Robust reverse bias safe operating area and improved electrical performance in 3300 V non-proportionally scaled insulated gate bipolar transistors

Robustness under high-temperature clamped inductive turn-off has been compared systematically among 3300 V scaled insulated gate bipolar transistors (IGBTs) with scaling factor k from 1 to 10 by technology computer-aided design simulations. Degradation of the reverse bias safe operating area (RBSOA) has been observed in proportionally scaled IGBTs, especially at a high turn-off current density. A non-proportional scaling method has been proven to be able to restore the robustness degradation with RBSOA close to the original k = 1 case. Moreover, the method for adjusting R pf (P-floating connecting resistor) adds more flexibility to device design and also improves the overall electrical performance of non-proportionally scaled IGBTs. The adjustment of R pf has been found to have a minimal effect on the RBSOA.


Introduction
The insulated gate bipolar transistor (IGBT) scaling principle was proposed for upgrading device performance. 1,2)][5] Besides a better trade-off relationship, an improved injection enhancement (IE) effect and a superior switching speed have also been demonstrated in scaled IGBTs that benefit from suppressed dynamic avalanche. 6)More aspects of scaled IGBTs need to be studied, including device robustness under various conditions, before real applications can be developed. 7,8)evice robustness under extreme application conditions has always been an important topic of power device research.][21][22] Deep failure analysis has been performed using both simulation and experiments to classify different failure modes. 23,24)n the other hand, the influence of the current filament effect on device stability and reliability has become a focus of research since it was first proposed in 2002. 25)It has been reported that the formation of a current filament usually causes a huge degradation of turn-off ruggedness because of current localization at several spots. 26)][29][30] It is believed that the gap in device robustness between experimental data and simulations is related to the current filament phenomenon.
In this paper, a comparison of RBSOA among scaled trench IGBTs has revealed the design concern behind the proportional scaling concept.We emphasize that the analytical methodology is based on acknowledgment of the gap between TCAD simulation, where current flows uniformly, and real devices, where current filaments appear.Despite that, the main purpose of this work is to study the change in device robustness in scaled IGBTs and also work out some design methods to compensate for the negative change.This paper is an extended version of a conference abstract. 31)ompared with the original abstract, a more comprehensive comparison of RBSOA and the E off -V ce,sat trade-off relationship for non-proportional scaling methods compared with original ones has been conducted.In addition, benefits for the electrical performance and also the influence of the R pf adjustment method on the RBSOA have been analyzed in detail.

Analysis method
A full proportional scaling method was applied to the top cell region in a 3300 V IGBT with scaling factor k = 1, 3, 5, 10.Structural and electrical parameters of reference (k = 1) and scaled (k = 3, 5, and 10) IGBTs are labeled and summarized in Fig. 1 and Table I.Corresponding half-cell device structures were generated with a TCAD simulation tool (note that the P-floating region is connected to the ground through an adjustable resistor R pf ).A typical inductive load circuit with a bus voltage of 2500 V was used to evaluate turn-off robustness with a relatively high on-state current density under the assumption that uniform current flowed through the device during turn-off.In other words, the halfcell structure in the simulation can be regarded as the cells in which the current filament is localized in a real device.Therefore, the current density of RBSOA in this work is about one order of magnitude higher than in the specification of real devices.In this study, overcurrent turn-off behaviors are simulated with a thermodynamic model.Other important physical models applied in the TCAD simulation include a doping-dependent mobility model, a carrier-carrier scattering model, a high-field saturation model, doping-dependent and temperature-dependent Shockley-Read-Hall recombination models with field enhancement and an impact ionization model (UniBo2).
We then introduce the critical temperature approach before comparing the RBSOA.Based on electrothermal simulation, device robustness under different on-state current densities (J on ) and ambient temperature (T amb ) conditions is compared.With certain J on , T amb is increased gradually until device failure occurs.For each fixed on-state current density J on , the lowest T amb at which failure happens is defined as the critical temperature (T Critical ).The value of T Critical changes as a function of J on .Then the T Critical values are plotted as a function of on-state current density J on to represent the RBSOA.

