betaFeedback

WorksPhysics and Astronomy

Application of exhaustive simulation flow for advanced performance prediction of monolithic active pixel sensors

Abstract

Monolithic active pixel sensor (MAPS) developments have pushed the detection performance in various directions, especially relative to timing where nanosecond-level precision is now considered. This evolution calls for a simultaneous upgrade of the simulation tools. We have developed a simulation flow that covers steps from the signal creation in the sensitive volume to the output of the pixel digital logic that performs the time-of-arrival and time-over-threshold (ToA/ToT) measurements. This approach adds several new features to the traditional use the of the TCAD - Allpix Squared duo, among which : the integration of the pixel wells from the layout in order to precisely describe the pixel key characteristics such as leakage and punch-through currents and the coupling of Monte Carlo simulations (Allpix Squared) with high precision electrical simulations (SPICE). The first (Allpix Squared) for the precise description of the current induced at the collection electrode and the second (SPICE) to guarantee high precision simulation of the front-end electronics using realistic signal events. Irradiation is also modeled, both from the charge propagation side (charge trapping) and from the front-end response side (high input signal discharge). We have applied this methodology to the MAPS developed in the context of the Belle II vertex detector upgrade. In this contribution, we detail the key features of the exhaustive simulation flow, present the outcome of the comparison with the TJ-Monopix2 measurements and discuss the interest of the methodology for the development of modern MAPS.

ᵒDepartment of Physics, University of Tokyo, Hongo 7-3-1, Tokyo 113-0033, Japan

ᵖInstituto de Fisica Corpuscular (IFIC), CSIC-UV, Catedratico Jose Beltran, 2. E-46980 Paterna, Spain

ᑫSchool of Physical and Chemical Sciences, Department of Physics and Astronomy, Queen Mary University of London, 327 Mile End Road, London, E1 4NS, United Kingdom

ᵣCollege of Physics, Jilin University, 2699 Qianjin Street, Changchun, Jilin, China

ˢUniversidade de Santiago de Compostela, 2010 Instituto Galego de Física de Altas Enerxías (IGFAE), Colexio de San Xerome, PZ Obradoiro, S/N. E-15782 Santiago de Compostela, Spain

E-mail: [email protected]

Introduction

Recent monolithic active pixel sensors (MAPS) developments [1] have pushed the detection performance in various directions, especially relative to timing where nanosecond-level precision is now consider. This evolution requires at the same time an upgrade of the simulation tools and methodologies used to develop these sensors, both on physics and integrated circuit design. In this contribution, we present and use the proposed simulation flow, from precise modeling of the 3D pixel model, irradiation modeling both from physics side and electrical side, up to reproducing accurate calibration measurements (before and after irradiation) using recently developed Monte Carlo coupled simulation within the Allpix Squared framework, allowing to precisely describe all the steps of the signal processing. First of all, we describe the integration of the nn and pp-wells in the pixel 3D model, then we present the currents evolution with irradiation, by comparing the simulated results to measurements. Afterwards, the front-end response degradation with irradiation is studied, before reproducing in simulation the measurements performed with 55Fe{}^{55}\mathrm{Fe} radioactive source, by applying the proposed exhaustive simulation flow.

