 for radius rjcO.1, 0.3, 1, 3 and 10 pm. The relatively high fixed charge density present at the SQ-Sic interface was numericaly investigated on unprotected cylindrical p-n junctions for iVrin the range O-5 x 1012cmh2. Fig. 6 shows both V,, and the normalized breakdown voltage versus background doping for rj= 1 pm. The effect of Nf is much more pronounced for lower doped substrates due to the reduced screening effect of the substrate. In the FP technique the effect of oxide thickness (I,,, FP length IdFP and oxide charge are analyzed in order to obtain the maximum I&. For a non-reach.through doping with pm, background junction Xj=l
Bnckground
doping, (cm a
R&ted
Mtrteriuls
6 ( 1997)
1500-1503
1.5 x lOi cms3, the optimal VBR=640 V is obtained by using d,, =0.4 pm, and I?.,,= 10 pm. An increase of some 60% for the breakdown compared with that for an unprotected cylindrical junction is shown. The normalized breakdown voltage is 0.65 V. The multiple FLR concept was studied by using a new effective method [7] to optimize the basic parameters, exerting influence on the breakdown capabilities: ring-to-ring spacing d,, ring width W,, and number of rings. This method consists of modeling the breakdown capabilities of a main junction protected by one ring. The breakdown voltage of such a structure is examinated as a function of d, and W,. These results are extended to a multiple ring system. As a result, a structure with five floating rings was optimized to give a breakdown at VDR = 863 V, showing an increase of 110% of the breakdown compared with that for a cylindrical unprotected junction. The normalized breakdown voltage is 0.9 V. The optimized widths of the rings are 8, 4, 2, 2 and 2 pm and the spacings are 0.5, 0.6, 0.7, 1.1 and 1.9 pm. The lateral spread of the structure extends some 30 pm in respect of the main junction. Fig. 8 shows the two-dimensional equipotential lines for the optimized five-FLR system at breakdown. The V,, sensitivity of both techniques was studied in respect of the oxide charge and compared with that for a cylindrical unprotected junction. The VDRdependence on Nr for the optimized FLR structure is plotted on Fig. 7 (solid line). Fig. 7 gives also a comparison of the normalized breakdown versus Nr for an unprotected junction (dashed line), an FP- (long dashed line) and an FLK-protected (dotted line) junction. The dccrcase m I;,, with the incrcasc in Nf is smaller for an FP junction in the wholc calcnlatcd range of charge dcnsi-
)
_-.
900 675
1I343
1.0
1.0
------ -_ ---.._.
lo+11 0.0 le+15
la+16 Background
le+17 doping, (cm -3
let18
)
b&6. Vu8verals doping; Nr is paramctcr,
Fixed charge
le+l2 density
?I, (cm ’
)
Fig. 7. V,, versusNr for the optimal FLR system. Comparison of the normalized V,~R between the FP- and five-FLR systems. and unprotected cylindrical junction.
E. SteJanos et a,? / Diatnandand
Related Materials 6 (1997)
150&1503
1503
0.0065 -i 4 II
0.000
8 ‘(
0.001
“I
“II
- ”
0.002
*“”
/ ‘(‘8
0.003
8”’
1
0.004
X axe, (cm)
Fig. 8. Two-dimensional
equipotential lines for a five-floating-rings protected junction at breakdown, Vnn= 863 V.
ties. The plate, by means of the field effect, spreads the maximum electric field away from the junction curvature and mcreases the depletion width. As a result, the electric field reduces and VnR increases. In the FLR structure, two adjacent ring junctions act on the maximum field locus by pushing it deeply in the bulk around the junction curvature. Our optimized five-ring structure was optimized for N,= 1 x 10” cmV2 and saves its excellent capabilities in breakdown the range 0-4x 10” cm’-2 by showing a negligible decrease. In order to expand the validity for larger Nr, additional optimization of FLR geometry has to be made by increasing the rings number and decreasing the spacings.