RBSOA degradation of proportionally scaled IGBTs
To compare RBSOA among proportionally scaled IGBTs, T Critical is plotted as a function of J on for k = 1, 3, 5, and 10 IGBTs in Fig. 2(a).Obviously, RBSOA degradation is observed in scaled IGBTs since the boundary of T Critical becomes smaller with a higher scaling factor k. The degradation is more prominent under high on-state current density conditions.For example, take J on = 6000 A cm −2 in Fig. 2(b): T Critical has dropped from 378 K to 348 K (30 K less), with an increase in scaling factor from 1 to 10.
To show more details behind the drop in T Critical , the switching waveforms for k = 1, 3, 5, and 10 IGBTs under identical conditions without device failure are compared in Fig. 3, where T amb is lower than T Critical .Note that the left axis is on a log scale to emphasize the difference in leakage current density after turn-off.As well as V ce and I c curves, the maximum temperature (T max ) inside the device is also added to Fig. 3.Even though the total power dissipation is expected to be lower in scaled IGBTs, the increase in T max during turnoff is higher.This may stem from the stronger localized thermal dissipation from the higher current density at the mesa region in scaled IGBTs.Moreover, because of the higher local temperature in the neighborhood of the reverse biased p-n junction, the leakage current after turn-off also becomes higher with scaling factor k. From these two    differences under identical switching conditions, it is natural to believe that the proportional scaling method weakens device robustness and makes device failure more likely.Deeper analysis of the failure mechanism in scaled IGBTs is necessary to work out the design aspects of some solutions.A typical turn-off waveform of (a) the turn-off period and (b) the failure period for a k = 1 IGBT under J on = 6000 A cm −2 , T amb = 378 K is shown in Fig. 4, and is a typical thermal runaway failure mode.The electron and hole current parts of the emitter electrode are also included for better illustration.After turn-off, a high leakage current flows through the device with the major part being hole current.We speculate that both thermal generation from the high temperature in the N-base region and impact ionization from the high electric field under reverse bias contribute to the high leakage current.For the particular case shown in Fig. 4, delayed thermal runaway failure occurs with a sharp rise in leakage current and temperature after about 176 μs.Finally, the trigger of the embedded parasitic NPN transistor leads to latch-up and destructive failure, as maked by the vertical dashed line in Fig. 4(b).The IGBT fails to sustain the switching voltage after latch-up occurs.Another important fact is that the electron current part increases exponentially before latch-up.This enhanced electron injection is related to the action of the embedded parasitic NPN transistor.
As further evidence, the time evolution of physical parameters during the failure period of Fig. 4(b) is presented in Fig. 5.In Fig. 5(a), transient carriers alter the electric field profile and gradually enhance impact ionization at the location close to the trench bottom region before failure.In

RBSOA restoration of non-proportionally scaled IGBTs
Our results indicate that the full scaling method assumed so far is not practical in terms of device robustness requirements.Here, we discuss some measures that restore the degradation of robustness.We speculate that the higher current gain of the embedded NPN transistor of scaled IGBTs is the cause of the robustness degradation.To improve robustness of the k = 10 IGBT, four different non-proportional scaling designs were tried, as shown in Table II.
Basically, different dimensions of the k = 10 IGBT are enlarged by a factor β to suppress NPN transistor current gain α npn .The corresponding schematic cross-sections of these modified non-proportionally scaled and original proportionally scaled k = 10 devices are shown in Fig. 6.
To check the restoration effect of these specialized design methods, corresponding T Critical values as function of the enlarging factor β are plotted in Fig. 7(a).In type A, the mesa width is enlarged to reduce the current density in the mesa region.As the enlarging factor β becomes higher, the T Critical value improves first but then drops.The possible reason is that the wider mesa width exposes the p-n junction to a higher electric field, so the higher impact ionization rate even leads to further robustness degradation.
As for type B, the trench extrusion distance is increased to keep the p-n junction and P-base region away from the trench bottom with its strong impact ionization and high temperature.First, the electric field at the p-n junction is reduced because the deeper trench acts as a more effective field plate, which is helpful for mitigating the leakage current.The longer distance also reduces the temperature of the P-base region and avoids the enhancement of current gain α npn from the longer carrier lifetime in the P-base region.So the type B method is quite effective in improving the T Critical value.
The situation is similar in type C, where the depth of the the P-base region is enlarged.From the aspect of a NPN transistor, the wider P-base region suppresses current gain since the wider base region has a lower base transport factor.Also the hole current density in the P-base region and parasitic resistance become lower, which both contribute to

