3, MARCH 2010
the thicknesses of the SJ and SPT IGBTs are 100 and 120 μm,
respectively. The SJ IGBT [Fig. 1(a)] structure is very similar
to the SJ MOSFET structure, i.e., the drift layer is made
of alternating p- and n-doped pillars. As with the FS IGBT
structure, the anode side of this device comprises a transparent
p-layer and a thin lightly doped n-buffer layer (Fig. 1).
In this study, we vary the doping concentration of the n- and
p-pillars of the SJ IGBT from 5 × 1013 to 1 × 1016 cm−3 . As
expected, the SJ effect allows the doping concentration of the
n- and p-pillars to be significantly increased compared to the
doping level of the drift region of a conventional IGBT. Indeed,
given that the n- and p-pillars are very thin, their maximum
doping that still allows the device to attain a breakdown voltage
Fig. 2. Electron flow lines while the SJ IGBT operates in the ON-state
of 1.2 kV is approximately 1 × 1016 cm−3 . In order to achieve a (I = 100 A/cm2 ). (a) High (1 × 1016 ), (b) medium (3 × 1015 ), and (c) low
completely flat electric field distribution across the drift region (1 × 1014 ) n- and p-pillar doping levels.
before avalanche occurs, (Ec ≈ 3 × 105 V · cm−1 ), the width
(half-cell pitch) of the device in this case is taken to be 3 μm,
with each pillar half-width of 1.5 μm. Fig. 1(a) shows the
FS IGBT structure that we have simulated. The n-drift region
doping concentration is 5 × 1013 cm−3 .
III. O FF -S TATE
In the blocking mode, the electric field distribution of the
FS IGBT has a trapezoidal shape, decreasing from the cathode
to the anode with no variation in the x-direction. In the case
of the SJ IGBT, the electric field is flat across the drift region
with a small increase around the reversed biased junction in
the middle of the device. It is therefore easy to see that the Fig. 3. Hole flow lines of the SJIGBT operates in the ON-state
breakdown voltage (which is the area under the electric field (I = 100 A/cm2 ). (a) High (1 × 1016 ), (b) medium (1 × 1015 ), and (c) low
curve when the maximum electric field is equal to the avalanche (1 × 1014 ) n- and p-pillar doping levels.
field) for the SJ IGBT is always higher than that of the FS IGBT
current being provided by the n-drift region. Therefore, the FS
for the same device dimensions. Therefore, in order to match
IGBT drift layer is fully modulated with plasma concentrations
the breakdown voltage of the two devices, we use a thicker drift
being well above the background doping.
region for the FS IGBT, i.e., the thickness of the FS IGBT is
For the SJ IGBT, in the ON-state, the p-emitter, n-buffer,
120 μm, whereas that of the SJ IGBT is 100 μm.
and p-pillar form a parallel network of narrow- to wide-base
The breakdown voltage of the SJ structures shows significant
p-n-p transistor structures depending on the pillar doping. The
dependence on the net charge balance of the drift region pillars
p-pillar provides a resistive contact to the collector (cathode).
[13]. The maximum voltage can be achieved if the charge of the
This is, in fact, the basis of the model that we developed and
pillars cancels each other perfectly.
will demonstrate later in this paper.
IV. F ORWARD C ONDUCTION
A. Pillar Doping Impact
In the ON-state, the behavior of the two devices is very
different. The doping charge level in the n- and p-pillars of In this section, we investigate and explain the effect of
the SJ IGBT changes the ratio between bipolar and unipolar varying the pillar doping concentration on the ON-state and
conduction within the drift region and, therefore, alters very switching performance. The modes of operation lie in three
significantly the speed and ON-state performance of the device. categories.
This is a unique effect that is uncharacteristic to any other power For high doping levels of the n- and p-pillars (above 3 ×
devices known in the field. 1015 cm−3 ), the unipolar effect across the drift region is more
When the gate–source voltage exceeds the threshold, an pronounced with a low IR drop and faster removal of the
inversion layer forms beneath the gate of the MOS structure. charge from the bottom side due to low p-pillar (collector) resis-
When VA is greater than approximately 0.7 V electron current tance. Fig. 2(a) shows the electron flow lines while the SJ IGBT
flows through the inversion layer into the base (i.e., n-drift (high n- and p-pillar doping levels) operates in the ON-state
region) of the p-n-p transistor, which, in turn, causes substantial (I = 100 A/cm2 ). A similar picture is observed for the holes
hole injection from the p+ anode contact layer into the n-drift traveling through the p-pillar of the drift region [Fig. 3(a)]. It is
region. The FS IGBT features a single (wide-base) p-n-p tran- obvious that the conduction of current is mainly unipolar, with
sistor with a short collector path. The p-emitter, n-drift, and the pillars primarily conducting majority carriers (holes in the
p-well form a wide-base p-n-p transistor structure with the base p-pillars and electrons in the n-pillars). Fig. 4(a) and (b) shows
596 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 3, MARCH 2010
Fig. 4. Minority carrier distribution along cross-cuts in the p- and n-pillars for
the SJIGBT medium (3 × 1015 ) n- and p-pillar doping levels.
