Journal of the Korean Physical Society, Vol. 52, No. 2, February 2008, pp. 487491 Design and Simulation of a Silicon Photomultiplier Array for Space Experiments H. Y. Lee, J. Lee, J. E. Kim, S. Nam, I. H. Park and J. H. Park Department of Physics, Ewha Womans University, Seoul 120-750 (Received 26 November 2007) A new type of photon sensor known as the silicon photomultiplier (SiPM) is a matrix of tiny avalanche photodiodes operating in the Geiger mode. The output of a SiPM sensor is the sum of signals from all `binary' diodes of the sensor, providing the capability for single-photon counting when illuminated with a low intensity of photons. All the features of the SiPM provide a promising photon sensor for space applications that require low power, small size and weight, less sensitivity to a magnetic eld, etc. For the image sensor to be applied in a future space telescope, we have designed a 4 4 array of SiPM sensors, each consisting of 32 32 photodiode micropixels with a size of 30 30 m 2. The design scheme and a simulation of the performance are presented. PACS numbers: 85.60.Gz, 29.40.Wk Keywords: Silicon photomultiplier, Photodiode, Geiger mode I. INTRODUCTION The silicon photomultiplier (SiPM) that was developed in 1998 and 2000 rst by Russian groups [1{3] is a promising device that will eventually replace the PMT (photomultiplier tube) in various scientic and industrial applications, including medical instruments [4]. The sensor consists of many subcomponents (micropixels) of silicon diodes with PN-junctions where a high electric eld of about 10 5 10 6 V/cm is formed by a reverse bias above the breakdown voltage of the junctions. Ionizations induced near the junction by incoming photon(s) initiate an avalanche process that results in a high amplication factor of up to 10 5 10 6. The avalanche discharge current is limited by a resistor connected in series with the junction, which is called a self-quenching process that stops the discharge at the breakdown voltage of the junction and restore the diodes back to the operating bias. As a result, each micropixel diode operates as a \binary device", producing a certain amount of signal independent of the number of photons incident on the pixel. A key operation requirement of the SiPM is that each micropixel should be electrically isolated from neighboring micopixels. The condition is implemented by using quenching resistors contained in every micropixel and by the geometrical gap between micropixel diodes. The readouts of all individual micropixels are combined into a common output so that a SiPM output is a linear superposition of signals from all the diodes of the sensor. E-mail: swnam@ewha.ac.kr; Fax: +82-2-3277-3415 -487- The SiPM provides many nice features. Many of them are from the nature of semiconductor sensors that use a small space for signal generation. Since the avalanche occurs in the thin layer of the PN-junction, the eect of the magnetic eld is negligible. The size and the weight of the sensor are much smaller than those of a PMT while keeping the level of signal amplication at a similar level. A moderate bias voltage (below 100 V) is enough for operation and results in a low consumption of electric power. The radiation hardness is also known to be very strong. The most attractive feature of the sensor is its excellent capability for single-photon counting, which is a direct result of the uniformity over all identical micropixels. The number of micropixels red at the same time is proportional to the total number of incoming photons in the condition of low intensity of the incident photon ux. All these features are just what make the sensor a very suitable device for space experiments. For example, the space telescope that is being developed for the measurement of weak uorescence signals generated from extensive air showers of high-energy cosmic rays requires photon sensors with characteristics such as these described above [5]. When a large number of sensor channels are needed in space applications for the image sensors, the preference of SiPM to PMT grows even bigger because of the lower cost in the fabrication. We have designed an array of SiPM sensors (commercially unavailable at the moment) for use in the future space experiment as a key component of the focal plane photo-receivers in various telescope designs [6]. The next section describes the design of a SiPM and the array. The simulated performance of the sensor is presented in the
