Inverse Low-Gain Avalanche Detector (iLGAD)

• iLGAD is a silicon detector featuring a continuous multiplication layer with readout segmentation shifted to the ohmic side for 100% fill-factor.
• The design addresses gain non-uniformity in conventional LGADs by ensuring particles experience consistent avalanche multiplication across the sensitive area.
• Prototype studies demonstrate robust timing performance, improved X-ray sensitivity, and enhanced radiation tolerance through optimized periphery engineering.

Inverse Low-Gain Avalanche Detector (iLGAD) denotes an LGAD-family silicon detector architecture in which the avalanche-multiplication region is kept continuous over the sensitive area while the readout segmentation is transferred away from the multiplication side. In the canonical formulation, iLGAD is a p-in-p LGAD with segmented $p^+$ electrodes on the ohmic side and a non-segmented deep p-well on the multiplication side, so that gain is intended to remain laterally uniform rather than collapsing in the inter-pad or inter-strip regions characteristic of conventional segmented LGADs (Currás et al., 2019). The concept emerged from the broader LGAD program on moderate-gain avalanche sensors for timing and tracking, and it has since developed into a family of devices spanning strip trackers, hybrid pixels, and soft-X-ray sensors with optimized entrance windows (Pellegrini et al., 2015).

1. Historical emergence and conceptual basis

The iLGAD concept was introduced as a response to a specific weakness of segmented LGADs. In a conventional LGAD, avalanche multiplication is produced by an optimized $p^+$ layer under the shallow $n^+$ diffusion, i.e. the basic APD-derived multiplication structure $\mathrm{n^{++}\!-\!p^+\!-\!p}$, or, in an explicit sensor stack, $n^{++}-p^+-p-n^{++}$ [(Pellegrini et al., 2015); (Cartiglia et al., 2013)]. The added $p^+$ implant creates a localized high-field region at a depth of about $1\text{–}5~\mu\text{m}$, with target gain typically in the range $10\text{–}100$, far below Geiger-mode devices such as SiPMs (Cartiglia et al., 2013). In that conventional geometry, electron multiplication dominates: at $270~kV/cm$, the ionization coefficients are approximately $\alpha_e \sim 0.7$ pair/$p^+$0m and $p^+$1 pair/$p^+$2m, so hole multiplication is comparatively weak (Cartiglia et al., 2013).

The timing motivation for the whole LGAD family preceded iLGAD-specific implementations. Early UFSD/LGAD modeling showed that internal gain could drive projected time resolution below $p^+$3 for suitable thickness and capacitance conditions, establishing why moderate avalanche gain mattered for 4D tracking (Cartiglia et al., 2013). The 2015 introduction of iLGAD then reformulated the architectural question: instead of segmenting the multiplication junction itself, the multiplication side could remain continuous and the segmentation could be moved to the backside ohmic contact, thereby addressing the spatial non-uniformity observed in strip LGADs (Pellegrini et al., 2015).

That shift became experimentally concrete in the first prototype studies from IMB-CNM. The 2019 strip-iLGAD work defined the device explicitly as a p-in-p LGAD in which the multiplication layer is continuous and the segmented electrodes lie on the opposite side, so that the device should ideally provide a constant gain value over the entire sensitive region, or “100% fill-factor by design” in the gain-response sense (Currás et al., 2019). The same design logic later motivated pixel-scale hybrids for Timepix readout and optimized iLGAD peripheries for X-ray use (Svihra et al., 2024, Doblas et al., 2022).

2. Device architecture and avalanche signal formation

In the standard iLGAD description, the multiplication layer remains on the junction side, but it is no longer patterned strip-by-strip or pad-by-pad. Instead, the segmented electrodes are implemented at the $p^+$4 ohmic side, while the multiplication side contains a continuous gain layer (Currás et al., 2019). The 2015 IMB-CNM proposal described this as a “new approach to the charge collection” in which segmentation is implemented at the ohmic contact, thus ensuring uniform multiplication at the $p^+$5 junction side (Pellegrini et al., 2015).

The underlying avalanche law remains that of LGADs generally: $p^+$6 with $p^+$7 the initial carrier population, $p^+$8 the ionization coefficient, and $p^+$9 the gain (Cartiglia et al., 2013). What changes in iLGAD is not the basic multiplication mechanism but the spatial relationship between the high-field region and the segmented readout.