Fig. 5. Minority carrier distribution switching off along the medium p-pillar
(1 × 1015 ) of the device.
the minority carrier distribution along cross-cuts in the p- and
n-pillars. Therefore, we confirm that current conduction is
mainly unipolar as the minority carrier concentrations are
very low.
On the opposite, if the n- and p-pillar doping levels are in the
range of 3 × 1014 to 3 × 1015 cm−3 , the current conduction is
bipolar in nature at the anode of the device turning into unipolar
at the cathode side. The extension of the plasma in the drift
region toward the anode depends on the pillar doping. Figs. 2(b)
and 3(b) show the electron and hole flow lines, respectively,
while the SJ IGBT operates in the ON-state (I = 100 A/cm2 ).
The unipolar effect results in an increase in the ON-state voltage
drop, due to high IR drops at the cathode side. Therefore,
for the same ON-state voltage drop as the equivalent FS IGBT, Fig. 6. Normalized peak charge at the anode side against switching off time
the plasma must now be increased and, therefore, the longer for the FS and the SJ IGBT 3 × 1015 n- and p-pillars doping levels.
the time needed to turn off the device. Fig. 4(a) and (b) also
shows the minority carrier distribution along cross-cuts in the
p- and n-pillars for medium doping. Therefore, we confirm
that current conduction is unipolar at the cathode side of the
device as the minority carrier concentrations are very low,
whereas the current conduction turns bipolar at the anode side
as the minority carrier concentration becomes higher than the
background doping.
At low doping levels (below 3 × 1014 ), the effect of the n-p
junctions in the drift region tends to fade away, and the bipolar
conduction tends to dominate the whole of the drift region and
the device behaves similarly to an FS IGBT. Figs. 2(c) and 3(c)
show the electron and hole flow lines while the SJ IGBT
operates in the ON-state (I = 100 A/cm2 ). Fig. 4(a) and (b)
shows the minority carrier distribution along cross-cuts in the
p- and n-pillars for low doping. Therefore, we confirm that
current conduction is bipolar throughout the drift region as the
minority carriers concentration becomes orders of magnitude
higher than the background doping.
In switching conditions (simulations), both devices (SJ IGBT
and FS IGBT) switch from an ON-state current density of
100 A/cm2 . Fig. 5 shows how the plasma in a 1 × Fig. 7. Static and dynamic characteristics of the SJ and the FS IGBT
as a function of the doping levels in the drift region for T = 300 K.
1015 atoms/cm3 pillar doping is removed from the drift region (a) ON-state voltage drop (for constant Eoff = 5 mJ/cm2 ). (b) Turn-off energy
after the gate signal is removed. Hence, for the device to turn losses (for constant Von = 1.2 V). (c) Degradation in breakdown for a 5%
off, the excess carriers need to be removed. Fig. 6 shows how charge imbalance.
the normalized peak charge (located at the anode side; see
Fig. 5) varies as a function of the turn-off time for the FS and Fig. 7(a) and (b) shows the ON-state voltage drop for the
the SJ IGBT 3 × 1015 n- and p-pillar doping levels. Hence, the same turn-off energy losses (which is equal to 5 mJ/cm2 ) and
reduced plasma in the drift and the p- and n-pillar junction helps the turn-off energy losses for the same ON-state voltage drop
reduce the plasma much faster. (which is equal to 1.2 V), respectively, as a function of the
ANTONIOU et al.: SUPERJUNCTION INSULATED GATE BIPOLAR TRANSISTOR OPTIMIZATION AND MODELING 597
pillar doping of the SJ IGBT. Similarly, Fig. 7(c) shows the can clearly be seen that for these operating conditions (I =
breakdown voltage degradation (for 5% charge imbalance) as 100 A/cm2 ), the SJ MOSFET performs worse than the SJ
the pillar doping increases. On the same graphs [Fig. 7(a) IGBT due to the unipolar conduction of current in the drift
and (b)], we plotted the corresponding values of the FS IGBT region [8]. Fig. 9 also shows the technology curve of the
with thicknesses of 100 and 120 μm (n-drift layer doping optimized SJ IGBT−ycell = 120 μm. This latter device length
5 × 1013 cm−3 ). This is done in order to investigate the impact achieves the 1.2-kV breakdown voltage despite the breakdown
that this difference in length has on the switching losses and on degradation at high pillar doping levels, and at the same time,
the ON-state voltage drop. As a result, at drift layer low doping, the switch-off energy losses are still considerably lower than the
where the SJIGBT behaves in a bipolar current conduction that of the FS IGBT (at the same ON-state voltage level equal
fashion, the reduction in power losses is due to the reduced to 1.2 V).
