Low-Gain Avalanche Detectors (LGADs) Overview

• LGADs are thin silicon detectors with a specialized p-type gain layer that facilitates controlled impact ionization, achieving gain values typically between 10 and 50.
• They deliver fast charge collection and timing resolutions below 30 ps in optimized conditions, making them ideal for high-energy collider timing and 4D tracking applications.
• Key challenges include ensuring uniform multiplication, addressing segmentation-induced fill-factor issues, and enhancing radiation hardness through innovations like carbon co-implantation.

Low-Gain Avalanche Detectors (LGADs) are thin silicon detectors in which a deliberately engineered gain layer provides moderate internal charge multiplication through impact ionization. In the silicon implementations discussed across the literature, LGADs are described as n-on-p sensors or, equivalently, as $n^{++}-p^{+}-p-p^{++}$ structures with an extra p-type multiplication layer beneath the collection electrode; typical gain values are in the range of about 10 to 50, sufficient to improve signal-to-noise ratio without operating as high-gain APDs (Grieco et al., 2022, Lange et al., 2017, Kramberger et al., 2017). Their combination of thin active thickness, fast charge collection, and controlled avalanche gain has made them the selected or baseline technology for precision timing layers such as ATLAS HGTD and CMS ETL/MTD, and a central candidate for 4D tracking (Grieco et al., 2022, Currás et al., 2019).

1. Device structure and avalanche mechanism

The canonical silicon LGAD architecture combines a thin depleted bulk with a localized high-field multiplication region. In the formulations reported by CNM, HPK, FBK, and related groups, the essential stack is $n^{++}-p^{+}-p-p^{++}$, where the shallow $p^{+}$ gain layer beneath the $n^{++}$ electrode raises the junction field enough for impact ionization when the device is reverse biased (Kramberger et al., 2017, Wu et al., 2022). In strip-LGAD descriptions, the epitaxial region is reported to sustain a relatively low field of about $30\ \text{kV/cm}$, while the gain layer reaches fields up to about $400\ \text{kV/cm}$, and the same work notes that charged particles in silicon produce roughly 70 electron-hole pairs per micron (Sun et al., 2024). Proton-irradiation studies of highly irradiated LGADs likewise describe the gain layer as the region in which the electric field can exceed $4\times10^5\ \mathrm{V/cm}$, enabling the controlled avalanche process (Sorenson et al., 2023).

A persistent design constraint is that multiplication must be confined to the intended planar region rather than to the device edge. Peripheral engineering is therefore integral rather than ancillary. An early IMB-CNM study formulates the relevant condition as

$V_{BD}|_{edge} > V_{BD}|_{planar},$

and compares floating guard rings, $n^{+}$ extensions, and deep N-type diffusion used as a JTE-like structure (Fernandez-Martinez et al., 2015). In that study, deep N-type diffusion is the preferred termination because it keeps the edge field below the planar junction field and supports breakdown beyond 1000 V; the same work also identifies P-Stop as the more robust anti-inversion strategy, while P-Spray is more compact but harder to control (Fernandez-Martinez et al., 2015). This peripheral design problem remains structurally linked to later discussions of inter-pad dead regions and fill-factor.

2. Signal formation and timing performance

The principal timing advantage of LGADs follows from the coexistence of short drift paths and internally amplified signals. A compact timing budget used in 50 $\mu$m Micron devices is

$n^{++}-p^{+}-p-p^{++}$0

with the same study emphasizing that time walk and jitter improve as the slew rate increases (Moriya et al., 2023). A related 45 $n^{++}-p^{+}-p-p^{++}$1m CNM analysis approximates electronics jitter as

$n^{++}-p^{+}-p-p^{++}$2

linking timing directly to rise time and signal amplitude (Lange et al., 2017).

Measured performance spans several implementation regimes. For 45 $n^{++}-p^{+}-p-p^{++}$3m CNM LGADs, time resolutions below 30 ps were obtained at the highest applied voltages before irradiation, with about 29 ps at 235 V for the medium-dose device and about 28 ps at 320 V for the low-dose device (Lange et al., 2017). In 50 $n^{++}-p^{+}-p-p^{++}$4m Micron Semiconductor Ltd. devices, the best reported timing resolution is 26.5 ps at 200 V and $n^{++}-p^{+}-p-p^{++}$5C, whereas the CNM references tested in the same study remained in the mid-60 ps regime under the stated conditions (Moriya et al., 2023). For the first iLGAD prototypes, the best region of the pulse reached about 20 ps and a gain of about 4.8 at 700 V, showing that the inverse architecture can retain competitive timing while modifying the segmentation concept (Currás et al., 2019).

