AstroPix: HV-CMOS Sensors for Gamma-Ray Detection

• AstroPix is a family of high-voltage CMOS sensors that integrate charge-collecting diodes and front-end electronics for precise gamma-ray and particle imaging.
• It is designed for dual applications in medium-energy gamma-ray astronomy (e.g., AMEGO-X) and collider calorimetry (e.g., ePIC), offering detailed spectroscopic and spatial resolution.
• Prototype generations demonstrate key performance metrics such as energy resolution, dynamic range, and low power consumption, driving further detector optimization.

AstroPix is a family of high-voltage CMOS monolithic active-pixel sensors developed for medium-energy $\gamma$-ray instrumentation and later repurposed for collider calorimetry. In the space-instrument context it is proposed for silicon tracker layers in observatories such as the All-sky Medium Energy Gamma-ray Observatory eXplorer (AMEGO-X), where the detector must combine two-dimensional hit localization, spectroscopic response in the tens-to-hundreds of keV regime, and stringent power limits. In collider R&D, AstroPix has been adopted as the fine-granularity imaging layer technology for the Barrel Imaging Calorimeter (BIC) in the ePIC detector, where the same monolithic architecture is used for three-dimensional shower imaging (Suda et al., 2024, Bok, 14 May 2026).

1. Scientific setting and detector role

AstroPix was conceived against the long-standing instrumental difficulty of the MeV band, described as the frontier of $\gamma$-ray astronomy for nuclear lines, continuum emission from jets and transients, and Compton-domain polarization. Missions such as AMEGO-X aim to survey $100~\mathrm{keV}$–$1~\mathrm{GeV}$, and the tracker in such a Compton/pair telescope must detect low-energy recoil electrons, measure their energy with $O(10\%)$ resolution at $\sim 60~\mathrm{keV}$, provide $O(100~\mu\mathrm{m})$ spatial resolution, and consume less than $1~\mathrm{mW/cm^2}$ or, in the AMEGO-X requirement as formulated in a later performance study, less than $1.5~\mathrm{mW/cm^2}$ (Apadula et al., 2022, Suda et al., 2024).

The core detector requirement is not solely positional. Compton reconstruction depends directly on energy measurement, and AstroPix papers repeatedly define energy resolution as

$\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$

or equivalently $\gamma$0. This requirement is tied to tracker-level event reconstruction rather than isolated spectroscopy. A related geometric constraint is the binary spatial resolution,

$\gamma$1

so that a $\gamma$2 pitch corresponds to $\gamma$3 (Suda et al., 2024, Apadula et al., 2022).

A second scientific trajectory emerged when AstroPix was incorporated into the BIC for the EIC. There the sensor is no longer a Compton tracker plane but a silicon imaging layer interleaved with Pb/SciFi sampling layers. In that role the emphasis shifts from recoil-electron spectroscopy to granular shower topology, electron/pion separation, and synchronization with calorimeter channels in a mixed readout environment (Bok, 14 May 2026, Hong et al., 8 May 2026).

2. Monolithic HV-CMOS concept and pixel architecture

AstroPix is a monolithic HV-CMOS MAPS architecture in which the charge-collecting diode and the front-end electronics are co-fabricated in the same silicon die. The basic pixel concept places each pixel in a deep n-well that also serves as the charge-collecting electrode, while a p-type substrate is reverse-biased to form a depleted region. Monolithic integration eliminates bump-bonding; front-end analog circuitry sits beneath or within the collecting structure, and the signal chain typically comprises a charge-sensitive amplifier, a shaping stage or low-pass network, and a discriminator (Suda et al., 2024, Apadula et al., 2022).

For depletion, AstroPix papers use the one-sided abrupt-junction form

$\gamma$4

and related variants. The design target in the AMEGO-X program is full depletion of a $\gamma$5-thick sensor so that charge carriers generated across the entire volume drift under the electric field, minimizing diffusion and recombination losses and supporting a per-pixel dynamic range of $\gamma$6–$\gamma$7 (Steinhebel et al., 20 Jan 2025, Steinhebel et al., 2024).

