SiD Calorimeter R&D Report

I. Overview of SiD Calorimeter R&D.

The ILC physics program places strong requirements on the performance of calorimeter systems. Many production channels to be studied require efficient identification and separation of jets in multi-jet final states (e.g. in e+e- → t tbar → 6 jets), and the measurement of jet energies and jet-jet invariant masses with unprecedented precision. Our goal is to measure jet energies at the level of σ/E ~3-4%. Event-by-event separation of hadronic decays of W- and Z-bosons and Higgs particles will be necessary to measure cross-sections and branching ratios at the few percent level of error. The calorimeter system is a critical element of the SiD detector concept design for achieving these goals. The calorimeter will also provide excellent electromagnetic energy resolution and efficient π0 reconstruction, charged hadron/photon shower separation, and tau lepton reconstruction.

The SiD detector design is based on an all-silicon tracking system which allows for a limited inner radius for the central calorimeter system. This, in turn, constrains the cost of the silicon-tungsten electromagnetic calorimeter, and limits the outer radius of the hadron calorimeter, thereby also constraining the size and cost of the superconducting coil. It is taken as a basic design parameter that both the electromagnetic and hadron calorimeters should lie inside the coil so as not to degrade their abilities to make particle-energy cluster associations. A schematic view of part of the SiD concept is shown in Fig.1, which indicates the dimensions and location of central and endcap components of the calorimeter system.

Fig. 1 Quadrant view of the SiD detector design concept.

The forward calorimeter systems for SiD will be covered in a separate report by the FCal collaboration, and so are not discussed here.

A promising approach for achieving the unprecendented jet energy resolution lies in the use of a Particle Flow Algorithm (PFA). We have invested considerable time and effort

1 in the development of such algorithm(s), and in understanding the implications of a PFA for the SiD detector design. This approach uses the tracking system to measure the energy of the charged hadron component of a jet and the electromagnetic calorimeter to measure photon energies. Accordingly, the neutral hadron energy is the only component of jets to be measured directly in the hadron calorimeter. Leakage from the rear of the hadron calorimeter could be estimated by the use of a “tail-catcher”, which would also be the first part of the muon system.

The measurement of jet energies using data from several subsystems leads to a very integrated view of the design of the SiD detector. The tracker must find the charged particles with high efficiency and ensure accurate projection of the tracks into the calorimeter to make the correct track/energy cluster associations. The electromagnetic calorimeter must be sufficiently fine grained to provide efficient charged particle/photon shower separation. The hadron calorimeter must be sufficiently fine grained to allow efficient track following for charged hadrons and their hadronic shower products, and to provide an approximately linear hits vs. energy for the digital hadron calorimeter case.

Simulations of PFA performance for light jet production at the Z-pole are beginning to show excellent performance. Intense work is underway to optimize the algorithms for the higher jet energies expected for higher center of mass energies. Prudence suggests we also remain open to the investigation of more traditional calorimetry or more radical, alternative ideas.

In the PFA approach, the electromagnetic energy resolution does not impose limits on jet energy resolution: the current Si-W design is expected to have an electromagnetic shower resolution of 17%/√E. However, the segmentation, both longitudinal and transverse, is a critical consideration for the use of PFAs for hadronic jets. Transversely, not only should the effective Moliere radius be kept small, but also the transverse cell size should be about half of this radius. Longitudinally, the active gap between absorber layers should be kept small to achieve a small effective Moliere radius, limiting transverse shower spread. A sufficient number of longitudinal layers need to be provided to contain electromagnetic showers and allow efficient recognition and reconstruction of these showers. The resulting highly segmented electromagnetic calorimeter, in addition to having the correct characteristics for use with PFAs, will then have outstanding performance for the reconstruction of electrons, photons, and tau leptons and for tracking charged hadrons and muons.

The choice of the silicon-tungsten combination for the electromagnetic calorimeter is a centerpiece of the original SiD design. The Moliere radius for tungsten is 9mm, and our present design has a pixel area of 12mm2. Longitudinally, there are 30 alternating layers of tungsten and silicon giving a total thickness at normal incidence of 29X0. The goal for the active gaps is 1mm width or less. Each group of approximately 1000 silicon pixels is read out via one KPiX-ASIC mounted on the wafer. The first version of the silicon wafers have been delivered and evaluated. The medium-term plan is to assemble a prototype calorimeter section based on the second generation silicon wafers in

2 conjunction with the 1024-channel KPiX readout. The electromagnetic calorimeter is discussed in detail in Section II.

The KPiX ASIC is being developed for the readout of the electromagnetic calorimeter, and the central silicon tracker for SiD, and has applicability to several technology options for the hadron calorimeter. The high energy density for electromagnetic showers requires a large dynamic range for the readout. The KPiX chip uses a novel method by which the feedback path on the front end amplifier can be switched between two capacitors, switching in the large (10 pF) capacitor only when it is required. This allows the amplified charge for smaller input signals to be well above the noise. There is an event threshold, which can hold off bunch crossing resets in order to allow a fairly long integration time of 1 s. The calculated noise level is about 1000 electrons, to be compared with a MIP signal charge of 25 times this. Charge digitization uses two overlapping 12-bit scales. The chip also allows up to four hits per bunch train to be stored for each pixel. Several prototype versions of KPiX have been fabricated and tested, with encouraging results for the present 64-channel version. A version of a 64-channel KPiX has also been developed to read out the gas electron multiplier based version of the hadron calorimeter active layers. Details of the KPiX development are given in section III.

The hadron calorimeter in our PFA-oriented, baseline design consists of 35-40 layers of alternating absorber plates and active sampling gaps. The critical feature of this calorimeter is the small transverse cell size, O(1-3cm), required for the imaging requirements imposed by a PFA. Depending on the approach taken, this device could be analog, semi-digital, or fully digital. There is a premium on a small active gap size as this has a large impact on the overall detector cost. The full depth of the baseline hadron calorimeter is 4λI. The basic mechanical structure is foreseen to have three main sections: a barrel and two endcaps. The barrel is subdivided along the beam direction into three sections, and there are twelve modules in azimuth.

The baseline material choice for the absorber is steel, although we have also considered tungsten and brass, with plates having thickness of 20mm or 1.1 X0. In contrast four technologies, three gaseous and one plastic, are under consideration for the active layers of the hadron calorimeter. These range from resistive plate chambers, gas electron multipliers, and micromegas for digital implementations, to small scintillator tiles for semi-digital and analog implementations. Prototypes of each technology have been built and evaluated.

We have had extensive discussions on the process for selection of the best technology for the SiD hadron calorimeter. We have enumerated a set of requirements for the calorimeter and selection criteria for the technologies, and the steps that should be followed to reach a decision. These requirements and criteria incorporate the demands of PFAs. For instance, our PFA studies are now focusing on center-of-mass energies beyond the Z-pole, and consequently, higher energy, more collimated, jets. This helps us

3 understand the requirement for the transverse cell size. Also our simulations have highlighted the pros and cons of neutron detection. The issue is whether we gain in the PFAs from the use of scintillator to see neutron clusters, or whether it is better to use a gas active medium to suppress the neutron signal and thereby simplify the track-cluster association in the PFA. Also at issue is the comparative ease of construction and cost differences between the various technologies. A full discussion of the hadron calorimeter is given in Section IV.

The option of installing a “tail-catcher” after the ~1.5 λI of the superconducting coil and cryostat, to potentially identify and measure the last few percent of hadron shower energies, is under discussion. Recent prototype results from CALICE data taking at CERN show improved energy resolution when the tail-catcher data is included. The technology for implementation of the tail-catcher would follow that of the muon system: currently scintillator strips and resistive plate chambers are being considered. The tail- catcher is also discussed in Section IV.

Basic simulation studies of hadron calorimetry are being carried out in the SiD context. The parameters studied include the absorber material, the active layer material, the longitudinal and transverse segmentation, the total thickness of the hadron section, the thickness (λI) per layer, the inner radius, and the magnetic field. The idea is to work from a basic understanding of single particle energy resolutions, through jets of known energy(s), to full dijet mass resolution in physics events. The single particle studies already reveal interesting and significant differences in energy resolution between the combinations of steel and tungsten absorber and scintillator and gas active media that will be essential input to decisions on the hadron calorimeter. Studies of jet energy resolutions and neutral energy measurement are ongoing. Section V discusses this simulation work.

Initially the study and development of PFAs for SiD was undertaken by a number of independent physicists at several different locations. In order to bring some structure into these studies, and provide a common basis for comparison of algorithms within the various PFAs, a framework has been produced by a collaboration between SLAC and ANL. Initial studies for SiD, as for the other concepts, have focused on the Z-pole region. Based on physics processes requiring the separability of W and Z bosons, the initial goal for the PFAs was set at achieving an energy resolution at the Z-pole of σ/E ~30%/√E. However, more recent considerations particularly at higher center-of-mass energies have revised this goal to σ/E ~3-4%. The current state-of-the-art for SiD PFAs is σ/E ~35%/√E. We are approaching the level of understanding of the PFA performance that will provide the basis for meaningful comparisons of technology choices for the hadron calorimeter,

DAQ, fiber cables, data rates?

Backgrounds in the electromagnetic calorimeter, mainly low energy photons from the beamcal, and photons from gamma-gamma interactions, produce an occupancy rate of at most 1 x 10-4/BX in the highly segmented silicon pixel detectors. This should be an

4 inconsequential effect in calorimeter pattern recognition, and the electronic buffer size is more than adequate to ensure full efficiency.

We are forming a SiD Engineering Group, with initial participants from SLAC and Fermilab, to carry out a preliminary engineering study of the SiD detector design. This study will consider the basic design of the calorimeter modules, materials, support of the barrel and endcap calorimeters and the superconducting coil, assembly procedures and magnetic force effects .This generic study will assume a calorimeter inside the coil and active gaps of order 1cm.

Stability, calibration, push-pull ?

In this report, we present the electromagnetic calorimeter and its readout in detail. For the hadron, we summarize the R&D relevant to the SiD calorimeter design and technology selection. The details of the hadron calorimeter and TCMT R&D will be discussed in the CALICE report submitted to this review. We include a section on basic simulation studies as these offer valuable insight into the relative importance of various materials and size parameters for the calorimeter design. Finally, we summarize the results from SiD-based PFA studies that also bear on the optimization of this design.

5 II. Electromagnetic Calorimeter

1. Introduction

In addition to its usefulness for reconstruction of hadronic final states using the PFA, as discussed in Section I, a highly-segmented, imaging ECal such as we propose will have important advantages for carrying out other critical elements of the overall ILC physics program. We outline some of these benefits in the next section. In general, we expect pattern recognition to be critical for identifying and quantifying final states, which may include hints of new physics in complicated high-energy events. An imaging ECal will have excellent capabilities for identifying and measuring EM showers in three dimensions, and separating these from other particles.

Hence, the Si/W ECal we propose is not only a necessary element of a detector which uses PFA methods for jet reconstruction, but its general ability to aid in the identification and measurement of many types of particles makes it a very powerful general-purpose tool for the ILC physicist. Even if one were to use a non-PFA jet reconstruction method, the benefits of an imaging ECal would still make it a very attractive choice for a multi- purpose ILC detector.

