Solid State Photomultiplier: Noise Parameters of Photodetectors with Internal Discrete Amplification

Transcription

1 Solid State Photomultiplier: Noise Parameters of Photodetectors with Internal Discrete Amplification K. Linga, E. Godik, J. Krutov, D. Shushakov, L. Shubin, S.L. Vinogradov, and E.V. Levin Amplification Technologies, Inc., New York, NY ABSTRACT We have designed and developed a new family of photodetectors with Internal Discrete Amplification (IDA) mechanism They operate as solid state photomultiplier devices at room temperature and may be used in numerous applications where high bandwidth of the detector is necessary in combination with maximum sensitivity and low excess noise. The photodetectors can operate in linear detection mode with gain-bandwidth product up to as well as in photon counting mode with count rates up to 10 8 counts/sec. The key performance characteristics exceed those of Photomultiplier Tube (PMT) and Avalanche Photodiode (APD) devices. The detectors have gain > 10 5, excess noise factor as low as 1.03, photoresponse rise/fall time < 300 ps, and timing resolution (jitter) < 200 ps. The combination of low excess noise at high gain and wide bandwidth, as well as scalability to large active areas, presents the main advantages of this technology over conventional photodetector solutions. Ultra low excess noise is one of the main features of the internal Discrete Amplification Detector (DAD), and in this paper its nature has been investigated more comprehensively. We investigated the behavior of the noise-factor and afterpulsing, and conclude that both have the same physical nature. Optical cross-talk between channels is shown to be responsible for the afterpulsing phenomenon, and, in turn, is the main source of excess noise. Thus, the noise characteristics of an DAD device and its timing resolution may be significantly improved as they are limited not by the discrete amplifier channel properties itself, but by the cross-talk, which strongly depends on the device design. Keywords: internal discrete amplification, discrete amplifier, DAD, photodetector, APD, PMT, Geiger mode, photon counting, avalanche amplification, gain, excess noise, noise factor I. INTRODUCTION High speed few-photon signal detection has remained a challenge for many years, and the need for a solution continues to grow with the increasing rate of development of various optoelectronics systems. The increasing requirements and accurately defining the signal pulses at simultaneous reduction of its intensity poses a challenging task for the detection and integration of low level optical pulses. Scintillation systems used in nuclear physics, high energy physics, and nuclear medicine imaging require fast high sensitivity detectors. Improvement in time characteristics and the ability to detect lower useful signals are also needed in communication systems. Single photon counting sensitivity is increasingly required in detection and measurement equipment. This creates a demand for high-speed photodetectors that are able to detect pulses consisting of several photons and even single photons. Devices based on the internal discrete amplification of a photosignal offer superior performance for such applications, as shown in [1], where the basic set of its features was given. II. PRINCIPLE OF OPERATION The unique set of Discrete Amplification Detector (DAD) parameters is due to use of the internal discrete amplification mechanism, providing high-quality internal amplification of an input signal. Its principle of operation is illustrated in Figure 1. Semiconductor Photodetectors III, edited by Marshall J. Cohen, Eustace L. Dereniak, Proc. of SPIE Vol. 6119, 61190K, (2006) X/06/$15 doi: / Proc. of SPIE Vol K-1

2 I(t) M I(t) Light pulse, photocurrent I(t) prior to amplification N spatially separated photoelectrons K>N of threshold avalanche amplifiers with gain M N charge packages, each of M electrons Sum output signal consists of NxM electrons Figure 1: Photodetector principle of operation: 1) dividing input signal into equal charge components, 2) calibrated amplification of each component, and 3) combining the components into output signal [1] It should be noted that amplifiers on the figure illustrate the physical principle of operation since no real electronic components are built in the device. Functionally DAD may be considered as consisting of the following parts: photoconversion part where light is converted to photocarriers as in any photodetector transporter part, that is responsible for gathering the photocarriers and delivering them to the corresponding amplifier amplifier part, consisting of a set avalanche threshold amplifier channels, distributed over all of the active area of the device reader part, where the charge packages from threshold amplifiers are combined to provide the output signal One can say the discrete amplifier is integrated in the photodetector, providing a new level of functionality 1. Discrete amplifier may be integrated in many types of semiconductor photodetectors (various active areas, spectral ranges, semiconductor materials, and so on) that allow one to say that DAD is not only a new device but a fundamentally different way of signal amplification that could be used to develop a new family of photodetectors. Depending on specific application requirements, these detectors would utilize various designs, methods of integration, and optimization techniques. In the case of perfect integration of discrete amplifier with photoconverter, the characteristics of the whole device are defined by the properties of its functional components. For example, spectral range and quantum efficiency are defined 1 This is similar to viewing a conventional APD as a simple PIN, integrated with an avalanche amplifier..but APD is known to be limited by its amplifier properties (noise, speed, gain). Integration of the discrete amplifier, free of these disadvantages, provides the new level of functionality for the whole device. Proc. of SPIE Vol K-2

