A Novel Strip Energy Splitting Algorithm for the Fine Granular Readout of a Scintillator Strip Electromagnetic Calorimeter

Transcription

1 1 3 A Novel Strip Energy Splitting Algorithm for the Fine Granular Readout of a Scintillator Strip Electromagnetic Calorimeter 4 Katsushige Kotera a, Daniel Jeans b, Akiya Miyamoto c, and Tohru Takeshita a a Department of Physics, Shinshu University, Asahi, Matsumoto, Nagano , Japan b Department of Physics, Graduate School of Science, The University of Tokyo, Hongo, Bunkyo-ku, Tokyo , Japan c High Energy Accelerator Research Organization (KEK), 1-1 Oho, Tsukuba, Ibaraki 35-81, Japan 5 January 9, 15 6 Abstract We describe an algorithm which has been developed to extract fine granularity information from an electromagnetic calorimeter (ECAL) with strip-based readout. Such a calorimeter, based on scintillator strips, is being developed to apply particle flow reconstruction to future experiments in high energy physics. The application of this algorithm to 1 GeV hadronic jets in an ECAL with 45 5 mm transverse segmentation improves the energy resolution from 3.6% to 3.%, to be compared to the resolution of.9% achieved by an ECAL with 5 5 mm segmentation. The performance can be further improved by the use of 1 1 mm tile-shaped layers interspersed between strip layers. 1 Introduction In the experiments being designed for next generation particle colliders, the particle flow approach (PFA)[1, ] is the leading candidate to provide unprecedented jet energy resolution. Especially at the electron-positron colliders, PFA achieves the jet energy resolution required to fully exploit the possibilities offered by the colliders well-defined initial and clean final states. In PFA, the energy of charged particles is measured by the tracking system, which has a much better momentum resolution than the energy resolution of calorimeters: typical resolutions are σ 1/pT 1 5 GeV 1 for the tracking system, and σ E /E 46%/ E(GeV) 1.6% for the HCAL response to single charged pions [3]. The calorimeters are used to estimate the energy only of neutral particles. In order to apply this approach, the calorimetric showers of each particle must be individually reconstructed. The granularity of calorimeter readout is therefore a key issue. Corresponding author (coterra@azusa.shinshu-u.ac.jp) 1

2 As an example, the sampling electromagnetic calorimeter (ECAL) with tungsten absorbers being designed for the International Large Detector (ILD, a detector being designed for use at the International Linear Collider (ILC) [3]) is optimized to have a transverse segmentation of 5 mm, corresponding to half of the Molière radius of tungsten, and - 3 longitudinal samplings in a total thickness of 3 X, giving a total of 1 8 readout channels. The effective Molière radius of the ECAL is around mm. A readout granularity finer than 5 mm does not result in significant performance gains for jet energies of 1 GeV and below, which correspond to the most relevant jet energies for the ILC physics program[1]. One technology being developed to effectively achieve this high calorimeter granularity is based on plastic scintillator strips individually read out by miniature photon detectors, for example pixelated photon detectors (PPD, also commonly known as SiPM) [5]. The use of long scintillator strips rather than 5 5 mm tiles simplifies the design of such an ECAL, and also reduces its cost, due to the reduced number of readout channels. Successive ECAL layers have orthogonally aligned strips, giving an effective granularity close to the strip width. The CALICE collaboration has developed and constructed ECAL prototypes based on this technology, using scintillator strips of length 45 mm and width 5 or 1 mm, individually read out by PPDs [6, 7]. This paper presents a reconstruction method which can be used to extract close to 5 5 mm effective granularity from such long scintillator strips, and reports on measurements of its performance using events fully simulated in ILD. Details of this detector are given in the next section, the reconstruction procedure is explained in section 3, and section 4 describes the calibration procedure. The performance of this method on the basic detector-related measures of position resolution and two-particle separation are discussed in sections 5 and 6, while the jet energy resolution, which affects the physics performance of the detector, is discussed in section 7. Finally we discuss the results in section 8 and summarize this study in section 9. Detector model Starting from the interaction point (IP), the ILD consists of a vertex detector, silicon tracking layers, a large time projection chamber (TPC) surrounded by additional silicon tracking detectors, a calorimeter system consisting of electromagnetic and hadronic sections, all placed within a solenoidal magnetic field of strength 3.5 T. The steel return yoke is instrumented to provide muon identification. The basic structure consists of a central, barrel, region aligned with the beam axis, closed by two endcaps in the forward regions. The ECAL barrel detector has an octagonal cross-section, a length of around 5 m, and an inner radius of 1.85 m. A cylindrical coordinate system with its axis (z) aligned with the beam line is used in this paper. More details of the ILD design can be found in [3]. The ILD is simulated in Mokka [8], a Geant4-based simulation tool [9]. Figure 1 left shows a multiple-jet event simulated in ILD. The ILD strip-scintillator ECAL (strip-scecal) is a sampling calorimeter. In the simulation model used in this study, thirty sensitive layers are interleaved with tungsten plates of thickness.1 (4.) mm in the inner twenty (outer nine) layers. The tungsten absorber layers correspond to a total thickness of X, while layers of readout electronics, copper heat transfer plates, and scintillators contribute less than a single additional X. The total thickness of the ECAL is around mm. The sensitive layers are tiled with 45 5 mm scintillator strips of thickness 1 mm. Strips are aligned orthogonally in successive layers. A dead volume of size mm 3 is implemented at the end of each scintillator strip to represent the volume occupied by the PPD, which leads to a constant term in the energy resolution. In the analyses presented in this paper, when scintillator tiles or strips with an area of less than 45 5 mm are used, this dead volume is scaled by the strip area 1. This avoids penalizing smaller cell sizes with an additional constant term to the energy resolution. Each strip is enveloped by a reflective film of thickness 1 This scaling is of course not possible in practice, although we note that techniques for the readout of scintillation light from the lower surface, which therefore avoid dead space due to the PPD, are currently being studied.

