1T semi-floating gate image sensor with enhanced quantum efficiency and high response speed

The active pixel image sensor based on a semi-floating gate transistor (SFGT-APS) has been proposed and investigated; however, the quantum efficiency (QE) and the sensitivity are too poor to meet low illumination intensity and high-speed application due to the shallow junction of photodiodes. In this work, we demonstrate a new structure, called buried photodiode semi-floating gate transistor active pixel image sensor (BSFGT-APS), which possesses enhanced QE, high sensitivity, and fast response speed. Moreover, BSFGT-APS has the same fill factor with SFGT-APS, which is 55% with a 1 μm2 photodiode in 0.13 μm process. The basic device characteristics were investigated by Sentaurus TCAD simulation, including potential of photodiode, transient response, dark current, conversion gain, and full well capacity.


Introduction
In the past few years, with the demand for image sensor reduction, there have been many attempts to reduce the size, among which reducing the number of transistors has been proved to be an effective method to reduce the size. [1][2][3][4][5][6][7][8] UTBB 9) and PISD, 10) two newly proposed 1T structures, have been studied and proved to be applicable in image sensors. Their principle mainly changes the threshold voltage of MOSFET on the buried oxide (BOX) through photoelectron collection at BOX/silicon substrate interface. Meanwhile, the active pixel image sensor based on semifloating gate transistor (SFGT-APS) has been proposed 11,12) and investigated. [13][14][15][16][17] It has a higher fill factor and a higher pixel density with a simple 1T active pixel structure and can realize the functions of the conventional 3T CMOS image sensor. 18,19) However, due to the shallow junction of photodiodes, the quantum efficiency (QE) and sensitivity are too poor to meet low illumination intensity and high-speed application. In addition, the manufacturing process is incompatible with self-aligned technology since the drain is covered by a polysilicon gate and extra masking steps are necessary, which increases complexity and cost. In this paper, a new structure, called buried photodiode semifloating gate transistor active pixel image sensor (BSFGT-APS), is proposed. Compared with SFGT-APS, this structure contains a large and deep floating P-well in the photodiode, and an additional narrow N+ guard layer is deposited on the floating P-well to expand the depletion region of the photodiode. Source/drain and N+ guard layer are formed by one-step injection. The characterization of this structure was investigated through Sentaurus TCAD simulation. Our simulation results indicated that the QE was reinforced and the overall response speed was promoted. Furthermore, transient response, dark current, conversion gain (CG), and full well capacity (FWC) were also investigated for this paper. Compared with the previous abstract for SSDM2019, 20) this paper has added the following content: more discussions and descriptions have been added to the simulation results of transient response, dark current, CG, and FWC. In addition, the potential change of photodiode has been discussed and the mechanism on QE and response speed promoted are explained.

Experimental methods
The plan view and cross-sectional view of the BSFG-APS are respectively shown in Figs. 1(a) and 1(b). The photodiode located at the left side is formed by the floating P-well surrounded by the deep N-well. The narrow N+ guard layer is deposited on the floating P-well. The semi-floating gate transistor (SFGT) located at right side contains the P+ doped polysilicon semi-floating gate (SFG) and the N+ doped polysilicon control gate (CG). There is a contact window on the side of the first SiO 2 dielectric layer through which the floating P-well is directly connected with the SFG. The second dielectric layer between the CG and the SFG is Si 3 N 4 . Source/drain and N + guard layer are formed by one-step injection. Figure 2(a) shows the equivalent schematic diagram of BSFGT-APS. The bit line is connected to the drain electrode; hence the drain current is read out as the output signal. The CG electrode is attached to the word line as a switch.
The BSFGT-APS uses the holes as signal charges and the SFG as a charge storage capacitor (C SFG ). The threshold voltage (V th ) of the SFGT can be modulated through collecting or eliminating the photo-generated holes in the SFG region. Then, the drain current as the output signal is changed with the different V th . Additionally, because the floating P-well is surrounded by N-well, photo-generated holes inside the floating P-well are confined in it, so the electrical crosstalk between the neighboring cells is reduced.

