0% found this document useful (0 votes)
42 views

Fast Recovery Diodes - Reverse Recovery Behaviour and Dynamic Avalanche

This document discusses the reverse recovery behavior of fast recovery diodes. It describes the differences between "snappy" and "soft" reverse recovery. For snappy recovery, the plasma zone that feeds the reverse current meets abruptly, interrupting the current. For soft recovery, sufficient plasma remains to avoid current interruption. The document presents simplified models to analytically investigate the reverse recovery process and parameters for achieving soft recovery. It examines reverse recovery and dynamic ruggedness conditions through qualitative and quantitative analysis.

Uploaded by

Yogi Saputra
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
42 views

Fast Recovery Diodes - Reverse Recovery Behaviour and Dynamic Avalanche

This document discusses the reverse recovery behavior of fast recovery diodes. It describes the differences between "snappy" and "soft" reverse recovery. For snappy recovery, the plasma zone that feeds the reverse current meets abruptly, interrupting the current. For soft recovery, sufficient plasma remains to avoid current interruption. The document presents simplified models to analytically investigate the reverse recovery process and parameters for achieving soft recovery. It examines reverse recovery and dynamic ruggedness conditions through qualitative and quantitative analysis.

Uploaded by

Yogi Saputra
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

PROC. 24th INTERNATIONAL CONFERENCE ON MICROELECTRONICS (MIEL 2004).

VOL 1, NIS, SERBIA AND


MONTENEGRO, 16-1 9 MAY, 2004 pp. 11-16

Fast Recovery Diodes - Reverse Recovery Behaviour


and Dynamic Avalanche
Josef Lutz

Abstract - This paper deals with the dynamics of the internal II. SNAPPY AND SOFT REVERSE RECOVERY
charge carriers during the reverse recovery process of a fast power
diode. For two conditions, simplified approximations are given
for analytical investigation, and parameters for achieving soft
The current waveform for a snappy diode is shown in
recovery are derived. Finally, both conditions are investigated Fig. 1a. In Fig. 1b the simulation of the internal behaviour
with regard to dynamic ruggedness. of the stored charge for a snappy diode is presented. The
line t0 shows the distribution of free carriers during on-
I. INTRODUCTION state. In the low doped base region, n ≈ p holds. During

Fast recovery diodes are necessary in every modern


power electronic circuit. Using modern switching devices, i
high current slopes di/dt occur, and it is essential that the t
ts tf
diode answers with soft recovery behaviour. It took a com-
paratively long time until sufficient solutions for soft-
recovery were found. Diodes with control of the axial life-
snappy
time profile [1] or control of the p emitter efficiency [2] are
meanwhile established. In the 600V - 1700V voltage range
they exhibit soft-recovery behaviour even under critical
conditions such as high di/dt’s and low currents. Also a
trench structures in the p-anode for improving the recovery
behaviour have been investigated [3]. However, for appli- Doping
cations in the voltage range >3000V, the existing solutions 1E18
Hole Density [cm-3]aa

are not sufficient.


