1. Introduction

The lateral double diffused metal–oxide–semiconductor transistor (LDMOST) has been widely used in smart power integrated circuits (SPIC). The main design target of LDMOSTs is the trade-off between specific on-resistance (R_on,sp) and breakdown voltage (BV), which remains a large challenge to the development and application of high voltage LDMOSTs^[1].

Normally, a higher BV requires a longer drift region length and a lower drift region doping concentration, which lead to a higher R_on,sp. The reduced surface field (RESURF) technique has been widely employed in LDMOST to realize a high BV. However, the BV of the RESURF LDMOST is very sensitive to the deviation of the voltage-sustaining region doping from the optimized value, and the R_on,sp is still high due to the long drift region with limited doping dose in it. The trench technique has been proved to effectively reduce the cell pitch by inserting an oxide trench in the drift region and thus reduce the R_on,sp without sacrificing the BV^[2–4]. Combining with the superjunction technique, a deep trench SOI LDMOST in Ref. [5] has achieved a significant improvement of the trade-off, but the BV is still greatly affected by the deviation of the voltage-sustaining region doping. In Ref. [6], a device with N/P pillars and field plates was reported to further improve the static performance. For this device, a high gate capacitance C_GD is induced by the vertical gate field plate and thus increases the gate driving loss. High-k dielectrics (Hk) are usually used as the gate dielectric in microelectronic devices^[7–11]. Besides, the device using Hk and Si as a compound voltage-sustaining region was proposed to improve the trade-off and achieve less sensitivity of charge imbalance^[12–15].

In this paper, an improved SOI trench LDMOST utilizing the double vertical Hk pillars (double Hk TLDMOST) is proposed for the first time. In the proposed TLDMOST, two Hk pillars, which are vertically inserted in the oxide trench, assist in depleting the drift region, and thus both the average electric field strengths in the drift region and the deep trench are enhanced. An ultra-low R_on,sp is achieved for the two Hk pillars to assist in depleting the drift region, and the drift region doping concentration could be increased. Therefore, the trade-off between the BV and R_on,sp can be significantly improved. In the following sections, the structure, the mechanism and the performance will be studied in detail.

2. Structure and mechanism

Figs. 1(a) and 1(b) show the structures of the previous and proposed TLDMOSTs, respectively. In the proposed structure, the double Hk pillars, Hk₁ and Hk₂, are vertically inserted in the oxide trench, and connected to the source and the drain electrodes, respectively. The t₁ and t₂ represent the thicknesses of Hk₁ and Hk₂, respectively. The W_Hk1 and W_Hk2 denote the widths of Hk₁ and Hk₂, respectively. The T_t and W_t are the thickness and width of the deep trench, respectively. The t_BOX and t_ox are the thicknesses of the buried oxide (BOX) layer and gate oxide, respectively. The N_d is the doping concentration of the drift region.

Fig. 1(c) illustrates the mechanism of the proposed double Hk TLDMOST under reverse bias. As for the vertical blocking voltage, the BV is determined by the maximum voltage drop across the SOI layer and BOX. By virtue of the back-gate effect, the holes and electrons are simultaneously accumulated on the opposite side of the BOX, which induce the additional electric field and significantly increase the total value in the BOX according to the Gauss’s law.

As for the lateral blocking voltage, the BV is influenced by both the path AG across the trench and the path ABCDEFG (A–G) around the trench, shown in Fig. 1(b). Considering the high critical electric field (E_c) of the insulator (e.g., E_c > 6 × 10 ⁶ V/cm for SiO₂, and E_c > 1 × 10 ⁶ V/cm for SrTiO₃^[16]), the breakdown hardly occurs in the trench. Therefore, the BV is mainly determined by the maximum voltage drop along the path A–G.

In the TLDMOST with single Hk pillar, Hk₁, inserted in the oxide trench (single Hk TLDMOST), since the permittivity of the Hk₁ is supposed to be much higher than that of the Si^[12–14], most electric displacement lines produced by the positive charges in the drift region and N⁺ drain region go through the Hk₁ instead of the Si, and terminate at the source electrode. This can be explained by comparing a static electric field to a current field^[12]. Compared with the drift region, the Hk₁ has a much higher conductivity, and it is obvious that a large portion of the current flows in the Hk₁. This makes only a few of the ionized donors in the drift region contribute to the electric field of the p-body/n-drift junction. Therefore, heavier doping concentration can be used in the drift region without producing a high electric field at the p-body/n-drift junction, and the improved average electric field in the drift region under the source can be achieved. However, in the drain side, a large amount of the electric displacement lines produced by ionized donors in the N⁺ drain region tend to gather at the drain side, causing a sharp increase of the electric field at the N⁺ drain side and a premature breakdown occurs.

To improve the electric field distribution under the drain, a second Hk pillar (Hk₂) connected to the drain electrode is introduced. Most of the electric displacement lines produced by the ionized impurities under the drain are via the Hk₂ with high conductivity, spread toward the bottom of the Hk₂, and finally terminate at the source electrode. This not only relaxes the gathering of electric displacement lines at the N⁺ drain, but also enhances the average electric field along the path E–G. From the above analysis, it is convincing that the proposed TLDMOST has a higher BV and a heavier N_d as well by comparing with the previous TLDMOST with the same geometry.