3 only by the photoconverter, while gain, speed, and noise-factor are defined by the amplifier properties. Achievement of the best performance parameters requires the optimization of the whole photodetector device III. KEY PERFORMANCE PARAMETERS The measured parameters of the photodetectors are as follows: Gain to Excess noise factor 1.02 to 1.08 Rise/Fall time 0.3 to 0.7 ns This unique combination of performance parameters enables reliable registration of single photoelectrons and measurement of few-photon pulse amplitude at a bandwidth of 1 GHz. The 1.02 value of the excess noise factor is unusually low and seems to validate the overall approach to decreasing noise while still attaining very high amplification gain. The unique combination of high gain, high speed and extremely low noise allows one to use discrete amplification to develop both high sensitive analog photodetectors and single photon counting photodetectors. In addition, it enables the creation of universal devices that could measure the amplitude of light pulses with unprecedented accuracy in analog mode, and to count individual photons at a low light intensity. In the analog mode of operation, theses photodetectors have significant advantages over PMTs. Due to a lower noise factor, and potentially significantly higher quantum efficiency, they could have much better threshold sensitivity, which approaches the ideal value limited only by the statistics of the optical signal itself. In the photon counting mode of operation, it is possible to obtain counting speeds as high as several hundred MHz due to the presence of numerous amplifier channels. The counting speed is about an order of magnitude faster than that of photon counting PMTs. As mentioned above, to achieve the best photodetector performance, optimization is required in order to improve its components and efficiently integrate them. Integration quality could be characterized by the following parameters (in addition to the DAD device key features as listed in [1]): a. Manufacturing deviation of channel-to-channel properties (increase noise-factor) b. Channel-to-channel cross-talk (increase noise-factor) c. Geometrical factor that determines the efficiency of photocarriers gathering from photoconverter to amplifiers (decrease the device detection efficiency) d. Dark count rate in excess of normal generation in corresponding semiconductor Investigation of these parameters is important in order to extract characteristics (physical limits) of the functional components from the parameters of the whole device. That would in turn allow further device optimization. In the following section we focus on the nature and optimization limits for the excess noise. The excess noise parameter is particularly important since noise suppression represents the key feature of the DAD device. IV. INVESTIGATION OF EXCESS NOISE AND AFTERPULSING Noise factor Excess noise factor is defined as F = M M = 1+ σ M and characterizes the excess noise, introduced in the process of amplification by the statistical fluctuations of gain (M). For the ideal, no-noise amplification, F=1. The higher the noise-factor, the lower the amplitude resolution and threshold sensitivity of the detector in analog mode 2. 2 Analog (proportional) mode involves measuring the amplitude of a signal (light pulse).. In the ideal photodetector, both amplitude resolution and threshold sensitivity are limited by the signal pulse photon statistics and not by the detector noise or gain. Proc. of SPIE Vol K-3