Other readout modalities confirm that the timing advantage is not restricted to single-pad CFD systems. In the HADES $n^{++}-p^{+}-p-p^{++}$6 upgrade program, FBK strip sensors operated at room temperature with NINO leading-edge discriminators and ToT-based walk correction achieved roughly 47 ps single-channel precision in 1.92 GeV proton-beam tests at COSY (Pietraszko et al., 2020). For the IHEP strip-LGADs with double-end readout, the best timing resolutions were 47.5 ps, 41.4 ps, and 37.5 ps for 1.0 mm, 0.5 mm, and 0.3 mm strip widths, respectively, with the narrower devices benefiting from lower capacitance, improved rise time, and better SNR (Sun et al., 2024). Collectively, these results indicate that sub-50 ps operation is routine in well-optimized geometries, while sub-30 ps remains achievable in thin, low-capacitance devices under favorable bias and temperature conditions.

3. Segmentation, fill-factor, and readout geometries

A central limitation of conventional segmented LGADs is the fill-factor problem. In the conventional architecture summarized in the iLGAD work, the multiplication layer is locally implemented beneath the active collecting electrodes, so a minimum-ionizing particle crossing the inter-pad or inter-strip region does not receive the same amplification. The result is a non-uniform response, explicitly described as severe degradation of timing and position resolution in those regions (Currás et al., 2019). The same paper reports the characteristic test-beam signature of this effect: a conventional LGAD strip detector shows two charge peaks, around 24 ke$n^{++}-p^{+}-p-p^{++}$7 for the interstrip region and around 77 ke$n^{++}-p^{+}-p-p^{++}$8 for amplified hits, whereas the iLGAD produces a single peak around 75 ke$n^{++}-p^{+}-p-p^{++}$9, consistent with uniform multiplication and a 100% fill-factor by design (Currás et al., 2019).

The term “dead region” requires technical precision. In irradiated FBK UFSD-3.2 arrays, the inter-pad dead region is associated with the JTE inserted to prevent breakdown at pad edges. The same study states that the JTE region collects charge but does not send it through the gain layer, so the issue is loss of multiplication rather than necessarily complete loss of charge transport (Darby et al., 2021). For Wafer 19 Type 10 arrays with 45 $p^{+}$0m bulk thickness and nominal IPD 49 $p^{+}$1m, the measured 50–50 IPDR before irradiation was about 80 $p^{+}$2m at typical biases; after neutron irradiation to $p^{+}$3 and $p^{+}$4, the effective dead width moved closer to the nominal value, with the higher fluence giving the best agreement (Darby et al., 2021). This is a reminder that the effective inactive width is field-dependent and not purely a mask-level geometric constant.

Several architectural responses have emerged. The abstract of a TCAD study on AC-LGADs describes a geometry with an un-segmented p-type gain layer and n-type N-layer, plus a dielectric layer separating the metal readout pads; that configuration is proposed to exploit intrinsic charge sharing and to provide spatial resolution on the 10’s of $p^{+}$5m scale, in contrast to conventional LGAD granularity limited to the mm scale by JTE-induced inactive space (Nizam et al., 2023). Strip LGADs pursue a different compromise: IHEP fabricated 19 mm long strips with widths of 1.0 mm, 0.5 mm, and 0.3 mm, achieving position resolution parallel to the strip direction better than 1 mm through double-end timing (Sun et al., 2024). These approaches address the same underlying constraint—how to preserve multiplication while reducing the segmentation penalty—but they do so through distinct electrostatic and readout strategies.

4. Radiation response and hardening strategies

The dominant radiation-induced failure mode in silicon LGADs is effective acceptor removal in the gain layer. This is formulated explicitly in the carbon-implantation study as

$p^{+}$6

and in gamma-irradiation and proton-irradiation studies through the equivalent depletion-voltage relation

$p^{+}$7

In all cases, the physical interpretation is the same: radiation reduces the active dopant concentration in the multiplication layer, lowers the peak field, and therefore reduces gain (Wu et al., 2022, Hoeferkamp et al., 2021, Sorenson et al., 2023). A thin-LGAD irradiation comparison further reports removal constants of order $p^{+}$8 for two producer families and concludes that pions are more damaging than neutrons at the same equivalent fluence (Kramberger et al., 2017).