4.
Collelllsion
Wc ilnillyX!d IllC Cll’CCl Ol‘ plillli~~ junction lacliUS. rf-cpitaxy doping :md SD,- 61-I-Sic“ inlcrlhcl: propcrlics on p. II junction breakdown capabilities by two-dimcnsional device simulation. The elficacity ofdevicc termination design has been examined for both field-plate- and five-floating-rings systems protected junctions. The results presented in the report clearly demonstrate the superiority of the FLR concept, and also the possibility to achieve near ideal efficiency of breakdown capabilities
for shallow junctions. hese results could serve as design constraints when designing power devices based on 6H-Sic.
This work is su uman Capital and No. ERBCWBG CT940696, and and Junta de Castilla y Leon.
and Related Materials 6 ( 1997) : 500- 1503
Numerical study of avalanche breakdown of 6 -Sic planar p-n junctions E . Stefanov a,*, L.. Baijlon b, J. Barbolla b a LAAS - CNRS, 7, Avenue Colonel Rock, 31077 Toulouse, Cedex. France b Department ofElectrici1.v and Electronics - Faculty Sciences*University of Vailadolid, 47071 YuNadufic?,Spain
of
Abstract
The effect of the junction curvature, oxide charge density and n-epitaxial layer doping on GH-SiC planar p-n junction breakdown capabilities is reported. The calculated breakdown voltage of a cylindrical unprotected junction is compared with that of fieldplate- (FP-) and multiple field-limiting-ring- (FLR-) guarded junctions with optimized geometries. 0 1997 Elsevier Science S.A. Kqwortk
Breakdown voltage;
GH-SIC,p-n junction; Junction termination extensions; Floating ring; 2-D numerical analysis
1. Introduction
The design of pawcr devices relies considerably on design of junctions that have the necessary high voltage handling cap;\bilitics. Recent technology dcvclol~~~~e~~ts to fabricate planar devices based on 6H-SK are encouraging [I J, and the opportunity to use some junction termination extension (57X) techniques directly from Si technology bccomcs realistic. Various JTF%for planar junctions [2-41 have been report4 for Sic dcviscs, but only 50% of the i&al plant breakdown voltage was observed. We report a numerical study of the efrect of the junction err1vature on breakdown voltage VU,for planar y-n junctions in 6H-Sic. The efl’ect of the junction radius rJ, background doping N,, and fixed charge density N, on V,, is analyzed for a cylindrical p-n junction with a gaussian profile. Two concepts for planar JTE (rJ= I pm ijnd NB= 1.5 x IO’”cm “-‘) are studied: ( 1) FP-: and (2) multiple-FLR-structures. Ninety percent of the ideal plane parallel junction breakdown voltage was obtained for an optimized FLR system with five rings. The two-dimensional off-state program POWER was used to simulate the breakdown. the
. T
The two-dimensional otY=statesimulator POWER [5] is a user-oriented program for power devices. The simu_m___l_*Corresponding niithor. ~25-9635i%‘/$l7.~ 0 1997 Elsevier ScienceS.A. All rights reserved. HI 50925-9635(97)0005G.3
lation is based on the solution of Poisson’s equation for an arbitrary two-dimensional structure: BEBI(/=-q(NI:
-N,;
i-p-n)-pr,
(1)
where $ is the electrostatic potential; /jr is the fixed charge density; E is the pcrmitivity, q is the elementary Clli\~@, Iz and 11 urc the electron and hole densilies. respectively. and N,{ and IV, arc the ionized donor and acceptor densities. The ionization integrals for lh! carricrs from thr rcsultcd clcctric field arc calculated. The criterion for achieving breakdown is when these integrals tend to unity. A special algorithm included in the program ensures fast and automatic search of breakdown voltage. The generation rates were calculated by using the average ser of the ionization parameters fol GH-SIC given by Ruff et al. [6]. The program has been extended by an efficient algorithm to reduce the number of iterations when adjusting the quasi-Fermi potential of floating guard rings.