Statement of main technology challenges

Si/W with integrated readout – meets requirements and is viable as a systems solution

Table E-I. Main parameters of the SiD silicon/tungsten ECal inner radius of ECal barrel 1.27 m maximum z of barrel 1.7 m

longitudinal profile (20 layers x 0.64X0 ) + (10 layers x 1.3X0 ) silicon segmentation 12 mm2 (hexagonal) pixels readout gap 1 mm effective Moliere radius 13 mm pixels per readout chip 1024

2. Physics Requirements

(Re-)statements of cherished beliefs (this is mostly in Section A)

We need to reconstruct hadronic final states because LHC can’t. We should strive to do it as well as possible since we don’t know what the New Physics is.

6 We still need to do the other “easy” measurements as well or better than LEP/SLD

The few mm segmentation possible with a Si-W ECal provides outstanding capability for reconstructing individual photons and for separating the photons from other photons and from charged hadrons. At the same time, good electromagnetic energy resolution is achievable --- designs to date have used sampling fractions which give EM resolution of approx 15%/sqrt{E}. So far, there is not a strong physics argument for providing better EM resolution than this. In addition, such an ECal will provide excellent lepton reconstruction. In particular, tau final states, so crucial for many signatures of new physics, will greatly benefit from such a highly-segmented ECal. In fact, these are imaging calorimeters. It is likely that we will continue to see benefits from these designs as the physics simulation studies become more detailed, for example in flavor tagging.

. Some of the “easy” things rely on the ECAL:

electrons: electron id, Bhabhas, Bhabha acollinearity and luminosity

photons: photon id;

Some other capabilities of imaging ECals:

pi0 reconstruction -- improvement of jet energy determination

charged particle tracking, esp V0's

photon vertexing

Re-emphasize raisons d’etre:

tau id, recon of certain final states, tau polarization

jets: separation of photons (pi0s) from charged hadrons; separation of photons from neutral hadrons

Specific physics examples – perhaps already in the Overview

7 3. Proposed Technical Solution

One of the outstanding technical questions is how to integrate a silicon detector wafer with its readout electronics. Since the number of detector pixels for these ECal designs is on order 50 million, a solution to the integration issue, along with the cost of the silicon detectors themselves, is likely to determine the overall viability of the Si-W approach. A few years ago, we proposed [mibtalk,reftalk] a possible solution to the integration problem and have received LCDRD support for three years to pursue this. The integrated approach can accommodate a design with high transverse segmentation (currently 12 mm2) and a small readout gap (currently 1 mm) to maintain a small Molière radius. In its application to the SiD concept ECal, this design includes 30 longitudinal layers, with 20 layers, each of thickness approx 5/7 X0, followed by 10 layers of thickness approx 10/7 X0.

The thrust of our project is to integrate detector pixels on a large, commercially feasible silicon wafer, with the complete readout electronics, including digitization, contained in a single chip which is bump bonded to the wafer. The starting point for our design uses a pixel size of 12 mm2, based on initial PFA requirements for photon- hadron separation. This gives N103 pixels per 6-inch wafer. We take advantage of the low beam-crossing duty cycle (10-3) to reduce the heat load using power cycling. Our design then has several important features:

1. The electronics channel count is effectively reduced by a factor N=1024. 2. A transverse segmentation down to a few mm can be naturally accommodated. 3. The cost, to first order, will be independent of the transverse segmentation. 4. Readout gaps can be small (1 mm), thus maintaining the small Molière radius intrinsic to tungsten.

The first property, we feel, is necessary for any realistic highly-segmented ECal. In this case, the electronics is likely to be relatively small fraction of the ECal cost. The third point makes the design flexible, so that one can optimize to meet the physics goals. The fourth is an optimization of the physics capability of the ECal at a given (barrel) radius. For example, the angle subtended by the Molière radius for an ECal at radius 1.27 m with our design is smaller than one with 3 mm readout gaps at 1.7 m. Hence, this has a significant impact on both performance and overall detector cost. We note that for a Si-W ECal, the features above remain unique to this R&D. Figure E-1 is a schematic of the readout gap cross section.

8 Fig E-1. Readout gap schematic. A few of the 1024 bump bond connections are shown for illustration,

Mechanical, thermal, power pulsing

4. Current R&D Status and Plans

The main goal of our R\&D collaboration is to fabricate a full-depth ECal module using the integrated technology outlined above, and to characterize the module in a test beam. The responsibilities for this R&D break down roughly as follows:

 Readout chip (KPiX) design: SLAC  KPiX checkout: SLAC and Oregon  Silicon detectors: Oregon  Bump bonding: Davis  Readout cable: Davis  Mechanical design: all

During the past year, the initial design of the KPiX readout chip was completed and was sent to industry. To date, four cycles of prototypes have been fabricated. The prototypes are 64-channel versions of the eventual 1024-channel design, implemented with 2 of the 32 32-channel rows. The most recent version is KPiX-4. The KPiX readout chip is described in more detail in Section D. Figure E-2 shows the KPiX-2 chip on the test bench (left). Figure E-3 illustrates one of the main novel features of the KPiX chip, namely that the range will dynamically switch. Thus it can handle the huge expected dynamic range for a tungsten ECal at the ILC. In the plot, as the injected

9 charge is increased, we see the range switch at about 700 fC. For 320 micron silicon, 1 MIP is equivalent to about 4.1 fC. Thus the upper end of the plot corresponds to about 2500 MIPs, more than the expected maximum signal for a 500 GeV electron at shower maximum in our 12 mm2 pixels (near the center of the barrel). Besides providing for the maximum charge, the range switching also maintains good signal/noise for MIPs.

Figure E-3. Output ADC counts vs input charge, indicating the dynamic range change at 700 fC.

Figure E-2. KPiX in the lab.

We have also made progress characterizing the prototype silicon detectors. In fact, we understand the main properties of these detectors, and have incorporated this understanding in the design of a new version of detectors, as discussed below, which will be used to outfit the full-depth ECal prototype.

A group at UC Davis has developed a novel readout cable concept which will help maintain the thin readout gap. Davis will also will provide the bump bonding for the initial prototypes.

Our work gives us confidence that we are on the right path and we hope to demonstrate the detector concept with prototypes in an electron test beam --- and to move on to the next phase: the development of a full-depth ECal module which includes the features required for a realistic ILC detector. This module could be part of an international test beam study.

10 4.1 Detectors

In 2004 we obtained a set of ten 6 inch Hamamatsu (version 1) prototypes. We have characterized these detectors The most important measurements for their eventual use in the ILC are the measurements of stray capacitance and leakage current. We have also investigated the use of a radioactive source for an absolute calibration. These results have been presented at previous meetings [LCWS05-strom}.

Pixel Capacitance and Trace Resistance

In almost all cases the noise of an individual pixel charge measurement in our detector will be directly proportional to the total capacitance seen at the input of the amplifier. The capacitance of a fully depleted pixel in detectors 320 $\mu$m thick is expected to be approximately 5.3 pF. The majority of the capacitance is due to the stray capacitance of the traces which connect the individual pixels to the bump-bonding array.

For the Hamamatsu process used in our prototype detectors, the thickness of the oxide to the second metal layer is approximately 0.9 m. In our detector we used 6 m thick traces, giving a theoretical capacitance of approximately 3.1 pF/cm. The total stray capacitance associated with a given pixel has two contributions. One contribution comes from the capacitance of the traces connecting the pixel to the bump- bonding array. The second contribution is due to any traces from other pixels which cross the pixel under test. The total stray capacitance is almost constant for many of the pixels as can be seen from Figure E-4. In region a it can be seen that pixels located closer to the bump-bonding array have a greater number of crossing traces than those further away. This gives an almost constant total measured (and calculated) stray capacitance. Typical stray capacitances in region a were 22 pF. An example run is shown in Figure E-5.

Figure E-4. Specific signal traces. Figure E-5.CV curve for pixel in region a.

11 A small fraction of the pixels have a very large number of crossing traces. These pixels are located in regions b and c of Figure E-4. For the first prototypes these pixels have capacitances of somewhat more than 100 pF. In the new version of the sensors we plan to reduce the stray capacitance in region c by narrowing the traces in the vicinity of the bump-bonding array.

Another important property of the detectors is series resistance of the traces. The noise contribution from this series resistance is proportional to Ctot R S where Ctot is the total input capacitance and RS is the series resistance. The contribution to the noise from the 2 input FET in a charge amplifier is proportional to where gm is the 3gm 2 transconductance of the input FET. Thus it is desirable to keep RS comparable to . 3gm 2 In our case we expect 籛300 . 3gm

From the measurement of trace resistance, we obtain 57� 2W / cm . This can be compared to an expected value of 47 W / cm for pure aluminum traces 1 m by 6 m. For the longest traces, of order 10 cm, the measured value implies a maximum resistance of 570 .

It would be desirable to reduce this trace resistance by making thicker traces. However, it is unlikely that the thickness can be increased much beyond its current value of 1m. Increasing the width of the trace is not helpful because it will increase the component of

Ctot from the traces connecting the pixels to the bump-bonding array almost linearly. Except in the region near the bump-bonding array, our trace width of 6 m is close to optimal.

Leakage Current

Leakage current can add an additional term to the electronic noise that grows with shaping time. Typical leakage currents in silicon pad detectors, such as the prototypes used here, have currents of a few nA$/$cm$^2$. The leakage current was measured during the capacitance tests and was found to be less than 2\,nA/pixel for pixels in the interior of the detector. In our tests the neighboring pixels and the guard ring were left ``floating''.For pixels on the edge of the detector, with the guard ring floating, the leakage current was less than 10\,nA/pixel. We expect the noise contribution for leakage current to be minimal; the expected contribution for a leakage current of 10\,nA and an integration time of 1\,$\mu$s is only 250 electrons.

Calibration

12 Calorimeters based on silicon are expected to be quite stable over time. The largest changes in calorimeter response will be due to changes in the electronics. The readout electronics are being designed with an internal calibration system that allows a wide variety of charges to be injected into each of the channels in the system. The accuracy of this system is expected to be limited by the knowledge of the values of the coupling capacitors incorporated into each of the channels in the readout chip. These capacitors are expected to be uniform, within a chip, with a spread of 1%. This spread is unlikely to have a noticeable contribution to the energy resolution of the calorimeter. Chip-to-chip variations could be larger.

One possibility is to calibrate each sensor after the readout chip has been bump bonded. A possible method for this calibration would be to use 60 keV photons from the decay of radioactive Am241. If the energy from these photons is fully contained in the silicon sensors they will give a signal of approximately 16,000 electrons. This is somewhat less than the MIP signal, but well above our noise floor.

The 60 keV photons will easily penetrate any mounting structures and printed circuit boards used in the testing and assembly of the calorimeter. However, the calibration must be done before the detector assemblies are placed between the tungsten sheets, as the photons will not efficiently penetrate the tungsten. We have used our laboratory electronics to measure the energy spectrum from the Am241 photons in the pixels as shown in Figure E-6.The width of the peak is consistent with the expected electronic noise.

Figure E-6. Signals from Am241 60 keV Figure E-7. Measured position photons observed in a typical of peak versus total pixel Hamamatsu silicon sensor pixel. capacitance.

As a demonstration of this technique we show the value of the photon Am241 peak versus pixel capacitance in Figure E-7. The peak shifts to lower values at large values of the capacitance because of the finite input capacitance of laboratory electronics. The line

13 corresponds to a “dynamic” capacitance of our laboratory electronics of Cdyn  790 pF which is consistent with the laboratory amplifier specifications.