That geometric inversion changes signal formation. In conventional LGADs, the segmented junction side favors fast electron-side response. In iLGADs, the readout side is the ohmic side, and the collected signal is correspondingly more hole-dominated. The 2015 paper stated explicitly that the collected signal is due to holes flowing back from the multiplication junction, and that the consequence is increased detection time because hole mobility is lower than electron mobility (Pellegrini et al., 2015). The first beam-and-laser prototype study refined this picture: the measured iLGAD waveform contains contributions from primary electrons and from secondary holes produced in the avalanche process, yielding a slower and more structured pulse than in a reference PIN detector (Currás et al., 2019).

For soft X-rays, the architecture was specialized further. In the 2023 soft-X-ray photodiodes, the iLGAD was built on p-type silicon with the gain layer placed on the backplane, i.e. on the same side as the X-ray entrance window. Under reverse bias, electrons drift toward the backplane and holes drift toward the front-side readout electrodes. The gain layer is a shallow boron implant confined within roughly $n^+$0 of the back surface, where the electric field exceeds about $n^+$1 (Liguori et al., 2023). In that configuration, the measured average gain was defined as

$n^+$2

using a matched no-gain diode from the same wafer as reference (Liguori et al., 2023).

A crucial refinement in the soft-X-ray regime is that multiplication depends on absorption depth. The 2023 study distinguished electron-triggered avalanches for photons absorbed beyond the gain layer from hole-triggered avalanches for photons absorbed before it, with fitted asymptotic multiplication factors satisfying $n^+$3 (Liguori et al., 2023). This depth dependence is not peculiar to photon science; it suggests a general iLGAD principle that pulse formation depends not only on lateral position but also on where the primary charge is generated relative to the continuous gain layer.

3. Fill factor, gain uniformity, and the inverse architecture

The central motivation for iLGAD is the fill-factor problem of segmented LGADs. In the timing-detector context of ATLAS and CMS, the baseline MTD concept used segmented LGAD pads of order $n^+$4, yielding timing around 20–30 ps but poor position resolution; moreover, a minimum-ionizing particle crossing the inter-pad region encountered little or no multiplication, degrading timing severely (Currás et al., 2019). More generally, conventional small-pitch LGADs require per-pixel or per-strip termination of the multiplication region, so the no-gain fraction grows as pitch shrinks (Svihra et al., 2024).

The first direct experimental demonstration of the iLGAD solution came from strip devices fabricated at IMB-CNM on $n^+$5-thick high-resistivity float-zone wafers (Currás et al., 2019). In beam tests of unirradiated 45-strip, $n^+$6-pitch detectors at room temperature, the conventional LGAD strip showed a bimodal charge spectrum at 120 V: a lower peak around 24 ke, corresponding to interstrip hits with no gain, and a higher peak around 77 ke, corresponding to hits under the multiplied strip region; the expected gain was about 3 (Currás et al., 2019). By contrast, the iLGAD strip showed one peak, around 75 ke, indicating that particles crossing anywhere in the sensitive region experienced multiplication (Currás et al., 2019). In the terminology of that paper, this is what “100% fill-factor by design” meant: not generic geometric activity, but laterally uniform avalanche-gain response across the full sensitive region (Currás et al., 2019).

The earlier 2015 TCAD study had anticipated the same effect. For strip geometries, it showed that in a standard LGAD the high-field region is narrow and concentrated at the strip center, whereas in the iLGAD the maximum electric field is uniformly distributed at the multiplication side. In MIP simulations, gain appeared wherever the MIP entered the iLGAD structure, while in the LGAD it was strong mainly at the strip center (Pellegrini et al., 2015). That paper also stressed the trade-off: iLGAD solves gain non-uniformity but does so with hole collection and therefore slower response (Pellegrini et al., 2015).

Later pixel-hybrid studies extended the same architectural claim to fine pitch. A $n^+$7-pitch, $n^+$8-thick Micron iLGAD bonded to Timepix3 showed a gain of

$n^+$9

with very good uniformity across the whole gain area and efficiency exceeding 99.94\% at 300 V and 983 $\mathrm{n^{++}\!-\!p^+\!-\!p}$0 threshold (Svihra et al., 2024). A related $\mathrm{n^{++}\!-\!p^+\!-\!p}$1-pitch, $\mathrm{n^{++}\!-\!p^+\!-\!p}$2-thick large-area iLGAD on Timepix4 showed an almost uniform gain of approximately 4 and efficiency of $\mathrm{n^{++}\!-\!p^+\!-\!p}$3 (Oppenhuis, 11 Sep 2025). These results indicate that the inverse architecture remains effective when scaled to small-pitch hybrid pixels, even though timing outcomes depend strongly on thickness and readout conditions.