3 Figure 1: Left: a view of ILD with a simulated multiple-jet event. The highlighted detector components are: 1. muon detector;. solenoid; 3. hadron calorimeter; 4. electromagnetic calorimeter; 5. TPC; and 6. vertex detector. Right: a 45 mm 5 mm mm scintillator strip and a PPD µm. Scintillator strips and PPDs are mounted on mm ECAL base units (EBU), printed circuit boards which also host the front-end electronics and LEDs used for calibration. The EBUs are arranged within mechanical structures in such a way as to avoid any projective cracks. Further technical details are available in [3]. Printed circuit boards and copper heat radiators are simulated in each detector layer. Four different scintillator tile configurations were used in this study: mm tiles ( 5 5 ); mm strips ( 45 5 ); alternating layers of 5 5 mm tiles and 45 5 mm strips ( alt5 ); and alternating layers of 1 1 mm tiles and 45 5 mm strips ( alt1 ). 88 Successive strip layers were always orthogonally aligned. A hadronic calorimeter based on forty layers of mm3 scintillator tiles interleaved with mm iron absorbers was simulated in this study Strip Splitting Algorithm A simple algorithm, the Strip Splitting Algorithm (SSA), has been developed to extract fine granularity information from the long strip geometry. Each strip is split into n virtual cells along its length; n is chosen to result in approximately square virtual cells; for example, a 45 5 mm strip is split into nine 5 5 mm cells. A simple procedure is used to distribute the total energy Estrip detected by the strip among its virtual cells. A weight wk is assigned to each virtual cell k, defined as the sum of the energies Ei of all strips in immediately neighboring layers which intersect with virtual cell k, when seen from the IP: wk = Ei, (1) (i=intersect) 3

