9 Sep 2008Paul Dauncey*MAPS/DECAL StatusPaul Daunceyfor the CALICE-UK MAPS group

Paul Dauncey

9 Sep 2008Paul Dauncey*MotivationNumber of charged particles is an intrinsically better measure than the energy depositedEnergy deposited (analogue ECAL) resolution ~50% worse than number of particles (digital ECAL) resolutionCan we measure the number of charged particles directly?Can we get anywhere near the ideal resolution?Average number of charged particles in an EM shower incident energyFluctuations around the average occur due to statistical nature of the showerAverage energy deposited in the sensitive layers number of charged particlesFluctuations around the average occur due to angle of incidence, velocity and Landau spread200.6X0 + 101.2X0

Paul Dauncey

9 Sep 2008Paul Dauncey*Digital ECAL conceptMake pixellated detector with small pixelsProbability of more than one charged particle per pixel must be smallAllows binary readout = hit/no hitEM shower density ~100/mm2 in core so need pixels ~ 50mmResults in huge number of pixels in a real ECAL ~ 1012 pixelsCannot afford to have external electronics with individual connections to so many channelsNeed readout integrated into pixelImplement as CMOS MAPS sensorIncludes deep p-well process to shield PMOS circuit transistorsVery high granularity should help with PFA tooRequires major systematic study; here concentrate on EM resolution

Paul Dauncey

9 Sep 2008Paul Dauncey*TPAC1.0 sensor168168 pixels = 28k total, each 5050mm2Two major pixel variants, each in two capacitor combinationsOnly one major variant worked well; preShaperBoth minor variants (Quad0 and Quad1) workedAll results shown are from this typeEvery pixel has 4 diodes, Q-preamp, mask and 4-bit pedestal trim, asynchronous comparator and monostable to give hit/no hit responsePixel hits stored with 13-bit timestamp on-sensor until end of bunch trainMemory for data storage inactive; 11% dead area in four columns

1cm1cm2.1mm250mm

Paul Dauncey

9 Sep 2008Paul Dauncey*Calibration using 55Fe55Fe gives 5.9keV photonDeposits all energy in ~1mm3 volume in silicon; 1640eIf within diode, then all charge registered in single pixel with no diffusionBinary readout mean measurement need threshold scanNeed to differentiate distribution to get signal peak in threshold units (TU)Derivative approximatedusing previousbin subtractionRate vsthresholdSignal peak ~200TU above pedestal; 1TU ~ 8e ~ 30eV deposited

Paul Dauncey

9 Sep 2008Paul Dauncey*Single pixel noise performanceAlso need threshold scan to see pedestal and noiseComparator fires on signal going high across threshold levelNo hits when far above or below thresholdWidth of distribution equivalent to noiseRMS ~ 5.5TU ~ 44e ~ 170eV on averageMinimum is ~ 4TU ~ 32e ~ 120eVTarget level was ~ 90eVNo correlation with position on sensorSpread not fully understoodQuad1 ~ 20% larger than Quad0Quad0Quad1ExamplesinglepixelAll preShaperpixels on examplesensor

Paul Dauncey

9 Sep 2008Paul Dauncey*Single pixel relative gainMeasured using laserSilicon transparent to 1064nm light so illuminate from back side of sensorFocus on epitaxial layerAgain need to do threshold scan and find edge to measure laser signalFixed laser intensity gives relative gain for individual pixelsCan do hundreds of pixels automaticallyGain uniform to 12%Quad1 ~40% more gain than Quad0Quad1 ~20% better S/N than Quad0Quad0Quad1

Paul Dauncey

9 Sep 2008Paul Dauncey*Charge spreadCharge diffuses to neighbouring pixelsReduces signal in hit pixelCauses hits in neighbouring pixelsNeed to make sure this is correctly modelledSimulation using Sentaurus packageFull 3D finite element model33 pixel array = 150150mm2 areaThickness of silicon to 32mm depth; covers epitaxial layer of 12mm plus some of substrateUse laser to fire at 21 points within pixelLaser spot size < 2mm, step size 1mmPoints numbered 0-20, 5mm apartSymmetry means these cover whole pixel surfaceMeasure signal using threshold scan in centre pixel and all eight neighboursNumbered Cell 1 to Cell 9798654321

Paul Dauncey

9 Sep 2008Paul Dauncey*Charge spread resultsSimulation reasonably reproduces the spatial dependenceSmall differences near diodes (points 9,13,14)Average signal over whole pixel ~ 35% of deposited signalTotal charge is 1300e so average ~ 450eAverage signal/noise ~ 10Worst case signal in central pixel is when hitting cornerGives ~ 24% of total charge so ~ 300e and S/N ~ 7arXiv:0807.2920

Paul Dauncey

9 Sep 2008Paul Dauncey*Effect of deep p-wellDevelopment included modification to foundry CMOS processDeep p-well INMAPS processingBlocks signal charge from being absorbed in pixel amplifier, etcDeep p-well essential for usable sensorAverage signal without deep p-well ~10% ~ 130eWorst case ~1% ~ 13earXiv:0807.2920