Device simulation is a powerful tool for designing a
soft recovery diode. But still not many papers have treated t0
the reverse recovery behaviour in an analytical way, alt-
hough, this is necessary for a complete understanding of
t1
simulation results and for evaluating the critical conditions. t2 t2
Moreover, if strong dynamical avalanche occurs, device t1
simulations show formation of current filaments, and it
1E14
becomes difficult to evaluate the results.
Therefore, this paper will try to give some useful ap-
proximations for the reverse recovery and dynamic rug- - t3 +
gedness of fast power diodes. In a first part, the internal p n t4 n
behaviour in the diode for both snappy and for soft recov-
ery behaviour is presented qualitatively. In the second part, Depth w
the internal phenomena are treated analytically A) during b
Fig. 1: Diode with snappy recovery behavior. Current (a), hole
the voltage build-up phase B) when the voltage is estab- density at different points in time (b)
lished and a tail current is flowing, but snap-off of the tail-
current must be avoided. In final part, condition A) and B)
commutation, holes are removed to the left side and elec-
are investigated with respect to dynamic ruggedness.
trons are removed to the right side, and the plasma zone is
feeding the current. t1 and t2 show points in time where still
Josef Lutz is Professor for Power Electronics and Electromag- a reverse current is flowing. Immediately after t2, both
netic Compatibility at the Faculty of Electrical Engineering, fronts of the plasma meet, suddenly the source for feeding
Chemnitz University of Technology, Reichenhainer Str. 70, D- the reverse current vanishes. The reverse current is inter-
09126 Chemnitz, Germany. E-mail: [email protected] rupted abruptly, leading to snappy recovery behaviour.
In the diode with soft recovery behaviour (Fig. 2a), NA, ND
p, n E0
there is still enough plasma in the diode at t3, and if the
width between pn-junction and plasma is sufficient to take |E| E(w) j = jn + jp
vl vr
the voltage until the end of the recovery process, no snap-
off will occur. p n+
j = jpl n≈p = n j = jnr

i -
n- +
t
ts tf wB
0
w
Fig. 3: Simplified model for the reverse recovery during the
soft voltage increase
holds. The boundary condition between the plasma and the
a removed zone is
j pl = j n + j p (1)
Doping
1E18 the difference in the hole current at the boundary is
j pl − j p = ∆j p = j n
(2)
From ∆jp, a movement of the plasma border to the right can
Hole Density

t0 be derived according to [4]. Simplifying the carrier distri-


bution in Fig 1 at t0 in a first step as homogenous with
t1 n = p = n , the velocity of the plasma boarder to the right
t2 t3
is
1E14 dw µn j
vl = = ⋅ (3)
t4
dt µ n + µ p q ⋅ n
p n- n+ In the same way, on the right side of the plasma zone
the boundary condition is
Depth
j nr = j n + j p (4)
b
Fig. 2: Diode with soft recovery behavior. Current (a), hole leading to
density at different points in time (b) µp j
vr = ⋅ (5)
Whether the recovery behaviour is soft or snappy de- µn + µ p q⋅n
pends on the final position of the plasma. It should be close In Silicon, µn ≈ 3µp holds, resulting in vl ≈ 3vr . The
to the right side. plasma boarder moves three times faster from the pn-
junction side than from the nn+ junction side. If n is con-
III. DYNAMICS OF THE INTERNAL CHARGE stant, the plasma boarders will meet at

For an analytical treatment, simplifications are made.


vl 3
wx = ⋅ wB ≈ wB (6)
The pn- and n+-junction are assumed as abrupt, and the vl + v r 4
plasma fronts are treated as abrupt. Since the amount of The electrical field is built up between the plasma
charge carriers removed by the current is much higher than and the pn-junction. Because of neutrality in the plasma
the recombination of charge carriers, recombination in this zone, the electrical field must be triangular as shown in Fig.
time interval is neglected. Under these approximations, two 3. The contribution of the right side to the voltage can be
situations shall be described neglected. Therefore the device can bear a voltage repre-
sented by the area under the E(w) with w = wx, before the
A.. Voltage built-up phase plasma boarders meet.

