Investigation of turn-on performance in 1.2 kV MOS-bipolar devices

In this paper, the turn-on characteristics of the 1.2 kV Trench IGBT (TIGBT) and the 1.2 kV Trench Clustered IGBT (TCIGBT) are investigated through TCAD simulations and experiments. The TCIGBT shows much lower turn-on energy loss (E on) due to higher current gain than an equivalent TIGBT and the negative gate capacitance effect is effectively suppressed in the TCIGBT by its self-clamping feature and PMOS action. In addition, the impact of 3D scaling rules on the turn-on performance of TIGBT and TCIGBT is analyzed in detail. Simulation results show that scaling rules result in a significant reduction of E on in both TIGBT and TCIGBT. Furthermore, the experimental results indicate that TCIGBT technology is well suited for high current density operations with low power losses. Compared to the state-of-the-art IGBT technology, an 18% reduction of total power loss can be achieved by the TCIGBT operated at 300 A cm−2 and 175 °C.


Introduction
The trends in the development of silicon MOS-bipolar devices have always been devoted to continuous increases in power density as well as power efficiency. Remarkable efforts have been made to improve the turn-off energy loss (E off ) versus onstate voltage drop (V ce(sat) ) trade-off. For example, the 3D scaling concepts on trench IGBT (TIGBT) [1][2][3] and trench clustered IGBT (TCIGBT) 4) have resulted in significant improvements in E off -V ce(sat) trade-off. Due to the enhanced thyristor effect, the scaled TCIGBTs show even lower V ce(sat) than the scaled TIGBTs. 4) Recently, it was revealed that the dynamic avalanche (DA) phenomenon posed fundamental limits on the operating current density, turn-off energy loss, dV/dt controllability and long-term reliability of TIGBTs. [5][6][7][8] In comparison, the TCIGBTs have been experimentally evaluated to show DA free behavior and low power losses at high current density operations. [9][10][11] In addition to the onstate and turn-off behavior, the turn-on performance is also important for MOS-gated bipolar devices to achieve high switching frequency operations with low power losses.
To lower the turn-on power dissipation (E on ) of TIGBTs, one common method is to reduce the gate resistance (R g ) to speed up the switching transients. However, the reduction of R g is limited by the turn-off behavior because small R g can result in additional turn-off energy losses due to DA. 6,9) Another method to lower the E on is to improve the turn-on dV ce /dt by reducing the Miller capacitance (C gc ), such as side-gate design, 12) split-gate design 13) and adoption of emitter dummy trenches. 14,15) However, such designs may enter into "self-controlled" mode (the turn-off surge voltage is independent of R g ) if the C gc is too low, 16) which limits the applications of TIGBTs. Furthermore, reducing C gc also increases the turn-off dV/dt, which enhances the DA effect during turn-off and short-circuit conditions. 6,17) Therefore, improving the turn-on performance of TIGBTs in terms of low E on and high dV/dt controllability by reducing C gc has to be compromised with their turn-off performance. Furthermore, the negative gate capacitance (NGC) effect during the turn-on transients of IGBTs cause strong oscillations under short-circuit conditions 18) and degrade the turn-on dI c /dt controllability, resulting in EMI noise in the applications. 19) To reduce or eliminate the EMI noise emission from IGBT modules, the NGC effect must be eliminated or suppressed without sacrificing other electrical characteristics. IGBT structures with P-float resistors 19,20) or hole path structure 21) can divert some of the holes and suppress the NGC effect, but at the cost of increased V ce(sat) due to weakened injection enhancement (IE) effect. Other IGBT cathode designs, such as side-gate design 12) and separate P-float design, 22) can also suppress the NGC effect by lowering the potential differences across the gate oxide. However, the impact of such designs on the turn-off behavior is not well studied.
In the previous work, 23) the turn-on behavior of 1.2 kV TIGBT and TCIGBT are studied via 3D TCAD simulations. 24) In this work, the superior turn-on performance of the 1.2 kV TCIGBT is validated through experiments. The TCIGBT shows significant improvement of E on due to the inherent thyristor actions. The negative gate capacitance effect is effectively suppressed by the self-clamping feature of TCIGBT and effective PMOS action. Furthermore, the impact of 3D scaling rules on the turn-on performance of TIGBT and TCIGBT is analyzed in detail. Finally, the measured total power losses of TCIGBT are compared to the state-of-the-art IGBT technology.  Table I, are kept identical herein to compare the turn-on performance. A double pulse inductive circuit as shown in  compares the turn-on waveforms at rated current I c = 100 A and identical R g condition. It can be seen that TCIGBT shows much higher turn-on dI c /dt and dV ce /dt than TIGBT. This is because, the thyristor structure within TCIGBT exhibits a higher current gain than the BJT structure within TIGBT, as expressed in Eqs. (1)-(4), where (1) and (2) are the turn-on dI c /dt and dV ce /dt of TIGBT, (3) and (4) are the turn-on dI c /dt and dV ce /dt of TCIGBT, g m is the transconductance of the MOSFET structure, V g_in is the gate input voltage, C gc is the gate-collector capacitance, and C ge is the gate-emitter capacitance.