The v3 generation established the canonical large-pixel architecture used across most subsequent reports: a $\gamma$8 matrix with $\gamma$9 pitch on a $100~\mathrm{keV}$0 die. One detailed module-performance report specifies fabrication in a $100~\mathrm{keV}$1 HV-CMOS process on a $100~\mathrm{keV}$2-thick, $100~\mathrm{keV}$3–$100~\mathrm{keV}$4 p-type substrate, with each pixel containing a charge-sensitive preamplifier followed by a discriminator. Above-threshold hits trigger an 8-bit ToA counter clocked at $100~\mathrm{keV}$5 and a 12-bit ToT counter clocked up to $100~\mathrm{keV}$6; hit packets contain chip ID, row/column address, ToA, and ToT, and the rows and columns are OR-summed to form a sparse-readout bus (Kim et al., 7 Nov 2025).

Beam-test descriptions of the same v3 generation present a closely related but not identical implementation: charge amplifier, CR-RC shaper, comparator, on-chip digitization, and a readout in which row and column hits are streamed continuously and reassembled offline by matching ToA and ToT consistency. This row/column factorization reduced routing complexity but introduced a known ambiguity that later versions aimed to remove (Kim et al., 4 Feb 2026, Suda et al., 29 Jul 2025).

3. Development lineage and prototype generations

AstroPix emerged from ATLASPix, and the earliest AstroPix literature is explicit that the decisive lesson from the ATLASPix heritage was the contrast between excellent intrinsic analog spectroscopy and inadequate digital spectroscopy. ATLASPix analog measurements yielded $100~\mathrm{keV}$7 at $100~\mathrm{keV}$8 and $100~\mathrm{keV}$9 at $1~\mathrm{GeV}$0, while the 6-bit digital ToT path produced energy resolutions exceeding $1~\mathrm{GeV}$1. This established the central design problem for AstroPix: the analog front end was already adequate for $1~\mathrm{GeV}$2-ray spectroscopy, but the digitization architecture had to be redesigned for photon-sensitive operation (Brewer et al., 2021, Brewer et al., 2021).

The first dedicated AstroPix generation, AstroPix V1, adopted larger pixels and a simplified architecture intended to reduce power and test spectroscopic digitization. Reported parameters include an $1~\mathrm{GeV}$3 array with $1~\mathrm{GeV}$4 pixels, 36 analog outputs, 36 digital comparator outputs, and a 12-bit time stamp plus 10-bit TOT per hit; power was reported as $1~\mathrm{GeV}$5 and was dominated by the front-end amplifier (Brewer et al., 2021, Brewer et al., 2021).

AstroPix2 and AstroPix3 introduced the larger-pitch, lower-channel-density direction associated with AMEGO-X optimization. In the 2024 performance evaluation, AstroPix2 is described as having a $1~\mathrm{GeV}$6 pitch and AstroPix3 as reaching the desired $1~\mathrm{GeV}$7 pitch for AMEGO-X. The same study reports AstroPix2 and AstroPix3 power consumptions of $1~\mathrm{GeV}$8 and $1~\mathrm{GeV}$9, respectively (Suda et al., 2024). A 2025 status paper on v3 gives a more detailed device-level picture: $O(10\%)$0 pixels, up to $O(10\%)$1 frontside bias, deep n-well implants of $O(10\%)$2, and analog and digital power contributions of $O(10\%)$3 and $O(10\%)$4 (Steinhebel et al., 20 Jan 2025).

AstroPix4 was introduced as a dedicated response-improvement iteration. It is a $O(10\%)$5 array of $O(10\%)$6-pitch pixels on a $O(10\%)$7-thick, medium-resistivity substrate, with reduced input capacitance, a $O(10\%)$8 clock, and a 4-bit flash TDC in the periphery. This version was explicitly framed as an intermediate step: it lowered the threshold sufficiently to observe the $O(10\%)$9 photopeak from $\sim 60~\mathrm{keV}$0 and reduced power to about $\sim 60~\mathrm{keV}$1, approximately half that of the previous AstroPix version, but it did not yet achieve the full AMEGO-X depletion and high-energy-range targets (Suda et al., 29 Jul 2025).