This work is presented in the context of the upgrade of the SuperKEK-B collider (Tsukuba, Japan), hosting the Belle II experiment for which a new fully pixelated vertex detector is being developed – the VTX. All the VTX detection layers will be equipped with the same MAPS: OBELIX (Optimized BELLE II pIXel sensor). Operation at room temperature after a NIEL fluence of 5×10145 \times10^{14} 1 MeV neq/cm2n_{\mathrm{eq}}/\mathrm{cm}^{2} and providing time-stamping at 50 ns as baseline or 3 ns with increased power dissipation are key specifications of OBELIX. This chip is derived from the TJ-Monopix2 prototype [2, 3]. The TJ-Monopix2 prototype [4] was originally developed for ATLAS ITk, and is manufactured in the TowerJazz 180 nm CMOS modified process [5], with a low dose n-type implant intended to uniformize the depletion. The sensor is a 512×512512 \times512 pixels matrix, with a 33.04 µm pixel pitch. Several versions have been fabricated, and a single sensor also contains several pixel front-end flavors. The measurements and simulations presented in this work correspond to the version featuring these characteristics : 30 µm epitaxial layer, "DC-cascode" analogue front-end. The pixel circuit features a time-of-arrival and time-over-threshold (ToA/ToT) logic. The ToT logic has 7 bits, with a 25 ns clock period. The TJ-Monopix2 prototype introduces an asymmetry in the deep p-well layer (shown in Figure 1a), to enhance the lateral electric drift in the surrounding area. This feature is called RDPW (Reduced Deep P-Well), as opposed to the typical FDPW (Full Deep P-well). The presented simulations are compared with measurements performed on the three following proton-irradiated sensors: W8R4, W2R17 and W8R6, irradiated respectively at fluences up to: Φ=1\Phi= 1, 2.5 and 5×10145 \times10^{14} 1 MeV neq_{\mathrm{eq}}/cm2^2[6]. The samples have undergone a long-term annealing of around one year between the irradiation and the measurements. There is an approximately 23 % doping difference in the low dose n-type implant: W8 samples are "low" and W2 is "high".

Image not extracted

Figure 1. (a) 2x2 pixels layout, (b) Cross-section model of a single pixel (not at scale).

Layout integration in the pixel 3D model

As introduced in section 1 and illustrated in Figure 1a, the TJ-Monopix2 prototype features an asymmetric deep p-well layer in the pixel area, that was introduced to enhance the lateral electric field, and consequently has an effect on some pixel properties, mainly the charge collection efficiency.

The regions where large area n-wells are placed have an impact on the in-pixel charge collection efficiency[4]. An impinging charged particle crossing the sensing volume will deposit energy all along its path, generating electrons/holes pairs. Depending on its position, a portion of the electrons may be generated directly in the n-well, thus being collected by the n-well itself, without being propagated in the p-type epitaxial layer. Depending on the depth of the epitaxial layer, this loss of charge can correspond to a few percent of the total deposited charge. For example, considering a 30 µm epitaxial depth, this loss can amount to approximately 4 %. Taking into account these effects in simulation requires integration of the pixel layout in the 3D TCAD pixel model. To do so, the layers of interest are extracted from the GDS (Graphic Design System) files to the TCAD Sentaurus Structure Device Editor (SDE) environment. Preliminary operations are mandatory to correctly prepare the layers of interest and to be able to manipulate them independently when creating the 3D pixel model. The resulting 3D model contains the following layers, crucial for the charge collection: deep p-well, p-well, n-wells, low-dose n-type implant, extra p-stop implants. The 3D pixel model created is presented in Figure 1b. To verify the effect of the deep P-Well on the efficiency of the charge collection, we compare the charge collection profile for the two FDPW and RDPW versions (Figure 2). In Figure 2a, the streamlines in black illustrate the difference in the electric field lines between the typical FDPW case (left) and the RDPW case (right).

Image not extracted

Figure 2. (a) Top views of the pixel, FDPW (left) and RDPW (right) versions. The red polygons represent deep p-well’s shape. The electric field streamlines are shown for the region of interest. A and B markers point the same position where the charge deposition is performed, for the two versions. (b) Charge collection profiles for the two versions, before irradiation.

The collection diode voltage is set to +1.15 V+1.15\ \mathrm{V}, the p-well and the substrate voltages are set to −6 V-6\ \mathrm{V}. A charge deposition of 63 e−/μm63\ \mathrm{e}^{-}/\mu\mathrm{m} (assumed to correspond to the charge deposited by a Minimum Ionizing Particle (MIP) in a 30 μm30\ \mu\mathrm{m} silicon thickness) is done through the sensor depth, for cases A and B, corresponding to FDPW and RDPW versions. Their charge collection profiles (seed pixel), before irradiation, are presented in Figure 2b. With the FDPW, the collected charge is 1175 e−1175\ \mathrm{e}^{-} with a rise-time (defined as the duration between 10%10\% and 90%90\% of the total charge collection time) trise,FDPW=6.14 nst_{\mathrm{rise,FDPW}} = 6.14\ \mathrm{ns}, while the RDPW improves the collected charge up to 1365 e−1365\ \mathrm{e}^{-} (≈16%\approx16\% better), with a rise-time trise,RDPW=5.88 nst_{\mathrm{rise,RDPW}} = 5.88\ \mathrm{ns}.

