Timepix3 Optical Camera (Tpx3Cam)

• Timepix3 Optical Camera (Tpx3Cam) is an event-driven imaging system that records optical events as sparse, time-stamped pixel hits using a Timepix3 hybrid pixel sensor.
• It employs per-pixel analog front-ends with independent time-of-arrival and time-over-threshold measurements, enhanced by an image intensifier for single-photon sensitivity.
• Tpx3Cam is applied in quantum optics, 3D TPC readout, neutron imaging, and ultrafast VMI, enabling advanced timing reconstruction and energy proxy analyses.

The Timepix3 Optical Camera (Tpx3Cam) is a Timepix3-based optical camera that records optical events as sparse, time-stamped pixel hits rather than as conventional frames. In its intensified form, it couples a Timepix3 hybrid pixel sensor to an image intensifier and relay optics so that individual optical photons can be converted into localized multi-pixel flashes, each tagged with pixel coordinates, time-of-arrival (ToA), and time-over-threshold (ToT). Across the literature, the same architecture has been used for single-photon quantum measurements, 3D optical readout of gaseous and liquid-argon time projection chambers (TPCs), neutron and $\alpha$-particle detection through scintillators, and high-rate velocity-map imaging (VMI) of charged particles (Nomerotski, 2019, Roberts et al., 2018).

1. Sensor architecture and optical chain

At the core of Tpx3Cam is the Timepix3 hybrid pixel ASIC, typically bump-bonded to a silicon sensor in a $256\times256$ array with $55\times55\ \mu\mathrm{m}^2$ pixel pitch. Each pixel contains its own analog front-end, discriminator, and local digital logic, and independently reports a hit only when its signal exceeds threshold. The hit payload comprises pixel coordinates together with ToA and ToT, yielding a data-driven, sparse, zero-suppressed stream. In the Timepix3 architecture, the ToA granularity is $1.56\ \mathrm{ns}$, the ToT granularity is $25\ \mathrm{ns}$, and the per-pixel dead time is $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$. Readout is performed with SPIDR electronics, which support maximum output rates of up to $80\ \mathrm{Mpix/s}$ and provide an external TDC with $0.26\ \mathrm{ns}$ granularity for synchronization (Nomerotski, 2019).

In the intensified configuration used for single-photon work, Tpx3Cam is coupled to a Photonis Cricket image intensifier comprising a photocathode, MCP stack, and fast P47 phosphor, with relay optics projecting the phosphor output onto the sensor. Reported photocathode choices include High-QE Red with $\mathrm{QE}\approx18\%$ at $800\ \mathrm{nm}$, Hi-QE Red with $256\times256$0 at $256\times256$1, and Hi-QE Green with $256\times256$2–$256\times256$3 QE at $256\times256$4. The P47 phosphor has a rise time of about $256\times256$5, a decay time of about $256\times256$6, and an emission peak at $256\times256$7, which matches the anti-reflection-optimized, high-QE silicon sensor response. The intensified chain is essential for single-photon optical operation because the bare Timepix3 sensor is not itself single-photon sensitive; in several studies, direct optical signals remained below threshold without intensification (Nomerotski et al., 2022, Lowe et al., 2020, Roberts et al., 2018).

2. Event-driven photon registration and timing reconstruction

The defining operational mode of Tpx3Cam is asynchronous hit logging. A photon converted at the intensifier photocathode produces an MCP avalanche and then a localized P47 flash, which illuminates multiple neighboring pixels on the Timepix3 sensor. Photon reconstruction is therefore based on clustering and centroiding rather than on single-pixel identification. In quantum-optics measurements, typical single-photon clusters spanned about $256\times256$8 pixels, while another single-photon characterization found a cluster-size distribution peaking at $256\times256$9 pixels, with a sub-peak at $55\times55\ \mu\mathrm{m}^2$0 pixels. Photon coordinates are commonly estimated with ToT-weighted centroiding, and the photon time is often assigned from the ToA of the pixel with the largest ToT in the cluster (Ianzano et al., 2018, Nomerotski et al., 2022).