In the KPiX readout chip planned for the final detector, the signal-to-noise for the Am241 peak will be about 8, which will broaden the peak considerably. Another important aspect of the planned readout electronics will be that a measurement of the charge be done relative to an external bunch clock rather than relative to the time of arrival of the photon, as was done in the laboratory. This will give an additional smearing of the observed spectrum of less than 5%. Thus we expect a total width for the Am241 60keV signal of approximately 15%.

The ADC in the planned detector readout will have a least-significant bit approximately equal to the expected noise. Therefore, if there were no systematics in the ADC it would be possible to calibrate each pixel to 1% with approximately 250 detected photons. For this calibration to be useful it will be necessary to relate the charge scale at 8 ADC counts to that at full scale readout. This is possible, but will require great care in the design of the calibration circuit on the readout chip. Somewhat easier, but still difficult, will be a wafer-to-wafer calibration at the sub-percent level. Here one can average over 1024 pixels/wafer. Again it will be necessary to relate the average charge scale at 8 ADC counts to the average full scale readout.

Cross Talk

We are continuing to study cross talk introduced by capacitive couplings between the channels. In general these have lead to cross talk at the 1% level or below. The cross talk is function of both the capacitive coupling and the properties of the readout electronics. While we have a qualitative understanding of the cross talk, we are continuing to work on a quantitative model and on incorporating the properties of the KPiX electronics into the model.

New Prototypes

14 4.2 KPiX – Gunther H

4.3 Readout cable and bump bonding

4.4 An ECal module

4.5 Test beam

5. The Uniqueness of our Approach

History Calice Convergence Differences There likely will be a Si/W ECal at the ILC – it is important to explore the attractive technical solutions (our approach is best at this point.)

15 III Readout Electronics for ECAL

1. Overview The ECAL system requires low power, compact, large channel-count readout electronics. The center piece of the system is the KPiX ASIC, a 1,024 channel integrated circuit with front-end processing and digitization. The KPiX ASIC will be bump-bonded directly onto Si-detectors in order to eliminate pitch-adapters or massive cabling. Figure 1 shows a view of the KPiX assembly. Up to 12 KPix/detector sets are interconnected using a flex- cable connected to digital concentrator boards located at the perimeter. About 50k KPiX ASICs are required to instrument the 500k channel ECAL barrel.

Metallization on detector from KPix to cable Bump Bonds

Tungsten

Kapton Data Cable KPix Si Detector

Kapton

Tungsten

Thermal conduction adhesive Figure 1: 1,024 channel KPiX ASIC bumpbonded to the detector.

16 To take advantage of the moderate occupancy in the ECAL and the 199 msec spacing between pulse-trains at the ILC, only signals above a programmable threshold are digitized and transmitted off chip before the next train arrives. Figure 2 shows a simplified block-diagram of one of the 1,024 channels on the chip. The detector signal is amplified and shaped and compared to a set of threshold voltages. Auto-range gain switching circuitry is included to achieve a total dynamic signal range of 15 bits. A signal exceeding the threshold is analog stored in an on-chip capacitor. Up to 4 events can be self-triggered and acquired in each of the 1,024 channels. That exceeds the minimum requirement of two during a pulse-train. The amplifier is reset before each beam crossing unless the signal exceeded the threshold, causing the resets to stop until storage of the signal is completed, equivalent to double-correlated sampling The analog stored signals of the selected gain range are digitized via 13-bit Wilkinson Analog-to-Digital converters (ADC). There is one ADC in each channel. Depending on the occupancy in a bunch train, up to 4k digitized events with associated time-stamps can be transmitted off chip via a single 25-Mb/sec output line. The KPiX is optimized to minimize IO’s. All the control and timing signals to configure and operate the KPiX are generated via an on-chip synthesized digital VHDL core. Only 4 signals are required to control and readout the ASIC, three of them single-ended, one differential. Power is minimized by reducing the current in the front-end during the 199ms bunch spacing by two orders of magnitude. The important feature is that the voltage remains at full value, only the current gets reduced. In that way, filter capacitors, on-chip of off, do not need to be recharged. On-chip calibration Digital-to-Analog converters (DAC) and associated control circuitry enable electronic calibration of arbitrary selectable combinations of channels. The gain of the amplifiers can be configured to accommodate a range of detector signal amplitudes. Circuits to compensate detector leakage currents are integrated on the chip.

The KPiX can be programmed to process positive or negative detector input signals. Neighbor triggering can be enabled via configuration to read out nearest neighbors of auto-triggered channels. This feature together with the above listed functionality enables the KPiX to be potentially used in other detector sub-systems, e.g. for the RPC’s and GEMS of the HCAL, and for the muon system RPC’s. In addition the chip can be configured to accept an external trigger instead of the on-chip channel-specific auto- trigger for use in e.g test-beams.

17 Low Gain Wilkinso Latch (4x) n Range Register Reset Range scaler Logic and logic Reset

High Gain (default) Range I Threshold Source Track

Analog Control Logic 1 Pulses to . Timing Latch, . Leakage Range Latch, . Current Servo Event and Event Analog 4 Threshold Counter Bunch Clock Cal Dac Reset

Cal Strobe Track Scaler

Timing Latches Figure 3: Simplified block diagram of one channel of the KPiX.

2. Status A 2x32 channel KPiX was designed and fabricated in the TSMC 0.25um technology. All functions were included on the chip including the complete digital core to support the full 32x32 final version. A 64-channel initial prototype was chosen in order to reduce development cost driven by the chip fabrication cost which scales with the total area of the ASIC. Figure 3 shows the layout of the device. Test boards were designed and fabricated, the test setup was commissioned and software to operate the chip and analyze its performance was written and debugged.

18 Figure 3: 2x32 channel KPiX version

At this time the chip was evaluated for two separate application: the ECAL application (auto-triggering, positive detector signals, ECAL gain settings) and the GEM test-beam application (negative input signals, external trigger, GEM gain settings). In addition to input signals from external pulsers and the internal calibration circuitry, the chip was also assembled to a Si-detector and Am241 and Sr90 source signals were recorded. Those measurement results are currently being analyzed.

In the following table the driving design requirements and the current measurement status is summarized.

Specification Required Tested/Measure d Configurable gain switching yes yes External Trigger yes yes Calibration yes yes Sparcified Readout Yes yes Operating clock frequency 25 MHz 25 MHz Auto-gain switching Yes yes Dynamic Range 15 bit > 14 bit Noise (ECAL mode) < 2,500 e < 3,000 e Max Signal (ECAL, 5 mm pixels) 10 pC > 10 pC Auto Self-trigger Yes Yes Cross-talk < 1% tbd Power (for 1,024 channels) < 40 mW ~ 20 mW Positive or negative input signals yes yes Noise (GEM mode) < 40,000 e < 40,000 e Nearest neighbor logic yes yes Leakage current compensation > 4 uA > 5 uA

19 In Figure 4 a calibration measurement result showing the low and high gain auto switching are shown.

Figure 4: Calibration injection transfer function, low and high auto-select gain range response.

A prototype of the flex-cable interconnecting up to 12 KPiX/detector assemblies has also been designed and is waiting to be fabricated. Tests to bump-bond KPiX chips to detectors are in progress.

3. Plans Evaluation of the 64-channel version is still in progress. Additional performance tests are planned, e.g. more cross-talk and noise measurements, forced neighbor trigger analog performance, and more measurements using different kind of radiation sources.

A measurement setup has been completed for a test-beam investigating the GEM mode. In June 07 the performance of the device is expected to be measured in the test beam.

20 Layout changes to further reduce voltage drops on power traces on the chip and cross-talk between channels have been completed. The current plan is submit another 64-channel device with subsequent evaluation before submitting the full 32x32 channel KPiX.

The current KPiX will be assembled to a detector and integrated with a flex-cable to investigate system performance

21 VI. Hadron Calorimetry

1. Introduction

The Hadron Calorimeter (HCal) is a sandwich of absorber plates with gaps instrumented with active detector elements. To satisfy the stringent imaging requirements of the PFA algorithm transverse segmentation is required to be small of order a few cm2 and each layer is read out separately. This in principle allows an analog or digital treatment of the signals.

The current baseline uses steel for the absorber and resistive plate chambers as the detector. Alternative detector possibilities, consisting of Gas Electron Multipliers (GEMs), Micromegas or scintillator are also under study. One of the design criteria for the HCal is to minimize the size of the active gap, because an increase in the gap size has a large impact on the overall detector cost. The current gap size is 12mm. The absorber consists of steel plates with a thickness of 20 mm or approximately 1.1 X0. The cell structure, which is identical for the barrel and the endcaps, is repeated 34 times, leading to an overall depth of the HCal corresponding to four interaction lengths. Tungsten has been and will also be considered as a absorber, but is currently not the baseline.

The inner radius of the HCal is 138 cm, the outer radius of the barrel is 233.7 cm and its length is 554.0 cm. The endcaps start at a distance of 179.65 cm from the interaction point. The SiD baseline design uses steel as absorber material in the HCAL. Steel with a radiation length Xo = 1.8 cm and an interaction λI = 16.8 cm offers the smallest X0/λI of all commonly used absorber materials. A small ratio permits a one radiation sampling, while keeping the number of active layers manageable for a hadron calorimeter with a depth equivalent to four interaction lengths at 90o to the beam pipe.

The first attempts at a mechanical structure foresee a barrel and two endcaps, which are inserted into the barrel structure. The barrel is subdivided along the beam direction into three sections (Fig. XX), and there are 12 modules to make a complete ring in azimuth.

22 Fig. XX Schematic of SiD Hadron Calorimeter (left), and single module (right).

The steel plates are held in place by a set of picture frames located at each end (Fig. XX). Readout cables, high voltage cables and the gas lines are routed to the outer radius of the barrel structure through the openings in the picture frames. In azimuth the barrel structure is subdivided into twelve modules. Each module weighs approximately ten tons and will be held in place through supports in the cryostat of the solenoid. Deflections have been calculated and do not exceed 0.5 mm at any point of the structure.

Prototyping and testing of active gap technology options has been ongoing for several years. The basic characteristics and initial viability of each option have been established through a series of cosmic ray and beam tests. For the gas-based technologies the next stage will involve a “slice test” of a number of RPC and GEM chambers with absorber layers. The scintillator based medium has been validated with test beam studies and focus has moved to production issues. Beyond this we plan to build one or more ILC prototypes of approximately 1m3 size to contain hadron showers, and read out using these technologies. This scale of testing, involving up to 400,000 channels is presently limited by funding in the US.

We have defined a set of criteria and requirements for the selection of the final active medium. We have established a time schedule for the convergence of the hardware tests

23 and the PFA results. We anticipate making a selection on the timescale of the SiD CDR – around middle to late 2008.