Turn-on performance of TIGBT and TCIGBT
The detailed derivation of the turn-on dI c /dt and dV ce /dt of TCIGBT can be explained as follows: the equivalent circuit model of the CIGBT 25) is essentially the use of a MOS gate structure to control a thyristor structure, which consists of a PNP transistor (P-anode/N-drift/P-well) and an NPN transistor (N-drift/P-well/N-well). During forward conduction, the MOS current serves as the base current of the PNP transistor, while the collector current of the PNP transistor serves as the base current of the NPN transistor. Therefore, the total collector/anode current follows a relationship of where a PNP is the current gain of the PNP transistor (Panode/N-drift/P-well), and a NPN is the current gain of the NPN transistor (N-drift/P-well/N-well). By differentiating both sides of Eq. (5), it produces: During the current rise phase of turn-on transient, the collector/anode voltage remains constant, and the gate current    During the collector/anode voltage fall phase, the gate voltage is fixed at the Miller plateau voltage (V plateau ). Thus, almost all the gate current is used to charge up the C gc at this stage If the FWD is an ideal diode with zero reverse recovery, the collector/anode current can be expressed as: Replacing the component of V plateau in Eq. (8) with Eq. (9), the fall rate of the collector/anode voltage (dV ce /dt) can be expressed as Eq. (4). Therefore, TCIGBT shows much lower E on than the TIGBT due to higher turn-on dI c /dt as well as higher dV ce /dt. As the FWD in the case of TCIGBT experiences a higher dV KA /dt as a result of the higher dV ce /dt, a higher surge I c is induced due to the larger reverse recovery current from the FWD. The high surge currents of TCIGBT can be effectively suppressed by replacing the silicon PiN diodes with the silicon carbide (SiC) Schottky diodes, which will be evidenced in the later analysis.   26) and a SiC diode 27) are used as FWD to compare the turn-on performance. It can be seen that the SiC diode reduces the surge current and improves the dV ce /dt in the turn-on transients. Moreover, compared to a state-of-theart TIGBT device (E on = 73 μJ A −1 at T j = 25°C), 28) the TCIGBT with silicon diode shows a 73% reduction of E on while the TCIGBT with SiC diode shows an 84% reduction of E on . Therefore, the TCIGBT technology provides a superior solution for improving the turn-on power dissipation of MOS-bipolar power devices.
To verify the consistency between the simulation results and measurement results, Fig. 6 compares the measured turnon waveforms and the simulated turn-on waveforms of a calibrated TCIGBT model. It can be seen that the simulated waveforms are close to the measured results.
In addition to the low E on performance, the TCIGBT design can suppress the NGC effect in the turn-on transients due to the self-clamping feature. The physics of the NGC effect can be explained as follows. In the conventional TIGBT, the hole current flows into the P-float region and evacuates along the trench sidewall. The P-float potential (V PF ), as shown in Fig. 7, is therefore raised during turn-on. If the voltage difference across the gate oxide (V PF -V ge ) is high, a reverse displacement current will be induced to charge the gate capacitance to a negative value, which has a significant impact on the dV ce /dt and dI c /dt controllability. 19) Figure 8 compares the V PF and the V ge of TIGBT and TCIGBT during turn-on transients. It can be seen that there is a significant potential difference between V PF and V ge in the TIGBT, which is caused by the hole current flow within the P-float region. However, in the case of TCIGBT, the floating P-base potential is clamped under a low value due to the selfclamping feature and the N-well acts as a barrier for the holes. Therefore, the hole current is evacuated through the mesa region directly via the PMOS structure formed by the p + -cathode/N-well/P-well, as shown in Fig. 7. Hence, the potential difference across the gate oxide is marginal and the NGC effect is therefore successfully suppressed. As shown, the gate charge (Q g ) is scaled in k3-TIGBT due to the scaled trench gates. However, the NGC effect is enhanced in the k3-TIGBT due to the IE effect. Figures 11 and 12 show the dV ce /dt and the dI c /dt of TIGBTs and TCIGBTs, respectively. It can be seen that the dV ce /dt and the dI c /dt of both k3-TIGBT and k3-TCIGBT are increased compared to that of conventional structures. Hence, the scaling rules result in a significant reduction of E on in both TIGBT and TCIGBT, as shown in Fig. 13. Due to the enhanced thyristor effect, 4) k3-TCIGBT shows the lowest E on in this comparison. The turn-on dI c /dt controllability with respect to the E on is shown in Fig. 14. As the NGC effect is enhanced by the scaling rules in TIGBT, the dI c /dt-E on trade-off of the k3-TIGBT is degraded compared to that of k1-TIGBT. In contrast, in the TCIGBT, E on can be reduced by the scaling  The Japan Society of Applied Physics by IOP Publishing Ltd design as same as the TIGBT. However, the suppression of NGC is kept even in the scaled TCIGBT due to the enhanced self-clamping feature. 4) Therefore, the dI c /dt-E on trade-off of the TCIGBT is slightly improved by the scaling design. At dI/dt = 5 kA μs −1 , k3-TCIGBT shows a 70% reduction of E on compared to that of k3-TIGBT.

