I. INTRODUCTION

As the growth of big data market and the demand for increased volume of the stored data increases, Silicon-Oxide-Nitride-Oxide-Silicon (SONOS) nonvolatile charge trapping memories (CTM) have been widely studied due to higher endurance, better scalability, and lower power consumption compared to the conventional floating gate memories ^(1-6). However, even though this SONOS structure has been broadly adopted and investigated in flash technology, the program and erase speed (P/E) is still one of the main issues needed to be improved. Specifically, when the tunneling oxide thickness becomes thinner for fast P/E operation, the increased direct tunneling leads to significant charge loss and finally results in degradation of retention characteristics. In order to solve this problem, several approaches on gate stacks have been proposed, such as bandgap engineered (BE) tunneling oxide ^(7,8), high-${k}$ dielectrics and metal gate ^(9,10), BE charge trapping layer (CTL) ⁽¹¹⁾, and in-situ silicon nanocrystals embedded in Si$_{3}$N$_{4}$ CTL ⁽⁶⁾. However, these proposed techniques in previous researches with controlling the oxide thickness concomitantly face with trade off issues such as relationship between P/E speed enhancement from thinner thickness and retention degradation ^(12,13).

More importantly, since the hole barrier between Si and SiO$_{2}$ (4.7 eV) is higher than the electron barrier (3.2 eV) between them during P/E operation ^(14-17), the erase speed is one of the main challenge of SONOS memory needed to be overcome ^(18-22). Even though this issue can be improved by applying higher erase voltage, this solution results in higher power consumption.

Therefore, another ‘intrinsic’ solution is still needed to improve erase speed. Considering the erase speed is mainly determined by the movement of hole, it can be intrinsic solution to manipulate the p-type doping concentration. Specifically, Tunnel FET (TFET), which has been widely researched for low power operation, have numerous p-type dopant in source side. Since TFET has about 100,000 times of hole concentration in source side compared to metal-oxide-semiconductor field-effect-transistors (MOSFET), it is surmised that the TFET based memory structure can remarkably improve the erase speed of memory device. Even though this doping technique can be a viable solution for erase speed enhancement, the approach on doping technique hasn’t been widely studied for improving erase speed.

In this work, the aim of this paper is to improve erase speed without making tunneling oxide thinner, by utilizing the structure of TFET. Erase speed enhancement by boron (B) doping is proposed and investigated in this paper. The relationship between B doping and erase efficiency is investigated through memory devices based on MOSFET and TFET structures.

This paper is organized as follows. First, the basic program and erase characteristics such as change in electron trapped charge density and hole trapped charge density are analyzed. Then, the improvement of erase efficiency from the TFET structure have been widely investigated with erase speed, change of threshold voltage, and the cross-sectional illustration describing the increased hole supplement. Finally, the effect of gate length (L$_{\mathrm{gate}}$) and source-to-gate overlap length (L$_{\mathrm{Source overlap}}$) on erase speed have been additionally analyzed for more detail investigation.

II. DEVICE STRUCTURE AND MODEL PHYSICS

Cross-sectional views of MOSFET based SONOS and TFET based SONOS are schematically shown in Fig. 1. As SOI SONOS devices ^(23-25), the devices in this work have gate, source and drain terminals, and no body terminal in order to reduce fabrication cost. Table 1 describes the film thickness and channel length for these devices. The blocking oxide and tunneling oxide are composed of SiO$_{2}$ films and their thickness is 9 nm and 3 nm, respectively. The charge trapping layer is composed of 6 nm thick Si$_{3}$N$_{4}$. In both types of devices, the channel length is 60 nm. The n⁺ source/drain and $p$ body in MOSFET based SONOS device have doping concentrations of 1${\times}$10$^{20}$ cm$^{-3}$ and 1${\times}$10$^{15}$ cm$^{-3}$, respectively. In TFET based SONOS device, which is suitable for low power operation, the $p$⁺ source and $n$⁺ drain have a doping concentration of 1${\times}$10$^{20}$ cm$^{-3}$, respectively, and the $p$ region of channel has a doping concentration of 1${\times}$10$^{15}$ cm$^{-3}$ ^(26,27). A drain-to-gate underlap (source-to-gate overlap) structure is applied to the TFET to suppress the ambipolar current, which undesirably increases the off state current.

Fig. 1. Schematic view illustrating (a) MOSFET based SONOS memory, (b) TFET based SONOS memory.

