Development of a Digital Hadron Calorimeter (DHCAL) Lei Xia Argonne National Laboratory HEP Division

RPC DHCAL Collaboration Argonne Burak Bilki Carol Adams Mike Anthony Tim Cundiff Eddie Davis Pat De Lurgio Gary Drake Kurt Francis Robert Furst Vic Guarino Bill Haberichter Andrew Kreps Zeljko Matijas José Repond Jim Schlereth Frank Skrzecz (Jacob Smith) (Daniel Trojand) Dave Underwood Ken Wood Lei Xia Allen Zhao Boston University John Butler Eric Hazen Shouxiang Wu Fermilab Alan Baumbaugh Lou Dal Monte Jim Hoff Scott Holm Ray Yarema IHEP Beijing Qingmin Zhang University of Iowa Burak Bilki Edwin Norbeck David Northacker Yasar Onel McGill University François Corriveau Daniel Trojand UTA Jacob Smith Jaehoon Yu IIT Guang Yang Daniel Kaplan RED = Electronics Contributions GREEN = Mechanical Contributions Yellow = Students BLACK = Physicist

Motivation: physics at the next lepton collider – Physics Benchmarks for the ILC Detectors Required: excellent Jet energy/mass resolution Solution: Particle Flow Algorithm (PFA)

Particle Flow Algorithms Need new approach Particle Flow Algorithms ECAL HCAL γ π+ KL The idea… Charged particles Tracker measured with the Neutral particles Calorimeter Particles in jets Fraction of energy Measured with Resolution [σ2] Charged 65 % Tracker Negligible Photons 25 % ECAL with 15%/√E 0.072 Ejet Neutral Hadrons 10 % ECAL + HCAL with 50%/√E 0.162 Ejet Confusion ≤ 0.242 Ejet 18%/√E Required for 30%/√E Requirements for detector system → Need excellent tracker and high B – field → Large RI of calorimeter → Calorimeter inside coil → Calorimeter as dense as possible (short X0, λI) → Calorimeter with extremely fine segmentation → Single particle energy resolution not critical thin active medium Opens up new possibilities

Digital Hadron Calorimeter (DHCAL) and Choice of active medium PFA hadron calorimeter  ‘Digital’ readout  implication for active medium Extremely fine segmentation (~1x1cm2 readout cell)  simple hit counting provide sufficient energy resolution  1bit readout/Digital Hadron Calorimeter (DHCAL) DHCAL  counting charged particles in hadronic showers (instead of measuring dE/dx in the active medium)  the active medium can have poor, or even no dE/dx resolution Thin active medium, embedded readout Cost … What RPC can offer Good position resolution  great for fine segmentation Excellent efficiency for charged particles  great for counting charged particles Large charge fluctuation for single MIP  no dE/dx resolution, poor particle counting on single pad  but this is exactly NOT needed for a DHCAL Can be made very thin Low cost RPC is a perfect choice for a Digital Hadron Calorimeter!

DHCAL R&D started with small prototypes Small size RPC prototype Studied several RPC designs and narrowed down to 1-gap design Measured RPC parameters: efficiency, multiplicity, noise, etc. Invented 1-glass RPC Small size DHCAL readout prototype Prototyped the whole DHCAL readout chain 4-asic, 256ch, 16x16cm2 active area FEB/pad board Thoroughly debugged Small DHCAL prototype – the slice test stack Up to 9 layers Tested with muon, positron and hadron beams Analysis done: 6 papers published on NIM, JINST Gave us valuable experience and confidence in building larger prototype

1 m3 – Digital Hadron Calorimeter Physics Prototype Description Readout of 1 x 1 cm2 pads with one threshold (1-bit) → Digital Calorimeter 38 layers in DHCAL and 14 in Tail Catcher, each ~ 1 x 1 m2 Absorber: 16mm Fe + (2mm Fe + 2mm Cu [cassette]) or 10mm W + (2mm Fe + 2mm Cu [cassette]), thicker Fe plates in Tail Catcher Each layer with 3 RPCs, each 32 x 96 cm2 ~500,000 readout channels Purpose Validate DHCAL concept Gain experience running large RPC systems Measure hadronic showers in great detail Validate hadronic shower models (Geant4) Status Started construction in 2008 Completed in January 2011 Test beam runs with Fe absorbers started in Oct. 2010 at Fermilab Finished Fermilab test beam by the end of 2011 Test beam runs with W absorbers at CERN in 2012 Critical for PFA validation 7 7