Results and discussion
Two-dimensional Sentaurus TCAD simulation was performed to investigate the performance of the BSFGT-APS. The time sequence and the voltage setting that were used for each operation are shown in Fig. 2(b). There were four phases in one operation cycle, consisting of reset, read1, exposure, and read2. The duration time of each operation was 1 μs. The source and substrate electrode were always connected to the ground.
The transient simulations were as follows: first, the reset operation was achieved with a drain voltage (V D ) of 0 V and a control-gate voltage (V CG ) of 3.3 V. The buried photodiode was operated in a forward-biased state since the N-well was connected to the SFGT's drain. The stored holes in the SFG were pushed away through the contact window and the floating P-well was refreshed and initialized. Then, in the read1 phase, the V CG and V D were set to 2.5 V and 3.3 V, respectively. The SFG voltage rose to a certain value due to the increased control-gate voltage. Then, the SFGT was turned on and the drain current was read out as the first signal of the double sampling. After the read1 phase, the V CG was changed to 0 V and V D remained 3.3 V, and the BSFGT-APS entered the exposure phase. Photo-generation holes were driven to the SFG through the contact window under the reverse-biased electric field, leading to a decreased V th . The illumination intensity and the exposure time determine the variable of the V th . Finally, the voltage setting was the same in the read2 phase as in the read1 phase, and the reduction of the V th in the exposure phase resulted in an increment of the drain current, which could be read out as the second signal of the double sampling. The contour of the total current density in the read2 phase is shown in Fig. 3. The difference between the first and second read out signals originates from the illumination, which can be calculated by a correlated double sampling 21) circuit to eliminate fixed pattern noise. 22) The Fig. 4 extracts the potential distribution of three phases: read1, read2 and exposure. The x-coordinate is the longitudinal depth of the photodiode (from the N + guard layer to the substrate), and the y-coordinate is the    electric potential. It can be seen that in the read1 and read2 phases, the floating P-well is almost fully-depleted, while in the exposure phase, the floating P-well is completely fullydepleted. The generation and collection of photo-generated carriers is mainly in the exposure phase, so this phase needs to meet the full depletion of photodiode. In the read1 and read2 phases, photo-generated carriers need not to be generated and collected, so the floating P-well do not have to be completely fully-depleted.
The transient response of the drain current varying with time is shown in Fig. 5(a). The read1 currents were the same value (∼1.66 nA μm −1 ) under different illumination intensities at a wavelength of 550 nm. The read2 currents were 7.7 μA μm −1 , 29.5 μA μm −1 , 56.9 μA μm −1 , and 79.2 μA μm −1 , corresponding to the 1 mW cm −1 , 2 mW cm −1 , 3 mW cm −1 , and 4 mW cm −1 illumination intensities, respectively. There was a current ratio of approximately 10 3 -10 4 between read1 and read2. Meanwhile, because of the large photocurrent, the drain current in the exposure phase was a bit larger than that of SFGT-APS, leading to an acceleration in the charging of the C SFG . Figure 5(b) shows the transient response of the SFG voltage (V SFG ) versus time with different light intensities. The voltage values of the SFG in read2 were larger than that in read1 at the same reading condition, which proves that the photo-generated holes were collected and stored in the SFG. In the exposure phase, the voltage in the SFG increased linearly with exposure time (T exposure ). The change ratio of the V SFG determines the exposure time and further decides the response speed of the BSFGT-APS. The change ratio of the V SFG can be calculated by the curve slopes in the exposure phase in Fig. 5(b). Under the 550 nm wavelength, the change ratio of the V SFG was 0.63 V μs −1 corresponding to 4 mW cm −2 illumination intensity. Compared to the change ratio of the V SFG for SFGT-APS, which is approximately 1.13 V ms −1 under the 4 mW cm −2 , the change ratio of the V SFG for BSFGT-APS was approximately 1 × 10 2 times higher than that of SFGT-APS. The increase of response speed is mainly due to the fact that the V CG (voltage applied to the CG) during exposure is larger than before, and the charge collection is faster under large electric fields in semi-floating gate. In the reset of the 2nd period, the declining curves were almost coincident under the different illumination intensities, which demonstrates that the reset speed was fast enough to push the storage holes away from the SFG.
