1. Introduction

In recent years, CMOS image sensors (CIS) have gained wide applications for various consumer electronics such as cell phones, digital cameras and other imaging devices due to the advantages of low power consumption, low cost and CMOS technology compatibility^[1-3]. However, as the pixel size is continuously scaling down to achieve higher integration density, the spatial spacing between adjacent pixels has also been scaled, resulting in a serious crosstalk issue by which the image resolution may be significantly reduced^[4]. Therefore, in order to achieve a high image quality in modern CIS, crosstalk suppression technologies should be further considered.

There are two mainly types of crosstalk mechanisms in CIS pixels. One is optical crosstalk, which corresponds to a photon exchange between neighboring pixels before electron/hole pairs are excited in the substrate region, the other is electrical crosstalk, which means photon generated carriers diffuse or drift to other pixels through the substrate or epitaxial (EPI) layer. Extensive efforts have been made to suppress the latter one including guard ring architectures^[5], heavy doping or a thinner EPI layer^[6], deep trench isolation (DTI) for the front side (FSI) or back side illuminated (BSI) technology^[7-9] and so on. They were reported to be effective in reducing the electrical crosstalk rate between neighboring pixels. However, major photoelectric performances such as the photosensitive fill factor, full well capacity (FWC), quantum efficiency (QE) or dark current may be sacrificed to different extents by using the solutions mentioned above, so the superiorities of the technologies were restricted.

In this work, based on the prevalent 4-transistor (4T) pixel structure, a novel concept of the design called an inverse U-shape buried doping (IUBD) architecture formed in the pinned photodiode (PPD) is proposed to solve the problem discussed above. No extra consumption of the fill factor is achieved by using the deep buried injections, which can suppress the electrical crosstalk and improve other major photoelectric performances, simultaneously.

2. Details of the proposed design

2.1 Mechanism of electrical crosstalk in PPD

The pixel we chose to investigate in this paper has a 4T structure. The schematic of a 4T pixel which consists of a PPD, a transfer gate (M1), a reset transistor (M2), an in-pixel amplifier (M3) and a selected switch (M4) is illustrated in Fig. 1(a). The excited carriers originated in the PPD will transfer to the floating diffusion node and change into a voltage signal, which can be ultimately output by a readout bus connected to the subsequent circuit. The PPD is a buried structure with an N-type region to drift the photon generated carriers dropped in the space charge region (SCR), as illustrated in Fig. 1(b). A P+ region is formed by a high dose surface implant to prevent the dark current noise typically originating from the interface states or trapped charges laid in the dielectric layers.

The photon absorption depth^[10] in a silicon material closely depends on the incident wavelength. Most of the carriers being excited by long wavelength illumination may drop deep into the EPI layer or high doping substrate. As illustrated in Fig. 1(b), the excess electrons in the substrate will be quickly recombined mainly by the Shockley-Read-Hall (SRH) mechanism due to the energy levels within the band gap created by defects^[11]. While the electrons in the EPI layer are treated as the minority carriers to diffuse following the concentration gradient $\triangledown n(x, y, z)$ in any direction with a lower recombination rate, and can be specified as the steady-state diffusion equation^[12]:

$ $$\begin{equation} \label{eq1} D_{\rm n} \nabla ^2n(x, y, z)=\frac{n(x, y, z)}{\tau }-G(x, y, z), \end{equation}$$ $

(1)

where $D_{\rm n}$ is the diffusion coefficient, $\tau $ is the lifetime of the excess electrons, and $G(x, y, z)$ represents the generation rate of electrons under long wavelength illumination. The maximum distance by which the excess electrons can diffuse in the EPI layer is given as:

$ $$\begin{equation} \label{eq2} L_{\rm n} =\sqrt {D_{\rm n} \tau } . \end{equation}$$ $

(2)

It can be revealed that, if the distance from the original position where an electron starts to diffuse to the nearest edge position of the SCR within the neighboring pixel is smaller than $L_{\rm n}$, the electron can be absorbed as the signal charge output by the neighborhood pixel, causing a crosstalk phenomenon. The conventional PPD architecture has a large crosstalk probability as the neighbor spacing shrinking.