RPC Construction RPC design RPC production Resistive paint spraying Glass Pad board Frame RPC design 2 – glass RPCs (chosen for construction) 1 – glass RPCs (developed at Argonne) Gas gap size 1.1mm Well arranged gas flow (by fishing line + sleeve spacer) Total RPC thickness < 3.4mm Dead area ~5% (frame ~3%, spacer ~2%) RPC production ~114 for DHCAL + 42 for TCMT + spares  at the end, produced ~ 205 RPC’s Resistive paint spraying Uses two-component artist paint Built dedicated spraying booth Accept plates with value 1 – 5 MΩ/□ Achieved a yield of ~60% overall Gap assembly Developed precision cutting/gluing fixtures Production rate 1RPC/day/tech 8 8

Quality Assurance Pressure tests Gap size measurement HV tests All RPC’s are tested with 0.3 inch of water pressure Gap size measurement Thickness of 1st batch RPC’s measured along the edges (Gap size away from the edges is assured by spacers) Gap sizes along edges vary within 0.1 mm Corners typically thicker (up to 0.3mm, but limited effect) HV tests ALL RPC’s tested up to 7.0 kV before placing readout board on top (nominal operating voltage is 6.3 kV) Cosmic ray test 9 of the 1st batch RPC’s were tested on a cosmic ray test stand for performance (efficiency, multiplicity, uniformity) All other RPC’s were later tested in vertical position with cosmic rays, after assembled into cassettes Overall yield: ~95% 9

Design consideration for readout system System is built around a custom ASIC (DCAL chip) Designed to handle huge number of channels + low channel density (1cm2 pad, 1m2 planes, 54 planes  497,664K channels) 64ch, handling 8x8 RPC active area System tailored for test beam / prototype tests, NOT for real colliding beam detector Facilitates all possible tests (test beam, cosmic ray test, noise runs, system diagnostics, etc) Avoided cutting-edge technology and fancy functionality Didn’t optimize for minimum power consumption, minimum data links, minimum thickness, etc. Compromises aimed at proving detector concept  key in getting system up quickly and running reliably Two basic running mode External triggering mode  primary method for beam events, also used for cosmic ray test 20-stage pipeline  2μs latency @ 100ns clock cycle Deadtimeless readout (within rate limit) Self triggering mode primary method for noise runs, also used for cosmic ray, beam runs Powerful running mode for monitoring RPC condition Equally powerful for cosmic ray running and some beam running as well

Readout system overview VME Interface Data Collectors – Need 10 Timing Module Double Width - 16 Outputs Ext. Trig In master IN 6U VME Crate To PC Data Concentrator Front End Board with DCAL Chips & Integrated DCON Chambers – 3 per plane Data Collectors – Need 10 VME Interface master IN Ext. Trig In To PC Timing Module Double Width - 16 Outputs Communication Link - 1 per Front-End Bd 6U VME Crate Square Meter Plane

FrontEnd/DCON board + Pad board The DCAL Chip Developed by FNAL and Argonne Input 64 channels High gain (GEMs, micromegas…) with minimum threshold ~ 5 fC Low gain (RPCs) with minimum thrshold ~ 30 fC Readout mode Triggerless (noise measurements) Triggered (cosmic, test beam) Versions DCAL I: initial round (analog circuitry not optimized) DCAL II: some minor problems (used in vertical slice test) DCAL III: no identified problems (final production verision) produced 11 wafers, 10,300 chips for DHCAL physics prototype Hits/100 tries Threshold (DAC counts) FrontEnd/DCON board + Pad board Build FE and pad boards separately to avoid blind and buried vias (cost and feasibility issue) Each board contains 1536 channels and 24 ASICs The data concentrator is implemented into the same board Glue the two boards together with conductive epoxy 12

Tests in the Fermilab Test Beam Fermilab Test Beam Facility 120 GeV primary proton beam 1 – 60 GeV/c secondary beam (pions, positrons, muons…) Broadband muon beam 3.5 second long spills every minute Muon data taking +32 GeV/c secondary and 3 m Fe absorber DAQ rates 500 – 2000/spill DAQ triggered by a pair of 1 x 1 m2 Scintillator paddles Secondary beam data taking DAQ rates 50 – 500/spill DAQ triggered by a pair of 19 x 19 cm2 Scintillator paddles 1 x 1 m2 Scintillator Paddle A 1 x 1 m2 Scintillator Paddle B DHCAL Tail Catcher Trigger DHCAL Tail Catcher Trigger Cerenkov 19 x 19 cm2 Scintillator A 19 x 19 cm2 Scintillator B