-488- Journal of the Korean Physical Society, Vol. 52, No. 2, February 2008 Fig. 1. Vertical cross section of a micropixel diode of a SiPM sensor (The plot does not re ect the real scale). Fig. 3. A corner of a sensor with 32 32 micropixels. Fig. 2. Horizontal structure of a micropixel design. A micropixel has a size of 30 30 m2, including the trench of 3 m width in the four sides. following section. II. STRUCTURAL DESIGN OF A SIPM AND THE ARRAY Our design is a 4 4 array of SiPM sensors. Each sensor in the array consists of 32 32 micropixels of diodes. Figure 1 shows the vertical structure of a micropixel formed in an epitaxial layer with a thickness of a few micrometers on a silicon-wafer substrate. The vertical structure has a n+/p+/p diode con guration. First, the epitaxial layer is a p-type layer doped with boron. As shown in the gure, a PN junction is constructed just below the surface of the layer. The n+ layer has a depth of 300 500 nm, and the p+ layer reaches 1 or 2 micrometer further down from the junction. The surface is covered with a thin layer of SiO2, except for the region for the metal contact needed for the signal readout. The contact is then connected immediately with a strip made of poly silicon, which provides a resistance for the quenching process. Since the poly silicon layer blocks incident light, the length of the resistor is designed to be the minimum possible while keeping the level of the resistance needed for the quenching process, which is 1 M. The boundary of the micropixels is trenched with a width of 3 um for the geometrical gap needed for the electric isolation of diodes. The trench is then lled with a polyimide material that prevents propagation of secondary photons generated in a micropixel into neighboring diodes during the avalanche process. The horizontal structure of the design is shown in Figure 2. In the gure, the white circle is the contact, hole for the metal contact, and the green shadowed region over the hole is the metal layer connecting the resistor strip and the PN junctions through the contact hole. The strip of the poly silicon resistor is then connected to the aluminum line for signal readout, combining all micropixels to a common output of the sensor. The p+ and the n+ layers shown in Figure 1 are also drawn in the gure. They are formed under the SiO2 layer with a the thickness of 1 m, which is not shown in the gure. The n+ implantation layer is larger than the p+ layer. The surface of the sensor is covered with an antire ective material. The structure connecting micropixels to a common readout pad is shown in Figure 3. This is a corner of a sensor with 32 32 micropixels. There are 32 aluminum lines connecting all the diodes to the readout pad. The geometrical acceptance is given by the ratio of the area that is transparent to incoming photons to the size of a micropixel. The area blocking the light is the resistor strip (10 %), the trench around the PN-junction (15 %), and the aluminum layer of the metal contact and signal readout (15 %). The acceptance of a mircopixel is 60 % in the design described above. Figure 4 shows the design of a 4 4 array of SiPM sensors and the readout pads in the four sides. Making an array of sensor pixels is useful when a large area has to be covered by the sensors without introducing insensitive regions. Possible dead regions between neighboring sensors can be avoided by overlapping the tilted sensor plane, which can be done more easily with a larger sensor
Design and Simulation of a Silicon Photomultiplier Array for { H. Y. Lee et al. -489- Fig. 4. A 4 4 array of SiPM sensors and the 16 readout pads. Each sensor has 32 32 micropixels. Fig. 6. Depletion region in the vertical structure. Fig. 5. 2-D doping concentration and pro le of the implantation layers in a micropixel. plane containing the array of unit sensors. The size of the array has to be a compromise considering the yield of the array in the fabrication. The design described above requires 6 layers of the photo-mask set for the lithography process of the fabrication. Our design is actually the rst step in obtaining the nal optimal parameters, like the size of a sensor, the number of micropixels, the size of array, etc. We included variations of the design parameters in the masks, including the size of the array and the width of the trenches, etc. The nal optimization will be made after a test with fabricated sensor. III. PERFORMANCE SIMULATION The micropixel design