The investigation for the behaviour of the stored plasma But in reality the carrier distribution in the on-state
in the phase of the voltage increase shall be done similar to is not homogenous. In a standard diode with highly doped
the model of Benda and Spenke [4]. Fig. 3 simplifies the emitters and with homogenous lifetime it is higher on the
situation for a selected time in Fig. 1. In the remaining pn-junction side (left side) than on the right side, as shown
plasma, j = jn + jp holds. At left side, towards the pn- in Fig. 1. If n on the left side is simplified by nl and on the
junction, the plasma is removed, for the current j = jpl right side by n r , the division of (5) by (3)will lead to
vr µ p n l j
= ⋅ (7) p= (9)
vl µn nr q ⋅ vsat(p)
Inserting this in (6) results in with vsat(p) approx. 1.107cm/s. The holes are of the same
1 polarity than the positive charged donors in the space-
wx = ⋅ wB (8) charge, resulting in an effective doping
1 + µ p nl µ n nr N eff = N D + p (10)
Simplifying the distribution in Fig. 1 with n l = 2 ⋅ n r and the gradient of the electrical field is
leads to wx = 0,6wB. This means the standard diode will
dE q
snap of even at a lower voltage. To avoid snap off, the = (N D + p ) (11)
width wB would have to be increased strongly. dw ε
This situation for a diode with low base doping ND is
1E+17
shown in Fig. 5. The voltage can be assumed as constant,
therefore the area under E(w) is constant. The hole density
standard diode
is one factor in dE/dw, and p represents a hole current, ex-
tracted from the remaining plasma. With (9), now (3) trans-
fers to
1E16
1E+16
dw µn p ⋅ v sat ( p )
vl = = ⋅ (12)
Soft recovery
dt µ n + µ p n
diode
1E+15

p n- n+
1E14
1E+14
-20 0 20 40 60 80 100 120

Fig. 4: On-state plasma distribution for a standard diode and for


a soft recovery diode. Emitter regions are neglected.

On the other hand, if the distribution in Fig. 1 is in-


a
verted, lower on the pn - side than on the nn+ - side, as
shown in Fig 4, the conditions are much better [5]. If n l
= 0,33 ⋅ n r , (8) leads to w x = 0,9 ⋅ wB , and soft recovery can
be achieved with a much thinner n--layer. This inversion of
the plasma is done in all modern soft recovery diodes as
presented by [1] - [3].
In real diodes, there are no abrupt, but diffused junc-
tions. The removal of the plasma will start earlier at the pn-
junction than at the nn+-junction, this is a advantage for the
reverse recovery behaviour. A low gradient at the nn+- b
junction may achieve that the final plasma zone remains Fig. 5: Snapp-off in the tail phase. Current (a), electrical field at
within the n+ zone as shown in Fig. 2. On the other hand, different points in time before snapp-off (b).
the assumed abrupt plasma borders are not real. Both devi-
ations may compensate, and (8) is a useful approximation As the space charge is widened, dE/dw gets lower, p
for the design. and j are decreasing. But if the space charge reaches the
end of the base, while a significant current is flowing, the
B. Tail current phase source of the current suddenly vanishes and the current will
then snap-off. The electrical field springs from a triangular
In Approximation B, it is assumed that the first phase to a trapezoidal shape.
of reverse recovery was soft, and the device has established To avoid this, the device must be capable of taking
the voltage. However, there is still plasma on the right side, the space charge at a given voltage without a bump of the
and a tail current is flowing. The tail current density j in the space charge to the n+-layer. The voltage limit, at which the
free part of the base region is carried by holes flowing with space charge reaches the n+-layer at a given background
the saturation velocity vsat(p), their density is doping ND and base-width wB, is
1 q ⋅ ND 2 4000
U sn = wB (13) static
3.3kV
2 ε
250kW/cm²
While the battery voltage Ubat is lower than Usn, no snap off
Domeij et al
will occur. This will be analysed more in detail in the paper 3000
Nagasu et al
[6] at this conference. It should be noted that (13) is the
1,7kV
same equation as applied for the punch-through of a thyris-