3. Results and discussion

The performances of the proposed TLDMOST are investigated and optimized by MEDICI^[17]. The models of CONMOB, SRH, AUGER, FLDMOB, and IMPACT.I are used in the simulations. The key simulation parameters are as follows: t₁ = t₂ = 18 μm, t_BOX = 1 μm, t_ox = 30 nm, T_t = 21 μm, W_Hk1 = 3.2 μm, W_Hk2 = 1 μm, W_m=1.4 μm, W_t = 10 μm, and the thickness of the SOI layer is 25 μm.

3.1 Static characteristics

Fig. 2 shows the potential contours of the proposed double Hk, single Hk and previous TLDMOSTs at breakdown. In Fig. 2(a), the double Hk TLDMOST exhibits a uniform potential distribution and realizes a high BV of 749 V. The single Hk TLDMOST suffers from a low electric field in the drift region under the drain. This is attributed to the amounts of electric displacement lines gathering at the point G owing to a high curvature, causing a high electric field and a premature breakdown at the point G. However, in Fig. 2(c), the very sparse equipotential contours in the lower section of the drift region under both the source and the drain are shown in the previous TLDMOST and correspond to a poor BV of 403 V.

Fig. 3(a) illustrates the electric field distributions around the oxide trench of the three TLDMOSTs. Due to the electric field modulation by the Hk₁ and Hk₂, it is obvious that four new electric field peaks are introduced at the points B, C, D, and E, whereas the electric fields at the corresponding points of the previous TLDMOST are very low. The single Hk TLDMOST exhibits a much more uniform electric field in the drift region under the source than that under the drain, as discussed earlier in Section 2. Besides, the electric fields at points A and G, which limit the breakdown voltage of the previous TLDMOST, are significantly reduced in the proposed double Hk TLDMOST. This avoids premature breakdown at the surface.

Fig. 3(b) shows the surface electric field distributions (y = 0.01 μm) of the double Hk, the single Hk, and the previous TLDMOSTs. The maximum electric field in the insulator trench of the double Hk TLDMOST reaches to 5.1 × 10⁶ V/cm compared with 3.3 × 10⁵ V/cm of the previous TLDMOST, but is still lower than the E_c of the oxide. That is to say, the breakdown will not happen in the deep trench.

According to Ref. [12], the ability for the Hk insulator to transmit the electric displacement is determined by the relative permittivity (k), the width and the thickness of the Hk pillar, which are the key parameters for the breakdown performance. There are some Hk materials which may be possible for the proposed TLDMOST, such as SrTiO₃, PZT, PbTe or an insulator with conducting particle. These dielectrics have a value of k in the range of 100–500 or even higher^{[14, 18–20]}.

Fig. 4 gives the BV and R_on,sp versus N_d in the previous TLDMOST and the proposed one for different values of k. In the proposed TLDMOST, the BV increases first and then decreases with the increase of k value for a given N_d. For example when N_d is 2 × 10¹⁵ cm⁻³, for k = 100, amounts of electric displacement lines produced by the positive charges of the depleted drift region under the source gather at the corner of the trench gate. Thus, the peak electric field there increases, and a premature breakdown occurs. When the permittivity of Hk increases, an increasing number of electric displacement lines produced by the charges mentioned above are absorbed by Hk₁. This relaxes the peak electric field at the corner of the trench gate and results in an improvement of BV. However, the electric field at point E increases with the permittivity of Hk₂ increasing. When the k value is too large, e.g. 700, a premature breakdown happens at point E.

As illustrated in Fig. 4, there is an optimal value of N_d for a certain value of k. The maximum BV of 749 V can be achieved corresponding to an R_on,sp of 67 mΩ·cm² for N_d = 2.4 × 10¹⁵ cm⁻³ and k = 500. As a consequence, the BV is enhanced by 86%, and the R_on,sp is reduced by 88% compared with the previous TLDMOST for the same geometry.

Furthermore, as indicated in Fig. 4, the proposed structure is more robust to the deviation of N_d than the previous one. For the latter, the electric displacement lines produced by the depleted N⁺ drain tend to concentrate at the point G at a value of N_d lower than the optimal value, whereas for the former, the gathering effect is significantly relieved by the Hk₂. Accordingly, the electric field peak at the point G is reduced, and the BV has less sensitivity to the N_d. When the N_d is higher than the optimal value, for the previous TLDMOST, the electric displacement lines produced by the ionized donors in the drift region tend to crowd at the p-body/n-drift junction and increase the surface electric field. However, for the proposed structure, most electric displacement lines produced by the aforementioned charges are prone to access the Hk₁. This makes only a few of the ionized donors contribute to the electric field of the p-body/n-drift junction. Therefore, the BV has no significant change with the variation of the N_d, compared with the previous TLDMOST.