4 Noise factor of 1.02 to 1.08 is observed in a typical DAD device. This result compares favorably with most widebandwidth high-gain analog devices with internal amplification. For example, photomultiplier tubes (PMT) typically have F>1.3, while avalanche photodiodes (APD) have even worse values even at significantly lower gains. Only VLPC [2] achieves compatible values of F, but it operates at cryogenic temperatures. While these absolute values of the excess noise factor are expressive, experimental measured dependence of noise factor on average gain in DAD devices was found to be qualitatively different from the Monte-Carlo simulations that predict both lower noise-factor and its further decrease with gain growth (to be published elsewhere). Noise-factor may be calculated from the experimental gain probability distribution shown in Figure 2. It was obtained from amplitude distribution of one-electron-initiated pulses, observed as the output of DAD device (dark count pulses). In addition to the main peak, there is an additional double-event probability peak on the distribution. With operating voltage increase this additional peak part also increases. The additional peak correspond to simultaneous firing of two amplifier channels and, especially for higher operating voltage levels, is higher than the statistical double-event probability of a Poisson distribution. This implies that the peak is caused not only by the by Poisson statistics, but also by the amplifier channel-to-channel cross-talk, in which a channel, while operating, could stimulate the firing of another channel Main peak Probability Doubleevent peak E E E E+05 Gain 4.64E E E+0 Figure 2: Gain probability distribution of a typical DAD device To investigate this effect, noise-factor as a function of gain was measured for three types of experimental DAD devices with the same active area size but different numbers of amplifier channels (Figure 3): Proc. of SPIE Vol K-4

5 1) one-channel device - circle markers 2) multi-channel device with greater distance between channels (90 channels) - triangular markers 3) multi-channel device with smaller distance between channels (170 channels) - square markers To understand whether the greater source of noise was widening of the main peak or the the appearance of the additional, excess noise factor was calculated both based on the full probability distribution (solid line), and from the main peak only after excluding the additional peak from distribution (dashed line). It could be seen that for the single channel device, where both cross-talk and channel-to-channel non-uniformity are absent, noise-factor is extremely small and decreases down to F=1.003 to with gain growth, which qualitatively corresponds to the results of modeling. This is the limiting value determined by the threshold avalanche amplifier itself. For a multi-channel device, at high levels of gain the double-event peak appearance is the main source of noise, and noise factor increases with gain growth. The rate of the noise-factor growth is higher for smaller distance between the channels. Excess noise caused by gain deviation within the main peak (dashed lines) does not increase with gain and appears to depend only on the number of channels in the device. It may be caused by the technological non-uniformity statistics between channels Noise Factor E E E E E E E E+05 Gain (main peak) Figure 3: Measured dependence of noise factor on gain at room temperature. Line with circle markers corresponds to the 1-channel device; line with square markers to the multi-channel device with smaller distance between channels, line with triangular markers to the multi-channel device with greater distance between channels. Dashed lines with filled markers correspond to the noise factor, calculated with rejection of additional peaks from the gain distribution. Based on these experimental results, we arrive at the following conclusions: The threshold avalanche amplifier channel itself operates with extremely low fluctuations of gain, with its noise factor level as low as to The true value may be even lower since in the experiment it was defined by the measuring equipment resolution. Proc. of SPIE Vol K-5

6 Fabrication non-uniformity increases noisefactor, but nor significantly The main source of noise is cross-talk between channels, leading to the appearance of double-event pulses Cross-talk increases with gain and decreases with greater distance between the channels. Afterpulsing Afterpulsing mean false pulses following the real signal pulse. This effect is present in almost all types of photon counters, but its nature is not always the same. For single-element semiconductor Geiger counters it is caused by the well-known trapping/detrapping effect, while for multi-element devices optical cross-talk, caused by hot-carriers photon emission, adds its contribution. Afterpulsing decreases the device timing resolution 3, especially for few-photon pulse applications. There are two characteristics of afterpulsing that are usually used. The physical one is afterpulsing probability P ap (or its distribution in time p ap (t)), which is defined as the probability of an afterpulse appearance after a primary pulse. More frequently used is the afterpulsing coefficient K ap (or its distribution in time k ap (t)), which is defined as the total average number of afterpulses per one primary pulse. Afterpulsing coefficient incorporates the probability of a chain of afterpulses, as each pulse may stimulate appearance of the next one. For Np observed primary pulses and Na afterpulses, Kap =Na/Np = Pap/(1-Pap). Cumulative distribution 4 of the afterpulsing probability in time for an experimental DAD device (with high afterpulsing) is shown in Figure % Afterpulse probability 10.00% 1.00% 0.10% 0.01% y = 4.37E-01e -1.27E+08x y = 1.46E-01e -1.27E+08x 0.00% 0.0E E E E E E E-08 Time, s Figure 4: Cumulative distribution of afterpulse probability for a DAD device. The top line correspond to higher operating voltage; the lower line to lower voltage. Exponential trends and their equations are also shown in the figure. The decay time = 7.9 ns. 3 If the internal amplifier own jitter is small enough and does not limit timing resolution 4 Pc() t = p ( τ ) dτ, thus total afterpulse probability Pap=Pc(0). t ap Proc. of SPIE Vol K-6