2.2 Proposed IUBD PPD architecture

2.2.1   Design of the architecture

Based on the mechanism analysis, in order to achieve crosstalk suppression in the PPD, the main starting point is to make the excess electrons dropped in the EPI layer be absorbed more efficiently or to reduce the minority carrier's lifetime so that the excess electrons can hardly diffuse to neighbors. Being different from the solutions mentioned in the previous section, based on the spirit of crosstalk reduction without sacrifices of other major pixel performances, a concept of deep buried implantation is introduced to establish a deep electric field by which the excess electrons dropped in the EPI layer can be drifted back to photosensitive SCR with little diffusion.

In order to obtain the electric field stated above, an extremely low P-type (LP) doping region, which has the concentrations orders of magnitude lower than the EPI layer, should be injected into the buried area beneath the original N-type (ON) region, as illustrated in Fig. 2(a), the shape of the LP region is like two "arms" in a two-dimensional cross section. Thanks to the low/high junction's behavior proved by the Gunn's theory^[13], a specific electric field $E_{\rm HL}$ with a direction from the LP region to the EPI layer can be established according to the doping gradient, which is given as:

$ $$\begin{equation} \label{eq3} E_{\rm HL} =\frac{k_0 T}{q} \frac{\ln \dfrac{N_{\rm EPI} }{N_{\rm LP} }}{z_{\rm HL} }, \end{equation}$$ $

(3)

where $k_{0}$ represents the Boltzmann constant, $T$ represents the temperature, $N_{\rm EPI}$ and $N_{\rm LP}$ are the doping concentrations of the EPI layer and the LP region, respectively, and $z_{\rm HL}$ represents the width of the drifting zone. It can be indicated that there should be a built-in potential $V_{\rm HL}$ across this drifting zone, which is given as:

$ $$\begin{equation} \label{eq4} V_{\rm HL} =\frac{k_0 T}{q} \ln \frac{N_{\rm EPI} }{N_{\rm LP} }. \end{equation}$$ $

(4)

Being different from the ordinary buried implanting, the LP region can be implemented by using two-step distinguishable high energy N-type implants with one identical mask to compensate the acceptor impurities in the EPI layer according to the principle of impurity compensation.

It can be known from Eq. (3) that making the doping concentration of the LP region as low as possible is an effective way to obtain a stronger drift force originating from the electric field. However, if the doping concentration of the LP region is so low that close to the critical N-type doping, this may cause a meaningless N-type stretch and reduce the depletion capacity of the overall N-type region, resulting in an obviously raising of the pinch-off voltage^[14] from $V_{\rm pinned}$ to $V'_{\rm pinned}$. Doing so, the bottom of the potential well in the PPD may get lower than the potential of the transfer gate (TG) channel stated in "on", then a potential barrier which can prevent the photon generated electrons transferring from the PPD to the FD is produced, leading to a part of the electrons residue in the PPD in the charge transferring period as illustrated in Fig. 3. This will result in an image lag issue. So taking account of the charge transfer efficiency, the LP region should not be inversed to N-type doping.

The LP design can make electrical crosstalk insensitive without any pixel area or fill factor penalization. In order to make further efforts on improving other major photoelectric performances such as FWC or QE, a buried N-type (BN) region can be implemented by a higher energy N-type implant with a medium doping dose. As illustrated in Fig. 2(b), the additional mask of the BN implant is in the interior of the peripheral LP "arms", resulting in a stretched N-type region inside of the LP region, and jointly constituting an inverse U-shape profile with the LP "arms". As illustrated in Fig. 2(b), the enhancement of the overall capacitance by introducing the new BN region can be composed by the bottom capacitance and stretched side wall capacitance. Therefore, the capacitance enhancement between the architecture with and without the BN region can be given as:

$ $$\begin{equation} \label{eq5} C_{\rm BN} =C_{\rm P} \Delta P+C_{\rm A} \Delta A, \end{equation}$$ $

(5)

where $\Delta P$ and $\Delta A$ respectively represent the increased side wall and bottom PN junction area due to the BN implant; $C_{\rm P}$ and $C_{\rm A}$ respectively represent the unit side wall and bottom PN junction capacitance of the BN region and can be given as:

$ $$\begin{equation} \label{eq6} C_{\rm P} =\left( {\frac{\varepsilon _{\rm r} \varepsilon _0 q}{2} \frac{N_{\rm LP} N_{\rm BN} }{N_{\rm LP} +N_{\rm BN} }} \right)^{1/2}\left( {V_{\rm bi1} -V_1 } \right)^{-1/2}, \end{equation}$$ $

(6)

$ $$\begin{equation} \label{eq7} C_{\rm A} =\left( {\frac{\varepsilon _{\rm r} \varepsilon _0 q}{2} \frac{N_{\rm EPI} N_{\rm BN} }{N_{\rm EPI} +N_{\rm BN} }} \right)^{1/2}\left( {V_{\rm bi2} -V_2 } \right)^{-1/2}, \end{equation}$$ $

(7)

where $\varepsilon_{\rm r} \varepsilon_{0}$ is the dielectric constant of silicon material, $q$ is the unit charge, $N_{\rm BN}$ represents the doping concentration of the BN region, and $V_{\rm bi1}$ and $V_{\rm bi2}$ represent the built-in potential respectively between two regions of the side wall junction and bottom junction. $V_{1}$ and $V_{2}$ represent the applied voltage respectively between each of the two regions.

It must be remarked that the signal charges, which can be read out in a 4T pixel, are the electrons that have been transferred to the FD node. The effective FWC should be defined as the maximum number of electrons that can be transferred out from the PPD. By introducing the BN region, the neutral zone in the new stretched N-type region will be difficult to exhaust, resulting in a raising of the pinch-off voltage. So according to the image lag issue mentioned previously, the doping dose of the new BN region should be lower.

The mechanism of the QE improvement induced by introducing the BN implant can be simply described as the result of the drifting volume for absorbing photon generated electrons is expanded by the vertical SCR expansion. Note that here QE should be exactly the internal quantum efficiency (IQE) in which the loss of incident light by the reflection of the lens or the dielectric layer on the top of the silicon surface could not be taken into account. IQE was proved to be positively correlated to the drift current density $J_{\rm drift}$ in the SCR, as stated in Eq. (8) for the one-dimensional case^[15]:

$ $$\begin{equation} \label{eq8} {\rm IQE} \propto J_{\rm drift} =-\int_{x_{\rm j} }^{x_{\rm j} +w+\Delta w} {G(x){\rm d}x}, \end{equation}$$ $

(8)

where

$ \Delta w=\left( {\frac{2\varepsilon _{\rm r} \varepsilon _0 }{q} \frac{N_{\rm EPI} +N_{\rm BN} }{N_{\rm EPI} N_{\rm BN} }V_{\rm bi2} } \right)^{1/2}, $

where $\Delta w$ represents the expanded vertical widths of the SCR with the BN region. It is observed that the IQE can be improved by a higher dose of BN doping. However, according to the image lag issue mentioned previously, the doping dose of the BN implant would not be so high. So there must be a tradeoff to consider for the BN design.

2.2.2   Implementation of the architecture

The proposed architecture involves multi-buried implantations, causing a number of noteworthy process implementation issues that should be analyzed.

First, the ion species of the N-type implant for the LP region creating must be selected with the atomic number smaller than ON-ion (generally arsenic) to possess a deeper incident range^[16], in order to make the most of the LP doping ions stopping in the buried position deeper than the bottom of the ON region and compensate the EPI acceptor impurities during the implanting process. So we select phosphorus as the LP doping ion species.