4 where the sum runs over strips, i, in immediately neighboring layers which intersect with the virtual cell k. The energy E k assigned to virtual cell k is then E k = E strip w k (j=all virtuals) w j, () where the sum j runs over all virtual cells of the strip being split. Figure shows a schematic of the SSA procedure. In the case of the alt1 model, the 1 1 mm tile is first split into virtual cells. The immediately neighboring strip layers are used to partition the tile s energy into these virtual cells using a very similar method to that defined above, making use of the strips orthogonal orientation. In a second step, the virtual cells originating from the 1 1 mm tiles are used to partition strips energy among their virtual cells, as described above. transverse longitudinal transverse Estrip k = 1,, 3, 4, 5, 6, 7, 8, 9 incident particle IP Figure : A cartoon illustrating the SSA procedure. The energy, E strip reconstructed in the central, longitudinal, strip is split among virtual cells (k = 1,, 9) by considering the energy in the orthogonally aligned transverse strips in neighboring layers. More details of the procedure are given in the text Events were analyzed using a particle flow reconstruction algorithm. In the results presented later in this paper, the PandoraPFA algorithm[1, 4] was used to analyze events, together with other standard ILD reconstruction programs (e.g. for tracking) in the MarlinReco[1] package. PandoraPFA uses as input the energy and position of calorimeter deposits, as well as reconstructed charged particle tracks. When SSA was not used, PandoraPFA was passed the central position and energy of each strip (this is the simplest possible approach), while in the case of SSA, the central position and assigned energy of each virtual cell was used. 4 Calibration Scintillator strips with an energy deposit were used to create calorimeter hits, whose energy was taken to be the energy deposited in the scintillator. No account was taken of non-linear or discretized behavior of the PPD response. In order to remove noise hits while keeping sufficient efficiency for minimum ionizing particle (MIP) detection, an energy threshold on each hit in the ECAL and HCAL was applied at a fraction of the most probable per-cell energy deposition of 1 GeV anti-muons ( MIP ). For the ScECAL, a threshold of.5 MIP was set for each strip, and an additional threshold of.3 MIP was set for virtual cells after the SSA procedure. For HCAL hits, a.3 MIP threshold was set, the default value used in PandoraPFA. The package versions defined in ILCSoft v1-16- were used. 4

5 The calorimeter was calibrated by studying the energy deposited by 1 GeV photons for the electromagnetic response, and 1 GeV neutral long lived Kaons (K L ) for the hadronic response. Particles were injected from the IP in a direction almost perpendicular to the beam-line (in order to avoid the central electrode of the TPC) and uniformly distributed in azimuth. The calibration factors used to convert between the energy deposited in the scintillator and that deposited in the whole calorimeter (including the tungsten absorbers and other passive materials) were chosen to give a mean energy, after PFA reconstruction, equal to the incident particle energy. Additional calibration factors used to account for different values of the hadronic and electromagnetic responses h/e, separately defined in the ECAL and HCAL, were optimized to obtain the best jet energy resolution, according to the standard PandoraPFA procedure [4]. The ECAL was re-calibrated for each ECAL configuration, while a single HCAL calibration was used for all configurations. Apart from this calibration, the same parameters of PandoraPFA were used for all ECAL geometries. 5 Position of clusters The precise reconstruction of cluster positions is important in a PFA analysis to ensure good matching between tracks and calorimeter clusters. In this study, 1 GeV photons were fired into the 45 5 ECAL from the IP, varying the ECAL injection position along the z direction. The injected position was taken to be the intersection on the ECAL front face of the line joining the reconstructed cluster position 3 and the IP. Figure 3 shows the difference between the position reconstructed by PandoraPFA and the true photon injection position, as a function of the injected position. The size of the vertical error bars reflects the width of the reconstructed position distribution. When SSA is not used, a strong position-dependent bias of up to 5 mm is observed, corresponding to cases when the photon shower passes near the ends or center of a strip (z = 68 mm corresponds to the center of strips aligned with the z axis). When SSA is used, these biases are almost completely removed, and cluster positions are reconstructed to better than 1 mm 4. Shift from injectiondum(mm) with SSA 45 5 without SSA Injection position in zdu(mm) Figure 3: Shift of reconstructed position for 1 GeV photons. The vertical error bars reflect the widths of the position shift distributions. 3 The position of a calorimetric cluster is defined as the energy-weighted mean position of its constituent hits. 4 A small z-dependent bias remains after SSA, due to the relative position within the virtual cell. The small consistent position bias visible after SSA is due to the fact that the chosen injection positions are all in the same relative position within the virtual cells. When integrated over z, the average bias becomes effectively zero. 5