2. Technologies under Consideration

Digital Hadron Calorimeter with Resistive Plate Chambers Resistive Plate Chambers (RPCs) with small readout pads are an ideal candidate for the active medium of a hadron calorimeter optimized for the application of PFAs. RPCs can provide the segmentation of the readout pads, of the order of 1 to 4 cm2, which is necessary to keep the ‘confusion term’ small. They can be built to fit small active gaps (less then 10 mm) to maintain a small lateral shower size and to keep the longitudinal extend of the hadron calorimeter as short as possible. Glass RPCs have been found to be stable in operation over long periods of time [1], especially when run in avalanche mode, and their rate capabilities are adequate for the ILC and for test beam studies of hadronic showers. RPCs are inexpensive to build since most parts are available commercially. Signals in avalanche mode are large enough (in the range of 100 fC to 2 pC) to simplify the design of the front-end electronics. Figure H1 shows a schematic diagram of a single-gap RPC. The chamber consists of two plates with high electrical resistance. Readily available window glass of thickness 0.8 to 1.1 mm is used. High voltage is applied to a resistive coating on the outside of the glass plates. The resistance of this coating must be low enough to allow for a fast local re-charge of the glass plate after a hit, and high enough to avoid screening of the electron avalanche in the gas from the readout pads located on the outside of the chamber. The glass plates enclose a gas volume in which ionization and electron multiplication takes place. Particles traversing the gas gap ionize the gas, creating an avalanche of electrons drifting towards the glass plate at positive high voltage. The signal is picked up inductively with pads located on the outside of the glass. Signal pads

Mylar Resistive paint 1.1mm glass 1.2mm gas gap -HV 1.1mm glass Resistive paint Mylar Aluminum foil

Figure H1 : Schematic diagram of a typical Resistive Plate Chamber

Detailed simulation studies have shown that the energy response is preserved with a simple one-bit (or digital) readout of the pads, see Fig. H2. The single particle energy resolution with one-bit readout is found to be significantly better at lower energies, due to its insensitivity to Landau fluctuations. At higher energies the resolution with analog readout is superior, due to the high density of particles in the core of the shower. However, for the application of PFAs, only the response in the energy range below say 20 GeV is important. Using a simple digital readout, pads with an area of 1 x 1 cm2

24 appear to provide the best possible resolution without unnecessarily increasing the channel number. /

 E Analog Digital (0.5x0.5) Digital (1.4x1.4) Digital (2.5x2.5) Digital (3.0x3.0)

E [GeV] Figure H2 : Single particle resolution versus particle energy for different readout segmentations and resolutions. (Plot by V.Zutshi, NIU)

Gas Electron Multipliers The essential requirements for DHCAL include a robust design with stable, reliable operation; a thin sensitive layer for compact calorimeter design; on-board amplification/discrimination/digitization for digital readout; high efficiency for minimum ionizing particle (MIP) tracking in a hadron calorimeter; flexible design for implementation of varying cell sizes; in addition to the basic requirements of minimal supports/intrusions for hermeticity, ease of construction, and cost containment. As a member of the SiD concept group, the University of Texas at Arlington (UTA) group have been exploring an implementation based on the Gas Electron Multiplier (GEM) [H5] technology developed at CERN by Fabio Sauli and the GDD Group [H6], shown in Fig. H3. As shown in Fig. H3a – c, thanks to its structure of having 70m holes separated by 140m, GEM detectors can provide wide range of granularity, starting from microstrips to macro-pads. Figure H3d shows a conceptual design of a GEM based hadronic calorimeter in which double GEM active layers are alternated with layers of absorber to form a sampling calorimeter.

25 Figure H3. (a) A field line diagram of a GEM foil. (b) A simulated electron avalanche process through a hole on a GEM foil. (c) A microscopic photograph of a GEM foil. (d) A schematic diagram of a conceptual DHCAL using double GEM foil sensitive layers.

In order to study the viability of using GEMs as a sensitive medium in a DHCAL, the GEM layer geometry has been implemented into Mokka [H7], a GEANT 4 [H8] based simulation package, replacing the scintillation counter sensitive layers in the TESLA TDR hadronic geometry (stainless steel/ scintillation counter) with the double GEM layer structure. All other detector structure as in TESLA TDR detector design [H9] was retained. Using this simulation package, nearly linear behavior was observed for a wide variety of single particle incident energies and for the number of hits above the threshold. Much more detailed simulation studies of GEM DHCAL have been documented in two master’s students’ theses [H10, H11].

Figure H4. (a) GEM DHCAL single pion energy resolution. (b) GEM DHCAL jet energy resolution using a “perfect” PFA.

Simulated performance studies show that GEM response for analog and digital calorimetry very close to one another. Figure H4a shows single pion energy resolution for GEM DHCAL (green) comparable to the TESLA TDR detector (red) except at low energies. On the other hand, the GEM analog mode resolution is significantly worse than other detectors. Figure H4b compares jet energy resolution for a GEM DHCAL using a “perfect” PFA (blue), which has a sampling term of 30%/E, to other detector technologies, using the single pion energy resolution obtained for GEM DHCAL. While the geometry is not that of SiD, the above studies give us confidence that GEM will perform well in a digital hadron calorimeter, providing the high jet energy resolution required for ILC physics.

26 Micromegas A micromegas detector was proposed few years ago by Y. Giomataris, P. Rebourgeard, J.P. Robert, and G. Charpak. As in the case of Gas RPC it is a very good candidate for the active medium of a fine granularity hadronic calorimeter. The Micromegas detector is a gaseous detector also, based on the micro-pattern detector technology and widely used by many experiments: COMPASS, CAST, NA48, n-TOF, ILC TPC project, and the T2K TPC project. A schematic view of the detector is shown in Figure H5: a commercially available fine mesh supported by pillars of 400 m diameter separates the drift gap (~ 3mm ) from the amplification gap (~ 100 m). A thin copper plate glued to the calorimeter absorber medium, defines the cathode, without adding extra space. Assuming a thin PCB for the anode pixel read-out (~0.8mm), this compact structure provides an economic active medium with a total depth of about 3.1mm, reducing the total radius of the HCAL. The cathode and the mesh are connected to high voltage (~800 and ~400 V respectively). The anode PCB is segmented into pads of 10x10mm2. This simple structure allows full efficiency for MIPs and thanks to the thin pillars provides a good uniformity over the whole surface. The rate obtained with Micromegas chamber is not constrained. Another advantage of such a chamber is the tiny size of the amplification avalanche, resulting in fast signals without physical cross talk.

Fig. H5 : Schematic view of a Micromegas chamber Fig H6 : T2K Micromegas TPC Prototype

A new promising technology ‘Bulk Micromegas’ has been recently developed in collaboration with the CERN PCB workshop. The basic idea is to build the whole detector in one process: the anode PCB plane with the copper pads, two photo-resistive films summing up to the correct pillar thickness, and the cloth mesh are all laminated together at high temperature, forming a single object. By a photolithographic method the photo resistive material is then etched which produces the pillars supporting the mesh. The new industrial process allows easy implementation and provides uniform, light, low cost and robust detectors. Employing the bulk technology, large detectors have been built for the TPC prototype of T2K experiment as shown in Fig. H6.

27 Scintillator/SiPM The scintillator-based hadron calorimeter option capitalizes on the marriage of proven detection techniques with novel solid-state photo-detector devices called silicon photomultipliers (SiPMs). SiPMs are multi-pixel photo-diodes operating in the limited Geiger mode. Their high gain (106), small size and low operating voltages (<100V) imply that they can be directly mounted or interfaced to the scintillator tile. Absence of fluids and gases and strong electric fields inside the detector aids longevity and operational stability. A scintillator hadron calorimeter will perform much better as traditional calorimetry than some of the other options under consideration. This is attractive since the performance of particle flow algorithms degrades as a function of the center of mass energy. Furthermore the flexibility offered by the scintillator response allows a trade-off between amplitude resolution and granularity to optimize overall cost and performance. A scintillator-SiPM hadron calorimeter can adequately fulfill three salient specifications of the Silicon Detector Concept: a) Compactness: The costs associated with the size of the magnet which encompasses the electromagnetic and hadronic calorimeters constrains the size of the calorimetric system and puts a premium on saving space used by the active modules. A scintillator-SiPM hadron calorimeter can have rather compact structure with the possibility of holding the inter-absorber gap to less than 6mm. The photo-conversion occurs at or near the scintillator tile, obviating the need for handling and routing large number of fibers. Furthermore the in situ use of these photodetectors opens the door to integration of the of the full readout chain which can be accommodated on rather thin PCBs. b) Operation in a high magnetic field: The large (5T) magnetic field expected in the SiD detector volume does not pose a problem for the scintillator-SiPM option. The SiPMs are immune to magnetic fields and no change in performance has been observed in tests up to 9T. c) Granularity: The SiD Concept proposes to use particle flow algorithms to achieve superior jet energy resolutions required to fully exploit the physics potential of an International Linear Collider. The successful implementation of this paradigm requires fine lateral and longitudinal segmentation in the calorimeters. A hadron calorimeter based on the scintillator-SiPM option is rather flexible in the choice of size of its readout granularity. Tile sizes in the range (6-9 cm2) with fibers interfaced to SiPMs have already been tested with satisfactory result. Furthermore encouraging results are emerging from R&D studies of direct i.e. fiberless coupling of the photodetectors to the tiles. This opens up the possibility of using even smaller tile sizes, if required. Preliminary test beam results and direct coupling studies (both described later) are encouraging and strongly suggest further examination of this option.

28 3. Prototype/Test Beam Results & Outstanding Issues/Next Steps

Resistive Plate Chambers Groups at Argonne National Laboratory, Boston University, University of Chicago, Iowa University, Fermilab and IHEP, Protvino have built and tested a number of prototype RPCs and developed an electronic readout system capable of dealing with the large number of channels of this type of calorimeter. Chambers based on different designs, such as featuring one or two gas gaps, two or one glass plate, high and low resistivity coatings, etc. have been investigated in detail, both in streamer and avalanche mode and also with a variety of gas mixtures. Measurements of the signal charge with a high-resolution analog data acquisition system have been performed [2]. A digital (one- bit) readout system capable of reading out multiples of 64 channels was developed and built at ANL. This system was used in detailed measurements of the MIP detection efficiency, the hit pad multiplicity, the noise rate and the rate capability. Measurements were performed with cosmic rays, radioactive sources and in the MT6 test beam at FNAL. These measurements involved either one, two or three RPCs at a given time, each read out with 64 pads with an area of 1 x 1 cm 2. All above measurements were consistent with expectation and confirmed the viability of RPCs as active medium of a digital hadron calorimeter. For more details on the measurements see [3] and the links therein and also [4]. The group is currently assembling a vertical slice test involving 10 RPCs and the complete readout chain, including pad boards, front-end boards, front-end ASICs, data concentrator boards, data collector boards and a trigger and timing unit. The test is a necessary step before commencing the construction of a 1 m3 prototype section. The latter will be ready for tests in 2008 and provide detailed measurements of hadronic showers. Beyond the prototype section, further R&D on the chambers and the readout system will be necessary to propose the best possible hadron calorimeter for the ILC detector. The following areas will need further exploration, where the experience gained in the mean time with the RPC stack in the Fermilab test beam will be crucial to define the details of this future research program: 1. Chamber design: test of thin chambers with only one glass plate or of other ‘exotic’ chamber designs. 2. Long term tests: testing and monitoring of chamber performance over longer periods (several years) to ensure stability over the expected life time of the ILC detector. 3. Neutron sensitivity: explore the possibility of increasing the sensitivity to low energy neutrons (if results from test beam show a need for this) 4. Finer segmentation of the readout (if results from test beam show a need for this) 5. Higher multiplexing of the readout system, to reduce cost and real estate. 6. Thinner readout boards, to reduce overall thickness of the active element. 7. Power pulsing of the front-end or other technologies to eliminate the need for active cooling. 8. Anything else which might be recognized as important from the experience gained in the test beam.