7 We can see that the afterpulsing probability of the DAD device is exponentially decreasing in time (delay, relative to the primary pulse) with a very small, in comparison with most photon counters, decay time, typically 3 to 15 ns. No additional afterpulsing components were observed from the nanosecond to the millisecond range. Overall afterpulsing probability is equal to the pre-exponential factor. This factor is seen to depend strongly on operating voltage. The afterpulsing decay time depends only on the device design. No temperature dependence of decay time was observed in the range -20 o C +20 o C. For a single-channel experimental device, no afterpulsing was observed. These results allow us to exclude the trapping mechanism and assume the model of optical cross-talk [3,4] as the only source of afterpulsing in the investigated DAD devices. In accordance with this model, hot carriers in a fired amplification channel emit photons, which in turn produce secondary photocarriers that are collected, with some probability, at the neighboring channels. Probability of an afterpulse in the same channel is negligible. If we adopt this model, afterpulsing Pap should be proportional to both the gain and the photon detection probability Pd of the device, as gain defines the number of emitted photons by a fired channel and Pd measures the efficiency of the secondary photocarriers collection. This dependence is confirmed by experimental results, as shown in Figure 5. 70% 60% y = 2.66E-06x Afterpulse probability, Pap 50% 40% 30% 20% 10% 0% 0.0E E E E E E+05 Pd * Gain Figure 5: Experimental dependence of Pap on the product of photon detection efficiency Pdet (for red light) and gain. Temperature dependence of Pap also follows the Pd x Gain dependence, providing additional validation to the optical cross-talk model. The coefficient Kdes=Pap/(Pd*Gain) is design-dependent and characterizes the design noise efficiency. As our experiments show, its value could be significantly improved. Proc. of SPIE Vol K-7

8 Correspondence between noise factor and afterpulsing The discussed earlier growth of noise-factor with gain could also be assumed to be caused by afterpulsing. It was shown that noise factor growth is caused by appearance of double-event pulses, with probability significantly exceeding its expected Poisson value. If some operating channel stimulates firing of another channel, and time interval between these events is shorter than the pulse-pair resolution time (Tr 1 ns) of the device, then a double-pulse event (output pulse with double amplitude) occurs. Total probability of double-pulse event may be estimated from the known cumulative distribution as P = P(0) P( τ ). dup c c r It is then easy to estimate the noise factor resulting from these afterpulsing-caused double-events, and to compare it with the directly measured corresponding values of noise-factor,. This comparison is shown in Figure Noise-factor, F E E E E E E+05 Gain Experimental Calculated Figure 6: Noise-factor dependence on gain. Comparison of experimental values of noise-factor with those calculated from Pap. We can see from Figure 6 that almost all of excess noise observed on the experimental DAD-device resulted from afterpulsing. Thus, it may be concluded that optical cross-talk is the main source of afterpulsing, which is, in turn, the main source of the increase in noise factor with the growth in gain. One of the conclusions is that further optimization of DAD-device design could lead to significant performance improvement, as the current parameters are far from their theoretical limits. Both the device noise characteristics and its timing resolution are still limited by the design-dependent effects. Proc. of SPIE Vol K-8