Performance degradation is substantial but not monotonic with fluence because higher reverse bias can partly compensate for gain-layer loss. In 45 $p^{+}$9m CNM LGADs irradiated with neutrons, similar sub-30 ps timing to the pre-irradiation case was retained after $n^{++}$0 by increasing the bias to about 430 V, whereas at $n^{++}$1 the best beam-test timing at 620 V was 57 ps because voltage stability was insufficient to recover the lost gain (Lange et al., 2017). A broader thin-sensor comparison up to $n^{++}$2 concludes that thin LGADs outperform standard $n^{++}$3m detectors and no-gain detectors up to about $n^{++}$4, and that larger initial gain prolongs useful operation (Kramberger et al., 2017).

Process-level hardening has focused particularly on carbon. In the IHEP-IME program, shallow carbon implantation combined with long-time annealing produced a best sample, W7-II, with an acceptor-removal constant of $n^{++}$5, about 3–5 times lower than typical boron-only devices in the comparison cited there (Wu et al., 2022). After $n^{++}$6, that 50 $n^{++}$7m device still delivered charge collection $n^{++}$8 fC and time resolution $n^{++}$9 ps at voltages below 400 V; 4 fC was achieved at 350 V and about 50 ps at 400 V (Wu et al., 2022). Proton-irradiation studies beyond $30\ \text{kV/cm}$0 are consistent with the same trend: FBK4 devices with carbon co-implantation show acceptor-removal constants of about $30\ \text{kV/cm}$1–$30\ \text{kV/cm}$2, compared with about $30\ \text{kV/cm}$3–$30\ \text{kV/cm}$4 for HPK2, although both families approach the edge of usable charge collection above $30\ \text{kV/cm}$5 under the tested biases (Sorenson et al., 2023).

Ionizing dose adds a separate surface-damage channel. In Co-60 gamma studies motivated by HL-LHC doses up to 2 MGy, the principal effects include increased surface leakage current, reduced punch-through voltage, degradation of inter-pad isolation, and a dose-dependent decrease of $30\ \text{kV/cm}$6, interpreted as a combination of oxide/interface damage and acceptor removal induced by secondary Compton electrons or photoelectrons (Hoeferkamp et al., 2021). Radiation hardness in LGADs is therefore not a single-parameter problem: gain-layer chemistry, surface isolation, edge design, and stable high-voltage operation all enter simultaneously.

5. Gain suppression, nonlinear response, and dynamic range

A recurrent misconception is that LGAD gain is determined only by reverse bias and temperature. Laboratory measurements with IR laser excitation and $30\ \text{kV/cm}$7Sr $30\ \text{kV/cm}$8 particles show that the measured gain also depends on the charge density projected into the gain layer (Currás et al., 2021). In that study, higher reverse bias increases gain and higher temperature decreases it, but denser ionization produces an additional suppression mechanism: a dense carrier cloud locally screens the electric field in the multiplication layer, reducing the impact-ionization probability and therefore the measured gain (Currás et al., 2021). The evidence is multi-pronged. With the IR laser, increasing the excitation from about 0.5 MIPs to 30 MIPs at fixed focus lowered the measured gain, and moving from defocused to focused illumination strengthened the suppression; with $30\ \text{kV/cm}$9Sr, tilting the detector to $400\ \text{kV/cm}$0 and $400\ \text{kV/cm}$1 reduced the projected density in the gain layer and improved both collected charge and timing relative to normal incidence (Currás et al., 2021).

The same work shows that test conditions matter quantitatively. For the tested 50 $400\ \text{kV/cm}$2m sensors, source-based time resolution was about 40–50% worse at low SNR and up to about 80% worse at higher SNR than laser measurements, reflecting both Landau fluctuations and density-dependent gain loss (Currás et al., 2021). This makes direct comparison between optical TCT data and particle data non-trivial unless the deposited-charge density is controlled or at least reported.