Paul Dauncey

9 Sep 2008Paul Dauncey*Simulation expectationShown at LCWS07 but with no verification of assumptionsNow have concrete noise values and measured charge diffusionCurrent extrapolation to real detector shows significant degradation of ideal DECAL resolution35% increase in errorNumber if pixels hit not trivially related to number of charged tracksDegradation arises fromNoise hitsDead areaParticles sharing pixelsParticles crossing pixels boundariesCharge diffusion to neighbouring pixelsImportance of various effects differs10GeV0.0320.044

Paul Dauncey

9 Sep 2008Paul Dauncey*Effect of noiseNoise adds hits to showers so increases NDepends very strongly on thresholdNeed to increase threshold above noise wallNoise has no effect for higher thresholdsGain spread ~12% is equivalent to threshold spread here so small effectResolution degradation ~ 10%If S/N can be improved, then get a plateau so noise has no effect on resolution

Paul Dauncey

9 Sep 2008Paul Dauncey*Effect of dead areaSensor has 11% dead region due to on-pixel memoryBands of 250mm wide spaced every 2.4mmShower width ~ 1cm so every shower sees several dead bandsAlways loses 11% of hits with small fluctuationsSince sE/E 1/N, impact is not largeGives 1/(0.89) ~ 1.06 effectHence ~ 5% degradationAssumes sensor large enough that edge effects are negligibleMay add ~ 5% more dead area in reality so ~ 2% more to resolution

Paul Dauncey

9 Sep 2008Paul Dauncey*Effect of hit confusion per particleArises from Particles close togetherParticles crossing pixel boundariesCharge diffusionOnly the last is known to be well modelledNeed to do neighbouring hit clustering to convert hits to particle countAlgorithm to use depends on effects which may not be modelled wellMajor study of clustering algorithms still to be doneCurrently gives ~ 20% degradation to resolution so dominatesEssential to get experimental data on fine structure of showers to know realistic resolution

Paul Dauncey

9 Sep 2008Paul Dauncey*Short term future plansDebugged version, TPAC1.1 due back on Sept 23All pixels uniform; Quad1 preShaper variantDecoupled power mesh, thought to cause pickup between pixels (and disrupted beam data)Adjusted pixel circuit layout to improve gain and S/NTrim setting has six not four bits to allow finer trim adjustmentOther small fixes, e.g. fix low level of memory corruption

9 Sep 2008Paul Dauncey*Long term future plansSubmitting a proposal this week for large sensor TPAC2450450 pixels and 2.52.5cm2; a factor ten in area; otherwise a scaled-up TPAC1Bid includes funding for 16-layer Si-W DECAL stack; 55 sensors = 12.512.5cm2 per layerSmaller than AECAL but OK for basic proof-of-principleTo pack sensors in the plane, will wirebond through slots in PCBAim for pixel-pixel gap between sensors to be only 500mm ~ 4% extra dead areaReal detector would bump-bond but we need to minimise engineering effort for this programmeA rough scheduleSensor design in 2009Stack assembly and system tests in 2010Beam test of stack in 2011BUT... not cheap, UK funding still very difficultExternal collaborators very much welcomeWould very significantly increase probability of approval if cost split with non-UK groups

Paul Dauncey

9 Sep 2008Paul Dauncey*ConclusionsDECAL seems possible in principleActual EM resolution which would be obtained depends heavily on details of showers and on algorithm for clusteringThe simulation has not been verified at small granularitiesEssential to get real data to compareWill have first look at showers early in 2009May have first look at EM resolution in 2011Approval very uncertain; collaborators very welcome!

Paul Dauncey

9 Sep 2008Paul Dauncey*Backup: Single pixel pedestalsPedestal given by mean of threshold scanPedestal spread is ~ 4 times noiseMust correct using trims to get sensible dataTrimming works reasonably well; down to RMS of ~ 4.5TUStill not completely below noise level so more trim bits would help

Paul Dauncey

9 Sep 2008Paul Dauncey*Backup: Pedestal and noise over sensor

Paul Dauncey

9 Sep 2008Paul Dauncey*Backup: pixel hit pickupFind different results for pixel if other pixels enabledPrevented pedestals from being determined until effect understoodPlots shown previously had most pixel maskedNot found before Dec 2007 beam test so data had bad trims; probably unusableProbably due to shared power mesh for comparators and monostablesIf >~100 pixels fire comparators at same time, power droops and fires other monostablesNot an major issue for normal use (once understood)Single enabled pixelAllpixelsenabled

Paul Dauncey

9 Sep 2008Paul Dauncey*Backup: DECAL 16-layer stackShould give definitive answer to whether DECAL concept is viable16 layers gives degraded resolution by factor ~ 2Funding not available for more layersHopefully extrapolate to realistic calorimeter sampling using simulation

Paul Dauncey