Voltage [V]
tor as given for example in [7] .
Tail current snap-off may occur especially in high 2000
voltage diodes (>2000V). In such applications, often low
di/dt is applied, but the circuit contains significant parasitic
inductances. In order to keep Usn high, the doping ND is not static
1000
allowed to be too low. However, this is in contradiction to
demands from cosmic ray stability. Hence, an optimal
trade-off must be found.
Approximation B has led to a design rule for wB and 0
ND, while approximation A led to a design rule for wB and 0 100 200 300 400 500
for the plasma shape. In reality, both cases must to be taken Reverse current density j [A/cm²]
into account and there are transitions.
Fig. 6: Onset of dynamic avalanche depending on the reverse
IV. REVERSE RECOVERY AND RUGGEDNESS current density
while, new diode designs are working safe far above [11].
Now, both approximations will be discussed regarding to Moderate dynamic avalanche is uncritical. Avalanche gen-
failure mechanisms. erates electrons, flowing through the space charge, there-
fore
A. Dynamic avalanche in the voltage build up phase N eff = N D + p − n av (15)
For the left side in Fig. 3, where the hole current and the hole density is compensated, hence dynamical
flows, (9) is valid for the density of holes, and (10) and avalanche is self-stabilising [12] But this holds only if there
(11) are valid too, but only until avalanche takes place. are no weak points in the diode design, especially at the
With the increase of dE/dw, E0 is increasing and may reach edge of the active area. This aspect is described in [13].
the avalanche field strength. Then dynamic avalanche, But if the current density in avalanche is increased,
caused by free carriers, occurs. the shape of the electrical field will change [14], as shown
Since the electrical field is triangular, the avalanche in Fig. 7. At 500A/cm², the field is still nearly triangular.
onset voltage may be expressed with But avalanche generation is not locally at the junction,
1 3 since carriers need a way for acceleration. Since holes are

1  8   q ⋅ N eff
4  4 flowing to the left, electrons are flowing to the right side,
U av = ⋅   ⋅   (14) the hole density increases close to the junction (a). At the
2 C   ε  5
for the ionization coefficients the approximation of Shields 3 10
und Fulop [8] with C = 1,8.10-35cm6V-7 at 25°C is used. If |E| J=1500 A/cm
2

(9) and (10) are inserted in (14), the onset voltage of ava- 2.5 10
5 a 2
J=500 A/cm
lanche depending on the current density can be calculated,
Electric field (V/cm)

as shown in Fig. 6. For ND typical values for a 1700V and a 2 10


5

3300V diode (ND = 4.3.1013cm-3 rsp. 1.7.1013cm-3) are used. b


The onset-voltage of avalanche is decreasing for a high 5
1.5 10
voltage device. For the 3.3kV diode, avalanche occurs al-
ready at 30A/cm². For 200A/cm², the onset of dynamic c
5
avalanche is already at 1100V and only a little above the 1 10
1700V device. The base doping is of second order, and free
carriers dominate. 5 10
4

Dynamic avalanche was first believed to be the failure limit


for fast recovery diodes, even a „silicon limit“ at 0
250kW/cm² was assumed. The alleged limit is also shown 0 10 20 30 40 50 60 70 80
in Fig. 6. But soon, diodes working clearly above this limit Position ( µm)
have been shown by Nagasu et al [9] and Domeij et al [10]. Fig. 7: Field shape at the pn-junction for dynamic avalanche
These operation points are also shown in Fig. 6. Mean- with increased current density.
junction holds NA, ND Epn E(w)
dE q p, n
= ( N D + p + p av ) (16) |E|
pav
nav nav
dw ε Enn+
Therefore, a higher dE/dw results as shown for j = n
p
1500A/cm². More away from the junction in region (b) p n≈p
holds n+
dE q -
= ( N D + p + p av − n av ) (17) n- +
dw ε
while pav is decreasing and nav is increasing. Threfore, in 0 wB
w
region (b) is a partial compensation, dE/dw is decreased.
The electrical field moves to an S-shaped form as shown in Fig. 8: Simplified drawing for a double-sided space charge
during strong dynamic avalanche
Fig.7 for j = 1500A/cm².
The voltage correlating to the area under E(w) is highly unstable with a time constant in the range of some
slightly lower for the higher current density. Therefore, the ns.
IV-characteristics has a region of weakly negative differen- In a more simplified way, the problem can be de-
tial resistance. Therefore, the current splits in areas with scribed as follows. If there are two electric field peaks as
lower density and in filaments with high current density.
But these filaments should not be inevitably destruc-
tive. The current density in the filaments is limited. The
temperature inside the filament will increase leading to a
locally increased avalanche voltage which counteracts the
filamentation. And the high local current density in a fila-
ment removes the stored carriers in the local region, coun-
teracting the driving force for dynamic avalanche. There-
fore, the filaments move, as can be seen in 2D-device simu-
lations, and dynamic avalanche can still be stable. But 2D-
simulations are a limited picture of reality, because fila-
ments are a 3D effect.
Up to now the right side of the plasma zone was ne-
glected. Fig. 8 shows schematically the situation at the nn+-
junction under very high dynamic avalanche. During dy-
namic avalanche, additional electrons nav are generated and
appear again in the zone between plasma and nn+-junction.
The density of negatively charged electrons can get higher a
5
than the background doping, so a space charge can be 2.5 10
formed carried by a negative charge. In this region holds
x=3200
dE q 5
= (n + n av − N D ) (18) 2 10
Electric field (V/cm)