Including the discussed deep p-well asymmetry into the 3D TCAD model of the TJ-Monopix2 allows to increase the accuracy of the simulations, thanks to better modeling of the electric field in the pixel epitaxial layer. This 3D pixel model is used in the following sections to perform I-V simulations (§ 3) and Monte Carlo simulations (§ 5).

Currents evolution with irradiation

The currents that impact performance of the pixel the most are the ones between the following electrodes: the substrate (SUB), the p-wells (PWELL) and the collection diode. The SUB and PWELL potentials are set at −6 V-6\ \mathrm{V} (before irradiation), while the CCE is biased by a reset diode at approximately VCCE=1.15 VV_{\mathrm{CCE}} = 1.15\ \mathrm{V}. The current contributions to be studied are the leakage current and the punch-through current. The leakage current IleakI_{\mathrm{leak}}, flowing towards the collection diode increases with irradiation and temperature. Characterization of IleakI_{\mathrm{leak}} is mandatory, as this requires modification of the DC-operating point of the front-end input node, as detailed in § 4. The punch-through current, IptI_{\mathrm{pt}} flows between SUB and PWELL when those electrodes are biased at different voltages. This current contribution decreases with irradiation (the resistance decreases). Before irradiation, measurements performed on the TJ-Monopix2 prototype have highlighted the impossibility to operate at a high (in absolute value) SUB voltage, while after NIEL irradiation it was possible to do so, due to effective build up of a positive charge at the edge of the PWELL, as in the double junction effect. TCAD simulations have been carried out to reproduce the I-V curves obtained in measurements with the samples W8R4, W2R17 and W8R6 (irradiation levels given in § 1). To include in the simulation the effects of irradiation on the evolution of currents, the University of Perugia model [7] is used. We compare in Figure 3 the results of the measurements and simulations. In Figure 3a, we give the evolution of IleakI_{\mathrm{leak}} in function of the temperature, given for both W8R4 and W8R6 samples.

Image not extracted

Figure 3. (a) IleakI_{\mathrm{leak}}, meas. and sim. (Perugia model), for different temperatures, PSUB set to -20 V, (b) IptI_{\mathrm{pt}}, meas. and sim., for different SUB bias voltages. PWELL set to -6 V in both cases.

We find that the simulations are in good agreement with the measurements. The leakage current behavior is well reproduced between 0.1 nA and 0.5 nA, especially for the most irradiated sample, W8R6. The radiation damage constant α\alpha is used as a figure of merit to compare the measurements with the simulations. It is defined by: α=Ileak/(Φeq×Vol)\alpha= I_{\mathrm{leak}/(\Phi_{\mathrm{eq} \times V_{\mathrm{ol}}})} where we consider the leakage current IleakI_{\mathrm{leak}} (calculated at T = 20°C), at the given fluence Φeq\Phi_{\mathrm{eq}} in the volume of the device VolV_{\mathrm{ol}} (we consider a full depletion of the epitaxial layer). In Table 1 are presented the calculated α\alpha values, for both measurements and simulations, showing a reasonable agreement between measurement and simulation, especially for the W8R6 sample. In Figure 3b we introduce the evolution of IptI_{\mathrm{pt}} depending