Fig. 5(a) illustrates the influence of t₁ and W_Hk1 on the BV when the t₂ = 18 μm and W_Hk2 = 1 μm. When there is no Hk₁, a large amount of electric displacement lines produced by the depleted drift region gather at point A, causing an electric field peak at point A and a premature breakdown. After the Hk₁ is applied, a portion of electric displacement lines are absorbed by the Hk₁, which causes a new electric field peak at point B shown in Fig. 1(b) and a reduced electric field at point A. With the value of t₁ (or W_Hk1) increasing, the number of electric displacement lines absorbed by the Hk₁ increases, and less ionized donors will terminate at the p-body/n-drift junction. Consequently, the electric field peak at the p-body/n-drift junction reduces, and the average electric field in the drift region under the source enhances, both of which suggest a higher BV. Once the value of t₁ (or W_Hk1) is larger than its optimal value, the BV reduces, largely on the grounds that excessive electric displacement lines crowd at point B inducing a high electric field.

The influence of t₂ and W_Hk2 on the BV at t₁ = 18 μm and W_Hk1 = 3.2 μm is depicted in Fig. 5(b), revealing that both t₂ and W_Hk2 have significant effects on the electric field distribution in the drift region under the drain. A small t₂ or W_Hk2 induces a high average electric field in the drift region between points G and F, whereas the drift region between points F and E sustains a low voltage drop. On the contrary, a large t₂ or W_Hk2 causes a large voltage drop between points F and E, whereas the average electric field in the drift region between points G and F is reduced, both of which are detrimental to the high BV.

Fig. 6 clearly manifests the dependence of the BV on the W_m at given dimensions of the two Hk pillars. The structure parameters of Hk₁ and Hk₂ are as follows: t₁ = 18 μm, W_Hk1 = 3.2 μm, t₂ = 18 μm, and W_Hk2 = 1 μm. When the W_m is larger than the optimal value, amounts of the electric displacement lines produced by the depleted drift region under the source are easily absorbed by Hk₁, causing a sharp increase of the electric field peak at point B and a premature breakdown occurs. The BV reduces if the W_m is lower than the optimal value. This is owing to the fact that most of the electric displacement lines produced by the aforementioned charge directly enter into the gate electrode, rather than the source electrode via the Hk₁, resulting in a high electric field at the corner of the gate.

3.2 Dynamic characteristics

Fig. 7 shows the gate charge characteristics of the proposed TLDMOST and the previous TLDMOST, where the inset illustrates the simulation circuit and VDD is set to 200 V. As revealed in the figure, the gate–source charge (Q_GS) of the proposed TLDMOST remains constant at different values of k, which is approximately identical to that of the previous TLDMOST. The proposed TLDMOST exhibits a gate–drain charge (Q_GD) of 3.2 × 10⁻¹⁵ C/μm, increased by 55% compared with the previous TLDMOST. The larger Q_GD is caused by the inserted Hk insulator. However, compared with the trench LDMOS in Ref. [6], the proposed structure exhibits a Q_GD value reduced by 88%.

Fig. 8 depicts the switching characteristics of the proposed TLDMOST and the previous TLDMOST. The gate resistance is 30 Ω. The drain electrodes of the devices are connected to a VDD of 200 V through a load resistor R_L. To obtain the same current density for both the TLDMOSTs, the values of R_L are chosen as 6.619 × 10⁷ Ω and 6.2 × 10⁷ Ω, respectively. In the proposed double Hk TLDMOST, with the value of k increasing, both the turn-on time and the turn-off time increase. The rise time of the proposed structure is 1.33 ns at k = 500, which is approximate to that of the previous structure. The fall time of the proposed structure is 60.4 ns, which is 55.4 ns slower than the previous structure. The longer fall time of the proposed TLDMOST is resulted from the added body capacitance C_DS by the Hk pillars.

4. Descriptions of device fabrication

Fig. 9 shows the key process steps to fabricate the double Hk TLDMOST. The deep trench can be formed by the DRIE in Fig. 9(a). The dry oxidation of the thin liner oxide is followed in Fig. 9(b) to reduce the interface states near the interface. In Fig. 9(c), the trench is filled with oxide and polished. Trench etching^{[6, 21, 22]} and deposition of the Hk are followed to form the double Hk pillars in Figs. 9(d)–9(e). In Fig. 9(f), the remaining process steps including the p-body implantation, gate etch and polysilicon deposition, P⁺/N⁺ region implantation, the contact etch, the metal deposition and etch are implemented by using the conventional processes.

5. Conclusions

This paper has presented an analysis of the specific on-resistance and the breakdown voltage of an improved TLDMOST with double Hk pillars. The double Hk pillars assist depleting the drift region, and modulate the electric field distribution, which lead to the increase of the drift doping concentration. As a consequence, compared with the previous TLDMOST, the proposed structure achieves a better trade-off between the BV and the R_on,sp. The simulation results indicate that the optimized proposed TLDMOST exhibits a BV of 749 V and an R_on,sp of 67 mΩ·cm² at k = 500, which breaks the silicon limit. Even though the turn-off time of the proposed structure is longer than that of the previous TLDMOST, it is acceptable in most applications that the switching frequency is not very high.