4. Tracking and timing performance

iLGAD timing performance has to be read against the conventional LGAD timing baseline. LGAD/UFSD studies established that internal gain improves timing primarily by increasing $\mathrm{n^{++}\!-\!p^+\!-\!p}$4, thereby reducing jitter, and projected sub-$\mathrm{n^{++}\!-\!p^+\!-\!p}$5 performance for gain-$\mathrm{n^{++}\!-\!p^+\!-\!p}$6 devices under favorable capacitance and thickness conditions (Cartiglia et al., 2013). iLGAD inherits the gain mechanism but not the same signal-formation geometry.

The first dedicated iLGAD strip-timing study therefore used a laser system rather than a beam-particle timing reference. A 1060 nm picosecond pulsed IR laser was split into two optical paths separated by about 52 ns; the device signal was amplified by a Miteq 1660 current amplifier with 60 dB gain and digitized by a 25 GSa/s oscilloscope (Currás et al., 2019). Timing extraction used a software CFD, and the time resolution was defined as

$\mathrm{n^{++}\!-\!p^+\!-\!p}$7

In the best CFD region—dominated by the fast initial electron-induced part of the pulse—the iLGAD achieved 20 ps at 700 V and room temperature, with gain 4.8 (Currás et al., 2019). The same work reported a preliminary spatial resolution of $\mathrm{n^{++}\!-\!p^+\!-\!p}$8 at 300 V, while explicitly noting that the beam-test electronics saturated above 100 V, so the value was not optimal (Currás et al., 2019).

Hybrid-pixel measurements have so far yielded slower timing, consistent with thicker hole-collecting implementations. The Timepix3-bonded $\mathrm{n^{++}\!-\!p^+\!-\!p}$9 iLGAD reached hit time resolution down to 1.3 ns without correcting for the time-walk effects (Svihra et al., 2024). The Timepix4-bonded $n^{++}-p^+-p-n^{++}$0 iLGAD showed raw timing of about 750 ps, improving from 746 ps at 250 V to $n^{++}-p^+-p-n^{++}$1 ps after per-pixel timewalk and clock corrections (Oppenhuis, 11 Sep 2025). Grazing-angle measurements further showed a best depth-sliced timing of 359 ps at $n^{++}-p^+-p-n^{++}$2 depth and 900 $n^{++}-p^+-p-n^{++}$3 threshold (Oppenhuis, 11 Sep 2025).

The Timepix4 study attributed this limitation primarily to the sensor rather than the ASIC. In that device, the amplified contribution arrives late relative to the weighting-field maximum near the readout implant, so the integrated signal can cross threshold before the multiplication-enhanced part contributes strongly (Oppenhuis, 11 Sep 2025). A plausible implication is that iLGAD timing depends not only on gain magnitude but on the temporal ordering of charge multiplication and current induction, especially in thick backside-multiplication geometries.

5. X-ray and soft-X-ray implementations

iLGAD has also become a photon-science detector technology. The 2022 X-ray periphery study argued that iLGAD is attractive for X-rays because it combines internal multiplication, low readout noise, 100% fill factor, and compatibility with thick silicon (Doblas et al., 2022). The emphasis there was not on the multiplication concept itself but on making the periphery robust under ionizing-dose-induced surface damage.

A parallel line of work optimized iLGADs for soft X-rays by combining the inverse geometry with a thin entrance window. In the 2023 SLS characterization, standalone $n^{++}-p^+-p-n^{++}$4-thick FBK iLGAD diodes with 4 mm$n^{++}-p^+-p-n^{++}$5 active area were measured from 200 eV to 1 keV (Liguori et al., 2023). For all sensor variations, the QE above 250 eV was larger than 55\%, specifically 55–67\% at 250 eV, and the charge collection efficiency was found to be essentially 100\%; the dominant residual loss was photon absorption in dielectric entrance-window layers with total thickness approximately 70–90 nm (Liguori et al., 2023). The same study showed that the average gain increases with photon energy because deeper absorption favors electron-triggered multiplication (Liguori et al., 2023).

For representative designs, the fitted asymptotic multiplication factors were markedly different for electron- and hole-triggered avalanches. For W17 (standard gain layer, medium dose), the fit gave $n^{++}-p^+-p-n^{++}$6 and $n^{++}-p^+-p-n^{++}$7; for W9 (shallow, high dose), $n^{++}-p^+-p-n^{++}$8 and $n^{++}-p^+-p-n^{++}$9; for W13 (ultra-shallow, high dose), $p^+$0 and $p^+$1 (Liguori et al., 2023). Those values quantify a basic iLGAD feature in the soft-X-ray regime: the same detector can respond differently depending on whether photons are absorbed before or after the gain layer.