NIEL (1 MeV neqn_{\mathrm{eq}}/cm2^{2})A (IleakI_{\mathrm{leak}}@20°C) (nA)EeffE_{\mathrm{eff}} (eV)α\alpha meas. (10−1710^{-17} A · cm−1^{-1})α\alpha sim.
1 × 101410^{14}0.033 ± 0.0011.65 ± 0.101.02 ± 0.042.04
5 × 101410^{14}0.214 ± 0.0081.53 ± 0.031.31 ± 0.051.07

Table 1. Pixel IleakI_{\mathrm{leak}}, and radiation damage constants, from the figure 3a fits

on the SUB voltage. The punch-through current starts flowing between SUB and PWELL when the two electrodes are biased at a different voltage. We can notice that higher irradiation levels reduce this current, making possible operation at SUB voltages down to -20 V (for the W8R6 sample). However, there is a discrepancy when comparing the measured IptI_{\mathrm{pt}} with the simulated one, suggesting that this specific case is slightly outside the scope of the Perugia model. Trying to understand the modeling of IptI_{\mathrm{pt}} have pushed the work in the direction of adding ad-hoc modifications to the Perugia model. Further optimization of the model, to better describe the effect of the positive charge build at the PWELL edge, is ongoing. Nevertheless, it should be noted that the 23 % relative ratio of the low-dose n-type implant doping, between the W8 and W2 samples, when taken into account in the TCAD pixel cell, reproduces the same trend as in measurements. The AC-coupled simulation is also performed to extract the detector capacitance: Cd=3.5C_{\mathrm{d}} = 3.5 fF (TCAD), matching the value obtained from layout parasitic extraction. The pixel TCAD simulation results are in good agreement with the measurements performed on the irradiated samples. For the purpose of modeling the effects of irradiation on the analogue in-pixel front-end, the simulated values of leakage current and detector capacitance are reused in the forthcoming steps of this exhaustive simulation flow presentation.

Front-end response degradation with irradiation

The irradiation increases the leakage current flowing towards the front-end input, which causes a large sensor discharge. This effect significantly shorten the current pulse (due to a faster reset of the input) and in turn reduces the input signal amplitude. We aim to reproduce the electrical behavior of the front-end in simulation, before and after irradiation. To do so, SPICE simulation of the front-end circuit is performed using Spectre (Cadence) [8]. We reproduce the same front-end bias conditions as in measurements. The effects of irradiation on the electrical behavior of the analogue pixel is modeled with a current source, representing the leakage current IleakI_{\mathrm{leak}}, as presented in Figure 4. Irradiation also causes surface damage (TID, Total Ionizing Dose) in the oxide layer (SiO2_2) by creating positive oxide charge (holes, being less mobile, can become trapped in the oxide). This can shift the threshold voltage of the transistors (negative shift for n-mos, positive for p-mos).

Image not extracted

Figure 4. Pixel analogue circuit with the injection circuit (highlighted by the red dotted rectangle).

We don’t model the surface damage in our simulation, as the samples have been proton-irradiated (thus mainly subject to bulk damage), and due to a lack of foundry models describing surface damage effects on the transistor behavior. The in-pixel charge injection circuit implemented in TJ-Monopix2 is used for the internal ToT calibration (Figure 4), to emulate a signal corresponding to the one created by a particle hitting the pixel. The amount of injected charge can be controlled thanks to the on-chip 8-bit DAC, working in the range from 0 to 200 DAC units, with the smallest step of 7.03 mV. To reproduce accurately the pixel-to-pixel response dispersion, originating from transistor-to-transistor mismatch during the fabrication process, we use the Monte Carlo mismatch dispersion values provided by the foundry, available in the TJ-180 PDK (Process Design Kit). The simulation temperature is set according to the measured one, as is IleakI_{\mathrm{leak}}: before irradiation, Tchip=20 ∘CT_{\mathrm{chip}} = 20\,^\circ\mathrm{C}, negligible leakage current (≈\approx pA), and after irradiation, Tchip=37 ∘CT_{\mathrm{chip}} = 37\,^\circ\mathrm{C}, and Ileak=1.2I_{\mathrm{leak}} = 1.2 nA. In Figure 5 are presented the comparisons between the measured and simulated ToT curves, respectively before and after irradiation. The following equation: ToT⁡(Qinj)=(aQinj+b−1)(Qinj−d)\operatorname{ToT}(Q_{\mathrm{inj}}) = \left(\frac{a}{Q_{\mathrm{inj}}} + b^{-1}\right)(Q_{\mathrm{inj}} - d) is used to fit the ToT as function of the input charge. The fitted parameter values are given directly on the Figure 5. Simulations reproduce fairly the behavior observed in the measurements. Before irradiation, at a charge injected corresponding to 140 DAC (max. DAC value reachable in measurement), we obtain a ToT of 37 in measurements, and 45 in simulation. After irradiation, at 140 DAC, measurement gives a ToT of 8, and 10 in simulation. The quantitative agreement is within 20 %, more importantly the effect of the high-sensor signal discharge is correctly modeled by integrating the leakage current into the electrical simulation after irradiation. It should be mentioned that the front-end DC-operating point is very sensitive to changes of the biasing currents, which makes it difficult to obtain the same exact result in simulation and in measurement. This model of the front-end, including both the impact of the irradiation and the mismatch dispersion of its response will be now used to perform Monte Carlo coupled simulation.