is the result of optimizing the fabrication parameters obtained by using the Silvaco software tools of ATHENA and ATLAS [7]. The tools are physics-based simulators that predict the electrical characteristics of the designed structure and conditions. These tools were used in developing other silicon sensors that were successfully applied for space measurements of high-energy cosmic rays [8, 9]. The rst step in the simulation process is to build a correct PN-junction that gives a high electric eld with a reserve bias across the junctions. The optimized result obtained by adjusting the parameters in modeling the implantation of the fabrication processes is shown in Figure 5. The vertical pro le of the doping concentration in Figure 6 (a) is for a micropixel with size of 30 30 m2. The depths of the n+ and the p+ implantations conducted in a p-type epitaxy layer are about 0.5 m and 1 m, respectively. The trench with a width of 3 m, the Si02 layer, and the metal lines are shown in the gure. The depletion depth and the vertical pro le of the electric eld in a micropixel are shown in Figure 6 and Figure 7, respectively. They are obtained with an operation bias of 20 volts. The maximum of the electric eld is about 106 V/cm at the junction. The height and the width of the electric eld near the junction is high enough to initiate an avalanche. It is checked with the I-V characteristics of the junction obtained in the simulation. As shown in Figure 8, the junction breaks down at about 12 V. The temperature dependence is also shown in the gure. We have estimated the photon detection e ciency of the sensor as a function of the wavelength of the incoming photons in the range of 300 to 900 nm. It was estimated by taking the ratio of the sensor anode current to the source photo current, which is the equivalent current that would be observed if all the light from the source were detected. The result is shown in Figure 9. For a compar-
-490- Journal of the Korean Physical Society, Vol. 52, No. 2, February 2008 Fig. 9. Estimate of the photon detection eciency as a function of the wavelength of the photon. The curves are normalized to their eciency at 900 nm. photons increased with smaller depth of the junction because shorter-wavelength photons have shorter penetration depths. Fig. 7. Electric eld strength along the vertical cut of a micropixel. IV. CONCLUSION A 4 4 array of SiPM sensors was designed, and the performance was studied using Silvaco simulation tools. An optimization of the fabrication parameters to make the electric eld of the PN-junctions high enough for the Geiger-mode avalanche process was obtained from the simulation. The geometrical acceptance of a SiPM is 60 % in the design. The characteristics of the sensors show of to be an ideal sensor for space missions that are under considerations. The rst batch of designed sensors is under fabrication with 4-inch wafers. ACKNOWLEDGMENTS This work was supported by Creative Research Initiatives (MEMS Space Telescope) of Ministry of Science & Technology/Korea Science and Engineering Foundation. Simulated I-V characteristic of a designed mi- Fig. 8. cropixel. REFERENCES ison of only the wavelength dependency, the two curves for the two designs of the junction depths, 0.50 m and 0.55 m, were normalized to the eciency at 900 nm. As expected, the detection eciency of shorter-wavelength [1] G. Bondarenko, B. Dolgoshein, V. Golovin, A. Ilyin, R. Klanner and E. Poppva, Nucl. Phys. B (Proc. Suppl.) 61B, 347 (1998). [2] G. Bondarenko et al., Nucl. Instr. Meth. A 442, 187 (2000); P. Buzhana et al., Nucl. Instr. Meth. A 504, 48 (2003).
Design and Simulation of a Silicon Photomultiplier Array for { H. Y. Lee et al. -491- [3] V. Saveliev and V. Golovin, Nucl. Instr. Meth. A 442, 223 (2000); V. Saveliev, Nucl. Instr. Meth. A 535, 528 (2004); V. Golovina and V. Saveliev, Nucl. Instr. Meth. A 518, 560 (2004). [4] Jae Sung Lee et al., J. Korean Phys. Soc. 50, 1332 (2007). [5] T. Ebisuzaki et al., Proceeding of 30th Intern. Cosmic Ray Conf. (Merida, 2007); http://jemeuso.riken.jp/ en/index.html. [6] I. H. Park, Nucl. Phys. B - Proc. Suppl. 134, 196 (2004). [7] ATHENA User's Manual and ATLAS User's Manual, Silvaco International (2002); http://www.silvaco.com [8] N. H. Park et al., J. Korean Phys. Soc. 49, 815 (2006). [9] I. H. Park et al., Nucl. Instr. Meth. A 535, 158 (2004); I. H. Park et al., Nucl. Instr. Meth. A 570, 286 (2007); N. H. Park et al., Nucl. Instr. Meth. A 581, 133 (2007).