.jpg){kind=logo}
CheeTah T3 Ultra is a hybrid pixel detector built from eight Timepix3 chips on one gap-free monolithic silicon sensor. It counts every electron with zero readout noise and time-stamps each one in 1.56 ns bins, so a spectrum image, a time-resolved spectrum and a coincidence histogram are all different binnings of the same recorded data.
The zero-loss peak may contain 105 to 106 times more electrons than the core-loss edge you are measuring, on the same detector, in the same exposure. Beam-sensitive specimens cannot be given more dose to make up the difference.
What can be measured is therefore set largely by the detector: its noise, point spread, dynamic range, speed and, increasingly, its ability to say when each electron arrived. Event-based detection changes all five.
Each pixel counts electrons above a threshold and rejects everything below it, including all electronic noise. Weak edges at low dose are limited by counting statistics only. Data are zero-suppressed at the source, so storage scales with the electrons recorded, not with channels × probe positions.
About 102 eV at 0.1 eV per channel for a monochromated low-loss spectrum with its ZLP; 512 eV at 0.5 eV per channel to record the C-K, N-K and O-K edges in one exposure; about 2 keV at 2 eV per channel. Chip boundaries cost six channels in total, well under 1 % of the range.
The non-dispersive direction retains momentum information for momentum-resolved EELS and phase-contrast imaging from the EELS signal. It gives the zero-loss peak room to spread, tolerates spectrometer aberrations and drift, and lets the same detector serve 4D-STEM and diffraction.
The two chips under the ZLP are rotated by 90°. A ZLP line that would otherwise load a single Timepix3 double-column bus now crosses all 128 double columns of each chip, so the full readout bandwidth of both chips is available to the peak.
No frames, no frame overhead. The scan unit defines probe position over time and each electron is assigned afterwards. Published work runs Lissajous scans at 5 to 120 kHz and a 2048 × 2048 scan at 250 ns per pixel in about 1 s, with the delivered dose known for every pixel.
Two TDC inputs (260 ps bins) put laser sync, photon detectors or the scan clock on the same clock as the electrons. The optional CERN PicoTDC adds 64 channels at 3.3 ps RMS for multi-detector coincidence and pump–probe experiments.
| Parameter | CheeTah T3 Ultra | Note |
|---|---|---|
| Readout ASIC | 8 × Timepix3 | 4 × 2 arrangement, CERN Medipix collaboration |
| Pixel matrix | 1024 × 512 | Monolithic sensor; two chips rotated under the ZLP |
| Pixel pitch | 55 µm | |
| Active area | 56.3 × 28.2 mm² (nominal) | |
| Sensor | 300 µm silicon | GaAs, CdTe and other options on request |
| Electron energy | 60–300 kV | Single-electron sensitivity down to 5 keV; best time resolution at 60–100 kV |
| Readout modes | Event mode (x, y, ToA, ToT) and frame mode | Selectable per experiment |
| Time-of-arrival bin | 1.5625 ns | About 1.4 ns measured after calibration |
| External TDC inputs | 2 × 260 ps | Laser sync, photon detector, scan clock |
| PicoTDC option | 64 channels, 3.3 ps RMS | Multi-detector coincidence and pump–probe timing |
| Readout noise | None | Per-pixel threshold, single-electron counting |
| Data format | Open .tpx3 event list, 8 bytes per event | Python-accessible; frame export to standard formats |
| Software & interface | SERVAL, Ethernet | HTTP interface for third-party software |
Each Timepix3 pixel handles on the order of a million hits per second before pile-up. In event mode, the Orsay group reports 1 to 5 pA as the maximum current with an unsaturated ZLP: adequate for monochromated low-loss and time-resolved work, low for core-loss mapping. Beyond the rotated chips, the ZLP can be spread across up to 512 rows, placed off the detector when only core losses matter, or recorded in frame mode, which has been used at 25 pA. The ToT value of each event flags pile-up so affected pixels can be excluded in analysis.
The 1.56 ns bin is the clock, not the resolution. After calibration the measured resolution is about 1.4 ns. At higher voltages the ionisation volume in the sensor dominates: about 0.8 ns at 60 keV, rising to 8.8 ns at 300 keV. For time-resolved EELS, work at 60 to 100 kV and expect nanosecond electron timing. The picosecond precision of the PicoTDC applies to the external references.
Single-particle cryo-EM delivers near-atomic structures, but the density maps do not say which element occupies a site.
Reconstructed EELS (REEL), developed at the Max Planck Institute of Biophysics, records low-dose STEM-EELS data cubes from many vitrified particles, determines particle poses from reference images, and reconstructs energy-loss images in 3D with those poses. The published method localised carbon, nitrogen and oxygen in macromolecular reconstructions at about 40 Å for the carbon edge.
A follow-up analysis identifies pose accuracy as a limit and proposes an integrated centre-of-mass (iCOM) phase-contrast reference computed from the momentum information in the EELS pattern itself. That reference reached 10 Å from 3,400 particles, against 27 Å from 16,648 particles for the earlier amplitude-contrast reference. The Ultra's 512 momentum-preserving rows, 1024 channels covering the C, N and O K edges in one exposure, and overhead-free readout match what this workflow requires.
REEL results with a CheeTah detector are, at the time of writing, a stated hypothesis of the MPI of Biophysics group and its partners, not a published result. ASI is a co-author on the poster and a partner in the work. Published REEL method: Pfeil-Gardiner et al., Nat. Methods 21, 2299 (2024).
Mechanical integration at the spectrometer camera position is defined per spectrometer model in the project phase. Contact us for a formal quotation, a compatibility assessment for your microscope, or a demonstration measurement on your samples.
Prices exclude customs duties and taxes. Specifications are typical values and subject to change.
Published EELS work carried out with CheeTah Timepix3 detectors.
Nanometre-resolution maps of calcium, phosphorus and organic carbon in cryofixed mussel tissue, monochromated at 100 kV. The authors attribute the higher signal-to-noise ratio to monochromated STEM-EELS combined with the Timepix3 detector.
Ruiz-Villaespesa et al., Commun. Biol. 9, 814 (2026)Electron–photon coincidences reveal which energy-loss pathways produce light, with nanometre spatial resolution, on hBN defects and Au/SiO2 nanospheres.
Varkentina et al., Sci. Adv. 8, eabq4947 (2022)Laser-synchronised electron energy-gain spectroscopy on a continuous-gun microscope, resolving optical resonances about 100 µeV wide.
Auad et al., Nat. Commun. 14, 4442 (2023)Laser-heated 2D semiconductors, membranes and metal films followed in temperature at nanometre and nanosecond resolution, with the full temporal evolution acquired in parallel.
Castioni et al., Nano Lett. 25, 1601–1608 (2025)Electron–photon delays longer than the setup response yield excitation lifetimes, measured at 20 to 40 ns for NV0 centres in nanodiamond.
Varkentina et al., Appl. Phys. Lett. 123 (2023)Hyperspectral EELS acquired at the rate of a single-channel dark-field detector, with the scan unit as leader and the detector as follower.
Auad et al., Ultramicroscopy 239, 113539 (2022)Full method descriptions, figures and 26 references are in the white paper Event-Based EELS with the CheeTah T3 Ultra.
Read the full technical white paper, or talk to us about integrating CheeTah T3 Ultra with your spectrometer and microscope, a demonstration measurement on your samples, or a quotation.