Recent proton-beam studies extend the same issue into the MeV-deposition regime relevant for stopped particles. Abstract-level reports from the CENPA Tandem accelerator describe measurements of LGAD gain as a function of bias voltage, incidence beam angle, and proton energy for MeV-range deposits, in the context of the PIONEER active target and its requirement to separate close-by hits over a wide dynamic range (Braun et al., 2024, Yang et al., 4 Feb 2025). The later abstract specifies FBK prototypes of 50 $400\ \text{kV/cm}$3m, 100 $400\ \text{kV/cm}$4m, and 150 $400\ \text{kV/cm}$5m active thicknesses and frames energy linearity as important for particle identification in ATAR (Yang et al., 4 Feb 2025). Independent 4H-SiC LGAD characterization with $400\ \text{kV/cm}$6 particles reports gain only about 2 to 3 and explicitly suggests that space-charge or saturation effects from the dense charge cloud may suppress avalanche gain, consistent with the silicon observations that multiplication is event-density dependent rather than purely device-static (Yang et al., 2024).

6. Applications, system integration, and material extensions

The dominant application domain of LGADs remains collider timing. ATLAS HGTD and CMS ETL adopted LGAD technology to mitigate pile-up at the HL-LHC, while earlier formulations described LGADs as the baseline sensing technology for MIP timing detectors in ATLAS and CMS (Grieco et al., 2022, Currás et al., 2019). Beyond the LHC, the HADES upgrade program used FBK strip LGADs to build a prototype reaction-time detector for proton and pion beams, achieving time precision below 50 ps at room temperature with leading-edge discrimination and ToT correction, while also targeting active areas much larger than the $400\ \text{kV/cm}$7 scale typical of the diamond sensors previously used there (Pietraszko et al., 2020). The strip-LGAD program for future colliders, especially CEPC, addresses a different systems-level constraint—channel density—by trading pixel granularity for 19 mm strips with double-end readout and still retaining about 37.5 ps timing and sub-millimetric longitudinal positioning (Sun et al., 2024).

A second application axis is beam monitoring in therapy. In FLASH-oriented radiotherapy measurements, two LGAD systems and silicon-diode references showed linear dose response for electron beams up to about 450 Gy/s and for proton beams up to about 12 Gy/s; above those ranges the response continued to increase with reduced slope, and no true signal plateau was observed up to 1800 Gy/s for electrons and 150 Gy/s for protons (Cadena et al., 4 Jul 2026). In 60 MeV proton therapy beam diagnostics, FBK strip LGADs resolved cyclotron microstructure and provided accurate flux measurements up to an instantaneous dose rate of about 7.5 Gy/s, with the main high-rate limitation identified as multiple protons crossing the detector within the $400\ \text{kV/cm}$8 ns signal window (Bellora et al., 17 Aug 2025). These studies show that LGADs are not confined to single-particle timing; they can also function as fast beam monitors on the microsecond scale.

A third direction is material diversification. The first operational 4H-SiC LGADs were reported with a circular mesa design, SiO$400\ \text{kV/cm}$9 passivation, a 75 $4\times10^5\ \mathrm{V/cm}$0m drift layer, and a gain-layer field of about $4\times10^5\ \mathrm{V/cm}$1 at 600 V in TCAD; experimentally, a 75 $4\times10^5\ \mathrm{V/cm}$2m diameter device reached a breakdown voltage of 1160 V and showed low-gain multiplication relative to a matched PiN diode, albeit with gain only about 2 to 3 under $4\times10^5\ \mathrm{V/cm}$3Po $4\times10^5\ \mathrm{V/cm}$4-particle excitation (Yang et al., 2024). Proton-irradiation studies of 4H-SiC LGADs up to $4\times10^5\ \mathrm{V/cm}$5 then showed progressive loss of rectification and gain, compensation of the gain layer, and transport degradation from radiation-induced defects; at the highest fluence no observable gain remained, but a measurable signal was still present, supporting continued interest in SiC LGADs for extreme environments (Satapathy et al., 30 Jul 2025). This suggests that the LGAD concept is no longer restricted to one semiconductor platform, even though silicon remains the most mature implementation.

Across these developments, the central research problems are stable: preserving a strong but localized multiplication field, minimizing segmentation-induced response loss, controlling gain suppression for dense ionization, and maintaining usable timing after irradiation. The literature shows that each of these problems is now being treated not as an isolated detector-physics issue but as a coupled optimization across electrostatics, fabrication process, readout architecture, and application-specific dynamic range.