Deep Trench MOSFET structures study for a 1200 Volts application

Deep Trench MOSFET structures study for a 1200 Volts application L.Théolier, K. Isoird, F. Morancho, J. Roig, H. Mahfoz-Kotb, M. Brunet, P. Dubreuil University of Toulouse, LAAS/CNRS 7 avenue du colonel Roche 31077 Toulouse Cedex 4, France Tel.: +33 5 61336390 Fax: +33 5 61336208 E-mail: ltheolie@laas.fr, kisoird@laas.fr, morancho@laas.fr, jroiggui@laas.fr, hkotb@laas.fr, mbrunet@laas.fr, dubreuil@laas.fr URL: http://www.laas.fr Keywords Power semiconductor device, MOSFET, Super Junction Devices, Traction application. Abstract In this work we studied some possible high voltage MOSFETs structures that can replace the IGBT in the railway traction converters. In this purpose, some high voltage power MOS structures are presented and theoretically compared using 2D simulations. Simulations results show that the DT-UMOSFET should be a good challenger to the 1200 Volts IGBT. Moreover, the influence of various parameters, like trench width, trench verticality or boron dose, on DT-UMOSFET static performances is shown. Introduction The aim of this work is to propose alternative solutions to the IGBT in the railway traction converters. Nowadays, it is necessary to increase the operation frequency of these converters to decrease the size of power modules. The technology has thus to be changed, especially because of the important switching losses of the IGBT. A solution could be the replacement of the IGBT technology by a new generation of high voltage power MOSFETs. In the 1200 Volts range, the on-resistance is the main drawback of the conventional MOSFET structures. Nevertheless, MOSFETs also exhibit some interesting advantages, such as short switching times and a good thermal stability. Furthermore, it is possible and very interesting to use the internal diode (P-body/N-epitaxial layer) of the MOSFET to replace the “IGBT/diode” module. New MOSFET technologies based on the drift region modification, either with deep trenches (OBVDMOSFET [1]) or with P regions (Floating Islands [2] or Superjunction [3]) have been recently proposed to reduce the specific on-resistance. The final objective of this work is to reach the static performance (“breakdown voltage versus specific on-resistance” trade-off) of a 1200 Volts IGBT, i.e. a 17 mΩ.cm2 specific on-resistance or a 100 A.cm-2 current density at 25°C with VCE = 1.7 V [4]. Studied Structures Principle of the new possible structures A solution to improve static performances of power MOSFETs is to realize deep trenches filled with a thick oxide and polysilicon connected to the source. When a positive drain voltage is applied, the structure becomes vertically depleted, but also laterally in the drift region. The width between two trenches of oxide being lower than the epitaxial layer thickness, the drift region is laterally depleted which allows an increase of the lateral component of the electric field, without its critical value being reached: for the same Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. breakdown voltage, the N- epitaxial layer doping concentration can be increased compared to the conventional VDMOSFET’s case. This trench concept was demonstrated in lower voltage range (100 V- 200 V): the OBVDMOSFET [1] is based on this concept (Fig. 1), but it is different from the trench gate concept [5]. Fig. 1: Schematic cross-section of a half-cell of an OBVDMOSFET. The Floating Island concept is based on the introduction of one (or several) P buried layer(s) in the N- epitaxial layer (Fig. 2.). This (these) P buried layer(s) is (are) located under the “P-well / N-” plane junction in order to divide, under reverse bias conditions, the maximal electric field in two (or several) parts, allowing the improvement of the breakdown voltage for the same N- epitaxial layer doping concentration. The FLIMOSFET [2], the FLYMOSFET [6] and the FITMOS [7], for instance, use this concept. Fig. 2: Schematic cross-section of a half-cell of a FLIMOSFET. The Superjunction concept has been proposed to overcome the ideal silicon MOSFET limit. This concept allows good performance in high voltage applications but the precise charge balance requirement and inter-diffusion problems makes the manufacturing of Superjunction devices complicated (Fig. 3). Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. Furthermore, the “multi-epitaxies and multi-implantations” for this voltage range make the overall process too complex and cost prohibitive. Nowadays, only 1200 Volts semi-SJ-VDMOSFETs are fabricated [8]. Figure 1 Fig. 3: Schematic cross-section of a half-cell of a SJ-VDMOSFET. The newly DT-UMOSFET concept (Deep Trench UMOSFET, Fig. 4), based on the VTR-DMOSFET [9], is a charge compensation structure like the SJ-VDMOSFET. The main differences with a SJ-VDMOSFET are the gate shape, which is a trench-gate MOS and the introduction of a deep trench in the N- epitaxial layer in order to realize the P-pillars necessary for the charge balance. These P-pillars are made by the diffusion of a P+ heavily doped polysilicon across a thin oxide in order to control the diffused dose. From a technological point of view, DT-UMOSFET has some advantages. On one hand, it includes only one epitaxial growth step, compared to SJ-VDMOSFET. On the other hand, the oxide thickness and the thermal process enable an accurate control of the boron dose, compared to VTR-DMOSFET. Moreover, this structure can exhibit the same conduction area, compared to a SJ-VDMOSFET, because the area required by the P-pillars and the trench oxide in the DT-UMOSFET (Wpox, Fig. 4) is approximately the same than the one required by the P-pillars in the SJ-VDMOSFET (Wp, Fig. 3). Fig. 4: Schematic cross-section of a half-cell of a DT-UMOSFET. Wp Wpox Wa Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. The P-Floating Islands MOS structure will not be studied in this voltage range because it is too complex to realize, as it actually needs at least 9 Floating Islands to exhibit the same performance as Superjunction devices with a 5 µm pillar width (Fig. 5). In order to determine the best solution, only the performance of the UMOSFET, the OBVDMOSFET, the SJ-VDMOSFET and the DT-UMOSFET will be compared. Fig. 5: Conventional VDMOSFET, vertical FLIMOSFET and vertical Superjunction MOSFET on- resistance limits in terms of “Ron.S/BV” trade-off. “n” is the number of Floating Islands (FLIMOSFETs), “W” is the P and N-layers width (Superjunction MOSFETs). Simulated structures The simulations have been performed with the ISE-TCAD software. Unless otherwise mentioned, all the structures have a cell size of 10 µm. We estimate current density at Vds = 1.7 Volts at 25°C. OBVDMOSFET Our first simulation was based on the state of the art. We chose to simulate a structure with these parameters: epitaxial layer doping concentration = 1x1015 cm-3, oxide thickness = 1µm, deep trench depth = 90 µm. The electric field distribution at breakdown is presented in Fig. 6-a. Breakdown occurs in the bottom of the trench, because of a potential constriction. In order to improve the breakdown voltage, a 15 µm oxide width, a 40 µm oxide thickness and a 12 µm conduction area are needed (Fig. 6-b). This oxide shape allows the electric potential redistribution in the N- epitaxial layer. The modified structure exhibits a 1320 Volts breakdown voltage and a specific on-resistance of 295 mΩ.cm2. However, to achieve these performances, a cell size of 50 µm is needed. a) b) Fig. 6: Electrostatic potential distribution at breakdown voltage: a) default structure b) optimized structure 300 V 0 V 0 V 1300 V Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. SJ-VDMOSFET We simulated a structure with N and P constant pillars doped to 6.5x1015 cm-3. The optimized structure exhibits a specific on-resistance (RON.S) as low as 19 mΩ.cm2 and a breakdown voltage (BV) of 1280V. This result shows the superiority of the SJ-VDMOSFET compared to the current high voltage semi- superjonction devices which show a RON.S of 54 mΩ.cm2 with a 1100V BV and 163 mΩ.cm2 with a 1400 V BV [8]. DT-UMOSFET Figure 7 shows the lateral doping concentration of the simulated structure where a N- epitaxial layer with a doping concentration of 7.5x1015 cm-3 and a P-pillar with a boron peak concentration of 5x1016 cm-3 at the oxide/silicon interface, are used. The total width of the trench is 4 µm for a depth of 95 µm. It is noteworthy that the boron peak concentration and the junction depth are chosen to fulfill the charge balance condition between the N and P-pillars. The optimized cell exhibits a RON.S of 20 mΩ.cm2 at a breakdown voltage of 1330 Volts. Fig. 7: Lateral doping concentration profile in the N and P pillars of the simulated DT-UMOSFET. The position “x = 0” corresponds to the centre of the cell (Fig. 4). Comparative performances of the different devices Table I: Comparative performances at 25°C UMOSFET OBVDMOSFET SJ-VDMOSFET DT-UMOSFET trench IGBT [4] Size of the cell (µm) 10 50 10 10 ? N- epitaxial layer thickness (µm) 115 120 92 92 ? N- epitaxial layer doping concentration (cm-3) 10 14 1015 6.5x1015 7.5x1015 ? Simulated Breakdown Voltage (V) 1220 1320 1280 1330 1200 Current density (A.cm-2) @VDS=1.7Volts 3.2 5.8 89 85 95 Table I summarizes the simulated characteristics and performances of the different structures and compares them with a simulated and optimized UMOSFET and a fabricated IGBT [4]. Two conclusions can be deduced: firstly, all the simulated structures exhibit better performances than the