Simulation and Implementation of a Filter-less CMOS Color Detector Graciele Batistell and Johannes Sturm Carinthia University of Applied Sciences
Villach, Austria Email:
[email protected],
[email protected] Abstract - An standard CMOS color detector is presented based on vertically and laterally arranged photodiodes, providing color separation based on lateral carrier diffusion and wavelength-dependent absorption-depth. The optical and electrical simulations are performed and the interferences caused by the Oxide/Nitride stack are analyzed. The proposed structure is implemented in a standard 130 nm CMOS technology. Three independent detector output signals with optimized spectral responsivities allow a discrimination between red, green and blue light spectral components. In the presented solution color separation can be achieved without modifications of the CMOS process or color filters. The color detector is therefore a low cost alternative solution for various color sensing applications.
1
INTRODUCTION
The growth of the mass market applications like Complementary metal-oxide-semiconductor (CMOS) cameras, Liquid-crystal display (LCD) displays, and mobile devices increased significantly the demand of low cost integrated sensors. Other possible fields of application are the ambient light sensing for general lighting, and screen back-light adaptive control depending on the daylight or to compensate aging effects. Color sensing in commercial and industrial applications is usually based on Bayer Color Filter Arrays (CFA) [1] or filter-less color sensors. In CFA’s the photo sensor has almost constant light sensitivity over a wide spectral range, while the color sensitivity is provided by filters added on top of the sensor. The use of CFA has some drawbacks. Due to the filters, just part of the incident light is transmitted to the photodiode, decreasing significantly the overall sensitivity of the sensor. Another factor to be considered is that at each pixel the information of a single color is read, while the other two must be obtained by interpolation algorithms. Furthermore CFA’s require complex filter assembly on top of the sensor which significantly increases the production costs. Alternatively to CFA’s, an implementation of filter-less sensor is proposed in the Foveon X3 Technology [2]. This technology is based on the wavelength dependent light absorption properties of silicon and has a stack of three optimized photodiodes, where each one absorbs the generated carriers related to different light colors. In this color sensor the red, green and blue light intensity can be measured, avoiding absorption losses and it achieves therefore better quantum efficiency than sensors with color filters. Even though the triple-junctions are usually available in modern CMOS technologies, the spectral responsivity of the 3 stacked photodiodes is
directly defined by technology parameters like doping concentrations and profiles and there is almost no room for optimization without expensive technology modifications. Also for older CMOS technologies a triple-well option is not always available. To gain higher flexibility for spectral response optimization, avoiding technology modification as well as triple-photodiode stacks and external color filters, an alternative sensor solution is presented in this paper. The proposed sensor structure consists of a vertically stacked double-photodiode and a single laterally arranged photodiode fully covered by metal. Color separation between the three photodiodes is achieved by lateral carrier diffusion together with wavelengthdependent absorption. In Chapter 2, the proposed color detector structure and its principle of operation is presented. Chapter 3 shows device and process simulations of the sensor structure including the spectral light response and the modeling and simulation of the oxide/nitride stack. The simulations are based on Synopsys TCAD tools. Finally in Chapter 4 the measurement results of the color sensor are reported. 2
PROPOSED COLOR SENSOR STRUCTURE
According light colorimetric theory, a color measurement requires at least three spectral independent sensor signals with opponent color sensitivity to accurately represent an R, G, B color space [3]. Ideally the three sensors should be matched to the spectral sensitivity of human eye photoreceptors. A main target of the proposed color sensor is the implementation in a standard, low-cost CMOS technology without any process modification. The theory of color separation and the proposed sensor structure, realized and optimized only by layout modifications, are presented in this chapter.
2.1
Color separation in standard CMOS
in Figure 2. In addition to the stacked double photodi-
In the proposed structure the color separation is based on the silicon properties of wavelength dependent light absorption in different depths, together with lateral carrier diffusion effects. A structure with two vertically stacked photodiodes is shown in Figure 1 (a) [5].