A central reconstruction issue is discriminator time walk: larger signals cross threshold earlier and therefore appear to arrive earlier, while also producing larger ToT. The standard correction is written as

$55\times55\ \mu\mathrm{m}^2$1

where $55\times55\ \mu\mathrm{m}^2$2 is obtained empirically, either per pixel or from a global lookup table. In the polarization-entanglement measurements, a ToT-based correction reduced intra-cluster timing spreads from about $55\times55\ \mu\mathrm{m}^2$3 to a few nanoseconds. In SPDC spectrometer measurements, the corrected two-photon coincidence distribution had rms $55\times55\ \mu\mathrm{m}^2$4, corresponding to about $55\times55\ \mu\mathrm{m}^2$5 rms per photon. A separate SPDC-based characterization that used the pixel with maximum ToT as the cluster time reported a single-detection temporal resolution of about $55\times55\ \mu\mathrm{m}^2$6 FWHM at $55\times55\ \mu\mathrm{m}^2$7 (Ianzano et al., 2018, Nomerotski et al., 2022, Vidyapin et al., 2022).

This timing behavior is often misunderstood. The ToA binning of about $55\times55\ \mu\mathrm{m}^2$8 is the sensor’s time granularity, not the full end-to-end timing resolution of an optical measurement. In practice, the latter depends on timewalk correction, phosphor response, photocathode and MCP statistics, cluster definition, and application-specific optics.

3. Conversion of ToA and ToT into depth, time of flight, and calorimetric proxies

In TPC readout, the Tpx3Cam data product becomes genuinely three-dimensional once ToA is combined with a drift model. In the CF$55\times55\ \mu\mathrm{m}^2$9 proof-of-principle demonstration, the longitudinal coordinate was reconstructed as

$1.56\ \mathrm{ns}$0

with the timing contribution to the depth resolution written as

$1.56\ \mathrm{ns}$1

Using $1.56\ \mathrm{ns}$2 and $1.56\ \mathrm{ns}$3 gave an ideal estimate $1.56\ \mathrm{ns}$4, while the reported practical $1.56\ \mathrm{ns}$5 resolution was “better than 1 mm.” In the ARIADNE liquid-argon TPC, PMT-detected prompt scintillation supplied $1.56\ \mathrm{ns}$6, the drift velocity was measured as $1.56\ \mathrm{ns}$7 at $1.56\ \mathrm{ns}$8, and ToT-based calorimetry yielded an energy resolution of approximately $1.56\ \mathrm{ns}$9. At meter scale, ARIADNE+ used four cameras over a $25\ \mathrm{ns}$0 active region and calibrated summed track intensity to the minimum-ionizing muon value $25\ \mathrm{ns}$1 with $25\ \mathrm{ns}$2 in one visible-light quadrant (Roberts et al., 2018, Lowe et al., 2020, Lowe et al., 2023).

In pulsed-neutron work, the same ToA information is converted into neutron kinematics through time of flight,

$25\ \mathrm{ns}$3

followed by

$25\ \mathrm{ns}$4

At CSNS BL20, this enabled wavelength-resolved imaging, arbitrary software selection of wavelength bands, and Bragg-edge imaging of steel, with edges observed at $25\ \mathrm{ns}$5, $25\ \mathrm{ns}$6, $25\ \mathrm{ns}$7, $25\ \mathrm{ns}$8, and $25\ \mathrm{ns}$9. In a separate thermal-neutron feasibility study using a $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$0 converter on LYSO, the reconstruction hierarchy was explicitly defined as hits, clusters, and events, with integrated ToT, hits per event, and clusters per event used for signal selection (Yang et al., 2021, Gao et al., 2024).

ToT plays a related but distinct role across these domains. It is routinely used as a proxy for local brightness or deposited charge, not as an absolute energy observable without calibration. In the CF$$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$1 TPC demonstration, the light-yield proxy was written as

$$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$2

and a clear peak in $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$3 was observed for $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$4 $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$5 particles at about $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$6. In LArTPC work, track-summed ToT was converted to $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$7 and then anchored to known muon stopping power.

