Speaker
Description
Timepix4 is the most recent application-specific integrated circuit (ASIC) developed by the Medipix4 international collaboration at CERN. It features a 448×512 pixel matrix with a 55 μm pitch. Its design supports integration with diverse semiconductor sensors, enabling its use in multiple domains such as X-ray spectroscopy, high-energy particle detection, and medical imaging [1].
Designed for compatibility with Through-Silicon-Via (TSV) technology, Timepix4 can be tiled seamlessly on all four sides, enabling coverage of large detection areas with negligible dead zones and achieving sub-200 ps timing resolution [1]. The architecture supports two readout modes: frame-based and data-driven. In frame-based mode, events exceeding a programmable threshold are counted at the pixel level, and the matrix is read out synchronously via the core clock. In data-driven mode, a pixel transmits a data packet after detecting a hit, capturing both Time-of-Arrival (ToA) and Time-over-Threshold (ToT) values. This mode extends the photon-counting capability with additional temporal and energy information. With support for hit rates up to 5.0×10⁹ hits/mm²/s and data transfer speeds up to 160 Gbit/s, it offers high-performance capabilities for demanding applications [1,2].
Since 2020, INFN has been a member of the Medipix4 collaboration. Within this framework, two INFN-CSN5-funded projects (MEDIPIX4 (2021–2024) and TIMEPIX4 (2025–2027)) have been launched to explore the applicability of readout chips in diverse domains, ranging from X-ray spectral imaging to nuclear medicine and radiation dosimetry. Ongoing work at INFN involves the characterization of Timepix4 assemblies coupled to different sensor materials, including silicon (Si), cadmium telluride (CdTe), and gallium arsenide (GaAs).
This presentation summarizes experimental work performed with a recently acquired 700 μm-thick GaAs detector assembly. Due to its higher atomic number and electron mobility than Si and lower fluorescence photon emission energy (<12 keV) compared to CdTe (<30 keV), GaAs provides superior detection efficiency (better than 70%) for X-ray energies up to 50 keV, making it well suited for mid-energy X-ray imaging applications.
At INFN Ferrara, the detector was extensively characterized under both frame-based and data-driven readout conditions. For frame-based operation, the setup includes an X-ray mammography tube with a tungsten anode. The Timepix4 assembly is positioned one meter from the source, aligned along the beam direction, and mounted on a dedicated holder equipped with a copper heat exchanger. Temperature regulation is provided by circulating chilled water (15 °C) through the exchanger via an external chiller. In data-driven mode, an X-ray fluorescence (XRF) setup is used. In this configuration, the Timepix4 detector is positioned perpendicular to the primary beam, which is directed onto target materials placed at a 45-degree angle along the beam path. Readout in both configurations is performed using the SPIDR4 electronics, with system control and data acquisition managed via DATAPIX4 [3], an in-house software developed at INFN Ferrara.
For energy calibration in data-driven mode, the XRF setup allows the conversion of ToT signals into the corresponding collected charge and, subsequently, into the energy of the incident X-ray [2]. Five different target materials were selected for this measurement, producing characteristic X-ray fluorescence photons with energies ranging from 9.89 keV (Ge) to 24.21 keV (In). This approach reduced the energy discrepancy from 32%, observed when relying solely on internal test pulses, to 6% (see left panel of Figure 1). Additionally, this readout mode enabled the investigation of charge-sharing effects across the pixel matrix as a function of photon energy, offering insights that support the interpretation of data acquired in frame-based mode. Right panel of Figure 1 shows the fraction of clusters with different sizes generated in the sensor as a function of photon energy. As the energy increases, the fraction of single-pixel (size-1) clusters decreases significantly, with a reduction of approximately 25% across the investigated energy range. However, analysis of the energy distribution of size-1 clusters (see left panel of Figure 1) reveals that 40% of the counts correspond to energies below half of the expected value, which is consistent with photon interactions occurring near pixel borders. This indicates a pronounced charge-loss effect that needs to be investigated.
The frame-based mode, optimized for high-rate acquisition, is particularly suitable for applications such as medical diagnostics. In this context, flat-field response uniformity is a key performance metric. Over a continuous 5-hour acquisition period, multiple flat-field images were acquired under identical conditions and compared, revealing relative variations smaller than 1% over the entire period for 99.4% of the pixels. A first radiography corrected for the flat-field and for defective pixels can be seen in Figure 2a-b, showing the quality of the GaAs assembly. The image was acquired at the LARIX-A laboratory of INFN Ferrara and University of Ferrara using a 700 μm GaAs sensor coupled to a Timepix4 chip, operated in frame-based mode (16-bit, 0.01 s frame duration) with a bias voltage of –350 V and a threshold of 3.36 keV. The X-ray tube was set to 25 kVp and 60 mA, resulting in a hit rate of approximately 15,000 hits/pixel/s over a 6 s exposure. The spatial resolution was evaluated using the slanted-edge technique at two different mean energies, 17 keV and 27 keV, by calculating the Line Spread Function (LSF) and the Modulation Transfer Function (MTF). For the higher energy acquisition, the full width at half maximum (FWHM) of the LSF is approximately 70 μm (see Figure 2c). In addition, a resolution test phantom (star pattern) was acquired, as shown in Figure 2d, where the lines are clearly distinguishable up to the Nyquist frequency (9.1 mm⁻¹). The results of preliminary tests on response linearity as a function of photon rate and detector efficiency, will also be presented and discussed.
[1] X. Llopart et al., Timepix4, a large area pixel detector readout chip which can be tiled on 4 sides providing sub-200 ps timestamp binning, Journal of Instrumentation, 17 (2022) C01044.
[2] A. Feruglio et al., Timepix4 Calibration and Energy Resolution Evaluation with Fluorescence Photons, Il Nuovo Cimento C, 47 (2024).
[3] V. Cavallini et al., DataPix4 A C++ framework for Timepix4 configuration and read-out, arXiv preprint 2503.01609 (2025).
| Workshop topics | Detector systems |
|---|