The drain current as the output signal was mainly impacted by illumination intensity and drain voltage, and therefore the relationship between drain current and illumination intensity under different wavelengths was investigated. As shown in Fig. 6(a), when the illumination intensity was low, the SFGT worked at the subthreshold state and the drain current displayed an exponential relationship with the illumination intensity. As the illumination intensity increased, a large    amount of the storage holes in the SFG transferred the SFGT into a saturation state. Thus, the output drain current showed a parabolic relationship with illumination intensity. The simulation results for drain current versus illumination intensity under different drain voltages were also extracted [shown in Fig. 6(b)]. The drain current increased as the drain voltage rose under the same illumination intensity. There are two primary reasons for this phenomenon: one is that according to the I D V D characteristics of the NMOS transistor, the current value is positively correlated with the drain voltage, and the other is that a high voltage in the exposure phase leads to a large electric field accelerating the movement of the photo-generated holes to the SFG, and therefore more photo-generated holes are collected in the same amount of exposure time.
Dark current is an important parameter for an image sensor, 23,24) which can influence the dynamic range and noise. Therefore, the dark current value was studied through the TCAD simulation. The thermal Shockley-Read-Hall (SRH) generation is the main leakage mechanism for dark current. So the model we used to calculated dark current is SRH generation recombination model. Meanwhile, Each dark current is normalized by the value in area PN (the interface of floating P-well and N-well) with trap area density of 100 (arb. unit). As mentioned above, the drain node was connected with the N-well of the photodiode and the high voltage applied to the drain causes reverse-bias of the photodiode, and thus the dark current corresponding to different temperatures were measured when the drain voltages were 2.5 and 3.3 V. As shown in Fig. 7, at a low temperature under 60°C, the dark current increased slowly, with the temperature rising. At a high temperature over 60°C, the dark current rose sharply with an increment of temperature. This phenomenon is attributed to the fact that the recombination-generation mechanism is dominant at low temperatures and the diffusion mechanism is dominant at high temperatures. In addition, due to the N+ guard on the top surface, the depletion region of the buried photodiode is prevented from contacting with the silicon-oxide interface, which is one of the main dark current sources; thus, the dark current can be effectively reduced.
FWC [25][26][27] and CG, 28,29) as basic parameters of the image sensor, were also investigated. Both are closely connected with the C SFG , which is determined by the ratio of the dielectric constant to the thickness of the Si 3 N 4 dielectric layer. The decrease in thickness of the Si 3 N 4 dielectric layer enhances the capacity of C SFG to store more photo-generated holes, leading to an increment of the FWC. As shown in Fig. 8 Where ε r is relative permittivity, ε 0 is vacuum dielectric constant, and the t ox is the thickness of dielectric layer, Q is the charge and U is the voltage. The capacitance is inversely   proportional to the thickness of the dielectric layer. The define of CG is the ratio of variable voltage in the SFG to the variable number of photo-generated holes, so the equation of CG is: It is obvious that the CG is proportional to the thickness of the dielectric layer. The C SFG determined the ability of stored holes to be converted into the SFG voltage. Therefore, the FWC and CG can be modulated by changing the thickness of the Si 3 N 4 dielectric layer.
The QE is defined as the ratio of the number of photogenerated holes to the number of input photons, 30,31) which was simulated under different wavelengths, as shown in Fig. 9, compared to the SFGT-APS, the QE of BSFGT-APS was enhanced in the visible spectrum, especially in longwavelength. The photodiode of BSFGT-APS possessed a larger and deeper depletion region than that of SFGT-APS, which allowed it to absorb the long-wavelength photons. Meanwhile, the depletion region between the narrow N + guard and the floating P-well also contributed to the absorption of short-wavelength photons. In addition, it has been reported that the pixel sensitivity is the fill factor multiplied by QE. The BSFGT-APS had the same fill factor as SFGT-APS, which was 55% with a 1 μm 2 photodiode region in 0.13 μm process. Therefore, the sensitivity of BSFGT-APS was also higher than SFGT-APS. This means that at a short exposure time, BSFGT-APS can produce much more photo-generated holes to collect and store in the SFG.

Conclusions
In this paper, we introduced the structure and performance of the BSFGT-APS. Based on the results of the simulation, we can conclude that compared with the SFG-APS, the BSFG-APS not only has the same fill factor but also possesses enhanced QE, higher sensitivity, and a faster response speed. The whole operation period was remarkably reduced and could be achieved at the microsecond level, making it feasible for high-speed application. Such a high-performance image sensor is very promising for future high-density and high-speed applications.