6 Two-particle separation To investigate the separation ability of the strip-scecal with SSA, di-muon events and π decays (for photon-photon separation) were studied. These simple events serve to illustrate and investigate the phenomena of spurious ghost clusters which can occur when two particles are simultaneously incident in a square area whose side is shorter than the strip length. The reconstructed invariant mass of the two photons from π decays provides a good test of both the photon separation and reconstruction ability. 6.1 µ - µ separation Di-muon events provide a sensitive system for measuring two-particle separation in different ScECAL designs. The ability to reconstruct two anti-muons of momentum 1 GeV produced at IP was studied as a function of the distance between their impact positions on the ECAL surface. One muon was injected into the ECAL at a fixed point, with polar angle almost perpendicular to the beam line and azimuthal angle φ = π/. The second was injected to impact the ECAL surface at some distance from this point, with δ rφ = δ z = /, where δ rφ (δ z ) is the separation between the ECAL entry points in the r φ (z) direction, and was scanned between 7 and 5 mm. Both injection positions were smeared by. mm on the ECAL surface in r φ and z. Four thousand simulated di-muon events were analyzed using SSA and PandoraPFA. eventsdumy - µ Fraction of µ alt 5 w/ SSA alt 1 w/ SSA 45 5 w/ SSA distance between µ - µ d(mm) Figure 4: Left: fraction of correctly reconstructed di-muon events as a function of the distance between the muons at the front face of the ECAL. Right: energy deposit of two electrons incident simultaneously on 45 5 strip-scecal reconstructed with SSA. Two of four peaks are ghosts Figure 4 left shows the fraction of such events in which the two muons were reconstructed without any additional ghost calorimeter clusters, as a function of the distance between the entry points of the two muons into the ScECAL. Apart from the 45 5 strip-scecal with the use of SSA, all types of ScECAL have over 9% correctly reconstructed events when the distance between the two muons is larger than 1 mm. In the case of the 45 5 strip-scecal with the use of SSA, the fraction depends strongly on the position, with a minimum at a distance of 3 mm. This is due to the possibility of two-fold ambiguities arising in reconstruction when two particles enter a square region with size smaller than the strip length, which can lead to the formation of ghost clusters. Figure 4 right illustrates a case in which two injected electrons produced two additional ghost clusters. At a separation of 3 mm in Fig. 4 left, around 5% of di-muon events in the 45 5 strip-scecal have a single additional reconstructed ghost cluster, and a further 5% have two ghost clusters. At distances below 3 mm, ghost clusters are more often combined 6

7 with the true muon cluster. At a separation of 7 mm, the correctly identified fraction drops to around 7% for all ScECAL configuration due to the merging of nearby clusters. The use of interleaved tile layers removes the ambiguities leading to ghosts, and dramatically improves the situation, resulting in a performance comparable to that of a tile-based ScECAL. 6. π reconstruction to study two-photon separation A π meson decays into two photons, which can be reconstructed using only calorimeter information. The π energy has a strong influence on the opening angle between the two photons, and variations in the decay angle give rise to different photon energies in the laboratory frame. Samples of π decays at different energies are therefore a powerful tool to measure the ECAL performance, both in terms of pattern recognition (the ability to identify two clusters), and energy resolution (by considering the invariant mass of identified clusters). The opening angle between photons decreases as the π energy increases: the most probable distance between photons from a 1 (3) GeV π decay on the ECAL surface is 53 mm (17 mm). A PandoraPFA photon cluster is required to satisfy several requirements on several observables, including the absence of associated tracks, the depth of the cluster start within the ECAL, and the angle between the reconstructed cluster axis and the line joining the cluster position to the IP [1]. Figure 5 shows the fraction of π events in which one, two, and more than two photon clusters are reconstructed in the 5 5 mm ScECAL, left, and alt1 ScECAL, right. At each energy 8 π s were injected from the IP almost perpendicular to the beam line and with a uniform distribution in azimuthal angle. In around 9% of events, at least one of the photons from the π decay converts before reaching the ECAL, and the number of events with no reconstructed photons is less than.5% for all π energies. The fraction of events correctly reconstructed with two photon clusters decreases strongly with increasing energy. One contribution to this decrease is the merging of nearby clusters, seen in the increase in the fraction of events in which a single photon cluster is reconstructed, understood as the effect of the decreasing distance between photons. An additional factor is the increase of events in which more than two photon clusters are reconstructed. This is seen in all geometries (5 5 mm and alt1 are shown in Figure 5), demonstrating that this is the effect of PandoraPFA clustering algorithms, rather than the effect of ghost clusters caused by the strip readout. 1 1 Fraction of events # photon = 1 # photon = # photon > Fraction of events # photon = 1 # photon = # photon > Energy of π (GeV) 1 3 Energy of π (GeV) Figure 5: The fraction of π events reconstructed with one, two, and more than two photon clusters, in the 5 5 mm ScECAL left and alt1 strip-scecal right geometries. No requirements are made on the energy or invariant mass of identified photons. The statistical uncertainties are not visible on this scale. 7