Image not extracted

Figure 5. ToT evolutions with injected charges, upper row: before irradiation, (a) measurement and (b) simulation, lower row: after irradiation, (c) measurement and (d) simulation

Coupling Monte Carlo physics and electrical simulations

Once the effects of the irradiation have been characterized and modeled in TCAD and SPICE simulations, we have reproduced 55Fe{}^{55}\mathrm{Fe} measured spectra with TJ-Monopix2 prototype in simulation. The methodology consists of using the latest Allpix Squared [9] developments - the integration of SPICE simulation within the Allpix Squared Monte Carlo environment. The simulation of the deposition, the propagation and the collection of the charge is performed by Allpix Squared, whereas the electrical simulation of the front-end is done by SPICE, within a single event. The SPICE simulation is fully integrated within Allpix Squared via the [NetlistWriter] module, to benefit from the accuracy of both simulators with a user-friendly approach. It should be also mentioned, that each pixel is assigned a mismatch ID value (corresponds to the seed of the Monte Carlo simulation) in order to guarantee that the same pixel will have the same mismatch value from one event to another while pixel-to-pixel dispersion is still considered (introduced in § 4). We compare the ToT histograms (front-end output), obtained in measurement and simulation, of a 55Fe^{55}\mathrm{Fe} calibration before and after irradiation in Figure 6a and Figure 6d. As illustrated in these two figures, measurements results can be well reproduced in simulation.

Image not extracted

Figure 6. (a), (b) and (c), before irradiation, respectively : the simulated and measured ToT histograms (front-end output), the simulated collected charge histogram (front-end input), and the 2D histogram of the simulated ToT depending on the simulated charge. (d), (e) and (f), after irradiation, same histograms

With the simulation, the collected charge (raw front-end input, without considering the front-end discharge effect) can be accessed. We present these histograms, before and after irradiation, respectively in Figure 6b and Figure 6e. Before irradiation, one can distinguish the 55Fe^{55}\mathrm{Fe} Kα_{\alpha} peak at ≈1600 e−\approx1600\ \mathrm{e}^{-}, contrary to after irradiation, due to the trapping of charges. The charge collected on the diode, and the resulting front-end analogue output (expressed in ToT with a 25 ns clock), both obtained in simulation, can be represented in a 2D histogram (Figure 6c and Figure 6f). This permits to determine to which ToT value corresponds the 55Fe^{55}\mathrm{Fe} Kα_{\alpha} peak (ToT = 57 before irradiation), and thanks to the ToT linearity curve presented in Figure 5b, to calculate the injection capacitance value Cinj,simC_{\mathrm{inj,sim}} following the same procedure as the one performed with the measurements. This ToT value corresponds to Inj55Fe=180I_{\mathrm{nj}^{55}\mathrm{Fe}} = 180 DAC, giving an injection capacitance of Cinj,sim=1600/180=8.9 e−/DACC_{\mathrm{inj,sim}} = 1600/180 = 8.9\ \mathrm{e}^{-}/\mathrm{DAC}, matching the one obtained with the measurements: Cinj,meas=9 e−/DACC_{\mathrm{inj,meas}} = 9\ \mathrm{e}^{-}/\mathrm{DAC}. The front-end loss of signal is a combination of trapping and sensor discharge caused by the high leakage current. By reaching a good matching between measurements and simulations of 55Fe^{55}\mathrm{Fe} source spectra, we confirm that the effects of the irradiation on the front-end behavior are correctly modeled in our coupled Monte Carlo simulation, emphasizing the interest of the presented exhaustive simulation flow.