In order to investigate the electrical characteristics of these two devices, physical models such as band-to-band tunneling (BTBT), trap-assisted-tunneling (TAT), Shockley-Read-Hall (SRH) recombination, and electric field saturation models are applied in the device simulation using Synopsys Sentaurus$^{\mathrm{TM}}$ TCAD tool. Furthermore, for the purpose of modeling Si$_{3}$N$_{4}$ as CTL, charge trap density of 3 ${\times}$ 10$^{19}$ cm$^{-3}$ is applied for Si$_{3}$N$_{4}$ which corresponds to charge trap density of Si$_{3}$N$_{4}$ in SONOS device ⁽²⁸⁾, and nonlocal mesh was also used for adjusting Fowler-Nordheim (FN) tunneling and direct tunneling.

Fig. 2. Program and erase pulse scheme.

Fig. 3. Electron trapped charge density change during program operation.

III. DEMONSTRATION OF ERASE SPEED ENHANCEMENT

As illustrated in Fig. 2, the program voltage of 16 V and erase voltages of -16, -18, -20 V were applied to gate for investigating the effect of doping concentration on the P/E speed. Fig. 3 and 4 illustrates the program and erase efficiency respectively by comparing MOSFET based SONOS and TFET based SONOS.

Even though TFET based SONOS has 3 times lower programing efficiency compared to MOSFET based SONOS, TFET based SONOS has 10$^{4}$ times higher erase efficiency as shown in Fig. 3 and 4. Specifically, a slight degradation of program can be explained by drain underlap and the dopant distribution. Since MOSFET based SONOS has n-type dopant in both source and drain region while TFET based SONOS has n-type dopant only in the drain region (which is located far away from charge trapping layer in the underlap structure), MOSFET based SONOS has slightly higher program efficiency compared to TFET based SONOS. In the same vein, the erase speed enhancement can be also explained by the dopant distribution. Since TFET based SONOS has a p-type doping concentration of 1${\times}$10$^{20}$ cm$^{-3}$ in the source region and whereas MOSFET based SONOS hasn’t, the TFET based SONOS is significantly superior to MOSFET based SONOS in erase speed. Furthermore, this erase speed enhancement can be achieved for various erase voltages such as -16 V, -18 V, and -20 V as shown in Fig. 4.

Fig. 4. Hole trapped charge density change during erase operation with erase voltages of (a) -16 V, (b) -18 V, (c) -20 V.

The transfer characteristics before and after erase operation was demonstrated in Fig. 5. For the fair comparison, threshold voltage was calculated when drain current is 10$^{-8}$ A/μm for MOSFET based SONOS and 10$^{-10}$ A/μm for TFET based SONOS, because TFET derives low current and the new threshold voltage is needed to be defined ⁽²⁹⁾. TFET based SONOS device showed threshold voltage shift about 0.61 V and this value is about 4 times of MOSFET based SONOS device. In addition, TFET based SONOS has additional advantage of lower off current and lower on current, which is suitable for low power application, as shown in Fig. 5.

Fig. 5. Transfer characteristics before and after erase operation by V$_{\mathrm{ERS}}$ = -20 V with (a) MOSFET based SONOS, (b) TFET based SONOS.

In summary, since p/i/n structure of TFET based SONOS is possible to supply significant amount of hole from source, erase speed can be significantly improved about 10$^{4}$ times with the advantage of low power operation for low power NAND flash application and low power NOR flash application.

IV. ADDITIONAL PHYSICAL ANALYZATION ON ERASE SPEED ENHANCEMENTT

In order to investigate the physical mechanism of erase speed improvement, hole current density during erase operation is investigated with MOSFET based SONOS and TFET based SONOS. As shown in Fig. 6, TFET based SONOS has about 10$^{6}$ times of hole current density compared to MOSFET based SONOS. Since TFET based SONOS can receive numerous amount of hole from p-type source, significant hole current can flow at the channel of TFET based SONOS. In addition, TFET based SONOS is possible to derive both of significant electron current from n-type drain and enormous hole current from p-type source, and this advantage is very crucial to TFET based memory application. Therefore, the doping technique is very strategic to memory operation.

Fig. 6. Hole current density during erase operation by V$_{\mathrm{ERS}}$ = -20 V with the cross-sectional view of (a) MOSFET based SONOS, (b) TFET based SONOS.