System-level applications followed quickly. A JUNGFRAU detector equipped with an iLGAD sensor was deployed at the EuXFEL hRIXS spectrometer and achieved $p^+$2 spatial resolution and a resolving power exceeding 10,000 at 47 kHz frame rate (Duarte et al., 15 Nov 2025). Intra-train-resolved CuO data at 928.5 eV, 1.1 MHz, and $p^+$3 showed a decrease in emitted signal by about 10\% over $p^+$4, indicating FEL-induced effects that had to be monitored during high-repetition-rate operation (Duarte et al., 15 Nov 2025). In a related direction, the Timepix3-bonded iLGAD showed that gain enabled useful X-ray spectroscopy down to 4.5 keV, with the Ti fluorescence line visible only in the multiplied region (Svihra et al., 2024).

6. Periphery engineering, irradiation, and unresolved constraints

Beyond fill factor, iLGAD development has been shaped by edge termination and radiation response. The first-generation X-ray-oriented iLGAD periphery (iLG1) was not designed for oxide-charge buildup at the Si–SiO$p^+$5 interface and suffered premature breakdown after irradiation (Doblas et al., 2022). TCAD-guided redesign first added floating rings and p-stops on the multiplication side, improving breakdown by about 50 V, but this was insufficient because the ohmic side also developed high-field peaks under irradiation (Doblas et al., 2022). The decisive advance was double-side optimization, culminating in iLG2, which used multiplication-side floating rings and p-stops together with an ohmic-side field plate, high-doped p-stops, and channel-stopper engineering (Doblas et al., 2022).

Fabricated on $p^+$6 high-resistivity p-type silicon $p^+$7, iLG2 showed $p^+$8, $p^+$9, leakage current about 10 nA, and breakdown around 450 V for a $1\text{–}5~\mu\text{m}$0 pad, with TCT-measured gain approximately 12 to 24 over 70–360 V (Doblas et al., 2022). After 10 MRad X-ray irradiation at 1.8 MRad/h, iLG2 showed only a slight leakage-current increase and retained essentially the same voltage capability, whereas iLG1 degraded to breakdown near 75 V (Doblas et al., 2022). That work established periphery engineering, rather than multiplication-layer redesign alone, as a prerequisite for ionizing-dose robustness in iLGAD X-ray sensors.

Radiation damage in the multiplication layer remains a broader LGAD-family problem. A compensated gain-layer proposal replaced the usual single $1\text{–}5~\mu\text{m}$1 multiplication implant by overlapping $1\text{–}5~\mu\text{m}$2 and $1\text{–}5~\mu\text{m}$3 implants so that the effective doping $1\text{–}5~\mu\text{m}$4 remained near the standard LGAD value while potentially becoming more stable under irradiation (Sola et al., 2022). That concept was not developed specifically for iLGAD, and the paper provided no irradiated hardware validation; however, a plausible implication is that continuous iLGAD gain layers could use the same compensation principle if acceptor and donor removal proved sufficiently balanced (Sola et al., 2022).

A second indirect but important result concerns the acceptor removal phenomenon. A 2023 defect study argued that LGAD gain loss under irradiation is better explained by formation of a donor-like $1\text{–}5~\mu\text{m}$5 complex than by older oxygen-centered assignments (Lauer et al., 2023). Since iLGAD relies on the same acceptor-based multiplication physics, that mechanism is likely to remain relevant. Likewise, a 2021 gain-suppression study in conventional LGADs showed that high local charge density can screen the gain-layer field and lower the measured gain (Currás et al., 2021). That paper did not study iLGAD, but a plausible implication is that dense localized excitation, including focused X-ray absorption, can perturb multiplication even when the gain layer is laterally continuous.

The main open design trade-off therefore remains unchanged across the iLGAD literature: the inverse architecture cures segmentation-induced gain collapse, but it often shifts the detector toward hole-dominated, thickness-sensitive signal formation. This is why several papers identify thinner sensors, SOI implementations, higher field near the implant, and optimized electronics as the route to better timing while preserving the uniform-gain benefit of the inverse geometry (Pellegrini et al., 2015, Currás et al., 2019, Oppenhuis, 11 Sep 2025).