wirestructure
and si
Pkmc parallel p --)Ijunctions with depths Xj, and planur cylindrical junctions with curvature radius rj of 0.1. 0.3, I, 3 and IOpm are studied. The doping profile in all p--r1junctions is approximated by a gaussian in both the vertical and lateral directions. a surface concentration ,V,= 1 x 10zocme3. The ratio of lateral to vertical diffusion for the planar junction is assumed to be unity. The background doping Nn is in
E. Srefarlov et nl. / Dianlodand
Related Materials 6 (1997) 1500-1503
the range 1 x 10”-1 x 1018cmm3 and the diodes are considered non reach-through. The fixed charge density Nr is taken into account. Two JTE techniques have been studied for breakdown voltage capabilities. Fig. 1 shows the device structure of multiple FLR protected planar p+n diode with substrate doping ND= 1.5 x lOi cmm3 and junction radius rj= 1 pm. The ring spacing d,(i) is the width of the IZsubstrate at the surface between two adjacent p-rings. The ring width W,(i) is the width of the ith ring at the surface. The main ,junction is reverse biased with applied bias I$,. The second JTE is an FP-guarded planar diode with the same junction radius and substrate doping as for FLR, and is optimized for both the field plate length and oxide thickness. Fig. 2 shows the calculated breakdown voltage VBR dependence on the background doping for plane parallel p-n junctions with gaussian profile. A comparison is given with the ideal V,, of an abrupt junction. Fig. 3 shows the maximum electrir field and depletion width at breakdown. When decreasing the junction depth Xj from 10 to 0.1 urn, the abrupt junction behavior is approached. In this case, the doping gradient in the depletion layer increases due to decrease of the gaussian standard deviation (the surface concentration is kept
1501 1e+02
1
,
ie+l5
,
, ,.,,.,,
1
le+16
,
,
11111,
,
,
le+17
Background doping, (cm -3
Fig. 3. Maximum electric field and depletion tiersus doping of plan; parallel junctions.
,
le-01
,,,,,,
le+l8
) width at breakdown
constant). As a result, one side of the junction becomes considerably more doped than the other, and the maximum electric field at the junction becomes higher when increasing the substrate doping. Fig. 4 shows the maximum electric field dependence on NB in the case of planar cylindrical junction with gaussian profile. The parameter is the curvature radius rj and the solid line is for plane abrupt junction. The critical field behavior differs from the one shown in Fig. 3, owing to the additional etkct of the junction curvature. In all casts. E!,, varies by less than 3 factur of three in going from rj=O. 1 urn to IO urn. I& of t is normalized in respect of I planar junction
l&b04
I@+01 -I-” It?+15
le+16
1@*17
btckground doping, (cm -’ ) Fig. 2. Vbi,,versus doping for a plane parallel junction
lecl8 Background dopiflg, (cm 2~)
Fig.4. Maximumelectric field versus doping
for in pia:r~’ junclwn.