Blue light
NWD
Red light
NWL - - - - -- - - - - - - - -
NWD
NWL
-
PPNW
NW P+
N+
PD1
P-sub
Carrier Generation
P-substrate
P+
(a)
Blue PD2
- - - - - - - - - - - -
- - -
-
-
-
P-substrate (b)
N-well Red
(a)
(b)
Figure 2. (a) Lateral carrier diffusion for short wavelength light. (a) Lateral carrier diffusion for long wavelength light. (b) Photodiodes spectral response. Figure 1. (a) Color sensor with double-photodiode. (b) Carrier generation rate versus Silicon depth. (c) Spectral response.
The first PN-junction (PD1) is formed between a shallow P+ implant and the N-well. The second PNjunction (PD2), which is vertically stacked below the first photodiode, is created between the N-well and the P-substrate. This configuration can be realized in any standard CMOS process. The Figure 1 (b) shows the wavelength dependent carrier generation rate. For blue color light with a wavelengths of about 400 nm to 500 nm, the light absorption coefficient of silicon is very high and the carrier generation mainly happens near the silicon surface. Therefore most of the carriers will diffuse or drift to the upper PN-junction PD1 which result in an increase of the photocurrent at PD1. For red light with wavelengths above 600 nm, there is high carrier generation even deep in the silicon, so the carriers are mainly collected by junction PD2 [5]. The two photodiodes, PD1 and PD2, have different spectral responses. PD1 presents a higher sensitivity for low wavelengths and PD2 has very broad response, as shown in Figure 1 (c). The current at the terminal PPNW is equal to the current of PD1 while the current at the terminal NW is equal to the currents of PD1+PD2. Even though the two diodes have quite different spectral responses, an accurate color measurement requires at least three spectral independent sensor signals to represent an R, G, B color space [3]. In the proposed structure a third spectral response is obtained by lateral carrier diffusion effects, as indicated
ode from Figure 1 (a), a third photodiode fully covered by metal is arranged beside it. As already discussed in Figure 1 (b) the photon absorption and carrier generation in different silicon depths is depended on light wavelength. The carrier drift and diffusion in semiconductors is defined by a local electrical field and a carrier density gradient. The carrier diffusion within the proposed color sensor for blue and red light is illustrated in Figure 2. When blue light is applied to the sensor, Figure 2 (a), the carriers are generated near the surface and are mainly collected by the photodiode NWL. When red light is applied, Figure 2 (b), a considerably higher number of carriers will diffuse also to photodiode NWD. The photodiode NWL has a wide spectral response, while NWD have an increased sensitivity for higher wavelengths, as shown in Figure 2 (c). 2.2
Proposed structure
The proposed structure is presented in Figure 3 (a). It is implemented in a P-substrate connected to ground through P+ implants (P sub ). An N-well (NWD) is covered by metal to shield the diode from direct light exposure. A wider N-well (NWL) without metal shield is placed in the center. A P+ implant (PPNW) is placed inside the uncovered N-well creating the doublephotodiode. The P+ and N+ are source/drain implants of a standard 4-metal 130 nm CMOS technology. The standard process flow without any technology modifications was used. The total width of the structure is 25 µm. The structure is laid out as square cell of 25 µm x 25 µm. The top view of the sensor unit cells is shown
in Figure 3 (b) where the green line represents the point of the cross-section.
NWD
Psub
N+
P+
N-well
P-well
NWL
PPNW
N+
P+ N-well
PPNW
NWL
N+
Psub
NWD
P+
N+
P-well
N-well
The optical intensity, the absorbed photon density, and the optical generation rate were calculated using Raytracer as optical model, which is optimized for high simulation speed. After the optical generation has been computed this value is added to the carrier continuity equation as a generation rate. After finding the initial solution, the quasi-stationary equations are solved by ramping the wavelength of an ideal monochromatic light source from 400 nm to 1000 nm. The SRH recombination model is used and the carrier mobility is defined as doping dependent. The complex refractive index in defined as wavelength dependent. All simulations were done with ideal electrical contacts biased at 0 V dc-voltage. All simulations are based on 2-dimensional solvers, due to the high simulation time of 3-dimensional models. The performed 2D simulation presents some differences of the real 3D structure. The correction of the responsivities is estimated by simple geometric comparisons, where the increase of the photodiode area are taken as linear increase of the responsivity. The area of the 3 photodiodes as well as the metal shield are corrected according to the 3D values. 3.2
Figure 3. (a) Cross-section of the proposed structure. (b) Top-view of the proposed structure.