GEM Prototypes and Test Beam Activities

29 Initial studies were conducted on signal characteristics and gain from a small prototype double GEM detector constructed using 10cmx10cm GEM foils purchased from the GDD group at CERN. The gain of the chamber was determined to be of the order of 3,500, which is consistent with measurements done on similar chambers by the CERN GDD group. The MIP efficiency was measured to be 94.6% for a 40 mV threshold, which agrees with a simulation of chamber performance. The corresponding hit multiplicity for the same threshold was measured to be 1.27. A gas mixture of 80% Ar/20% CO2 has been shown to work well, and give an increase in gain of a factor of 3 over the original 70% Ar/30% CO2 mixture. A minimum MIP signal size of 10 fC, and an average size of 50 fC were observed from the use of this new mixture. The prototype system has proved very stable in operation over many months. The second generation of studies has been done using the same prototype detector, using a nine 1 cm×1 cm cell anode pad layout. Using multi-pad readout we have studied the crosstalk between neighboring cells. A large peak from the signal on the central pad of the 3 x 3 arrangement causes up-down peaks of the second trace which is the crosstalk signal on an adjacent pad. We have reproduced these peaks using direct signal generator pulse injection. We have also used collimated gamma rays from a 137Cs source to study signal sharing between adjacent pads. Since it is highly desirable to transport digital signals from the calorimeter, rather than small analog signals, UTA has been working with ANL RPC, FNAL and other collaborators for digital ASIC readout scheme. The details of this ASIC chip are described elsewhere. Each channel of an ASIC would comprise a charge amplifier, discriminator, and register. Our goal is to have the electronics “layer” not make a significant contribution to the overall thickness of the active layer. Subsequently, we have designed and built prototype GEM calorimeter modules with appropriate designs for signal readout with noise shielding, module-to-module interconnections for power in and signals out, and spacers between large GEM foils to prevent discharges from spiraling low energy charged particles in a magnetic field. This also required us to develop our own techniques for producing GEM foils of the size needed for calorimetry, with an eye towards future mass production of foils. For the large size foil production, we have worked with 3M incorporation and received several batches of 30cm x 30cm GEM foils made to our specifications. Each foil has been subjected to extensive high voltage testing [12-14] since the impact of leakage current to the overall gain of the chamber is significant [15]. Fig. H7a shows the behavior of the current from each of the twelve HV sectors for a typical uncoated foil. The currents for all sectors settle down to less than 5 nA in a short period of time. 3M also experimented with various types of coating for the foils e.g. an organic polymer and a gold layer. The current behavior results from a gold-coated foil are shown in Fig. H7b. The results show erratic behavior and higher residual currents than obtained with the uncoated foils. Similar results were obtained from other coatings. Consequently, we have decided to use only uncoated foils for prototype GEM chambers. 3M Corporation are currently working on a modification of the GEM foil production procedure to make ~1m long x 30cm wide foils. The full ~1m would be obtained by repeating this pattern three times down the length of the roll, with a gap between each ~1m length. The first long foils are expected in late 2007. We will carry out the usual high voltage tests on these foils prior to assembling a first long (1mx30cm) unit chamber.

30 Figure H7. (a) Leakage current (nA) vs time after HV turn on for an uncoated 3M GEM foil (b) Leakage current (nA) vs time after HV turn on for a gold coated 3M GEM foil.

We have assembled a number of double-GEM chambers using the 3M 30cm x 30cm foils. The layer structure of these chambers is shown in Fig.H8a and an anode board is shown in Fig. H8b. The anode board shown is a simplified version of the anticipated 9+ layer board we will use in a final design. The periphery of the board has locations for connectors for the 32-channel amplifier boards which use the QPA02 chip, kindly produced for us by Fermilab PPD Electronics.

GEM chamber results from test beam runs at KAERI (Korea) and Fermilab will be presented in the CALICE section of the review.

With our colleagues at ANL, we are preparing a Vertical Slice Test at the MTFB area at Fermilab in summer 2007. Chamber construction for this test is underway, and uses on-board digital ASIC chips (DCAL). This prototype chamber has been tested with a Sr90 source and seen to work well. A similar approach will be used for the chambers to be used to test the KPiX chip readout. Variations on the rear side pad connections for the DCAL version are anticipated to simplify the front-end board manufacture and reduce the cost. Similarly, we may change the layout of the KPiX board to position the chip within the 30cm x 30cm area, rather than occupying an additional piece of board on one side. We have been considering a potentially interesting and cost saving alternative to the standard GEM foil technology. Recent work at the Weizman Institute has shown that a so-called “thick-GEM” (THGEM) can, in a single layer achieve multiplication levels Figure H8. (a) A schematic diagram of 30cmx30cm double GEM chamber (b) 30x30 anode board.

31 typical of at least a double-GEM device. A THGEM is essentially a circuit board, clad with copper on both sides through which holes have been drilled. A typical configuration might be a 0.4 mm thick board with 0.3 mm diameter holes spaced 1 mm apart. We have been using in-house micro-drilling facilities to make small thick-GEM boards. A number of small boards have been made to test the drilling procedure, and the side-to-side registration.

Once we have received and tested the long GEM foils from 3M, we will build the first full length prototype chambers for the 1m3 GEM stack. We anticipate the first chamber to be ready for testing in later in the year 2007. Each 1mx1m active layer will consists of three 1mx30cm unit chambers due to the size of the 3M large foil. We plan to build minimum 40 layers of these chambers to instrument a 40 layer, 1m3 stack. This prototype detector will be exposed to test beams as part of the CALICE prototype stack. While this step depends heavily on the availability of funds, we anticipate this to occur later in 2008. We will continue to pursue the development of thick-GEM’s as they still present an attractive alternative to regular GEM foils. We will build small prototypes using our in- house produced boards. We will also explore the possibilities of making larger boards using the facilities of the local electronics industries. We aim to have a medium-size prototype thick-GEM chamber assembled by Fall 2007 for testing and performance comparison with our regular GEM chambers.

Micromegas In the context of the hadronic calorimeter R&D, work has started on a project with a bulk Micromegas as the active medium. The R&D program in its first phase examines detector properties and design and fabrication issues with small prototypes read out by external electronics. In the second phase, we aim to build a prototype fulfilling the HCAL requirements with the electronics on board and the optimized thickness. We have built three 6x16 pad Micromegas chamber prototypes (Fig. H9) read out by 6 Gassiplex chips mounted on a dedicated card produced by Saclay. The aim is to study the detector behavior with cosmic rays and radioactive sources. The electronics card located on chamber side provides analog read out, allowing gain measurements and cross talk studies. The gas system provides multiple gas mixtures to allow for gain optimization.

Figure H9 : The 6x16 pads Micromegas detector

32 The phase two prototype, under design now, includes a PCB of 80x320 mm2 read out with four HARDROC chips developed by the Orsay group. The chips are mounted on the detector, daisy chained together, providing both analog and digital output. Special attention is given to the PCB board that can be common with a Gas RPC detector. With 16 such boards we plan to build a Micromegas plane of 640x640 mm2. This prototype will be put to a beam test. Finally, the construction of a 1m3 prototype is foreseen, in collaboration with three European groups: the Russian IHEP of Protvino, the Saclay group developing the Micromegas chambers and the CIEMAT group of Madrid for the mechanics and electronics aspect.

Scintillator/SiPM HCal and TCMT Results The primary goals of the CALICE collaboration are the demonstration of technologies under consideration for high-granularity calorimetry at the ILC, and to use test beam data to validate and improve the existing simulation models for hadronic showers, which is essential for the optimization of the ILC detectors. Many members of the SiD concept group are also CALICE collaborators. Results from the CALICE experiment are reported here to demonstrate the strengths of the scintillator/SiPM option and appropriateness for SiD. Detailed discussions are deferred to the CALICE results. More information about the CALICE collaboration is provided at their web site, available at http://polywww.in2p3.fr/activites/physique/flc/calice.html. The core of the CALICE test beam prototype consists of a 30-layer Si-W electromagnetic calorimeter with 1x1 cm2 cells, a 38-layer steel-scintillator withSiPM readout hadron calorimeter with a high granularity core (3x3 cm2 tiles), and a tail catcher/muon tracker (TCMT), also based on steel/scintillator plus SiPM readout. Some accessory components were drift chambers used for tracking; a Cerenkov detector used for beam particle ID; and scintillator planes used mostly for triggering: finger counters, muon counter and a veto counter for double beam particles. The combined CALICE prototypes have been exposed to beam at CERN in the summer and fall of 2006. Figures H11 and H12 show typical calibrated energy distributions associated to MIP particles (muons) going through an HCal tile or a TCMT strip, respectively. Pedestal distributions over an interval of 26 hours so stable behavior, with average RMS variations within 0.23%.

H10: Pedestal stability H11: HCal Tile MIP Dist. H12: TCMT Strip MIP Dist.

33 TCMT strip efficiency is estimated from the response of each tile to software triggered muons, while rejection is estimated using pedestals measured for each strip. Figure H13 shows the variation of efficiency (blue) and rejection (red) as a function of the energy threshold used for a single TCMT strip.

H13: Strip Eff. vs Rej. H14: Uniformity of Efficiency

The next plots show the effect of including data from the tail catcher in the final energy reconstruction, for pion beams with incident energies of 20 GeV. As shown in Fig. H15 there is a clear anti-correlation between the energy deposited in the TCMT and that in the ECal plus HCal. Figure H16 shows that properly combining the TCMT energy (solid line) with that of ECal plus HCal (dashed line) reduces the low-energy tail, improving the hadron energy measurement. The small peaks below 100 are due to muons, probably

H15: TCMT energy vs ECal+HCal Energy H16: 20 GeV response w and w/o TCMT coming from in-flight decays of beam particles. As expected, (although not shown here for space limitations) we observe that the TCMT effect is more evident for higher energies. These preliminary results show that the scintillator/SiPM technology demonstrates the stability, MIP sensitivity, and energy response required for PFAs. Further studies and integration of the performance characteristics into simulations are required to fully evaluate the scintillator/SiPM option. Attention has recently turned to optimizing detector. The CALICE prototype scintillator tiles and strips incorporate a WLS fiber to transmit light to the SiPM. Construction and cost would be greatly reduced 1) if the scintillator and SiPM were directly coupled and 2) if the scintillator, SiPM, and digitization electronics were fully integrated on a PCB. Studies are underway at the University of Colorado and Northern Illinois University to investigate the uniformity of response for direct coupling. Direct coupling

34 refers to SiPMs that are positioned directly on the surface of the scintillator plane such that the face of the SiPM touches the scintillator surface. Measurements at Colorado examine the uniformity of the output signal for two cases: 1) a 5x6x0.5cm3 cast scintillator tile with the signal collected by a green fiber embedded in the scintillator and then transmitted to the SiPM and 2) a same-sized extruded scintillator tile with the SiPM located at the center of the tile face. As can be seen in the Fig. H17, for both cases the output signal due to a radioactive source is quite uniform across the face of the scintillator.

H17: The figure on the left is cast scintillator where the light is collected by a fiber and transmitted to a SiPM. The figure on the right is the case where the SiPM is touching the face on an extruded scintillator tile.