The DHCAL in the Test Beam Fermilab test beam run dates DHCAL layers RPC_TCMT layers SC_TCMT layers Total RPC layers Total layers Readout channels 10/14/2010 – 11/3/2010 38 16 54 350,208+320 1/7/2011 – 1/10/2011 8 46 350,208+160 1/11/2011 – 1/20/2011 4 42 50 387,072+160 1/21/2011 – 2/4/2011 9 6 47 53 433,152+120 2/5/2011 – 2/7/2011 13 51 470,016+0 4/6/2011 – 5/11/2011 14 52 479,232+0 5/26/2011 – 6/28/2011 11/2/2011 – 12/6/2011 460800 Run I Run II Run III Run IV Run V ~ 480K readout channels ~ 35M events DHCAL Tail Catcher (TCMT)

Some event display One muon 60 GeV pion 32 GeV positron Occasionally, neutral hadron

Energy reconstruction in the DHCAL Data Consists of hit patterns of pads with signal above 1 threshold and their time-stamps with 100 ns resolution Incident particle energy reconstruction To first order E ∝ N N = ∑layer Ni … total number of hits Correction for contribution from noise E ∝ N - Nnoise Nnoise … accidental hits Correction for variation in chamber inefficiency E ∝ ∑layer Ni ·(ε0 /εi) – Nnoise ε0 … average DHCAL efficiency εi … efficiency of layer i Correction for variation in pad multiplicity E ∝ ∑layer Ni ·(ε0 /εi) ·(μ0 /μi) – Nnoise μ0 … average pad multiplicity ` μi … average pad multiplicity of layer i Second order corrections Compensate for e/h ≠ 1 Saturation (more than 1 particle/pad) …

General DHCAL Analysis Strategy Noise measurement - Determine noise rate (correlated and not-correlated) - Identify (and possibly mask) noisy channels - Provide random trigger events for overlay with MC events (if necessary) Measurements with muons - Geometrically align layers in x and y - Determine efficiency and multiplicity in ‘clean’ areas - Simulate response with GEANT4 + RPCSIM (requires tuning 3-6 parameters) - Determine efficiency and multiplicity over the whole 1 x 1 m2 - Compare to simulation of tuned MC - Perform additional measurements, such as scan over pads, etc… Measurement with positrons - Determine response - Compare to MC and tune 4th (dcut) parameter of RPCSIM - Perform additional studies, e.g. software compensation… Measurement with pions - Compare to MC (no more tuning) with different hadronic shower models - Perform additional studies, e.g. software compensation, leakage correction…

Estimation of contributions from noise Data collection (in general) Trigger-less (all hits) mode for noise, cosmics Triggered (record hits in 7 time bins of 100 ns each) for noise, cosmics, testbeam → Only hits in 2 time bins used for physics analysis Noise measurement These results from trigger-less mode In quantitative agreement with measurements with random trigger Results Noise rate measured to be 0.1 – 1.0 Hz/cm2 Rate strongly dependent on the temperature of the stack (FE-electronics embedded into stack and generates heat) Tail catcher in 4/2011 run DHCAL in 4/2011 run DHCAL in 10/2010 run Noise rate [Hz/cm2] 0.1 0.5 1.0 Nnoise/event in DHCAL + Tail Catcher (2 time bins) 0.009 0.05 0.09 Nnoise/event in DHCAL + Tail Catcher (7 time bins) 0.033 0.165 0.33 1 hit corresponds to ~ 60 MeV! Contribution from noise negligible for most analysis

Tracking Clustering of hits Loop over layers 1 cluster 2 clusters Performed in each layer individually Use closest neighbor clustering (one common side) Determine unweighted average of all hits in a given cluster (xcluster ,ycluster) Loop over layers for layer i request that all other layers have Njcluster ≤ 1 request that number of hits in tracking clusters Njhit ≤ 4 request at least 10/38(52) layers with tracking clusters fit straight line to (xcluster,z) and (ycluster,z) of all tracking clusters j calculate χ2 of track request that χ2/Ntrack < 1.0 inter/extrapolate track to layer i search for matching clusters in layer i within record number of hits in matching cluster