3
The simulated current density of the structure for a wavelength of 800 nm is presented in Figure 4. Although the light is only falling on the double photodiode structure, there is a significant electron current density at the covered N-well due to lateral carrier diffusion, as discussed in Figure 2.
DEVICE SIMULATION
This chapter presents the the methods and tools used for the investigation of color detector structure as well as the simulation results. 3.1
Simulation results
NWD Psub NWL
PP
NWL Psub NWD
Methods and tools
The photodiode device and process simulations were performed with Synopsys TCAD Environment. Synopsys TSUPREM-4 is used to perform a 2-dimensional simulation of CMOS production process. Steps like ion implantation, oxidation, deposition and etching are simulated. The existing process data of the 130 nm CMOS technology was used, in order to reproduce a realistic behavior of the color detector structure. The optical and electrical simulations are performed with Sentaurus Device using the drift-diffusion transport model, that is defined by the basic semiconductor equations. The electrostatic Poisson and the carrier continuity equations for electrons and holes are solved using the Newton method.
Figure 4. Simulated electron current density.
The detector responsivity simulations are presented in Figure 5. As expected, three independent spectral responses are obtained for NWL, PPNW and NWD. The responsivity maximum shifted to low wavelengths for PPNW, to high wavelengths for NWD and with a broad spectral response for NWL. By a scaled subtraction of NWD and PPNW from NWL
Simulation Results 0.25
simulation results with Raytracer, the solid lines represent the simulated responsivity obtained with BPM optical solver.
PPNW NWD NWL
0.2
0.15
0.2 0.1
Responsivity [A/W]
Responsivity [A/W]
Simulation Results PPNW NWD NWL
0.25
0.05
0 400
500
600
700 Wavelength [nm]
800
900
0.15
0.1
1000
0.05
Figure 5. Photodiodes responsivities simulated with Raytracer optical solver. 0 400
a responsivity with a maximum at about 600 nm can be achieved as shown in Figure 5. Due to metalization processes, an oxide/nitride layer stack is created on the sensor surface, which is not modeled by the Raytracer optical solver. Therefore an optical interference is expected in the real device, caused by different complex refractive indexes. Even though the presented structure contains just double and lateral photodiodes, the responsivities are comparable to stacked photodiodes presented in [6], [7], and [8]. The structure dimensions are optimized to obtain the appropriated ratio between the three responsivities. The total sensor area can be scaled with minor changes of the responsivity. Since only standard process steps are used, the oxide and oxide/silicon-interface charges cause no leakage or inversion problems. To avoid edge effects, the silicon substrate depth was increased to 100 µm for simulation. 3.3
Oxide/nitride Stack
The modeling and simulation of the color sensing structure previously presented reproduces the characteristics of the produced sensors, except for the diffraction and interference effects caused by the oxide/nitride stack created during the metalization processes. The oxide/nitride stack can be observed on Figure 4. The simulation without the interference caused by the oxide stack is sufficient when investigating structures with application in sensing wide spectrum range light, as for example sensing the color of the ambient light. But considering applications where single wavelengths or narrow bands are to be sensed a more precise simulation is required. A better model of the color sensor can be achieved by using the BPM optical solver, the simulation result is presented in Figure 6. While dashed lines are showing
500
600
700 Wavelength [nm]
800
900
1000
Figure 6. Comparison of photodiodes responsivities simulated with Raytracer and BPM optical solvers.