Conclusion

We presented the main features of the proposed exhaustive simulation flow. The integration of the n/p-wells into the 3D TCAD pixel model helps to refine the electric field conditions of the sensing volume directly impacting the charge collection properties. The I-V simulations were performed to reproduce leakage and punch-through current measurements on irradiated sensors. Thanks to the Perugia model, we obtained a correct description of the evolution of the leakage current as a function of the temperature and irradiation levels. We modeled the effect of the irradiation on the front-end electrical behavior with a current source, reusing the simulated I-V results, which allowed to reproduce the sensor discharge similar to the measurements. By embedding SPICE simulation into Allpix Squared, we extended the framework to electrical simulations and successfully reproduced 55Fe calibrations both pre- and post-irradiation. By comparing measurements with simulations we validated the methodology and demonstrated the interest of the proposed simulation flow. This methodology aims to be applied to the ongoing sensor R&D programs, and the future generations of MAPS.

Acknowledgments

This work has received the support from the European Union’s Horizon 2020 Research and Innovation programme under Grant Agreements no 101004761 (AIDAinnova), Multilateral Scientific and Technological Cooperation in the Danube Region (MULT 03/23), Horizon 2020 ERC-Consolidator Grant No. 819127, TY-FJPPN (Toshiko Yuasa-France Japan Particle Physics Network), the MCIU with funding from the European Union NextGenerationEU (PRTR-C17.I01) and Generalitat Valenciana (GVANEXT), Project ASFAE/2022/016.

References

  1. [1]W. Snoeys, Monolithic CMOS Sensors for high energy physics — Challenges and perspectives, Nucl. Instrum. Meth. A 1056 (2023) 168678.
  2. [2]G. Rizzo et al., The DMAPS upgrade of the Belle II Vertex Detector, Nucl. Instrum. Meth. A 1072 (2025) 170164.
  3. [3]C. Bespin et al., Development and characterization of a DMAPS chip in TowerJazz 180 nm technology for high radiation environments, Nucl. Instrum. Meth. A 1040 (2022) 167189.DOI
  4. [4]K. Moustakas, Design and Development of Depleted Monolithic Active Pixel Sensors with Small Collection Electrode for High Radiation Applications, Ph.D. thesis, Bonn U., 2021.
  5. [5]W. Snoeys et al., A process modification for CMOS monolithic active pixel sensors for enhanced depletion, timing performance and radiation tolerance, Nucl. Instrum. Meth. A 871 (2017) 90.DOI
  6. [6]G.F. Benfratello, Characterization of irradiated monolithic CMOS active pixel sensors for the upgrade of the Belle II Vertex Detector, Ph.D. thesis, Pisa U., 2025.
  7. [7]F. Moscatelli et al., Combined Bulk and Surface Radiation Damage Effects at Very High Fluences in Silicon Detectors: Measurements and TCAD Simulations, IEEE TNS 63 (2016) 2716.DOI
  8. [8]Cadence Design Systems, Inc., Spectre Circuit Simulator, version 23.1 ed., 2024.
  9. [9]S. Spannagel et al., Allpix2: A modular simulation framework for silicon detectors, Nucl. Instrum. Meth. A 901 (2018) 164.DOI

Paper details

Contents