Calibration constants, etc… Calibration factors = mean of multiplicity distribution/(average over detector) = ε·μ/ ε0·μ0

Simulation Strategy Comparison Parameters GEANT4 Experimental set-up Beam (E,particle,x,y,x’,y’) Measured signal Q distribution Points (E depositions in gas gap: x,y,z) GEANT4 RPC response simulation Hits Parameters Distance cut dcut (within which only 1 avalanche) Charge adjustment Q0 (if needed) Exponential slopes a1, a2 (of signal spread in pad plane) Ratio R (between 2 exponentials) Threshold T (of discriminator) Hits Comparison DATA With muons – tune a1, a2, R, T, (dcut), and Q0 With positrons – tune dcut Pions – no additional tuning 21

x = Mod(xtrack + 0.5,1.) for 0.25 < y < 0.75 y = Mod(ytrack – 0.03,1.) for 0.25 < x < 0.75 Scan across one 1 x 1 cm2 pad Note These features not implemented explicitly into simulation Simulation distributes charge onto plane of pads…

Efficiencies, multiplicities Select ‘clean’ regions away from - Dead ASICs (cut out 8 x 8 cm2 + a rim of 1 cm) - Edges in x (2 rims of 0.5 cm) - Edges in y (6 rims of 0.5 cm) - Fishing lines (12 rectangles of ±1 cm) - Layer 27 (with exceptionally high multiplicity) Measured average response Add in dead regions Note: Simulation of RPC response tuned to this data Tail towards higher multiplicity reproduced with 2nd exponential

Analysis of Secondary Beam Data: Topological Particle ID Muons: All active layers have aligned clusters with no more than two consecutive layers with nonisolated clusters. Pions: At least one track segment in the interaction region that spans at least four layers and is not compatible with the beam direction. If such a track segment is not found, at least one pair of track segments that span three layers with at least 20o angle in between. Or remaining events with rrms > 5 cm. Positrons: rrms < 5 cm. Where rrms is defined as,

Unidentified μ's, punch through Results - October 2010 Data CALICE Preliminary Gaussian fits over the full response curve Unidentified μ's, punch through 25

Pion Response N=aE As expected, linear up to > 25 GeV CALICE Preliminary (response not calibrated) N=aE Standard pion selection + No hits in last two layers (longitudinal containment 16 (off), 32 GeV/c (calibration off) data points are not included in the fit. As expected, linear up to > 25 GeV 26

Energy Resolution for Pions (response not yet calibrated) CALICE Preliminary (response not yet calibrated) B. Bilki et.al. JINST 4 P10008, 2009 MC predictions for a large-size DHCAL based on results from small scale prototype. 32 GeV data point is not included in the fit. Standard pion selection + No hits in last two layers (longitudinal containment) Measurements confirm prediction 27

(response not yet calibrated) B. Bilki et.al. JINST 4 P04006, 2009 Positron Response CALICE Preliminary (response not yet calibrated) N=a+bEm Non-linearity Due to saturation (more than 1 hit/pad) Can be improved with software compensation Not detrimental: this is a hadron calorimeter! Data (points) and MC (red line) from small prototype test and the MC predictions for a large-size DHCAL (green, dashed line). Measurements confirm prediction 28

Positron energy resolution CALICE Preliminary (response not yet calibrated) Un-corrected for non-linearity Corrected for non-linearity

The DHCAL at CERN Some repairs done before CERN test beam Some RPC’s are taken out to fix gas leakage – they are repaired, retested and insert back to the DHCAL A small number of RPC’s are replaced due to low efficiency 2 more DHCAL layers were produced They are made out of the spare parts 54 layers (96 x 96 cm2) of RPCs with 1 x 1 cm2 readout pads Up to 497,664 readout channels (world record for calorimetry and RPC systems) Transport to CERN Built spring-damped transport fixture All RPCs survived intact Installation at CERN 39 layers into Tungsten (1 cm ~ 3 X0 plates) absorber structure 15 layers into Steel (2 cm and 10 cm plates) tail catcher

Beams at CERN PS Trigger recorded so far (million events) SPS Testbeam Covers 1 – 10 GeV/c Mixture of pions, electrons, protons, (Kaons) Two Cerenkov counters for particle ID SPS Covers 12 – 300 GeV/c Mostly set-up to either have electrons or pions (18 Pb foil) Trigger recorded so far (million events) Testbeam Configuration Muons Secondary beam Total CERN2 DHCAL + TCMT 5.6 23.4 29.1