R pf adjustment in non-proportionally scaled IGBTs
In addition to the comparison of RBSOA, the justification for better electrical performance is also demanded for the nonproportional scaling methods.First of all, it has been confirmed that different non-proportional scaling methods have a minimal effect on breakdown voltage.The trade-off relationships of non-proportionally and proportionally scaled IGBTs are compared in Fig. 8.The turn-off power dissipation E off is compared as function of on-state voltage drop V ce,sat by varying the doping concentration of the P-collector region under normal operating conditions with an on-state current density J on = 80 A cm −2 .
Thanks to the much stronger IE effect that originates from the higher aspect ratio of the mesa region, both type B (k = 10, β = 5) and type B + C (k = 10, β = 5) devices enjoy an even better E off -V ce,sat trade-off relationship than the original k = 10 device.Nevertheless, the higher carrier density at the cathode side from the stronger IE effect impairs the switching speed and also causes dynamic avalanche to occur more easily.It would be ideal to be able to regulate the IE effect from the requirement of device design flexibility while maintaining a decent enough RBSOA in non-proportionally scaled IGBTs.An effective method is to adjust the p-floating resistor R pf .The following discussion about the adjustment of R pf is based on a baseline P-collector doping concentration of 8.0 × 10 17 cm −3 .
To demonstrate overall performance improvement in non-proportional scaling methods, six different R pf values (R 1 > R 2 > R 3 > R 4 > R 5 > R 6 ) are chosen to compare the electrical performance, as shown in Fig. 9.In Fig. 9(a), the turn-off power dissipation E off becomes lower, and the onstate voltage drop V ce,sat increases due to the weaker IE effect.Compared with the two dashed lines which mark the E off and V ce,sat values of the k = 1 baseline device, the green area represents the R pf range for which the type B + C (k = 10, β = 5) device has both better E off and V ce,sat .In addition, the dV/dt Critical value that represents the controllable range of dV/dt during turn-off is demonstrated in Fig. 9(b).Benefits from the better uniformity of the electric field profile in type B + C (k = 10, β = 5) devices include a wide range of possible R pf values to yield both a faster switching speed and lower on-state voltage drop in non-proportional scaling devices.Up to now, the essential design flexibility from R pf adjustment has been proved to be capable of enhancing the overall electrical performance.
From another perspective, adjustment of the R pf value might have a negative effect on the E off -V ce,sat trade-off relationship as shown in Fig. 10.In the type B + C (k = 10, β = 5) device, degradation of the E off -V ce,sat trade-off relationship is observed if R pf gets lower.Originally, the better trade-off relationship comes from the higher carrier density at the emitter side under the on-state (the IE effect).Thanks to the better IE effect, the carrier density at the collector becomes lower under the same V ce,sat condition, which yields a smaller collector current tail, and thus a lower  energy loss, during turn-off under the same switching conditions.On the contrary, the adjustment of R pf mitigating the IE effect leads to a higher carrier density at the collector side, thus generating a bigger collector current tail and a higher turn-off energy loss.According to the simulation results, a type B + C (k = 10, β = 5) device still enjoys a better trade-off relationship compared with the original k = 1 case as long as the value of R pf is larger than R 5 .
Furthermore, we need to discuss the influence on RBSOA of adjusting R pf .A comparison of RBSOA of a type B + C (k = 10, β = 5) IGBT with various R pf values is given in Fig. 11.It is surprising that the change in R pf has almost no effect on T Critical , regardless of the on-state current density.To recover the physical mechanism, the carrier density along the channel region is plotted under on-state currents of 80 A cm −2 and 6000 A cm −2 in Figs.12(a) and 12(b).Under the lower on-state current condition, the carrier density at the cathode side is easily altered by the R pf value, whereas this difference disappears when the overall carrier density is elevated to much a higher value.As a consequence of the identical carrier density, no difference in power dissipation occurs during turn-off.As a result, no change in device robustness can be observed under extreme high onstate current conditions.Hence we can conclude that the R pf adjustment approach has almost zero effect on the RBSOA.