Second, an appropriate dose for N-type implantation which dominates the doping concentration in the LP region is needed to control the region without inversing. An empirical implant dose with a magnitude of 10$^{11}$ cm$^{-2}$ is obtained from the deep sub-micrometer (DSM) bulk CMOS technology to completely compensate the acceptor impurities in the EPI layer with a concentration of 10$^{15}$ cm$^{-3}$. Therefore, Figure 4 simulates the "peak" LP doping concentration produced by N-type implantation in the EPI layer versus the implant dose increasing from 1 $\times $ 10$^{11}$ to 9 $\times $ 10$^{11}$ cm$^{-2}$ by the TCAD process tool. It is observed that, as the implant dose increases, the P-type doping level is going to be lighter and inversed to N-type approximately at the dose of 3.2 $\times $ 10$^{11}$ cm$^{-2}$. So by taking account of the general dose implantation deviation existing in an actual ion implanter, the maximum LP implant dose we select should be a little lower than 3 $\times $ 10$^{11}$ cm$^{-2}$ to ensure there is no inversing in an actual implanting process. In addition, the minimum LP implant dose of 2 $\times $ 10$^{11}$ cm$^{-2}$ is carefully designed simultaneously to establish a sufficient built-in electric field for the effectively photon generated electrons absorption. Therefore, the LP implant dose window is determined between the value of 2 $\times $ 10$^{11}$ and 3 $\times $ 10$^{11}$ cm$^{-2}$.

Third, due to the buried implanted impurities that will diffuse around after annealing, the mask open areas of the IUBD insertions ($X_{\rm LP}$ for LP insertion and $X_{\rm BN}$ for BN insertion shown in Fig. 5) and the spacing between them should be all optimized to ensure that the LP region not be eaten up by the latter diffused BN region, and to maintain the depletion capacity of the overall N-type region. Based on the analysis above, for a certain PPD with a size of 2.3 $\times $ 2.3 $\mu $m$^{2}$, here we set $X_{\rm LP}$ and $X_{\rm BN}$ to 0.3 $\mu $m and 0.8 $\mu $m with a proper spacing of 0.2 $\mu $m, respectively. A layout design of the proposed PPD pixel is shown in Fig. 6 with the in-pixel transistors.

Based on the discussions of the design and the implementation of the IUBD architecture, the technology flow and process parameters to implement the architecture beneath the ON region of a conventional PPD is listed in Table 1.

According to the parameters, a two-dimensional IUBD PPD model combined with the TG transistor is simulated by the TCAD process tool, as shown in Fig. 7. The doping concentration comparisons between the conventional and the IUBD PPD along the AA' and BB' cross sections are respectively shown in Figs. 8(a) and 8(b). It is observed that the "peak" doping concentrations of LP1 and LP2 regions are both lower than the EPI layer by almost two orders of magnitude, knowing that an approximately 100-200 mV built-in potential can be established for the electrons drifting back to the photosensitive area according to Eq. (4). The BN region successfully follows the doping gradient of the ON region and stretches a small distance without disturbing the peripheral LP region to maintain the pinch-off voltage at a similar level as the conventional PPD by taking the image lag issue into account.

3. Results and discussion

Based on the DSM CIS dedicated technology, the neighborhood IUBD model for evaluating electrical crosstalk and other performances is developed by the TCAD process tool, as shown in Fig. 9. The operating voltage is 3.3 V. The FD nodes are respectively linked to PPD1 and PPD2 by a TG transistor with the gate length of 0.35 $\mu $m to supply bias for PPD resetting and charge transferring. The simulated size of the IUBD PPD is 2.3 $\times $ 1.0 $\mu $m$^{2}$. The spacing between the neighboring IUBD devices is 1.0 $\mu $m. We suppose that no other pixel devices or circuits exist in the area between PPD1 and PPD2 in order to guarantee that the model is in the worst case for electrical crosstalk, since there are no electrical or physical isolations. In order to facilitate the comparison, there are an additional three models created with different structures, as listed in Table 2.

Optical simulations are made by using a finite difference time domain (FDTD) tool to be coupled to TCAD numerical simulations^[17]. The information we want to extract from the optical simulations is the optical generation field defined as the photon generation rate of electron/hole pairs in silicon and given by:

$ $$\begin{equation} \label{eq9} G_{\rm opt} =\frac{-{\rm Re}\left( {\nabla \cdot \boldsymbol{P}} \right)}{h\nu }, \end{equation}$$ $

(9)

where $\boldsymbol{P}$ is the Poynting vector with the unit of W/m$^{2}$, and $h \nu$ is the photon energy. Then, $G_{\rm opt}$ is injected in the carrier's continuity equations as a generation term for the subsequent simulations.