dw ε x=6000
if the total electron density is >ND. A field peak Enn+ may 5
1.5 10
be built up, at high nav it may reach the avalanche field
strength. A high current density filament is found when the
final plasma removal occurs while there is still dynamic 1 10
5

avalanche at the pn junction [15]. This can lead to a space


charge controlled current with a high electric field and im-
pact ionisation also at the nn+ junction, i.e. a double-sided 5 10
4

dynamic avalanche. Fig. 9 shows a simulation under this


condition. This hammock-shaped electric field distribution
is similar to the field distribution at second breakdown as 0
described by Egawa in [16]. The Egawa-instability is as- 0 50 100 150 200 250 300 350 400
sumed to be destructive. In a mathematical way, this is ex- w [µ m]
plained in [15]: Applying Wachutkas model for the de- pn-junction nn+-junction
struction limit of GTOs [17] to fast recovery diodes using b
the boundary conditions in [4] leads to the conclusion, that Fig. 9: Egawa instability at strong dynamic avalanche. a) cur-
dynamic impact ionization at the pn-junction should be rent filament at the end of the plasma removal b) electric Field
stable, whereas impact ionization at the nn+-junction is inside (x=3200) and outside the filament (x = 6000).
shown in Fig. 8, there are two possibilities to find a way to V. CONCLUSION
the steady-state condition; a) if dynamic avalanche at the
junction ceases, nav vanishes and the field at the nn+- In order to avoid snap off in the first phase and to
junction is removed before the plasma is finally removed. avoid destructive conditions caused by dynamic avalanche,
b) if there is a high field at the nn+-junction, when the rules for a minimal base width and for the shape of the
plasma boarders meet, the device must pass the Egawa- plasma in the on-state have been derived. To avoid snap off
instability. The possibility of either occurring depends on in the tail phase, a rule which contains wB and background
the position of the final plasma in the diode, and the pro- doping ND is given which is similar to an equation for
cess discussed here has clear similarities to the process punch-through in thyristors. Respecting both rules can lead
which leads to snappy or soft reverse recovery behaviour as to diodes with soft recovery and high dynamic ruggedness
discussed in section III.A. Again, a minimal width of the in the voltage range equal and above 3.3kV.
space charge on the junction side is required to take the The process at the nn+-junction is of main importance
voltage before the plasma fronts meet. Based on simula- in all cases. The Egawa-instability was found as limiting
tions and proved by experimental results, in [15] a rule to mechanism for the ruggedness of fast recovery diodes. Fur-
calculate the minimum value of wB required for a non- ther design measures at the nn+-junction such as buffer
destructive reverse recovery with dynamic avalanche in structures and others are possible to improve soft recovery
dependence to the reverse voltage peak is given and dynamic ruggedness.
2 1 7
w B = 2 3 ⋅ C 6 ⋅ V pk 6 + w plas (19) REFERENCES
with wplas ≈ 100µm. Vpk is a voltage peak immediately after
IRRM, depending on di/dt and Lpar. If (19) is respected, the [1] J. Lutz, U. Scheuermann: "Advantages of the New Controlled
Axial Lifetime Diode", PCIM Proceedings (1994)
occurrence of the Egawa-instability during reverse-