length, whereas, the ON state voltage drop remains almost equal Note that the optimum SJ IGBT doping regions are identified
to that of the FSIGBT. for a given width of the n- and p-pillars. Changing the width
Therefore, we can specify two regions where the SJ IGBT will alter the absolute values of these regions, but we expect
can operate with better performance than the FS IGBT. a similar behavior. Nevertheless, to take full advantage of the
1) At drift layer pillar doping levels of 6 × 1013 to 2 × SJ effect, the width of these pillars needs to be kept to a
1014 cm−3 , where the charge imbalance effects are negli- minimum—a limit imposed by the fabrication technology.
gible and the reduction in switching losses at Von = 1.2 V This paper does not aim at addressing fabrication issues. For
are about 50%. The SJ IGBT behaves as an FS IGBT with the current status of SJ technology, one can refer to [14].
a reduced cell thickness due to the SJ technology present
in the drift region to almost flat. Therefore, for the same V. T EMPERATURE I MPACT
Von = 1.2 V, the SJ IGBT needs less plasma in the drift
The performance of the SJ IGBT is also examined under
region so the switching losses are much lower than the
an increased operating temperature. In order to assess its per-
FS IGBT. The presence of the p- and n-pillar junction in
formance, we compared it to an FS IGBT under the same
the drift region also helps collect the plasma much faster.
operating conditions. Fig. 10 (similarly to Fig. 7) shows the SJ
2) At drift layer pillar doping levels of 3 × 1015 to 5 ×
IGBT switching behavior for an increased operating tempera-
1015 cm−3 , where the unipolar conduction of the current
ture equal to 400 K. The increase in the ON-state voltage drop
is dominant and both the switching losses and ON-state
(for the same energy losses equal to 5 mJ/cm2 ) for the SJ and
voltage are reduced. The charge imbalance effects are
FS IGBTs are 25.3% and 9.4%, respectively. This is expected
present, but the voltage degradation is still less than 20%.
as the ON-state unipolar transport is affected due to mobility
At the second optimum region, the background doping
degradation while the bipolar gain in the drift remains almost
helps the unipolar conduction to become more efficient
unchanged with the temperature (as the ambipolar mobility
at the top of the device. Hence, although the plasma is
degradation in temperature is somewhat canceled out by the
confined at the anode side [as shown in Fig. 5(a) and (b)],
lifetime increase in temperature).
the reduced resistance of the pillars compensates, leading
to lower ON state losses. Hence, the switching losses are
also reduced as the plasma concentration is low and the VI. M ODEL S TRUCTURE
conduction of majority carriers is dominant in the device. A well-established model for the standard IGBT is an in-
Furthermore, we also consider the behavior of the SJ trinsic MOSFET in series with a wide-base p-n-p bipolar
MOSFET with a pillar structure similar to that of the SJ IGBT junction transistor (BJT) [Fig. 11(a)]. Improvements to this
and a pillar doping concentration equal to 5 × 1015 cm−3 . model include the introduction of a JFET connected in series
Figs. 8 and 9 show the ON-state output characteristic and the to the MOSFET. This component models the neck region (with
energy losses of such a device, respectively. The corresponding adjacent depletion regions to the p-wells), which forms at the
optimized SJ IGBT results are plotted on the same graphs. It cathode side of the device in the drift region. The industry
598 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 3, MARCH 2010
Fig. 12. Complete PSPICE model for an SJ IGBT. Added components to the
Kraus model (narrow-base p-n-p BJT and JFET) are included in the dashed
boxes.
Fig. 13. Simulated (DESSIS) output characteristics for the SJ IGBT and the Fig. 14. Simulated (DESSIS) output characteristics for the SJ IGBT (pillars
PSPICE-based Kraus and power BJT models. doping equal to 3 × 1015 atoms/cm3 ) and the PSPICE-based complete model
for different Vg ’s.