300 GeV pion showers

A few event variables The interaction layer Longitudinal barycenter 100 GeV π simulation The interaction layer Algorithm tuned with Monte Carlo events Defined as first of two consecutive layers with more than 3 hits Longitudinal barycenter with Ni ... hits in layer I Hit density R with sgn(Ni ) = 1 for Ni > 0, = 0 for Ni = 0 80 GeV ‘pion’ beam Muons Pions

Event selection General cut: 1 cluster in layer 0 with less than 12 hits Particle selection: Particle Cerenkov   BC R IL N0 μ >20 <3.0 - >0 >10 e± C1·C2=1 <8 >4.0 for E>12 GeV >4 for π- C1+C2=0 >2.0 – 5.0 >2 for E>3 GeV π+ C1=0 and C2=1 (p ≤ 10 GeV/c) (p > 10 GeV/c) p BC … Longitudinal barycenter R … Average number of hits per active layer IL … Interaction layer N0 … Hits in layer 0

Response at the PS (1 – 10 GeV) Fluctuations in muon peak Data not yet calibrated Response non-linear Data fit empirically with αEβ β= 0.90 (hadrons), 0.78 (electrons) W-DHCAL with 1 x 1 cm2 Highly over-compensating (smaller pads would increase the electron response more than the hadron response) Remember: W-AHCAL is compensating!

Resolution at the PS (1 – 10 GeV) Resolutions corrected for non-linear response Data fit with quadratic sum of constant and stochastic term Particle α c Pions (68.0±0.4% (5.4±0.7)% Electrons (29.4±0.3)% 16.6±0.3)% (No systematics yet)

Comparison with Simulation – SPS energies Data Uncalibrated Tails toward lower Nhit Simulation Tuned with Fermilab data Rescaled to match peaks Shape surprisingly well reproduced

Response at the SPS (12 – 300 GeV) Fluctuations in muon peak Data not yet calibrated Response non-linear Data fit empirically with αEβ β= 0.85 (hadrons), 0.70 (electrons) W-DHCAL with 1 x 1 cm2 Highly over-compensating (smaller pads would increase the electron response more than the hadron response)

Further R&D Large single glass RPC High rate RPC ANL invention: thinner, can go to smaller pad sizes Built and tested 2 large single glass RPCs (32x48cm2) Showed good efficiency and uniformity High rate RPC Several approaches are going in parallel: low resistivity material, new RPC design Making progress on every approach: new glass sample, new Bakelite board, new RPC R&D towards a realistic DHCAL Need: low power, better timing resolution (~1ns), better mechanical design, gas recycling, HV distribution Have some ideas already In discussion with new collaborators

Summary Digital Hadron Calorimeter (DHCAL) concept has been proven by a large DHCAL physics prototype Test beam at Fermilab with Fe absorber is completed, data analysis made a lot of progress already Test beam at CERN with W absorber is ongoing, and will finish next week. First look at the data suggests that the data quality is very good Further R&D has started

Additional slides

Cassette Assembly Assembly Cassette Testing - FEB’s are placed onto RPC’s directly (no gluing), positioning and contact is assured by pressure asserted from cassette - Cassette is compressed horizontally with a set of 4 (Badminton) strings - Strings are tensioned to ~20 lbs each, very few broken strings Cassette Testing - Cassettes were tested with CR before shipping to test beam 38+14 cassettes assembled 42

Calorimeters/configurations during Fermilab test beam DHCAL 38 layers, each 1 x 1 m2 1.2 X0 Fe (+ Cu) absorber plates 350,000 readout channels Tail Catcher 14 layers, each 1 x 1 m2 8 1.7 X0 Fe (+Cu) + 6 5.9 X0 Fe (+Cu) absorber plates 129,000 readout channels Silicon – Tungsten Electromagnetic Calorimeter 30 layers, each 18 x 18 cm2 Variable thickness Tungsten absorber 10,000 readout channels Minimal Absorber Structure 50 layers, each 1 x 1 m2 0.27 X0 Fe (+Cu) absorber plates 461,000 readout channels

Events taken with Minimal Absorber Stack 50 layers Cassette contain 2 mm Cu and 2 mm Fe 0.27 X0/layer → 13.4 X0 total 0.038 λI/layer → 1.9 λI total Beam 1,2,3,4,6,8,10 GeV/c secondaries 8 GeV e+ 16 GeV/c π+ 8 GeV/c π+