[2] T. Laska, L. Lorenz, A. Mauder: „The Field Stop IGBT Con-
recovery can be avoided. cept with an Optimized Diode“, Proceedings of the 41th
PCIM, Nürnberg 2000
B. Voltage peaks by snap-off in the tail phase [3] M. Nemoto et al, Proceedings of the ISPSD, Osaka (2001)
[4] H. J. Benda, E. Spenke: “Reverse Recovery Process in Silicon
In section III.B, it was shown that snap-off of the cur- Power Rectifiers”, Proceedings of the IEEE, Vol 55 No 8
rent can occur also at a late point in the tail-phase. In this (1967)
moment, there is no avalanche in the diode, although some [5] M. Rahimo, N. Shammas, IEEE Trans. Ind. Appl., Vol 37,
dynamic avalanche might be happened before. Now the No. 2 p. 661 (2001)
[6] H.P. Felsl, E. Falck, M. Pfaffenlehner, J.Lutz: “The Influence
electric field (see fig. 5b) is far beyond the avalanche con-
of Bulk Parameters on the Switching Behavior of FWD's for
dition. But a current snap-off will cause a voltage peak, Traction Application”, Proceedings of the MIEL (2004)
and if there is snap-off from a still high tail current and if [7] Benda/Govar/Grant: Power Semiconductor Devices, New
Lpar is significant, the induced voltage peak will reach the York 1999
static avalanche voltage. [8] W. Fulop, Solid-State Electronics, Vol. 10 (1967)
Also in static avalanche, generated holes are removed [9] M. Nagasu et al, Proceedings of the PCIM 98 Japan, 175
to the pn-junction side and electrons to the nn+-junction (1998)
side. At the nn+-junction, similar to (9) a electron density [10] M. Domeij, B. Breitholtz, M. Östling, J. Lutz, Apllied Phys-
nav can be calculated. For example, a avalanche current ics Letters Vol. 74 No 21, 3170 (1999)
[11] M. Rahimo et al, to be published in Proceedings of the
density of 28A/cm² is sufficient to compensate a base dop-
ISPSD (2004)
ing ND = 1,7.1013cm-3 in a 3.3kV diode, and dE/dw at the [12] H. Schlangenotto, H. Neubrand: „Dynamischer Avalanche
nn+-junction will change its sign. Such current densities at beim Abschalten von GTO-Thyristoren und IGBTs“, Archiv
the nn+-junction can easily be reached, since in modern der Elektrotechnik 72, S. 113-123 (1989)
planar devices avalanche starts first at one point at the edge [13] J. Lutz, M. Domeij: “Dynamic avalanche and reliability of
of the device. In this case, a second voltage peak will grow high voltage diodes”, Microelectronics Reliability 43, 529-
at the nn+-junction, and a failure mechanism with a field 536 (2003)
shape similar to Fig. 9b can occur. This mechanism was [14] J. Oetjen et al: „Current filamentation in bipolar devices
originally analysed by Egawa [16] for the case of static during dynamic avalanche breakdown“, Solid State Electron-
ics 44, 117-123 (2000)
avalanche.
[15] M. Domeij, J. Lutz, D. Silber: “On the destruction limit of Si
To avoid this failure mechanism, it is necessary to fol- power diodes during reverse recovery with dynamic ava-
low the rule (13) for avoiding snap-off in the tail phase. lanche”, IEEE Trans.Electr.Dev. Vol. No 2, 486-493 (2003)
Increased n-doping before the nn+-junction - as possible by [16] H. Egawa, IEEE Transactions on Electron Devices, Vol. ED-
buffer layers or by a shallow gradient of the doping at the 13, No. 11, p. 754 (1966)
nn+-junction - promise to enlarge the stability against ava- [17]G. K. Wachutka, IEEE Transactions on Electron Devices,
lanche. However, this must be further investigated. Vol. 38, p. 1516 (1991)

You might also like