8 Events/.34 GeV/c 6 4 Model "5 5" = 5 GeV = 1 GeV = GeV = 3 GeV = 4 GeV a) Events/.34 GeV/c 6 4 Model "45 5"+SSA = 5 GeV = 1 GeV = GeV = 3 GeV = 4 GeV b) dum (GeV/c ) M γγ M γγ dum (GeV/c ) >.1 GeV/c having clusters M γγ π w/ SSA alt5 w/ SSA alt1 w/ SSA 5 5 c) 1 3 Energy of π (GeV) ) (GeV/c duµ µ 45 5 w/ SSA alt5 w/ SSA alt1 w/ SSA Energy of π (GeV) σ d) σdu(gev/c )dus34 Figure 6: Reconstructed invariant mass M γγ at different π energies, for events with two reconstructed photon like clusters in a) the 5 5 tile-scecal, and b) 45 5 strip-scecal, and c) the fractions of such events, in the four ECAL types, having M γγ >.1 GeV/c, and d) the mean (µ) and width (σ) of a Gaussian fit to the reconstructed M γγ distributions as a function of the π energy. Vertical bars show statistical uncertainties only Figure 6 shows various characteristics of π events in which exactly two photon-like clusters were identified. Figure 6a (6b) shows the reconstructed invariant mass of such events at different π energies in a 5 5 tile-scecal (45 5 strip-scecal with SSA). Figure 6c shows the fraction of π events which have exactly two identified photon-like clusters with a reconstructed invariant mass greater than.1 GeV/c, for the four ScECAL geometries considered. The use of alternate tile layers improves the performance compared to a purely strip-based geometry, as was also seen in the case of di-muon events. The fact that the alt1 model gives statistically significantly better results than alt5 at energies of 1 and GeV is unexpected, and not yet fully understood. Figure 6d shows the means and standard deviations of Gaussian fits to the invariant mass spectra for π of different energies, reconstructed in different ScECAL geometries. SSA was used in all models except the 5 5 tile-scecal. These studies show that some level of π reconstruction is possible at energies of up to 3 GeV using the clustering algorithms implemented in PandoraPFA. No large performance differences were seen between the four considered ECAL geometries. 8

9 Jet energy resolution The reconstruction of the energy of two-jet events is an important benchmark test of the PFA. Samples of 1 4 e + e q q (q = u, d, s) events were fully simulated in ILD at center-of-mass energies of 91.,, 36, and 5 GeV, and then reconstructed using MarlinReco, including SSA and PandoraPFA. To evaluate the jet energy resolution, only events in which both quarks were produced in the barrel region ( cos(θ) <.7) were considered. The total reconstructed event energy E jj is defined as the sum of the energies of all particles reconstructed by PandoraPFA. The distribution of this event energy is used to determine RMS9(E jj ), defined as the root mean square of the smallest range of the E jj distribution which contains at least 9% of events. The single jet energy resolution RMS9 is estimated by dividing RMS9(E jj ) by. This RMS9 measure is useful to describe the generally non-gaussian energy distributions with small, but long, tails obtained by PFA, since it makes no assumptions about the shape of the distribution, and also avoids giving undue weight to events far in the tails. At lower jet energies, where single particles are well separated in the calorimeters, the energy resolution is dominated by the single particle response of the calorimeters. For higher energy jets, in which the average distance between particles in the calorimeters is smaller, the so-called confusion term becomes important. Confusion is due to the mis-association of fragments of charged particle showers to neutral clusters, and vice versa, which leads to a over- or undercounting of energy, leading to a degradation in jet energy resolution. The relative contributions of the single particle calorimetric response and confusion in hadronic jets become comparable at energies of around 1 GeV [1]. 7.1 Energy spectra Figure 7 shows the total reconstructed energy (E jj ) at a center-of-mass energy of GeV when using the 5 5 and 45 5 Sc-ECAL models, with and without the use of SSA. The shape of the energy spectrum for the 45 5 Sc-ECAL is noticeably improved by the use of SSA, and comes close to that of the 5 5 model. Events/4 GeV w/o SSA 45 5 w/o SSA Reconstructed energym (GeV) Figure 7: Reconstructed energy in q q events at GeV, for the 5 5 mm ScECAL and the 45 5 mm strip-scecal with and without the use of SSA. 9