When the BPM optical solver is used the expected interferences can be observed. The average values of both simulations are comparable. The modeling of the interferences become very complex, since the thickness of the isolation layers has a large variation over process and the interferences are strongly dependent on the thickness of these layers. Furthermore the BPM simulation is very sensitive to boundary conditions. 4
MEASUREMENTS
A testchip was produced in a 130 nm CMOS technology, consisting of a 10 x 10 photodetector array with all detectors connected in parallel as shown in Figure 7. Each unit cell has dimension of 25 µm x 25 µm (See Figure 3 (b)) leading to an area of 250 µm x 250 µm for the complete sensor. 4.1
Measurement setup
The measurement setup includes a Newport Apex 70613NS illuminator and an Oriel Cornerstone 130 monochromator. The photodiode currents are measured by Keithley sourcemeters. A Newport power meter Model 2936 is used to measure the optical power falling on the sensor as a reference in order to calculate the sensors responsivity. 4.2
Measurement results
The measured spectral responsivities of the 3 photodiodes are presented in Figure 8. The thick lines show the measurement results while the thin lines are the simulation results. The average response shows good
tion, is presented. The color sensor structure includes three photodiodes with spectral independent reponsivities, produced in a 130 nm standard CMOS technology with no process modifications. Device and process simulations were performed, showing good agreement of the photodiode responsivity between simulation and measurement. It could be demonstrated, that the proposed structure provides three independent spectral responses, distributed over the whole visible wavelength range, which can be used for R,G,B color recognition. Therefore the structure is a potentially competitive solution for low-cost integrated color sensors. Acknowledgments
Figure 7. Picture of the Testchip
agreement between simulation and measurement, even though the measured optical interferences do not match exactly to the simulations. This mismatch is caused by the the large process variation of the oxide and nitride layers and the strong dependence of the interference simulation on the layers thickness.
The authors would like to thank Infineon Technologies Austria AG for assistance and support in testchip production. This work was supported by FFG (FIT-IT) funded project COSMOS. References [1] B.E. Bayer, Color Imaging Array, US Patent 3,971,065, 1976.
Responsivity V1 0.3
0.25
[2] R. B. Merrill, Color Separation in an Active Pixel Cell Imaging Array Using a Triple-Well-Structure, US Patent 5,965,875, 1999
PPNW NWD NWL
[3] N. Ohta, A. R. Robertson, Colorimetry - Fundamentals and Applications, John Wiley & Sons Ltd, 2005.
Responsivity
0.2
0.15
[4] H. Zimmermann, Integrated Silicon Optoelectronics, Springer Verlag, Berlin, p.329, 2000.
0.1
0.05
0 400
500
600
700 Wavelength [nm]
800
900
1000
Figure 8. Comparison of measured photodiodes responsivities and BPM optical solver.
The process variation of the responsivities was not analyzed. Nevertheless a variation due to the oxide interferences has a minor effect on the average responsivity, since the absorption of the oxide/nitride layers is small and the sensor currents are proportional to an integration over the wavelength. The measured PN-junction capacitances are between 12 pF and 16 pF for a reverse bias voltage of 2 V. The series resistance of the diodes is between 45 Ω and 60 Ω. The photodiode dark current is below 24 pA. 5
CONCLUSION
A new method for color separation, based on lateral carrier diffusion and wavelength-dependent absorp-
[5] J. Sturm, S. Hainz, G. Langguth, H. Zimmermann, Integrated Photodiodes in standard BiCMOS technology, Proceedings of SPIE 4969, 109, 2003 [6] A. Polzer, W. Gaberl and H. Zimmermann. FilterLess Vertical Integrated RGB Color Sensor for Light Monitoring presented at the 34th MIPRO, Opatija , 2011. [7] K.M. Findlater, D. Renshaw, J.E.D. Hurwitz, R.K. Henderson, M.D. Purcell, S.G. Smith and T.E.R. Bailey.A CMOS Image Sensor With a Double-Junction Active Pixel IEEE Transactions on Electron Devices, 2003 [8] G. de Graaf and R.F. Wolffenbuttel. Illumination Source Identification Using a CMOS Optical Microsystem IEEE transactions on instrumentation and measurement, 2004.