NIU has corroborated the Colorado results, observing reasonable uniformity (~50% variations) for a source scan of a 0.6mm thick scintillator tile. NIU has also tested direct coupling in a cosmic ray test stand. Figure H18 shows the locations of the directly coupled SiPM on a 3x3cm2 tile during cosmic ray tests, the same CPTA sensor was moved to each spot for three data runs. A clear Landau distribution (Fig. H19) above pedestal was evident for the three cosmic rays runs; the maximum response (as given by the most probable value) was at the center position with the edge and corner yielding 92 and 71% of the central position signal. The light yield as given by a Hamamatsu SiPM (Fig. H14) was around 20PE. The signal is far from pedestal making it possible to reach ~100% registration efficiency at all positions. These promising measurements of signal uniformity will continue over the summer of 2007 with different tile configurations and sensors.

H18: Direct coupling locations H19: Direct coupling response

35 Colorado, NIU, and Fermilab will collaborate on full integration of the scintillator/SiPM technology. NIU and Fermilab have designed an integrated PCB which will house the scintillator and SiPM and carry SiPM signals to an edge connector. This board will be populated after direct coupling optimization and exposed to beam at Fermilab in late 2007 or early 2008. Work is already underway on a fully integrated board and the associated electronics. These new electronics, designed by Fermilab, will be suitable for both calorimetry and muon tracking.

4. Steps to Technology Decisions and Conclusions To ensure timely selection of a robust and high performance hadron calorimeter, the SiD collaboration has instituted a five-year R&D and selection plan. The main issue will be the quality of overall ECal and HCal physics performance versus the cost of the HCal. Both the performance and cost are mainly driven by the longitudinal and transverse segmentation. The plan assumes a conceptual design report will be required in 2008 and an engineering report in 2011. The plan has been discussed widely within SiD and establishes the requirements for the HCal, the performance criteria for the technologies, and steps forward. The HCal provides essential information for the identification and reconstruction of jets and other physics objects via a PFA. The six basic requirements for this device strongly reflect the needs of PFA, these include efficient tracking of charged particles throughout the volume; sufficient depth such that any energy loss in the coil or energy measured in the outer detectors does not significantly impact jet energy resolution; sufficiently small cell size to permit separation and association of closely spaced energy clusters with the correct tracks; shower sampling that does not significantly degrade jet energy resolution; an outer radius that limits the cost of the solenoid and muon system to reasonable levels – requiring the radial size of each active layer to be as small as possible; and sufficient rate capability so as not to lose information, particularly in the forward directions. There are a number of additional specific factors that will also be considered in the final SiD HCal selection. As shown in the table below, these can broadly be described as performance criteria, technology issues, and costs.

Performance Criteria Technology Issues Cost MIP efficiency/pad Maturity and history Development costs Hit multiplicity/pad Reliability Assembly and test costs Uniformity of response across active Component availability Active layer % of total element Ease of calibration Active layer thickness Total HCal cost Track/cluster separability Smallest element size Response to neutrons Ease of assembly, testing, installation, commissioning PFA jet resolution at Z-pole, 200 Recovery time after i) hits & ii) Gev, 500 GeV significant beam event Rate of discharge (gas) Ageing affects Sensitivity to magnetic field Technical risk

The performance criteria will ultimately be gauged by a complete PFA simulation which will include all performance aspects of the technology. The PFA performance will also

36 indicate if a technology can meet the six basic requirements. The technology issues are many and will all be considered in an overall estimate of technical risk and cost. The overall strategy reflected in the R&D and selection schedule proposes an initial demonstration and test of each technology, followed by the construction of prototypes for each technology, and, ultimately, selection of a technology for an ILCD prototype. Many of the performance criteria and technological issues have already been examined for each technology. This document represents the first formal report of these findings and has been required before management endorses construction of large prototypes. By late 2007, the first GEM and RPC slice test results; mature scintillator/SiPM test beam results and comparisons with simulations; the initial results on scintillator/SiPM direct coupling; and advanced PFA simulations for each technology will be available. These will form the basis of a SiD review in late 2007 which will establish the leading candidate or candidates for inclusion in the SiD conceptual design report and for further large prototype studies. Due to limited funding and the CDR/EDR time scale, technical choices may be based on simulations and partial prototype. Because of the incomplete nature of the technical information, the collaboration may continue to study more than one option during the CDR and EDR process, a down-select to one technology will be made as soon as prudent. A second comprehensive review will be held in mid 2008 to confirm and finalize the technology choices for the CDR and review module construction to be carried out while the EDR is written.

References

[1] See e.g. Proceedings of the VII Workshop on Resistive Plate Chambers, Clermont-Ferrand, France (October 2003), Nucl. Instr. And Meth. A533, (2004). [2] “Resistive Plate Chambers for Hadron Calorimetry: Tests with Analog Readout”, G.Drake et al., to appear in Nucl. Inst. and Meth. [3] http://www.hep.anl.gov/repond/DHCAL_US.html [4] Document submitted to this review by the CALICE collaboration [5]R. Bouclier, et al., “The Gas Electron Multiplier (GEM),” IEEE Trans. Nucl. Sci. NS-44, 646 (1997); F. Sauli “GEM: A new concept for electron amplification in gas detectors,” Nucl. Inst. Meth., A386, 531 (1997). [6]CERN Gas Detector Development group, http://gdd.web.cern.ch/GDD/ [7]Mokka [8]S. Agostinelli et al. (GEANT4 Collaboration) Nucl. Instrum. Methods A506 (2003) 250, http://geant4.web.cern.ch/geant4/ and references therein for the various models implemented. [9]TeV Energy Superconducting Linear Accelerator (TESLA) Technical Design Report, Edited by F. Richard, J. R. Schneider, D. Trines, A. Wagner, March (2001); http://tesla.desy.de/new_pages/TDR_CD/start.html [10]V. Kaushik, “Performance of Novel Digital Hadron Calorimeter Using Gas Electron Multiplier (GEM) and the Energy Flow Algorithm Development,” MS Thesis, University of Texas at Arlington, UTA- HEP/LC-004, Unpublished (2004). [11]S. Habib, “Simulation Studies of a New Digital Hadron Calorimeter, Using Gas Electron Multipliers (GEM),” MS Thesis, University of Texas at Arlington, Unpublished (2003). [12]3M GEM foil Testing procedure [13]HV testing results [14]Another result [15]Heather’s HV impact study

37 V. Calorimeter simulation results

1. Introduction

Since a primary role of the hadron calorimeter is the identification and energy measurement of neutral hadrons, it is essential for calorimeter design purposes to study the impact of variations of longitudinal and transverse segmentation, calorimeter depth, absorber material, magnetic field and other parameters on performance. Neutral hadrons have been studied through simulation in an effort to aid design decisions. All simulations used GEANT4, with the LCPhys physics model. First, single particles were studied in (semi)infinite sampling calorimeters, varying design parameters and studying responses, resolution and transverse spread of the showers. These results were then used for calibrations, allowing the study of full detector simulations. For four variations of hadron calorimeter designs, the jet energy resolution in mono-energetic jets and the dijet mass resolution in 500 GeV ZZ events were also studied.

38 2. Single particles in uniform (semi)infinite detectors

To study the response of neutral hadrons in sampling calorimeters, simulations have been performed using semi-infinite (1000 layer) calorimeters. The absorber, active medium, and B-field were varied, with the mean and rms of responses observed. These responses were also studied as a function of depth of the calorimeter, analog/digital readout and incident angle of the particle. The hadron calorimeter simulations had 1cm by 1cm transverse segmentation, and studies of the effect of segmentation were done by combining cells. For each data set, k0L’s, neutrons, and antineutrons were generated at a fixed energy and angle, with the mean and rms of the response recorded. The mean and rms varied according to particle type, so a “combined” neutral hadron was defined as (.5k0L + .25n + .25nbar). A sample plot showing the resolution of this “combined” neutral hadron for four detector variations is shown in Fig. 1. The parameter space is enormous, but enough variations were examined to see trends. Comparisons were made by fixing all parameters but one and varying that parameter. A summary of observations follows:

B-field: No significant variation in response or resolution was observed between 0 and 5 Tesla. However, the radial and transverse spreads have not yet been studied. Fig 2.

Incident angle: Results indicate that significant response corrections are needed. At 45 degree incidence, corrections are ~5% for plastic and ~12% for gas active media. Fig 3.

Segmentation: 20 – 30% worse resolution for 3cm by 3cm segmentation vs. 1cm by 1cm. (Digital) Fig. 4.

Comparison of analog vs. digital modes: The results seem to depend on absorber type and lambda/layer. Below ~ 10 GeV, digital resolutions are better. For W absorber with .08 lambda/layer, and plastic active medium, ~20% improvement in resolution at higher energies is seen using analog mode. Fig. 5. (Note: analog responses used the straight sum of all energy deposits above threshold. More sophisticated algorithms may yield more improvement.)

Depth: Resolution is energy and active medium dependent. For 50GeV a 20-30% worse resolution is seen from a 5-4 lambda HCal. Fig. 6.

Absorber: W and Pb show similar and worse resolution than SS or Cu. Shower spread smaller in W than SS. The energy resolution for Wcan be improved by using a smaller lambda/layer, at cost of shower transverse spread. Fig. 7.

Active medium: Plastic shows better resolution than gas at lower energies. However, plastic shows a worse transverse spread than gas. Fig 8.

39 Note: These plots point out the model change in GEANT4 LCPhys physics list between 10 and 15 GeV. While the absolute scale may be suspect, comparisons between detector configurations should be valid.

Infinite Hcal: (sigma/mu)*sqrt(E)

1.35

1.30 WScint 1.25 WRPC SteelScint 1.20 SSRPC 1.15

1.10

1.05

1.00

0.95

0.90

0.85

0.80

0.75

0.70

0.65

0.60

0.55

0.50

0.45

0.40

0.35

0.30 5 10 15 20 25 30 35 40 45 850 "Combined" neutral hadron Energy Fig 1 – Resolution vs energy for 4 HCal configurations

40 Mean #hits vs "combined" neut had energy (rms/mean)*sqrtE vs "combined" nh energy

# hits 1.08 270 260 1.06 250 SSRPC: B = 5T 1.04 SSRPC: B = 5T 240 SSRPC: B = 0 SSRPC: B = 0 230 1.02

220 1.00 210 200 0.98 190 0.96 180 170 0.94 160 0.92 150 140 0.90 130 0.88 120 110 0.86 100 90 0.84 80 0.82 70 60 0.80 50 0.78 40 30 0.76 20 0.74 10 0 0.72 5 10 15 20 25 30 5 10 15 20 25 30 "combined" neutral hadron energy "combined" neutral hadron energy

Fig. 2 – Response comparison, B=0,5 tesla

Mean #hits vs "combined" neutral hadron energy

600

550

SSRPC:90 degrees 500 SSRPC:45 degrees

450

400

350

300

250

200

150

100

50

0 5 10 15 20 25 30 35 40 45 50 Fig. 3 – Response comparison for 45 and 90 degree incidence

41 Finite detector:(sigma/mu)sqrt(E) Finite detector:(sigma/mu)*sqrt(E)

2.1 2.1 WScint1X1cm SteelScint1X1cm 2.0 2.0 WScint3X3cm SteelScint3X3cm 1.9 WScint5X5cm 1.9 SteelScint5X5cm

1.8 1.8

1.7 1.7

1.6 1.6

1.5 1.5

1.4 1.4

1.3 1.3

1.2 1.2

1.1 1.1

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3 10 20 30 40 50 10 20 30 40 10 50 "Combined" neutral hadron Energy "Combined" neutral hadron Energy Fig. 4 – Effect of segmentation on neutral hadron resolution

