Characterisation of Silicon Photomultipliers for the T2K Experiment, 18th May 2010 Martin Haigh, University of Oxford
Outline Brief introduction to the T2K experiment. Overall configuration and goals. ND280 near detector. Structure and properties of the MPPC Silicon Photomultiplier. Characterisation measurements and physics models. MPPC Monte Carlo results. Comparison to data. Extrapolation to predict MPPC performance and saturation behaviour. 2
The T2K Experiment T2K (Tokai to Kamioka) is an experiment based in Japan to study ν oscillations. Proton collisions with a solid target produce pions. Pions focused by magnets and decay π μ+νμ producing a beam of muon neutrinos. Measure νμ disappearance and νe appearance at Super Kamiokande, 295 km away. Improve on CHOOZ θ13 sensitivity by a factor ~10. 3
The ND280 Detector ND280 is a near detector for T2K, 280m from beam source. Measures the unoscillated ν flux to reduce systematic errors in oscillation fit. Fiducial volume consists of fine grained detectors and time projection chambers (TPCs), plus a Pi zero detector. Surrounded by electromagnetic calorimeters (ECALs). All detectors except TPCs are plastic scintillator based, read out with Y11 WLS fibres coupled to SiPMs (MPPCs supplied by HPK). Shroud Connector Scintillator bars MPPC μ-coax interface board Ferrule Foam spring WLS fibre 4
The MPPC The Multi Pixel Photon Counter (MPPC) is an electrically parallel array of avalanche photodiodes (APDs) on a single silicon wafer. Each pixel contains a p n junction operated above breakdown voltage. A single incoming photon triggers a Geiger avalanche in a pixel. Quenched by a built in resistance in series with the APD junction. Similar performance to PMTs: Gain ~ 5x105 1x106 and PDE ~30%. Dynamic range up to 1000s of photons. 1.3 mm Timing jitter ~250ps. Also some advantages: Epoxy shield Low cost (~ 10 / piece). Ceramic package Low operating voltage ~ 70V. Insensitive to magnetic fields very useful for HEP detectors. 5
MPPC Characteristics Several noise effects: Silicon operated at Troom => get dark noise avalanches. Rate ~500kHz over whole device. Carriers trapped on impurities can cause delayed afterpulses (AP) in the original pixel. Avalanches can produce photons which trigger neighbouring pixels (cross talk or CT). Finite pixel recovery time recharging rate after trigger limited by quenching resistor. Means that response to large light signals (Nphot x PDE ~ Npix) is nonlinear. Dark rate, CT/AP probability and PDE all vary with bias voltage. Need to understand for simulation/calibration. Crosstalk: single large pulse Afterpulse: delayed secondary See noise cascades. Afterpulse delay up to ~ 100ns Light signals and dark noise look identical 6
Overall device response Measure charge output over a gate period ~100s of ns. Histogram of output charges (at low light level) shows peaks corresponding to number of pixels fired. 0 pe 1 pe 2pe Peak populations differ from Poisson distribution because of crosstalk/afterpulsing. Use population of zero peak with no light to calculate dark rate. With light on, use to calculate PDE. Deviation of data from a Poisson distribution with same zero probability gives contribution from CT+AP, but can't separate the two effects. Gain given by peak separation. Linear in overvoltage: G = C/e x (Vbias Vbd), where C is the pixel capacitance. 7
Dark noise and PDE M. Ziembicki et al., Warsaw Tech. Dark rate found to vary linearly in overvoltage and exponentially in temperature. Temperature scaling in good agreement with theoretical number of carriers N~e ε/2kt. Find dark rate varies by a factor ~2 between individual sensors. PDE voltage dependence measured using collimated LED pulse calibrated with PMT. Also measured λ dependence with spectrophotometer, calibrated with PIN diode. Y. Kudenko et al., INR Moscow 8
Afterpulsing Afterpulsing studied in detail using waveform analysis: Measure time from an initial trigger pulse to the next pulse. Fit distribution to a hypothesis for afterpulse probability as a function of time elapsed. Fitted well by using 2 time constants for AP probability. Probability ~ overvoltage2. Time constants don't depend strongly on V. F. Retiere et al., TRIUMF 9
Crosstalk Crosstalk measured separately for different pixels by focusing a laser pulse on a single pixel. Corner/edge pixels have fewer neighbours => lower CT probability. Difference in CT probabilities helps constrain the model to use. A. Vacheret, Imperial Tried modelling CT with Poisson distribution for number of CT, or just one CT per avalanche. Nearest neighbour only, or some probability of hitting more distant pixels. Find that simplest model nearest neighbour with up to one pixel fired fits data best. 10
Monte Carlo Framework 11
Data/MC comparison Compare MPPC MC results to gated charge measurement with 540 ns gate, made at Imperial College. Check absolute agreement of MC with data Don't change parameters measured in lab for each device feature. Had to measure dark rate for specific device DCR/kHz=577.3*(Vbias Vbd) 72.1 No absolute light calibration => calibrate based on data for 1.32 V. Peak widths σi= (σ02 + i σpe2) also fixed by data at 1.32 V. Good data/mc agreement for a range of light levels and operating voltages. 12
Data/MC comparison Compare MPPC MC results to gated charge measurement with 540 ns gate, made at Imperial College. Check absolute agreement of MC with data Don't change parameters measured in lab for each device feature. Had to measure dark rate for specific device DCR/kHz=577.3*(Vbias Vbd) 72.1 No absolute light calibration => calibrate based on data for 1.32 V. Peak widths σi= (σ02 + i σpe2) also fixed by data at 1.32 V. Good data/mc agreement for a range of light levels and operating voltages. 13
Timing and energy resolution Use MPPC response to estimate incident light level (i.e. energy) by dividing by the mean response per photon. Best theoretical energy resolution is the Poisson uncertainty on the number of incident photons, σ(nphot) = Nphot. Excess noise factor F is ratio of measured spread to this theoretical prediction. Timing resolution depends on time distribution of incident light. Ran MC for Y11 decay time of 7 ns. MPPC pulse modelled as an exponential decay, τ=13.5 ns. Resolution < 2 ns for MIP level signal. 14
Voltage recovery MPPC essentially an array of RC series circuits. Pixel voltage drops to Vbd when it fires. For pixel connected to voltage source, overvoltage recovers on timescale RpixCpix=13.5 ns. Ignore bias source and amplifiers, and just consider MPPC and local capacitor CG. Fairly simple equivalent circuit. Analytic solution for pixel voltages Vi(t) is a simple exponential to leading order. CG Cpix RG Rpix... Cpix Rpix Npix 15
Use full MC package with recovery model to predict saturation behaviour for the device. Decay time of Y11 fibre is 7ns same timescale as sensor recovery. Recovery from capacitor instead of voltage source => saturation for large signals, even for wide time distribution. Predict fairly linear response over T2K physics range. Mean response / pe Saturation Simple τ = 13.5ns model Local capacitor model τlight = 50ns τlight = 7ns τlight = 0 ND280 physics range Incident photons 16
Summary SiPMs are capable of similar performance to PMTs, and have a number of important advantages relevant for HEP applications. Characterisation measurements have been carried out to study device properties. Behaviour as a function of bias voltage now well understood. A Monte Carlo for the devices has been developed and is now in a mature state. Agreement between data and MC is good. Have also been able to use the MC to estimate device resolution. Recovery effects specific to the ND280 front end board have been studied. Need a comprehensive set of real measurements to validate the simulation. Characterisation studies and simulation work being prepared for publication. Some simulation work already in PD09 proceedings: M. Haigh et al., Monte Carlo simulation of MPPC photosensors for the T2K experiment. 17