4. Reported performance envelope

The reported performance of Tpx3Cam is strongly configuration-dependent, because optics, scintillator or phosphor choice, intensifier photocathode, clustering method, and object-plane magnification all change the effective instrument response. Representative figures reported across the literature are summarized below (Nomerotski, 2019, Nomerotski et al., 2022, Vidyapin et al., 2022, Yang et al., 2021, Gao et al., 2024, Lowe et al., 2023).

Characteristic	Reported value	Context
Sensor format	$$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$8 pixels, $$t_{\mathrm{dead,pixel}}=\mathrm{TOT}+475\ \mathrm{ns}$$9 pitch	Core Timepix3 geometry
ToA granularity	$80\ \mathrm{Mpix/s}$0, often stated as about $80\ \mathrm{Mpix/s}$1	Sensor-level timing bin
Pixel dead time	$80\ \mathrm{Mpix/s}$2	Timepix3 per-pixel limit
Readout bandwidth	up to $80\ \mathrm{Mpix/s}$3 or $80\ \mathrm{Mpix/s}$4	SPIDR/Tpx3Cam streaming
External timing	$80\ \mathrm{Mpix/s}$5 or $80\ \mathrm{Mpix/s}$6	SPIDR/TDC markers
Single-photon timing	$80\ \mathrm{Mpix/s}$7 rms after correction; $80\ \mathrm{Mpix/s}$8 FWHM in another characterization	Processing- and setup-dependent
Example object-plane spatial scales	$80\ \mathrm{Mpix/s}$9; $0.26\ \mathrm{ns}$0 over $0.26\ \mathrm{ns}$1; $0.26\ \mathrm{ns}$2 neutron imaging; $0.26\ \mathrm{ns}$3 with $0.26\ \mathrm{ns}$4 optics	Application-specific

These values are not interchangeable as if they belonged to a single operating point. The $0.26\ \mathrm{ns}$5 neutron-imaging result, for example, came from centroiding with a thin scintillator and beamline TOF reconstruction, whereas the $0.26\ \mathrm{ns}$6 figure arose from mapping a $0.26\ \mathrm{ns}$7 TPC quadrant onto the same $0.26\ \mathrm{ns}$8 sensor. Likewise, single-photon timing depends not only on the Timepix3 binning but also on intensifier afterpulsing, cluster-time assignment, and ToT-based timewalk correction. One single-photon study reported afterpulsing within $0.26\ \mathrm{ns}$9 pixels in $\mathrm{QE}\approx18\%$0 of cases and within $\mathrm{QE}\approx18\%$1 in $\mathrm{QE}\approx18\%$2 of cases, while an SPDC-based efficiency calibration measured an average system efficiency of $\mathrm{QE}\approx18\%$3 at $\mathrm{QE}\approx18\%$4.

5. Experimental domains and demonstrated uses

The same hit format—$\mathrm{QE}\approx18\%$5—has supported a wide range of measurement modalities.