10 Dependence on strip length The dependence of the jet energy resolution at a center-of-mass energy of GeV on the strip length is shown in Fig. 8. The same strip width of 5 mm was used in all models. The strong degradation in performance with increasing strip length seen when SSA is not used is almost completely mitigated by the use of SSA. The difference in jet energy resolution between an ScECAL using 5 5 tiles and one using strips of length up to 6 mm is almost negligible 5. RMS9/E(%) ScECAL w/o SSA ScECAL w/ SSA Length of strip (mm) Figure 8: Estimated single jet energy resolution RMS9/E of jets produced in q q events at a center-of-mass energy of GeV Comparison of 5 5, 45 5, and ScECAL models Figure 9 shows the jet energy resolution as a function of jet energy for a 45 5 strip-scecal without and with SSA, and and 5 5 tile-scecal models. At smaller jet energies, below 1 GeV, the jet energy resolution is dominated by the intrinsic single-particle resolution, giving a resolution which improves with increasing energy. At energies above 1 GeV, the confusion between charged and neutral calorimeter clusters becomes significant, leading to a degradation of resolution with increasing energy for all ScECAL geometries. The jet energy of the 45 5 strip- ScECAL is significantly improved by using SSA, especially at and above 1 GeV, indicating that confusion is decreased by SSA. A tile of has the same area as a 45 5 strip, however it is clear that the jet energy resolution when using the strip geometry is significantly better (by up to.5 percentage points (p.p.) ), demonstrating the real merit of a strip-based geometry used in conjunction with SSA. The degradation in jet energy resolution between the 5 5 and 45 5 (with SSA) models is rather small, less than.5 p.p. for 45 GeV and 5 GeV jets, and around.1 p.p. for jets between 1 and GeV. 7.4 Jet energy resolution of strip-tile-scecal As discussed in section 6.1, the major problem faced by a strip-based ECAL is the formation of ghost clusters. Therefore, the use of interleaving tile layers is expected to improve the jet energy resolution. Figure 1 compares the jet energy resolutions of 45 5, alt5, and alt1 models. The performance of the alt5 and alt1 models are almost identical. At jet energies of up to 1 GeV, they strips. 5 Note that this assumes a uniform response along the strip length. Response non-uniformities will favor shorter 1

11 4 RMS9/E(%) w/o SSA w/o SSA Energy of One Jet du(gev) Figure 9: Single jet energy resolution of as function of jet energy for different tile- and strip-scecals give almost the same performance as the 5 5 model, while at higher energies the performance lies approximately half way between the 5 5 and 45 5 models. 4 RMS9/E(%) w/ossa alt5 w/ossa alt1 w/ossa Energy of One Jet du(gev) Figure 1: Single jet energy resolution as a function of jet energy for 45 5, alt5, alt1, and 5 5 ScECAL models Discussion SSA successfully extracts fine granularity information, at an effective scale close to that of the strip width, from a strip-based calorimeter. This motivates the development of scintillator stripbased calorimeters, with their advantages of less readout channels and lower cost, for a wide range of experiments. The small degradation in jet energy resolution, of around. p.p., when going from a 5 5 tile-scecal to a 45 5 strip-scecal can be largely recovered by the use of tile layers interleaved between the strip layers. These tile layers prevent the formation of ghost clusters, as has been demonstrated in the reconstruction of a simple di-muon system. Tile layers with a granularity of 1 1 mm have been shown to work well. The use of such a tile size is technically feasible. 11