4. Reported performance metrics

AstroPix performance has been reported through single-pixel spectroscopy, pixel-by-pixel calibration, radioactive-source measurements, proton beam tests, and irradiation campaigns. The most useful interpretation is prototype-specific, since depletion depth, substrate resistivity, bias voltage, readout mode, and the very definition of “dynamic range” differ across reports.

Prototype and paper	Reported spectroscopic result	Other reported metric
AstroPix2 (Suda et al., 2024)	Dynamic range from $\sim 60~\mathrm{keV}$2 to $\sim 60~\mathrm{keV}$3; energy resolution meeting the AMEGO-X target value at $\sim 60~\mathrm{keV}$4	Chip fully operational after $\sim 60~\mathrm{keV}$5; gain decrease by approximately $\sim 60~\mathrm{keV}$6
AstroPix3 (Suda et al., 2024)	Mean energy resolution of $\sim 60~\mathrm{keV}$7 (FWHM) at $\sim 60~\mathrm{keV}$8; $\sim 60~\mathrm{keV}$9 of pixels satisfying the target value	Dynamic range from $O(100~\mu\mathrm{m})$0 to $O(100~\mu\mathrm{m})$1
AstroPix v3 (Steinhebel et al., 20 Jan 2025)	Energy resolution $O(100~\mu\mathrm{m})$2 FWHM at $O(100~\mu\mathrm{m})$3	Depletion $O(100~\mu\mathrm{m})$4 at $O(100~\mu\mathrm{m})$5
AstroPix4 (Suda et al., 29 Jul 2025)	$O(100~\mu\mathrm{m})$6 at $O(100~\mu\mathrm{m})$7 and $O(100~\mu\mathrm{m})$8 at $O(100~\mu\mathrm{m})$9	Dynamic range from $1~\mathrm{mW/cm^2}$0 to $1~\mathrm{mW/cm^2}$1; depletion depth $1~\mathrm{mW/cm^2}$2 at $1~\mathrm{mW/cm^2}$3

Beyond source-based spectroscopy, the v3 proton beam study at Fermilab extracted an effective depletion depth from the most probable MIP energy deposition. Using a mean MPV of $1~\mathrm{mW/cm^2}$4 for $1~\mathrm{mW/cm^2}$5 protons and $1~\mathrm{mW/cm^2}$6 in silicon, the study obtained

$1~\mathrm{mW/cm^2}$7

which was interpreted as confirmation that AstroPix v3 operates in a partially depleted regime at $1~\mathrm{mW/cm^2}$8 (Kim et al., 4 Feb 2026).

Beam tests relevant to calorimetry emphasize efficiency and imaging rather than spectroscopy. In 2025 tests at KEK PF-AR and CERN PS T10, AstroPix-v3 reached a maximum hit efficiency of $1~\mathrm{mW/cm^2}$9 at a bias voltage of $1.5~\mathrm{mW/cm^2}$0 under pion-dominated beam conditions, exhibited residuals of $1.5~\mathrm{mW/cm^2}$1–$1.5~\mathrm{mW/cm^2}$2 pixels in both $1.5~\mathrm{mW/cm^2}$3 and $1.5~\mathrm{mW/cm^2}$4 after alignment, and showed clear electron/pion discrimination through hit multiplicity and radial shower size (Hong et al., 8 May 2026).

The power narrative is similarly version-dependent. Space-oriented requirements generally state a ceiling of $1.5~\mathrm{mW/cm^2}$5 for AMEGO-X, while reported prototypes range from $1.5~\mathrm{mW/cm^2}$6–$1.5~\mathrm{mW/cm^2}$7 for v3 module implementations to about $1.5~\mathrm{mW/cm^2}$8 for AstroPix4. A plausible implication is that AstroPix development has been driven by a simultaneous optimization of depletion, noise, timing, and power, rather than by a single monotonic figure of merit (Suda et al., 2024, Kim et al., 7 Nov 2025).

5. Instrument integration and system-scale use

AstroPix has progressed from single-chip characterization to multi-chip detector assemblies. A module-performance study enumerates four v3 configurations: a single chip, a quad-chip array, a three-layer stack of quad-chips for the AstroPix Sounding Rocket Technology dEmonstration Payload (A-STEP), and a nine-chip linear module intended as the fundamental unit of the BIC imaging layer. In quad-chip tests, more than $1.5~\mathrm{mW/cm^2}$9 active-pixel yield was achieved in all four chips, while nine-chip modules reached at least $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$0 yield in eight of nine chips (Kim et al., 7 Nov 2025).