3.1 Electrical crosstalk rate

For the evaluation of the electrical crosstalk rates of the four models, we have made one of the neighboring PPDs (for example PPD1 in Fig. 9) under a monochrome incident beam while the other under darkness. The integration time and the beam intensity are respectively set to 10 $\mu $s and 0.05 W/cm$^{2}$, ensuring the excited electrons can fill the capacity to saturation. The direction of the incident beam is vertical to the silicon surface. Reading the electrical crosstalk rates, we have excited the models with various wavelength simulation conditions. Doing so, the crosstalk charges can be read at the dark pixel just before the saturation of the illuminated pixel.

Figure 10 shows crosstalk rates (the ratio of the number of collected charges in the dark pixel to the number of collected charges in the illuminated pixel when the illuminated PPD just achieves saturation) comparison of the four models and the corresponding crosstalk suppression ratios (the ratio of the decreased crosstalk rate in optimized models to the crosstalk rate in conventional PPD) of the three optimized models. Note that, all the models show extremely low crosstalk rates at wavelengths below 450 nm (blue light), rapid growths at wavelengths between 450 nm and 650 nm (green light) and stability linear growths at wavelengths beyond 650 nm (red light). The reasons can be explained as the following: the absorption depth of the blue wavelength is near the surface of silicon, so the excited electrons can be rapidly collected by the electric field in SCR with little diffusion; as the wavelength increases, the absorption depth also increases, resulting in a large generation of electrons in the EPI layer and a rapid growth of the crosstalk rate; most electrons excited by the red wavelength are probably quickly recombined by the SRH mechanism in the high doping substrate and can rarely contribute to the crosstalk charges, resulting in a gently linear growth of the crosstalk rate. Figure 11 shows the simulation curves of the SRH recombination rate along the $Y$-axis of the model 1, it can be observed that the red wavelength causes a much larger SRH recombination rate than green and blue especially at the tails of the curves. That means the excited electrons by red wavelength illumination can hardly diffuse out from the substrate.

As shown in Fig. 10, it can also be observed that clear improvements of crosstalk performance have been achieved gradually by introducing the proposed optimization designs from model 2 to the IUBD at any wavelength. Ultimately for the IUBD architecture, the crosstalk rate has been suppressed by 60%-80% compared to the conventional PPD ahead of the model 2 and model 3 at wavelengths beyond 450 nm. The phenomenon indicates that the introduction of the BN implant can further reduce the electrical crosstalk rate attributed to the SCR expansion.

Figures 12(a) and 12(b) respectively show the electron current density distributions of both the model 1 and the IUBD at the wavelength of 700 nm when the illuminated PPDs achieve 50% saturation. The clear U-shape distribution of the electron current density with the depth of about 2 $\mu $m can be observed in the IUBD, indicating that an electric field introduced by the LP region is successfully created, and electrons dropped in the EPI layer can be drifted back to the upper SCR. The electron current density at point A near the edge of the dark pixel SCR is 0.028 mA/cm$^{2}$ in the IUBD, which shows as more than 20 times lower than 0.66 mA/cm$^{2}$ in the model 1, indicating the electrons dropped in the EPI layer of the IUBD PPD diffuse with more difficultly to the neighboring SCR than that of the conventional one, and the feasibility of the proposed crosstalk suppression design is proved.

3.2 Full well capacity

Figure 13 shows the comparisons of the four models' FWCs and pinch-off voltages. It can be observed that the FWCs of both model 2 and model 3 have no significant growth compared to model 1, keeping the number of less than 5000e$^{-}$. The pinch-off voltages are also maintained at about 1.2 V. It is proved that there is no influence on the depletion capacity of the ON region by introducing the LP implants. A remarkable growth of FWC happened in the IUBD, the growth rate reaches to 107.5% compared to model 1. However, the growth rate of the pinch-off voltage is only 32.5%, much smaller than that of FWC. It is proved that, by introducing the BN implant, FWC get greatly improved, which is attributed to the increasing of the overall junction capacitance. However, the depletion capacity of the whole N-type region is not out of control thanks to the good design of the junction depth and the concentration of the BN implant.