Domain	Representative result	Source
Quantum optics	CHSH $\mathrm{QE}\approx18\%$6; spatial mapping over $\mathrm{QE}\approx18\%$7 region pairs; wavelength precision $\mathrm{QE}\approx18\%$8 and timing precision $\mathrm{QE}\approx18\%$9 in SPDC spectroscopy	(Ianzano et al., 2018, Nomerotski et al., 2022)
Optical TPC readout	First 3D optical readout of a TPC in $800\ \mathrm{nm}$0 CF$800\ \mathrm{nm}$1; later $800\ \mathrm{nm}$2 dual-phase LArTPC operation with four cameras and $800\ \mathrm{nm}$3 spatial resolution	(Roberts et al., 2018, Lowe et al., 2023)
Neutron and $800\ \mathrm{nm}$4 detection	$800\ \mathrm{nm}$5 neutron imaging after centroiding; thermal-neutron rate $800\ \mathrm{nm}$6 after filtering matching $800\ \mathrm{nm}$7 simulation; $800\ \mathrm{nm}$8-event timing about $800\ \mathrm{nm}$9	(Yang et al., 2021, Gao et al., 2024, D'Amen et al., 2020)
VMI and coincidence imaging	Coincidence VMI of electrons and ions on one detector; $256\times256$00 and $256\times256$01 FEL measurements; simultaneous-hit separation down to roughly $256\times256$02 with GPU-accelerated centroiding	(Zhao et al., 2017, Bromberger et al., 2021, Gabalski et al., 5 Mar 2026)
Ultrafast nanoparticle ion imaging	High-occupancy ToF+VMI at ELI with sustained user-level throughput $256\times256$03 and correction of $256\times256$04 timestamp shifts	(Ševaev et al., 7 Oct 2025)

This diversity reflects a structural feature of the platform rather than a collection of unrelated case studies. Once the camera output is treated as an asynchronous event stream, the same reconstruction primitives—clustering, centroiding, ToT-based timing correction, coincidence filtering, and object-plane calibration—can be adapted to very different physical observables.

6. Advantages, limitations, and development trajectory

Several advantages recur across the literature. In TPC applications, Tpx3Cam provides native 3D information because $256\times256$05 and $256\times256$06 come directly from the sensor while $256\times256$07 is inferred from ToA, avoiding multiple wire planes and ambiguity resolution. The data stream is intrinsically sparse and zero suppressed, which reduces event size and supports continuous, triggerless operation. Cameras can be mounted outside the cryostat or vacuum vessel, simplifying maintenance and upgrades. In quantum and VMI experiments, the same architecture supports simultaneous multi-region analysis, multi-hit operation, coincidence filtering, and software-defined time windows. In large-area optical TPC concepts, scalability is pursued through tiling cameras and standardizing THGEM modules, while direct VUV imaging has been demonstrated alongside wavelength-shifted visible-light readout (Roberts et al., 2018, Lowe et al., 2023, Gabalski et al., 5 Mar 2026).

The limitations are equally consistent. A common misconception is that Tpx3Cam is intrinsically a single-photon camera; in practice, single-photon sensitivity in the optical band is achieved through the appended intensifier, and the effective quantum efficiency is then dominated primarily by the photocathode. Another misconception is that ToT directly measures energy; in all cited use cases it is an energy proxy or brightness proxy whose absolute interpretation depends on optical coupling, gain stability, threshold settings, and calibration. High occupancy introduces cluster overlap, pile-up, and local dead-time effects; intensified configurations add dark counts, afterpulsing, phosphor persistence, and vignetting; and very high instantaneous rates can expose readout artifacts such as the $256\times256$08 timestamp shifts reported in ultrafast nanoparticle imaging. The practical timing resolution is also broader than the nominal Timepix3 binning because of timewalk, phosphor rise and decay, diffusion in TPC media, and optical blur (Vidyapin et al., 2022, Nomerotski et al., 2022, Ševaev et al., 7 Oct 2025).

The development trajectory is correspondingly technical rather than conceptual. Reported directions include integrating PMT-detected prompt light for absolute $256\times256$09 and absolute $256\times256$10 in TPCs, deploying multiple Tpx3Cams on larger THGEM areas in liquid argon, improving VUV optics to reduce vignetting, using optimized photocathodes for wavelength-matched operation, refining GPU-based sparse clustering for high-rate VMI, and moving toward next-generation sensors such as Timepix4 and more integrated intensifier-sensor packages. This suggests that the central idea of Tpx3Cam is stable: a sparse, per-pixel, time-stamped optical event stream. Current research is focused on extending that idea to larger fields of view, higher occupancies, better calibrated timing, and more application-specific optical front ends (Roberts et al., 2018, Lowe et al., 2023, Gabalski et al., 5 Mar 2026).