Moreover, the relationship between erase efficiency and body doping concentration of MOSFET based SONOS device is also investigated to discover the specific reason of erase speed enhancement as shown in Fig. 7. For all cases with erase voltage of -16, -18, -20 V, MOSFET with higher body doping concentration has higher erase efficiency. In particular, erase efficiency is significantly improved when the body doping concentration becomes bigger than 5${\times}$10$^{19}$ cm$^{-3}$. This is because when the body doping concentration becomes bigger than 5${\times}$10$^{19}$ cm$^{-3}$, not only more hole supplement occurs, but also bandgap narrowing of silicon channel ⁽³⁰⁾ also helps more hole movement. Specifically, when the bandgap narrowing occurs ⁽³⁰⁾, the bandgap of silicon becomes narrower and energy of conduction band decreases. Therefore, the hole in the conduction band of channel easily moves to charge trap layer (CTL), and the erase speed is greatly improved when the body doping concentration becomes greater than 5${\times}$10$^{19}$ [Fig. 7]. Therefore, our aforementioned analysis ‘The erase speed improvement at TFET based SONOS originates from considerable amount of hole supplement from source can be a reasonable analyzation.

Fig. 7. Relationship between erase efficiency and body doping of MOSFET based SONOS device with erase voltage of (a) -16 V, (b) -18 V, (c) -20 V.

Furthermore, this erase speed improvement achieved by increased body doping can apply in three-dimensional (3D) vertical NAND (VNAND) application. By adjusting the doping concentration of channel, it is expected that the 3D VNAND structure is also possible to improve erase efficiency. In particular, erase speed of the cells which are not located in the edge of VNAND string can be effectively improved by this doping technique.

Fig. 8. Analysis on the effect of gate length on the erase speed in TFET based SONOS.

Fig. 9. Effect of scaling on the potential of CTL in TFET based SONOS during erase operation.

Regarding the scaling issues, it has been demonstrated that the scaling of memory devices deteriorates the erase speed efficiency [Fig. 8]. This can be explained by the relationship between scaling and gate controllability. From the previous research ⁽³¹⁾, it is well known fact that the scaling with short L$_{\mathrm{gate}}$ decreases gate controllability. Namely, the decreased gate controllability from short L$_{\mathrm{gate}}$ deteriorates the potential of CTL during erase operation [Fig. 9], and therefore the erase speed of scaled memory device becomes worse [Fig. 8].

Regarding device geometry of source-to-gate overlap in TFET, which has been widely investigated for the suppression of ambipolar current ^(32,33), erase speed is slightly increased as L$_{\mathrm{Source overlap}}$ increases [Fig. 10]. Since source side in TFET has higher hole current density as demonstrated in Fig. 6(b), the TFET device with higher L$_{\mathrm{Source overlap}}$ has higher erase efficiency. In addition, the energy band diagram during erase operation [Fig. 11] also demonstrates that the TFET has higher erase efficiency and stores more amount of hole during erase operation.

Fig. 10. Analysis of erase speed enhancement depending device geometry of TFET (Source overlap length (L$_{\mathrm{Source overlap}}$) will be illustrated in Fig. 12).

Fig. 11. Comparison of energy band diagram of MOSFET based SONOS (black line) and that of TFET based SONOS (red line) during erasing operation. The abbreviated letters S, O, N, O stand for Si (channel), SiO$_{2}$ (tunneling oxide), Si$_{3}$N$_{4}$ (CTL), SiO2 (blocking oxide) respectively.

Even though the increase of erase speed from source overlap (1.12 times) is somewhat small compared to that from boron doping technique at source region (10$^{4}$ times), it is remarkable that the source overlap can improve not only ambipolar current ^(32,33) but also erase speed at the same time.

Fig. 12. Schematic diagram illustrating both of hole supplement (source) and electron supplement (drain) in TFET based SONOS.

In summary, the erase speed enhancement by incorporating TFET structure is achieved by the asymmetric structure of TFET, namely structure with electron supplier of n$^{+}$ type drain and hole supplier of p$^{+}$ type source [Fig. 12]. Therefore, TFET can be a viable solution for the future memory device with enhanced erase speed and source overlap in TFET can contribute to not only suppression of ambiploar current but also enhancement of erase speed as well.

V. CONCLUSIONS

In this paper, we have demonstrated erase speed enhancement at Tunnel FET (TFET) based SONOS device. From the several analyzation on physical mechanism of erase speed enhancement, it has been demonstrated that this erase speed enhancement originates from significant amount of hole supplement through p-type source. Even though TFET based SONOS has 3 times lower programing efficiency compared to MOSFET based SONOS, TFET based SONOS has 10$^{4}$ times higher erase efficiency in terms of trapped charge. In addition, the transfer curve showed the threshold voltage shift with TFET based SONOS is about 4 times compared to threshold shift with MOSFET based SONOS. In essential, TFET based SONOS is possible to flow both of significant electron current from n$^{+}$ drain and enormous hole current from p$^{+}$ source, and this advantage is very strategic to memory application. Therefore, the doping technique is very important in operation of memory devices and TFET based memory can be a promising candidate for the future low power application with higher erase speed.