Figure 14 shows the residual electron densities comparison between the model 1 and the IUBD along the B-B' section marked in Fig. 7 after reading out the operation. It can be observed that, the maximum residual electron density in the conventional PPD drops from 2.05 $\times $ 10$^{16}$ to 2.22 $\times $ 10$^{8}$ cm$^{-3}$ after transferring, indicating that the electrons have been completely transferred out from PPD. Note that the maximum residual electron density in the IUBD PPD after transferring drops to 1.23 $\times $ 10$^{9}$ cm$^{-3}$, which can be compared to the level of model 1. Similarly, a conclusion of completely transferring in the IUBD pixel can be derived. It is proved that the introduction of the BN implant does not make the bottom of the potential well in the PPD be lower than the potential of the TG channel stated in "on", effectively stopping the image lag from happening. As a result, the design of the BN implant achieves FWC enlargement without any image lag issue.

3.3 Internal quantum efficiency

IQE which represents the ratio of the number of photon generated electrons absorbed in the PPD to the total number of incident photons in the silicon substrate is simulated both in the model 1 and the IUBD. The incident wavelength ranges from 400 to 850 nm with steps of 50 nm.

Figure 15 shows the IQE improvement comparing the IUBD with model 1. Note that clear improvement of the IQE can be observed in the IUBD at the wavelengths between 400 and 600 nm (approximately blue or green light), especially at 450 nm, the improvement rate can reach to 8.63%. It is indicated that the SCR expansion induced by the introduction of the BN region plays a remarkable role for the absorption of the electrons excited in this wavelength range. However, IQE at the wavelength beyond 600 nm does not increase as much as the blue or green wavelength since the electrons excited by the deep-penetrating photons have been quickly recombined by the SRH mechanism in the substrate as the theory analyzed previously, and can hardly be absorbed by photosensitive SCR.

3.4 Dark current

The multiple high energy doping implants introduced in the IUBD design may cause a number of extra lattice defects in bulk silicon, especially the surface^[16], even after successive multi-step annealing flows. The inherent surface P+ implant of the conventional PPD process followed by all the IUBD implants can effectively suppress the dark current originated from the surface defects. Even though there is still a dark current originating from the deep defects in the EPI or the substrate, we can adopt a widely used technique to eliminate the residue dark current portion by fabricating a number of dark pixels in the peripheral of the effective pixel array with an identical fabricating process^[18]. By subtracting the dark signal reference produced in the dark pixels from the read-out signal with the subsequent digital signal processor (DSP), the dark current can be eliminated.

4. Comparison

According to the simulation results, a comparison of the characteristics between the conventional and the IUBD design of PPD is listed in Table 3. Note that the wavelengths we choose for the comparisons of crosstalk rate and IQE are respectively from 550 to 750 nm with steps of 100 nm and from 450 to 650 nm with steps of 100 nm, since optimization of performances by the IUBD design can typically be reflected in these wavelength ranges.

The proposed design is representatively compared in Table 4 with the most prevalent DTI technology, which has been demonstrated as one of the best crosstalk suppressors due to its perfect barrier property compared to other technologies^[7-9]. As it can be seen in the table, the crosstalk rate suppression of the IUBD is greatly larger than the DTI design at the typical wavelength of 620 nm under both the conditions of the DTI depth increased by 30% or 50%. There is no FWC improvement in the DTI design since there is no modification in the side wall or bottom junction capacitances. The QE can be improved in both designs approximately by similar levels. So the superiority of the proposed design is exhibited.

5. Conclusion

In this work, we have proposed the design of an inverse U-shape buried doping PPD architecture for electrical crosstalk suppression and performance optimization in CMOS image sensors without any extra fill factor consumption. The photoelectric characteristics of the novel architecture are carefully studied. The purposes of the design are demonstrated by FDTD simulation results, indicating that the electrical crosstalk rate between the neighboring pixels is substantially reduced in a specific range of the green or red spectrum. Other major photoelectric performances such as IQE and FWC are clearly improved by different levels. Furthermore, the image lag performance can be sustained to a perfect low level, as in the conventional design. The comparison with the relevant reference shows the superiority of the proposed design. This IUBD architecture is expected to improve the photoelectric performances of modern ultra high resolution image sensors.