A-STEP is the first explicit space-environment technology demonstration for AstroPix. It places three layers of AstroPix quad chips in a sounding-rocket payload, with each layer holding one $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$1 quad-chip. The payload design specifies nominal $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$2 operation, on-board threshold and calibration loading, Ethernet-via-telemetry data streaming at $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$3, and a total payload power of $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$4. Simulations of the three-layer geometry predicted a total isotropic trigger rate of about $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$5 and a three-layer coincidence rate of about $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$6 (Violette et al., 2024).

ComPair-2 scales the same technology to a balloon-borne telescope prototype. Its tracker consists of ten segments, each with 95 AstroPix quad-chips, and the readout chain uses FPGA-based front-end electronics. In this setting AstroPix is described as operating with an effective spatial sampling of $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$7, ToT digitization with $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$8 steps, and a global time stamp of $\mathrm{FWHM}(E)=2.355\cdot \sigma(E),$9 resolution (Caputo et al., 2024).

In the BIC for ePIC, AstroPix is integrated with Pb/SciFi sampling layers rather than acting as a stand-alone tracker. The baseline design uses four AstroPix imaging layers interleaved with five Pb/SciFi sampling layers, followed by a bulk Pb/SciFi section, for a total depth of approximately $\gamma$00. A status report states that the full barrel is planned to tile roughly $\gamma$01 AstroPix chips over about $\gamma$02, and that beam tests at KEK PF-AR demonstrated preliminary correlations between beam-impact positions and fired AstroPix pixels, confirming spatial imaging capability (Bok, 14 May 2026).

6. Open technical issues, interpretive cautions, and future directions

Two recurrent interpretive issues accompany AstroPix. The first is the distinction between target specifications and demonstrated performance. Multiple papers describe the AMEGO-X target as full depletion of a $\gamma$03-thick sensor, dynamic range of $\gamma$04–$\gamma$05, and energy resolution better than $\gamma$06 at $\gamma$07, yet several measured devices remain partially depleted: v3 values include $\gamma$08 in A-STEP descriptions, $\gamma$09 at $\gamma$10 on a low-resistivity test substrate, $\gamma$11 at $\gamma$12 in beam-test summaries, and $\gamma$13 effective depletion from MIP data (Violette et al., 2024, Steinhebel et al., 20 Jan 2025, Hong et al., 8 May 2026, Kim et al., 4 Feb 2026). This suggests that reported v3 achievements should be read as intermediate milestones rather than as evidence that the full depletion target has already been universally achieved.

The second issue is the relationship between analog front-end quality and digitized spectroscopic quality. Early ATLASPix-derived studies showed that the intrinsic analog chain already exceeded the baseline $\gamma$14 at $\gamma$15 requirement, while the digital ToT implementation was unusable for spectroscopy because 6-bit quantization drove the resolution above $\gamma$16 (Brewer et al., 2021). Later AstroPix generations therefore increased ToT granularity, added local counters, considered per-pixel ADC/SCA schemes, and planned per-pixel threshold trimming. AstroPix4 also introduced per-pixel comparator thresholds, hit buffers, and time-stamp improvements intended to eliminate the row/column hit ambiguities seen in earlier versions (Apadula et al., 2022, Suda et al., 29 Jul 2025).

The current R&D path is correspondingly explicit. Reported future work includes higher-resistivity substrates, improved guard-ring layouts, dynamic-feedback-capacitance CSA designs to avoid saturation above $\gamma$17, further input-capacitance reduction, power reduction toward or below $\gamma$18, proton/electron environmental validation up to $\gamma$19–$\gamma$20, and scaling to tiled multi-module systems for both AMEGO-X-class telescopes and ePIC calorimetry (Steinhebel et al., 2024, Steinhebel et al., 20 Jan 2025, Suda et al., 29 Jul 2025). In that sense AstroPix is best understood not as a single detector instance but as an evolving HV-CMOS platform whose space-astrophysics and collider-calorimetry branches continue to inform one another.