High current density operation
The continuous increase in power density is important for the development of power semiconductor devices to achieve low cost and design optimization for power electronic systems. Higher power densities require reliable operations at higher operating current densities with low energy loss per chip area. However, it is found that the DA phenomenon limits the reduction of E off of conventional TIGBT and scaled TIGBTs. 9) Moreover, experimental results confirm that the DA phenomenon is enhanced at high current density operations, 9) which limits the maximum operating current density and affects the long-term reliability. 5,7) Therefore, the DA must be suppressed or eliminated when increasing the operating current density of the MOS-bipolar power switches. Due to DA-free performance, 8,11) the current density of TCIGBTs can be continuously increased without DA-related concerns. Figure 15 shows the I-V and current saturation characteristics of the fabricated 1.2 kV       FS-TCIGBT at T j = 175°C. It can be seen that the FS-TCIGBT shows a low V ce(sat) of 1.87 V at J c = 300 A cm −2 and T j = 175°C as well as a low current saturation level due to the self-clamping feature. Figures 16(a) and 16(b) show the measurement results of the impact of current density on the total power loss of FS-TCIGBT at T j = 25°C and T j = 175°C, respectively. A commercial FS-TIGBT 28) is used as a benchmark and the power losses are calculated at I c = 100 A and identical maximum dV ce /dt conditions. As the P-anode dose of TCIGBT has not been optimized, the turnoff loss of TCIGBT is higher than that of TIGBT, but the onstate losses at different current densities are lower than that of TIGBT. Therefore, it can be seen that the FS-TCIGBT shows much lower power losses than the benchmark FS-TIGBT. For example, an 18% reduction of total power loss can be achieved by the FS-TCIGBT operated at J c = 300 A cm −2 and T j = 175°C. In addition, the FS-TCIGBT does not show a significant increase in total power losses with the increase in operating current density. Therefore, the FS-TCIGBT is well suited for high current density operations with low energy losses and DA-free performance. Figures 17(a) and 17(b) compares the total power loss of scaled TIGBTs and scaled TCIGBTs at J c = 200 A cm −2 and J c = 500 A cm −2 , respectively. The E on and E off are calculated at the minimum R g that can achieve DA free. As shown, scaling rules result in a 10% reduction of total power loss in both TIGBTs and TCIGBTs. k3-TCIGBT shows a 40% reduction of total power loss compared to k3-TIGBT at  The Japan Society of Applied Physics by IOP Publishing Ltd T j = 425 K. Therefore, the energy efficiency of TCIGBTs can be further improved by the employment of 3D scaling rules.

Conclusions
The turn-on behavior of TIGBT and TCIGBT are studied using TCAD simulations as well as experiments. The E on is significantly reduced in TCIGBT due to higher current gain and the negative gate capacitance effect is effectively suppressed by the self-clamping feature. Moreover, the scaling rules on TIGBT and TCIGBT result in a significant reduction of E on . The k3-TCIGBT shows high power efficiency due to the enhanced thyristor effect as well as DA-free performance. Finally, experimental results show that TCIGBT shows an 18% reduction of total power loss compared to the state-of-the-art TIGBT even when operated at J c = 300 A cm −2 and T j = 175°C.