UMOSFET. N P Oxide Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. Secondly, only SJ-VDMOSFET and DT-UMOSFET seem to be good alternatives to the IGBT. However, as it is discussed previously, the fabrication process for DT-UMOSFET is expected to be less complex and low cost compared to SJ-VDMOSFET’s one. So, this promising structure (DT-UMOSFET) will be considered in detail in the following sections. DT-UMOSFET detailed study Some parameters sensitivity simulations will be discussed in order to evaluate the static performances evolution. At first we present impact of epitaxial layer doping concentration and diffused boron dose on breakdown voltage and on-state current density. Subsequently we observe influence of parameters which control principally the charge balance condition. These parameters will include the active area width (Wa), the diffused boron dose (DB) and the trench shape (depth, width and verticality). Other parameters have been studied like the gate shape, its placement or the P-body dose. But they are not presented here because they have little influence on the on-state current or on the breakdown voltage. The simulated breakdown voltage variations versus unbalanced diffused boron dose in percent are presented in Figure 8. The simulations were made with a 7 µm width base cell and a 3 µm opening trench based on the laboratory possibilities in order to reach a better current density. In the case of higher epitaxial doping concentration, which is the most interesting case to increase the on-state current density (Fig. 8-b), it can be observed that a light unbalanced boron dose would result in a dramatic decrease in the breakdown voltage. Because of this strong sensitivity to the boron dose we will choose weaker epitaxial doping concentration to fabricate DT-UMOSFETs exhibiting the best trade-off between the breakdown voltage and the current density. a) b) Fig. 8: a) Simulated breakdown voltage variations (BV) vs unbalanced diffused boron dose (DB) in percent, b) Simulated current density variations (J) vs unbalanced diffused boron dose (DB) in percent. Influence of active area width and diffused boron dose on the electric behavior The simulations were made with a 7 µm width base cell, a 3 µm opening trench, an epitaxial layer doping concentration of 1.4x1016cm-3 and a constant boron diffusion length at 0.6 µm in order to reach a better current density. The simulated breakdown voltage variations (BV) versus active area width (Wa) and diffused boron dose (DB) are presented Figure 9-a and the simulated current density variations (J) vs active area width and diffused boron dose are presented Figure 9-b. It can be observed in the first picture that the breakdown voltage is very sensitive to both parameters; this is principally a charge unbalance issue. For example; for a fixed boron dose, a variation in the active area width (Wa) will change the number of N Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. charge carriers. Then it can induce an unbalanced charge in either side. In the same way, for a fixed value of Wa, a variation of the diffused dose induces an unbalanced charge. In both cases, this charge unbalance will result in the variation of BV given in Figure 9-a. However, the variations of the active area width can be corrected by adjusting the diffused boron dose so we get the best results on the line defining the balance charge. Boron dose and active area width have a poor influence on the current density. Figure 9-b shows a light current density variation when active area increases. The best static performances are obtained for an active area width above 2 µm, and the diffused boron dose higher than 4.1012 cm-2. a) b) Fig. 9: a) Simulated breakdown voltage variations vs active area width (Wa) and diffused boron dose (DB). b) Simulated current density variations vs active area width (Wa) and diffused boron dose (DB). Influence of the trench shape on the electric behavior The simulated breakdown voltage variations versus the active area width and the deep trench depth are presented Figure 10. The depth of the deep trench is not a critical parameter like the P-well or the gate shape. To ensure a higher breakdown voltage the trench depth must be at least equal to the thickness of the epitaxial layer. The oxide trench penetration into the substrate does not damage the breakdown voltage. It is an important result for the process: the epitaxial layer thickness fixes the breakdown voltage like in the punch-through structures. Fig. 10: Simulated breakdown voltage variations vs active area width (Wa) and deep trench depth. Substrate penetration Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. The simulated breakdown voltage variations versus the active area width (Wa) – obtained in the middle of the trench depth – and the trench slope defined by the difference in width between the top (WT) and the bottom (WB) of the trench Fig. 11-a, are presented Figure 11-b. The chosen values for trench slope in this simulation are based on study results made in our team on realizing high aspect ratio deep trenches for integrated MIM capacitors applications [10]. This study used a time multiplexed Inductively Coupled Plasma etcher and a Bosch etch-process where an etching gas (SF6) and passivation gas (C4F8) are used alternatively [11]. Results showed that deep trenches of about 33 aspect-ratio are realizable with a slope of about ± 0.5 µm where the slope sign depends strongly on etching conditions (platen power, pressure, passivation/etching time ratio, …) [12-14]. Our simulation results show that verticality of the trench is a critical parameter, particularly the overetching in the bottom of the trench. The asymmetry is not yet confirmed, but we think it is due to the doping concentration profile variation between the substrate and the P-body. Consequently, the trench verticality is essential to insure equivalent performance with the SJ- VDMOSFET. a) b) Fig. 11: a) DT-UMOSFET schematic cross-section, b) simulated breakdown voltage variations vs active area width (Wa) and trench slope. Conclusion In this work, we study different concepts in order to propose alternative solutions to the 1200 Volts IGBT technology. Simulations confirm that conventional VDMOSFET (or UMOSFET) have poor performances in this range of voltage. The few improvements obtained with the OBVDMOSFET dismiss this technology for 1200 Volts applications because of its complexity. Superjunction concept based devices seem to be the best candidates. In comparison with the IGBT, the SJ-VDMOSFET and the DT-MOSFET exhibit equivalent static performances, but the DT-UMOSFET is more advantageous because of its lower complexity. The optimized structure needs to reach an on-state current near 130A.cm-2 at VDS = 1.7 Volts, so a specific on-resistance of 13mΩ.cm2. The study has shown that the electric performances are very sensitive to the charge unbalance, but the sensitivity can be reduced by decreasing the doping concentration at the expense of the on-state current. However to fabricate a DT-UMOSFET exhibiting the highest current density, we must control accurately the trench shape (width, slope, depth) to ensure the charge balance. The study needs to be completed with termination, dynamic and thermal simulations. Moreover, the internal diode needs to be optimized in order to keep good performance of the “DT- UMOSFET / internal diode” compared to those of “IGBT / diode” module. However, the realization of Overetching in the bottom of the trench Authorized licensed use limited to: Jiangnan University. Downloaded on October 18, 2008 at 04:08 from IEEE Xplore. Restrictions apply. this structure remains difficult because the performances are very sensitive to the technological parameters’ variations. The deep trench process, using boron diffusion across a thin oxide to realize P- pillars, is now in progress in order to validate the concept of the DT-UMOSFET: we plan to fabricate 1200 Volts Deep Trench diodes and UMOSFETs in the near future. References [1] Liang Y.C., Gan K.P., Samudra G.S.: Oxide-bypassed VDMOS (OBVDMOS): an alternative to superjunction high voltage MOS power devices, Electron Device Letters, Vol. 22, pp. 407-409, 2001. [2] Cézac N., Morancho F., Rossel P., Tranduc H., Peyre-Lavigne A.: A New Generation of Power Unipolar Devices: the Concept of the FLoating Islands MOS Transistor (FLIMOST), ISPSD’2000, pp. 69 - 72, May 2000. [3] Chen X.B., Mawby P.A., Board K., Salama C.A.T.: Theory of a novel voltage-sustaining layer for power devices, Microelectronics Journal, pp. 1005 - 1011, 1998. [4] Iwamoto H., Kawakami A., Satoh K., Takahashi H., Nakaoka M.: New generation 1200V power module with trench gate IGBT and super soft recovery diode and its evaluations, IEE Proc. Electric Power Application, Vol. 147, N°3, pp. 153 - 158, May 2000. [5] Ueda D., Takagi H., Kano G.: A new vertical power MOSFET structure with extremely reduced on-resistance, IEEE Transactions on Electron Devices, Vol. 31, N°1, pp. 2 - 6, 1984. [6] Alves S., Morancho F., Reynès J-M., Margheritta J., Deram I., Isoird K., Tranduc H.: Technological realization of low on-resistance FLYMOS™ transistors dedicated to automotive applications, 11th European Conference on Power Electronics and Applications (EPE’2005), September 2005. [7] Takaya H., Miyagi K., Hamada K., Okura Y., Tokura N., Kuroyanagi A.: Floating Island and Thick Bottom Oxide Trench Gate MOSFET (FITMOS) – a 60V Ultra Low On-Resistance Novel MOSFET with Superior Internal Body Dio