Finite detector:(sigma/mu)*sqrt(E) Finite detector:(Sigma/mu)*sqrt(E)

1.8 1.8 WScint digital SSScint digital 1.7 WScint analog 1.7 SSScint analog

1.6 1.6

1.5 1.5

1.4 1.4

1.3 1.3

1.2 1.2

1.1 1.1

1.0 1.0

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3 10 20 30 40 50 10 20 30 40 12 50 "Combined" neutral hadron Energy "Combined" neutral hadron Energy Fig. 5 – Resolution of analog vs digital response

42 Finite detector: (sigma/mean)*sqrt(E)

1.50

1.45 SSRPC: 0.12 lambda/layer 1.40 CURPC: 0.10 lambda/layer WRPC: 0.074 lambda/layer 1.35 PBRPC: 0.10 lambda/layer 1.30

1.25

1.20

1.15

1.10

1.05

1.00

0.95

0.90

0.85

0.80

0.75

0.70

0.65

0.60 0 5 10 15 20 25 30 35 40 45 50 "combined" neutral hadron energy (GeV) Fig 6 – Resolution for different absorbers

(sigma/mu)*sqrt(E)

1.40 SSScint: Infinite 1.35 SSScint: 68 layers 1.30 SSScint: 51 layers 1.25 SSScint: 42 layers SSScint: 34 layers (finite) 1.20

1.15

1.10

1.05

1.00

0.95

0.90

0.85

0.80

0.75

0.70

0.65

0.60

0.55

0.50

0.45

0.40

0.35 5 10 15 20 25 30 35 40 45 1350 "Combined" neutral hadron Energy Fig. 7 – Resolution vs depth of HCal

43 Transverse distance containing 75% of hits Transverse distance containing 90% of hits

cm SSScint: 0.12 lambda/layer cm SSScint: 0.12 lambda/layer 22 SSRPC: 0.12 lambda/layer 44 SSRPC: 0.12 lambda/layer WScint: 0.074 lambda/layer WScint: 0.074 lambda/layer 21 WRPC: 0.074 lambda/layer 42 WRPC: 0.074 lambda/layer

20 40

19 38

18 36 17 34 16 32 15 30 14 28 13 26 12 24 11 22 10

20 9

8 18

7 16

6 14 10 20 30 40 50 10 20 30 40 50 "combined" neutral hadron energy (GeV) "combined" neutral hadron energy (GeV) Fig. 8 – Transverse spread of neutral hadron showers

3. Calibrations

Calibrations were performed for each of four detector configurations: (W absorber, .08 lambda/layer + SS absorber, .118 lambda/layer) x (plastic, gas). Mean responses from isolated detector simulations on each HCal configuration plus isolated ECal simulations were used. Because of the non-linear responses, the calibration uses linear extrapolations between data points, and an iterative algorithm to combine energy from different detector components. The algorithm estimates the hadron energy adding the detector contributions, then estimates the contribution from each detector as a fraction of the response to the energy estimate. Three iterations were found to be sufficient. Full detector simulations at the Z-pole were then used to provide a detector response -> neutral hadron energy mapping, with particle type and energy distributions from a physics process. Results are shown in fig. 9. The calibrations could be improved, most notably by correcting for depth of interaction (losses beyond the HCal) and using a neutral hadron energy distribution from a 500 GeV cm physics process for the inversion. The weighted neutral hadron energy distributions in Z-pole events and 500 GeV ZZ events are shown in fig. 10.

44 SSRPC:Zpole Event nhad dE/rootE SSScint:Zpole Event nhad dE/rootE

900 Event/NHadron dE0 over rootE 1,100 Event/NHadron dE0 over rootE Entries : 7969 Entries : 7964 1,000 800 Mean : 0.014433 Mean : -0.013701 900 700 Rms : 0.69928 Rms : 0.55402 NaN : 1064 800 NaN : 1062 600 SumOfWeights : 6905.0 700 SumOfWeights : 6902.0 500 gauss 600 gauss 400 amplitude : 778.12±11.8 500 amplitude : 972.29±15.1 mean : 0.037268±0.007723 400 mean : 0.021356±0.005439 300 sigma : 0.62063±0.00575 sigma : 0.44711±0.00436 300 200 χ² : 3.4512 χ² : 3.7807 200 100 100 0 0 -5 -4 -3 -2 -1 0 1 2 3 -4 -2 0 2

WRPC:Zpole Event nhad dE/rootE WScint:Zpole Event nhad dE/rootE Event/NHadron dE0 over rootE Event/NHadron dE0 over rootE 1,000 1,000 Entries : 7969 Entries : 7970 Mean : -0.023948 900 Mean : -2.0516E-5 900 Rms : 0.60082 800 Rms : 0.71357 800 NaN : 1063 NaN : 1064 700 700 SumOfWeights : 6906.0 SumOfWeights : 6906.0 600 600 gauss gauss 500 500 amplitude : 873.98±13.8 amplitude : 859.28±13.2 mean : 0.032795±0.005911 400 mean : 0.022063±7.8739E-3 400 sigma : 0.47947±0.00483 300 sigma : 0.62311±0.00592 300 χ² : 5.2969 χ² : 4.4395 200 200 100 100 0 0 17 -4 -2 0 2 4 -5 -4 -3 -2 -1 0 1 2

Fig. 9 – Effective resolution of neutral hadrons in Zpole events

Fraction of neutral hadron energy vs energy

0.085 ZZ500 normalized neutral hadron wted energy distribution Zpole normalized neutral hadron wted energy distribution 0.080 0.075 ZZ500 normalized neutral hadron wted energy distribution 0.070 Entries : 24242 0.065 Mean : 35.219 Rms : 30.783 0.060 SumOfWeights : 1.0000 0.055 Zpole normalized neutral hadron wted energy distribution 0.050 Entries : 23040 Mean : 9.9825 0.045 Rms : 7.4775 0.040 SumOfWeights : 1.0000 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000 0 10 20 30 40 50 60 70 80 90 100 Neutral hadron energy

Fig. 10 – Fraction of neutral hadron energy vs energy in Zpole and 500 GeV ZZ events

4. Perfect pattern recognition

45 The calibrations were used to analyze full detector simulations with perfect pattern recognition, i.e. all calorimeter hits are correctly associated with the final state particle from which they originate. Energy sums in mono-energetic jets were studied in 2 light quark (uds) events at 91, 200, 500 GeV cm energy. Results are shown in fig. 11. For comparisons, the quantity alpha = (rms90/mean90)*sqrtE is shown, where 90 means the 90% of the events with the smallest rms were used. The transverse spreads were also studied in these events. A cylinder was defined through the HCal for each neutral hadron. For a given radius of the cylinder, the number of hits from the neutral hadron inside the cylinder divided by the total number of hits in the HCal from the neutral hadron (summed over all neutral hadrons) was used to define an efficiency. Similarly, the number of hits inside the cylinder from charged hadrons over the number of hits inside the cylinder (again summed over all neutral hadrons) defined a purity. The results are shown in fig. 12 and13. Also studied were full detector simulations of ZZ events at 500 GeV, where one Z->nu nu, and the other to 2 light quarks. Using perfect pattern recognition for both the neutral hadrons and the photons, and requiring 2 jets from the jet finder, both the dijet mass and delta (dijet mass – generated Z mass) are shown in fig. 14 and 15. The direction of both quarks was required to have abs(cos(theta)) < 0.8 to minimize losses down the beampipe. The effective resolution for neutral hadrons in these events is shown in fig. 16. For a significant difference in neutral hadron resolution, little or no difference is seen in the width of the dijet mass distribution. The same events were analyzed with a different level of cheating. Instead of perfectly associating the hits with the correct particle, first all the hits in the calorimeters were clustered using a weighted density clustering algorithm. (Directed Tree ). The clusters were then associated with the particle contributing the most hits to the cluster. Dijet mass and delta mass are shown in fig. 17and 18.

Perfect Pattern recognition PFlow PFlow PFlow

Alpha: 1,200 WScint 1,400 WScint Alpha: 1,200 WScint Alpha: Entries : 6862 1,150 Entries : 5457 1,150 Entries : 4908 Mean : 88.696 1,300 Mean : 196.25 Mean : 486.61 1,100 .171 1,100 Rms : 2.4922 Rms : 4.2169 .168 Rms : 13.122 .343 1,050 1,050 SSScint 1,200 SSScint SSScint 1,000 Entries : 6621 Entries : 4895 1,000 Entries : 3196 Mean : 88.696 950 .162 1,100 Mean : 196.22 950 Mean : 487.94 Rms : 2.0988 Rms : 3.4351 .156 Rms : 10.056 .277 900 900 WRPC 850 1,000 WRPC 850 WRPC Entries : 6876 Entries : 6397 Entries : 4499 800 Mean : 88.482 Mean : 194.82 800 Mean : 485.59 .204 900 750 Rms : 2.6903 Rms : 4.8800 .216 750 Rms : 12.994 .363 700 SSRPC SSRPC 700 SSRPC Entries : 6803 800 650 Entries : 6288 650 Entries : 3951 Mean : 88.565 .194 Mean : 195.23 Mean : 487.84 600 Rms : 2.5026 700 Rms : 4.1783 .196 600 Rms : 10.419 .300 550 550 600 500 500 450 450 WScint 500 400 SSScint 400 WRPC WScint 350 400 WScint 350 SSRPC SSScint 300 SSScint 300 WRPC 250 300 WRPC 250 SSRPC 200 SSRPC 200 200 150 150 100 100 100 50 50 18 0 0 0 70 80 90 100 160 180 200 220 350 400 450 500 550 Fig. 11 – Energy sums in uds events at 91, 200, 500 GeV

46 Efficiency/Purity vs cylinder radius(cm) in 45.5 GeV jets

1.00

0.95

0.90

0.85

0.80 SSScint91TransPos:Efficiency 0.75 SSRPC91TransPos:Efficiency 0.70 SSScint91TransPos:Purity SSRPC91TransPos:Purity 0.65

0.60

0.55

0.50

0.45

0.40

0.35

0.30

0.25

0.20

0.15

0.10

0.05 0.00 33 0 5 10 15 20 25 30 35 40 45 50 Fig. 12 – Efficiency/Purity curves for neutral hadrons using a fixed radius cylinder at 91 GeV

Eff/Pur vs cylinder radius(cm) in 100 GeV uds jets Eff/Pur vs cylinder radius(cm) in 250 GeV uds jets

1.00 1.00

0.95 0.95

0.90 0.90

0.85 0.85

0.80 0.80

0.75 0.75

0.70 0.70 SSScint200TransPos:Efficiency SSScint500TransPos:Efficiency 0.65 SSRPC200TransPos:Efficiency 0.65 SSRPC500TransPos:Efficiency SSScint500TransPos:Purity 0.60 SSScint200TransPos:Purity 0.60 SSRPC200TransPos:Purity SSRPC500TransPos:Purity 0.55 0.55

0.50 0.50

0.45 0.45

0.40 0.40

0.35 0.35

0.30 0.30

0.25 0.25

0.20 0.20

0.15 0.15

0.10 0.10

0.05 0.05 0.00 0.00 35 0 10 20 30 40 50 0 10 20 30 40 50 Fig. 13 – Efficiency/Purity curves for neutral hadrons using a fixed radius cylinder at 200,500 GeV