Conclusions
Established from the assumption of uniform current flow, considerable degradation of RBSOA has been observed in proportionally scaled IGBTs from TCAD simulation results.After comprehensive analysis of the physical mechanism behind the failure mode, the degradation is concluded to originate from the higher current gain of the parasitic NPN bipolar transistor.Certain non-proportional scaling methods have been demonstrated to be able to restore RBSOA degradation in scaled IGBTs while maintaining good static and dynamic characteristics.In non-proportionally scaled IGBTs, the adjustment of R pf has been considered to be an indispensable way to meet the demands of device design.The favorable range of R pf values that generates better electrical performance than traditional ones has been discussed.In the end, the influence from different R pf values on RBSOA has been found to be neglectable.

Fig. 2 .
Fig. 2. (a) Critical temperature (T Critical ) as function of on-state current density (J on ) of k = 1, 3, 5, and 10 IGBTs and (b) the degradation of T Critical as a function of scaling factor k under J on = 6000 A cm −2 .

Fig. 3 .
Fig. 3. Comparison of leakage current density in I c after turn-off and T max change of k = 1, 3, 5 and 10 IGBTs under J on = 6000 A cm −2 , T amb = 347 K, V cc = 2500 V and the gate resistance Rg = 0.1 Ω. V ce is also shown.

Fig. 4 .
Fig. 4. (a) Typical turn-off waveforms and (b) failure waveforms of a k = 1 IGBT at J on = 6000 A cm −2 , T amb = 378 K.Note that the time range is different in (a) and (b).

Fig. 5 .
Fig. 5. Time evolution of (a) impact ionization rate, (b) temperature, (c) hole current density and (d) electron current density in cross-sections of a k = 1 IGBT corresponding to the waveforms in Fig. 4(b).

Fig. 5 (
Fig.5(b), positive feedback among electric field, impact ionization, carrier density and temperature moves the hot spot towards the emitter side.Moreover, the higher hole current density and temperature rise in the P-base region enhance electron injection from the NPN bipolar transistor, corresponding to the exponential rise of I E-e shown in Fig.4(b).The gradually enhanced behavior of the NPN transistor eventually leads to latch-up and destructive failure, which is clearly observed in Fig.5(d).

Fig. 8 .
Fig. 8.Comparison of turn-off power dissipation (E off ) versus on-state voltage drop (V ce,sat ) under T amb = 300 K, J on = 80 A cm −2 for different P-collector doping concentrations.

Fig. 9 .
Fig. 9. Comparison of (a) E off and V ce,sat and (b) dV/dt Critcal and V ce,sat for different R pf values.

Fig. 11 .
Fig. 11.Comparison of T Critical as function of on-state current density (J on ) for k = 10 (type B + C, β = 5) IGBTs for different R pf values.

Fig. 12 . 6 ©
Fig. 12.Comparison of the carrier density profile for different R pf values under an on-state current density (J on ) of (a) 80 A cm −2 and (b) 6000 A cm −2 .

Table I .
Structural and electrical parameters of original and scaled trench gate IGBTs.

Table II .
The change of structural dimensions in different types of non-proportional scaling methods.