12 The use of mm tiles, which have the same area as the 45 5 mm strips currently being used in a prototype ScECAL, and therefore the same density of readout electronics, are certainly technically feasible. Studies of reconstruction performance with such larger tiles are continuing. The use of scintillator-based 5 5 mm layers is technically difficult at present, but a different technology, such as the silicon readout ECAL being developed by CALICE [11], could be used. The Mokka simulation model used in this study describes the detailed geometry of scintillator strips, dead volumes due to the reflector, PPD packages, radiators, circuit boards, and mechanical structures. However, effects such PPD saturation, cross-talk and dark noise phenomena, finite photo-electron statistics, and non-uniform scintillator response have not been simulated in the results presented in this paper. These effects may have some influence on the absolute energy resolutions reported in this paper, but the effect on the relative performance improvements achieved by SSA are expected to be minimal. 9 Summary An algorithm ( SSA ) to extract fine granularity from scintillator strips was developed and tested. The reconstructed position of clusters was significantly improved with SSA: the maximum shift in the reconstructed position of 1 GeV photon clusters from the true position is improved form 5 mm to sub-mm by using SSA on a 45 mm strip-scecal. The Strip-ScECAL with SSA also shows comparable particle separation performance to the 5 5 mm tile ScECAL in muon pair and π events. Ghost clusters created when two particles impinge on a square area with a size less than the strip length are effectively removed by interleaving strip layers with square tile layers. The energy resolution for jets of up to 5 GeV obtained with several ScECAL geometries was compared. The differences in the obtained jet energy resolutions when using a 5 5 tile- ScECAL and 45 5 strip-scecal with SSA reconstruction ranged from.15 p.p. to.5 p.p.. This difference can be removed for jet energies below 1 GeV, or decreased to.1 p.p. for jet energies in the range 15-5 GeV, by alternately replacing strip layers with tile layers. Acknowledgments The authors would like to thank to ILD group for providing essential simulation (Mokka) and analysis tools (Marlin and PandoraPFA), and for useful discussions. John Marshall has given invaluable advice on the tuning and calibration of the PandoraPFA algorithm. Members of the CALICE collaboration, in particular the CALICE-ASIA group, and of the ILD-ASIA group, have also contributed to many important discussions. This work is supported in part by grant-in-aid for specially promoted research: a global research and development program of a state-of-the-art detector system for the ILC of the Japan Society for Promotion of Science (JSPS). References [1] M. Thomson, Particle Flow Calorimetry and the PandoraPFA Algorithm. NIM, A611, 5 (9). [] Jean-Claude Brient and Henri Videau, The Calorimetry at the future e + e linear collider. arxiv:hep-ex/4 (). [3] Ties Behnke et al., The International Linear Collider Technical Design Report - Volume 4: Detectors. arxiv: (13). [4] J. Marshall and M. Thomson, The Pandora Particle Flow Algorithm. Proceedings of CHEF13, Paris. arxiv: (13). 1

13 [5] K. Kotera, Scintillator Strip ECAL Optimization. Proceedings of LCWS13, Tokyo. arxiv: (14). [6] K. Francis et al., Performance of the first prototype of the CALICE scintillator strip electromagnetic calorimeter. NIM, A763, 78 (14). [7] K. Kotera, Performance of the CALICE Scintillator-Based ECAL Depending on the Temperature. Proceedings of LCWS11, Granada. arxiv:11.698v (1). [8] P. Mora de Freitas and H. Videau, Detector simulation with MOKKA / GEANT4: Present and future. Proceedings of LCWS, Seoul. C-8-6.5, p63-67 (3). Mokka Homepage, mokka.inp3.fr (14). [9] S. Agostinelli et al., Geant4 a simulation toolkit. NIM, A56, 5 (3). [1] Marlin Homepage, (14). [11] J. Repond et al., Design and electronics commissioning of the physics prototype of a Si-W electromagnetic calorimeter for the International Linear Collider. JINST 3 P81 (8). 13