47 PPR PFlow: Cut jet-jet mass

520 500 SSRPC2jetZZdebug2.aida SSScint2jetZZdebug2.aida 480 gauss 460 gauss_1 440 420 400 SSRPC2jetZZdebug2.aida 380 Entries : 2614 360 Mean : 88.433 340 Rms : 7.6533 320 SSScint2jetZZdebug2.aida 300 Entries : 2625 Mean : 88.142 280 Rms : 7.6898 260 gauss 240 amplitude : 348.15±11.09 220 mean : 89.190±0.074 200 sigma : 3.5309±0.0853 χ² : 4.8162 180 160 gauss_1 amplitude : 362.77±11.61 140 mean : 89.021±0.073 120 sigma : 3.3704±0.0819 100 χ² : 5.0316 80 60 40 20 0 24 10 20 30 40 50 60 70 80 90 100 110 120 130 Fig. 14 – Dijet mass distribution using “perfect” pattern recognition

PPR PFlow: cut delta jet-jet mass

600 SSRPC2jetZZdebug2.aida SSScint2jetZZdebug2.aida 550 gauss gauss_1 500

450 SSRPC2jetZZdebug2.aida Entries : 2610 Mean : -2.1479 400 Rms : 3.1326 SSScint2jetZZdebug2.aida 350 Entries : 2614 Mean : -2.4010 Rms : 3.1625 300 gauss amplitude : 441.23±12.7 250 mean : -1.8337±0.0444 sigma : 2.1280±0.0438 χ² : 4.8417 200 gauss_1 amplitude : 454.60±13.1 150 mean : -1.9451±0.0452 sigma : 2.0362±0.0418 χ² : 6.3073 100

50

0 25 -30 -25 -20 -15 -10 -5 0 5 10

Fig. 15 – Perfect pattern recognition dijet mass – generated Z mass

48 ZZ 500 GeV: Neutral hadrons - delta E/sqrtE

1,100 SSRPC SSScint 1,000 gauss gauss_1 900 SSRPC 800 Entries : 12040 Mean : -0.14873 Rms : 0.76333 700 SSScint 600 Entries : 11996 Mean : -0.15324 Rms : 0.65823 500 gauss amplitude : 731.72±8.62 400 mean : -0.11165±0.005954 sigma : 0.64391±0.00475 300 χ² : 2.7885 gauss_1 200 amplitude : 949.80±11.3 mean : -0.085978±0.004534 100 sigma : 0.48835±0.00365 χ² : 4.7251 0 -7 -6 -5 -4 -3 -2 -1 0 1 2 3

Fig. 16 – Effective resolution of neutral hadrons in 500 GeV ZZ events

dijet mass

240 SSRPC SSScint 230 SSRPC 220 SSScint 210 200 SSRPC 190 Entries : 2639 180 Mean : 90.290 Rms : 8.8266 170 SSScint 160 Entries : 2639 150 Mean : 90.594 140 Rms : 8.8444 130 SSRPC 120 amplitude : 166.43±5.13 mean : 90.619±0.11 110 sigma : 5.6297±0.130 100 χ² : 2.8306

90 SSScint 80 amplitude : 163.26±5.12 mean : 90.988±0.12 70 sigma : 5.7217±0.136 60 χ² : 3.0335 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 110 120 130 48140 GeV Fig. 17 – Dijet mass cheating on cluster association

49 delta Z mass

300 SSRPC 280 SSScint SSRPC 260 SSScint

240

220 SSRPC Entries : 2639 200 Mean : -0.53388 Rms : 5.7619 180 SSScint Entries : 2639 160 Mean : -0.23016 Rms : 5.8494 140 SSRPC amplitude : 188.47±5.92 120 mean : -0.49920±0.08703 sigma : 4.2671±0.102 100 χ² : 4.0868

80 SSScint amplitude : 177.62±6.246 mean : -0.29028±0.092283 60 sigma : 4.3595±0.124 χ² : 5.5577 40

20

0 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 4930 GeV Fig. 18 – Cluster association cheating dijet mass – generated Z mass

5. Summary

Simulations of single neutral hadrons in isolated detectors have been studied varying several design parameters, yielding an overall picture of tradeoffs in resolution and shower spread. With a limited set of parameters, the effect of intrinsic HCal resolution on jet energy and dijet mass resolution has been quantified. With both perfect pattern recognition and real clustering, no significant gain in dijet mass resolution is seen with better intrinsic neutral hadron resolution. Attempts to quantify the effect of shower spread on confusion in PFAs have been made, but the algorithms themselves will probably be needed.

50 VI. Use of PFAs in SiD Development

1. Introduction

The SiD detector is designed to take advantage of jet reconstruction using the Particle Flow Algorithm (PFA) approach. A PFA attempts to reconstruct all of the individual particles in an event according to their detector signatures – tracks for charged particles and calorimeter clusters for photons and neutral hadrons. Separation and identification of calorimeter energy deposits is, therefore, crucial for the PFA approach to work. Charged hadrons normally comprise ~65% of the typical jet energy, photons ~25%, and neutral hadrons the remaining ~10%. By separating the calorimeter clusters made by these particles, better jet energy resolution and also dijet mass resolution can be obtained by optimal use of detector components – tracking detectors for charged particles, the ECAL for photons, and the ECAL and HCAL for neutral hadrons. A PFA can be used to optimize choices for detector components – mainly the ECAL and HCAL. The SiD will take advantage of the PFA approach to optimize many of the detector parameters, resulting in improved jet reconstruction performance over existing detector designs along with optimized cost.

2. PFAs in Physics Processes

51 The goals for PFA performance have been expressed in many ways. For an ILC detector based on PFA jet reconstruction, the goal is to be able to separate on an event-by-event basis W and Z bosons using all decay modes including the dijet mode. For no loss in luminosity, this means that the W and Z masses must be measured with a resolution of ~3 GeV, or to ~3-4%. Typical calorimeters designed without PFA optimization cannot achieve this goal – even if they are perfectly compensating. Since the performance goal depends on (QCD) jet reconstruction, the optimization of the detector will ultimately depend on jets in simulation.

The PFA approach for the SiD detector has been studied with this goal in mind. The dijet mass resolution performance of the SiD has been studied as a function of several variables – notably the solenoidal B-field, the inner radius of the ECAL, jet energy, and jet multiplicity. Our approach has been to start with the simulated process e+e- -> ZZ where one Z -> 2 jets and the other Z -> nunu. This process produces jets of energy ~120 GeV – the typical jet energy of many 4-jet processes at e+e- annihilation at 500 GeV. The dijet mass can be determined without any jet combinatoric ambiguity for these events. Then, the all-hadronic decay mode is studied, where ZZ -> 4 jets. Thus, the detector is filled with more 120 GeV jets. Another test is e+e- -> ttbar – 6 jets with reduced jet energies. Finally, we can test the limitations of PFAs with e+e- -> qqbar at 500 GeV. With this analysis, we can investigate the jet energy dependence of the dijet mass resolution. As the jet energy increases, confusion in the PFA eventually dominates the mass resolution, degrading the PFA performance.

3. Tools for PFA Development

To develop the PFA, several tools are needed which are common to all detector models and PFA implementations. A common basis for PFA applications is required – in particular, a common standard detector calibration is needed along with a standard Perfect PFA definition in order for useful comparisons to be made between PFA algorithms and detector models.

Calorimeter Calibration A method of calorimeter calibration has been developed which could be used in a real test beam for a detector prototype. The PFA approach results in separated charged hadrons, photons, and neutral hadrons. Therefore, a full calibration should include beams of these species if possible. Neutral hadron beams are especially useful. In simulation, pions (+-), photons, and a mixture of n, nbar, and Klong are used to calibrate the ECAL and HCAL. The calibration method gives corrected energies for cheated (all hits) clusters. Additional corrections are needed for the application of real cluster algorithms, analysis techniques, etc. A standard calibration method exists and SiD variants all use this method to provide standard calibrations for detector components.

Perfect PFA Definition For PFA analysis, it is important to have a standard definition of the perfect application of the PFA – no confusion in the identification of particle-cluster associations. It is

52 especially important for the useful cases in which perfect definitions are used – for example, when perfect charged particle tracks are used to test the PFA performance in the calorimeter only. When used with the standard detector calibration procedure, resolutions of individual particle energy distributions match those of the particles used in the calibration method. The Perfect PFA provides a check not only of the optimal PFA performance, but also a check of the calorimeter calibration method.

Cluster Algorithms Many cluster algorithms have been developed for use in PFAs. More importantly, most of these have standardized inputs and outputs, so easy evaluation and comparison of multiple cluster algorithms can be done. Procedures exist to illustrate purity and efficiency of cluster algorithms, so that optimal choice can be made of algorithms to use in various stages of the PFA. For example, when choosing a cluster algorithm for track- cluster matching, overall purity performance is preferred over efficiency, while for photons, a more balanced performance is desired.

4. PFA Approaches for SiD

Several PFA approaches have been investigated for the ILC detector concepts. These can be divided into basically 2 types – a modular approach designed to perform particle identification and separation with methods optimized for each particle species, and cluster algorithm approaches where a specific cluster algorithm is used to define all particle clusters in the calorimeter.

One approach to PFA development is to use an optimized single cluster algorithm to try to form clusters of all particles in the event. These cluster algorithms must be able to reconstruct clusters of many varied types – mips, electromagnetic showers (photons and electrons), and hadronic showers (both neutral and charged). Cluster algorithms based on hit density, fragment distances, and shower shapes have been developed for this type of PFA.

Another approach is to develop individual algorithms optimized for the particular particle being tested. This type of PFA has a natural modular structure with each step consisting of a cluster algorithm and topological evaluation algorithms optimized for a particular particle species. The modular structure allows easy comparisons to be made at each stage and for fast identification of PFA performance.

5. PFA Results

The following plots show results for application of a PFA on single particle species. The ultimate performance limit of the PFA can be illustrated for each type of particle in this way and estimates of the minimum confusion introduced by the PFA can be made. Plots for charged hadrons (pions), photons, and neutral hadrons here.

Single Particle PFA Response Charged Hadron/Cluster Association

53 Photon Algorithm Performance Neutral Hadron Algorithms

Initially, PFAs were developed using simulated e+e- -> Z production at the Z pole (91.5 GeV). No jet algorithms were needed and the PFA performance could be shown as an energy sum. Since there is no ambiguity in the choice of jets used in the dijet mass calculation, the dijet mass resolution and the energy sum resolution are ~equal for these events. Plots for ZPole performance of PFA if any.

PFA Results for Total Energy Sum at ZPole

The dijet mass is the most important result of PFA performance for the e+e- collider at 500 GeV CM. Dijet results are shown here for events with 2 jets, 4 jets, and 6 jets at 500 GeV CM. When plotted as a function of jet energy, the increasing contribution of confusion in the real PFA is seen.

Dijet Results @500 GeV vs jet energy

6. Future SiD Optimization with PFAs

Based on the performance obtained so far, PFAs will be used to optimize several parameters of the SiD detector, helping to choose these parameters for the final SiD design. In particular, the PFA performance can be used to evaluate various choices of inner radius (distance from IP to inner face of ECAL), and B-field (central solenoidal field strength). Along with the cost parameters, the PFA can then be used to optimize the size of the SiD detector. Also, various technologies for an optimized HCAL can be tested with PFA applications. For example, the PFA performance of analog scintillator designs can be compared to digital gas calorimeters. Some plots illustrating beginning studies of these things here.

IR B-Field (H)CAL Technology

54