E. Stt$erov
1502
et cd. / Dimondnttd
breakdown voltage in plane parallel gaussian doped junction with the same junction depth rj= Xj. Both & and normalized breakdown voltage versus the substrate doping is given in Fig. 5 for radius rjcO.1, 0.3, 1, 3 and 10 pm. The relatively high fixed charge density present at the SQ-Sic interface was numericaly investigated on unprotected cylindrical p-n junctions for iVrin the range O-5 x 1012cmh2. Fig. 6 shows both V,, and the normalized breakdown voltage versus background doping for rj= 1 pm. The effect of Nf is much more pronounced for lower doped substrates due to the reduced screening effect of the substrate. In the FP technique the effect of oxide thickness (I,,, FP length IdFP and oxide charge are analyzed in order to obtain the maximum I&. For a non-reach.through doping with pm, background junction Xj=l
Bnckground
doping, (cm a
R&ted
Mtrteriuls
6 ( 1997)
1500-1503
1.5 x lOi cms3, the optimal VBR=640 V is obtained by using d,, =0.4 pm, and I?.,,= 10 pm. An increase of some 60% for the breakdown compared with that for an unprotected cylindrical junction is shown. The normalized breakdown voltage is 0.65 V. The multiple FLR concept was studied by using a new effective method [7] to optimize the basic parameters, exerting influence on the breakdown capabilities: ring-to-ring spacing d,, ring width W,, and number of rings. This method consists of modeling the breakdown capabilities of a main junction protected by one ring. The breakdown voltage of such a structure is examinated as a function of d, and W,. These results are extended to a multiple ring system. As a result, a structure with five floating rings was optimized to give a breakdown at VDR = 863 V, showing an increase of 110% of the breakdown compared with that for a cylindrical unprotected junction. The normalized breakdown voltage is 0.9 V. The optimized widths of the rings are 8, 4, 2, 2 and 2 pm and the spacings are 0.5, 0.6, 0.7, 1.1 and 1.9 pm. The lateral spread of the structure extends some 30 pm in respect of the main junction. Fig. 8 shows the two-dimensional equipotential lines for the optimized five-FLR system at breakdown. The V,, sensitivity of both techniques was studied in respect of the oxide charge and compared with that for a cylindrical unprotected junction. The VDRdependence on Nr for the optimized FLR structure is plotted on Fig. 7 (solid line). Fig. 7 gives also a comparison of the normalized breakdown versus Nr for an unprotected junction (dashed line), an FP- (long dashed line) and an FLK-protected (dotted line) junction. The dccrcase m I;,, with the incrcasc in Nf is smaller for an FP junction in the wholc calcnlatcd range of charge dcnsi-
)
_-.
900 675
1I343
1.0
1.0
------ -_ ---.._.
lo+11 0.0 le+15
la+16 Background
le+17 doping, (cm -3
let18
)
b&6. Vu8verals doping; Nr is paramctcr,
Fixed charge
le+l2 density
?I, (cm ’
)
Fig. 7. V,, versusNr for the optimal FLR system. Comparison of the normalized V,~R between the FP- and five-FLR systems. and unprotected cylindrical junction.
E. SteJanos et a,? / Diatnandand
Related Materials 6 (1997)
150&1503
1503
0.0065 -i 4 II
0.000
8 ‘(
0.001
“I
“II
- ”
0.002
*“”
/ ‘(‘8
0.003
8”’
1
0.004
X axe, (cm)
Fig. 8. Two-dimensional
equipotential lines for a five-floating-rings protected junction at breakdown, Vnn= 863 V.
ties. The plate, by means of the field effect, spreads the maximum electric field away from the junction curvature and mcreases the depletion width. As a result, the electric field reduces and VnR increases. In the FLR structure, two adjacent ring junctions act on the maximum field locus by pushing it deeply in the bulk around the junction curvature. Our optimized five-ring structure was optimized for N,= 1 x 10” cmV2 and saves its excellent capabilities in breakdown the range 0-4x 10” cm’-2 by showing a negligible decrease. In order to expand the validity for larger Nr, additional optimization of FLR geometry has to be made by increasing the rings number and decreasing the spacings.
4.
Collelllsion
Wc ilnillyX!d IllC Cll’CCl Ol‘ plillli~~ junction lacliUS. rf-cpitaxy doping :md SD,- 61-I-Sic“ inlcrlhcl: propcrlics on p. II junction breakdown capabilities by two-dimcnsional device simulation. The elficacity ofdevicc termination design has been examined for both field-plate- and five-floating-rings systems protected junctions. The results presented in the report clearly demonstrate the superiority of the FLR concept, and also the possibility to achieve near ideal efficiency of breakdown capabilities
for shallow junctions. hese results could serve as design constraints when designing power devices based on 6H-Sic.
This work is su uman Capital and No. ERBCWBG CT940696, and and Junta de Castilla y Leon.