9 Applications that require lower afterpulsing (noise factor) For most tasks, noise factor values as low as of 1.02 to 1.08 that are typical for DAD devices, are quite sufficient. For such values the detector amplitude resolution as well as its threshold sensitivity are defined mainly by the photon statistics of the light signal and not by the detector own noise. However, there are some important applications where few-photon light pulses, not synchronized in time, need to be detected, and the requirement is to minimize dark count rate (false detections) while preserving high detection probability of light pulses. In this case, the task is to suppress (discriminate against) the dark count of a silicon device that is caused by thermal generation, and thus approach, without cooling, effective dark count values as low as those of vacuum devices. The noise factor here is a critical parameter, as only its ultra-low value allows one to effectively discriminate (exclude) against dark count single-electron pulses from the signal multi-electron pulses by their amplitudes. Estimation of the discrimination level that is necessary to provide the requiring dark count rate suppression were given in [1], but afterpulsing was not considered. The calculated dependence of dark count rate on the discrimination level is shown in Figure 7 for various values of afterpulsing probability. It is seen from the figure that without afterpulsing, discrimination of only single-event pulses leads to the decrease of effective dark count rate by four orders of magnitude. Such a low discriminator value has almost no effect on the device sensitivity, i.e., the probability of light pulse detection. With greater afterpulsing (or noise factor increase) one has to set higher discriminator to provide the same dark count rate suppression, thus decreasing threshold sensitivity of the device. Even Pap=1% leads to more than doubling of the necessary discrimination level. Comparing this figure with the experimental dependence, one can estimate afterpulsing probability of the device used in that experiment, as 10-15% Dark count rate, cps Discriminator level No afterpulsing afterpalsing Pap=1% Pap=15% Figure 7: Calculated dependence of dark count rate on the discrimination level. Discrimination level is normalized to 1-electron pulse amplitude. Proc. of SPIE Vol K-9

10 For such tasks further improvement of noise-factor in DAD devices is desirable as an alternative to cooling. The other type of applications where decrease of afterpulsing is important, is timing measurements with low light-level signals. Though the timing resolution of 200 to 300 ps typical for DAD devices is acceptable value for most applications of this type, its further improvement is also desirable. V. CONCLUSIONS Integration of discrete amplifier in a photodetector results in a new family of room-temperature photodetectors with qualitatively different features. DAD devices offer superior performance for applications, requiring wide bandwidth registration and measuring of ultra-weak light signals in visible to near IR spectral range. Investigation of noise-factor and afterpulsing behavior in experimental DAD devices allow one to conclude that they are caused mostly not by noise properties of discrete amplifier channels themselves, but by optical cross-talk between channels. Currently, both noise and timing resolution of the devices are limited by the design-dependent parameter. The combination of high gain (greater than 10 5 ), low excess noise factor (1.02 to 1.15), and high speed permits the use of these photodetectors in a wide variety of applications requiring the detection low level light signals. Though the current design already provides better excess noise and speed compared with conventional detectors, both parameters could be significantly improved by further design optimization. REFERENCES 1. K. Linga et al, Ultra low noise photodetectors with internal discrete amplification, Proc. SPIE, Vol. 5726, pp , Petroff M.D., Stapelbroak M.G. Photon-counting solid-state photomultiplier. IEEE Trans. Nucl. Sci., 36, , J. Jackson et al, Toward integrated single-photon-counting microarrays, Opt. Eng., Vol. 42, pp , A. Lacaita et al, On the bremsstrahlung origin of hot-carrier-induced photons in silicon devices, IEEE Elec. Dev. Lett.., Vol. 40, pp , K. Linga et al, High Performance universal analog and counting photodetectors for LIDAR Applications, Proc. SPIE, Vol. 5887, pp to , R.J. McIntire, Multiplication noise in uniform avalanche diodes, IEEE Trans. Electron Devices, 13, pp (1966) 7. R.J. McIntire, A new look at impact ionization Part I: A theory of gain, noise, breakdown probability, and frequency response, IEEE Trans. Electron Devices, 48, pp , A.T. Young, Photomultipliers: Their cause and cure, in Methods of Experimental Physics, 12, Part 1: Optical and Infrared, pp. 1-74, Academic, New York, S. Takeuchi et al, Development of a high-quantum efficiency single photon counting system, Appl. Phys. Lett., 74, pp , J. Kim et al, Noise-free avalanche amplification in Si solid state photomultipliers, Appl. Phys. Lett., 70, pp , J.N. Hollenhorst, A theory of multiplication noise, IEEE Trans. Electron Devices, 37, pp , 1990 Proc. of SPIE Vol K-10