Engineering and Technology
Photon-trapping-enhanced avalanche photodiodes for mid-infrared applications
D. Chen, S. D. March, et al.
Mid-wave infrared (MWIR, 2–5 µm) photonics underpins applications in sensing, spectroscopy, medical diagnostics and communications, and is expanding via silicon photonics integration. A key receiver metric is signal-to-noise ratio (SNR), yet MWIR photodetectors suffer from weak signal levels and high dark current due to narrow-bandgap absorbers (e.g., HgCdTe, InAs/GaSb and InAs/InAsSb superlattices). Avalanche photodiodes (APDs) can improve SNR via internal gain but introduce excess noise and are especially impacted by dark current in the MWIR. The central challenge is reducing dark current without sacrificing quantum efficiency (QE). Thinning the absorber suppresses dark current but reduces absorption. This work targets maintaining high QE with an ultrathin absorber by employing photon-trapping structures to couple normally incident light into lateral waveguide modes, thereby enabling low-dark-current, high-efficiency MWIR SACM (separate absorption, charge and multiplication) APDs.
Conventional MWIR photodetectors such as HgCdTe require cryogenic cooling due to high dark current arising from Shockley–Read–Hall generation, tunneling, and thermal generation in narrow bandgaps. Alternative absorber systems include InAs/GaSb type-II superlattices (T2SL) and InAs/InAsSb type-II strained-layer superlattices (T2SLS), yet dark current remains a key limitation. Prior work demonstrated the first 2 µm AlInAsSb SACM APDs with dark current comparable to HgCdTe at nearly 100 K higher temperature. Photon-trapping using photonic crystals or metal plasmonic structures has enhanced absorption in other wavelength regimes but had not been demonstrated for MWIR APDs. Simulations suggested that submicrometre metal grating arrays could raise QE by over four times by diffracting normal incidence into lateral waveguide modes and leveraging plasmon-enhanced transmission. AlInAsSb multiplication layers are known to provide low excess noise with k ~ 0.01, comparable to silicon.
Device design: A SACM APD on GaSb employs a narrow-bandgap AlInAsSb absorber (Eg ≈ 0.58 eV) thinned to 200 nm to suppress dark current while supporting waveguide modes, and a wide-bandgap AlInAsSb multiplication layer (Eg ≈ 1.16 eV) optimized for high gain and gain-bandwidth product. A p-type charge layer establishes strong field contrast between absorber and multiplication regions to avoid premature breakdown and yield a moderate punch-through bias. A continuously graded Al0.3–0.7InAsSb layer smooths the bandgap transition for efficient electron injection and high transit-time bandwidth. Photon-trapping structures: Two-dimensional rectangular lattice metal grating arrays are placed atop the device to couple normal-incidence light into lateral modes. Optimized grating parameters: spacing ≈ 598 nm (measured optimum ~596 nm), duty cycle ≈ 0.82, metal thickness ≈ 380 nm. Mechanisms include diffractive coupling into absorber waveguide modes and plasmon-enhanced near-field transmission, boosting lateral coupling efficiency beyond dielectric gratings. Epitaxial growth: Layers grown by molecular-beam epitaxy at 460 °C on n-type GaSb, using solid-source effusion cells (Al, Ga, In, Be, GaTe) and valved crackers (As, Sb). AlInAsSb was formed as a digital alloy of stable binaries via repeated shutter sequences (AlSb, AlAs, AlSb, InSb, InAs, Sb soak) lattice matched to GaSb. Device fabrication: Circular mesa devices defined by photolithography and citric acid wet etching. Sub-micrometre metal grating arrays fabricated by electron-beam lithography and lift-off. Metal stack: 10 nm Ti (adhesion) + 370 nm Au. Characterization and analysis: I–V characteristics measured with Keithley 2400; temperature-dependent measurements in a liquid-nitrogen-cooled cryostat with HP 4145. Unity gain determined by fitting excess noise to photocurrent changes (excess-noise-based punch-through gain determination). Excess noise factor F(M) measured under 2 µm illumination and compared with the local-field model for various k values. C–V measured with HP 4275A LCR meter at 1 MHz to verify punch-through and depletion behavior. EQE at normal incidence measured at punch-through using a xenon light source, monochromator and lock-in amplifier; referenced to a calibrated strained-layer InGaAs photodiode. Edge-coupled EQE measured by coupling a polarized 2 µm CW laser via a lensed fiber into a cleaved device edge; responsivity and EQE computed from measured photocurrents and optical power. Frequency response characterized for devices of various areas; 3 dB bandwidth and gain-bandwidth product extracted; microwave S-parameters fitted to an equivalent circuit to estimate RC-limited bandwidth.
• Photon-trapping boosts external quantum efficiency (EQE) of a 200-nm absorber SACM APD from ~7% (planar) to ~22% at 2 µm under normal incidence; simulated EQE ~36–38% for optimized gratings. • Edge coupling into the absorber waveguide yields EQE ~24% without anti-reflection (AR) coating; simulations indicate potential ~35% with AR. • Excess noise is very low with measured k ≈ 0.01 under 2 µm illumination, consistent with AlInAsSb APDs and comparable to silicon. • High gain: maximum multiplication gain ~700 measured at 240 K under low optical power (~10 nW). • Dark current density is substantially suppressed by the ultrathin absorber: ~1 × 10^-6 A cm^-2 at 180 K, nearly two orders lower than prior 2 µm AlInAsSb APDs with 1-µm absorbers and nearly three orders lower than state-of-the-art 2 µm HgCdTe APDs at comparable temperatures. • High-speed performance: 3 dB bandwidth ~7 GHz and gain-bandwidth product (GBP) > 200 GHz for a 10 × 35 µm^2 device, both exceeding previous 2 µm APD records by >4×. • Grating-spacing dependence of EQE follows Bragg condition behavior: peak around ~596 nm spacing; deviations reduce EQE, matching simulations. • RC modeling indicates an RC-limited bandwidth of ~55.9 GHz for the 10 × 35 µm^2 device, implying additional bandwidth limitations (beyond transit time) due to RF losses in the conductive GaSb substrate.
The study addresses the MWIR APD tradeoff between dark current and quantum efficiency by thinning the absorber to suppress dark current while recovering absorption through photon trapping. The metal grating arrays diffract normal-incidence light into laterally guided modes and enhance near-field coupling via plasmonic effects, enabling a 3× improvement in EQE over a planar device despite a 200 nm absorber. Combined with the intrinsically low excess noise (k ≈ 0.01) of AlInAsSb multiplication layers, the devices achieve high gain (~700 at 240 K) with significantly reduced dark current, thereby improving SNR by ~70× over earlier 2 µm AlInAsSb SACM APDs and ~20× over state-of-the-art HgCdTe APDs under low optical intensity. The ultrathin absorber also supports short transit times, yielding ~7 GHz bandwidth and >200 GHz GBP. The observed area dependence and S-parameter analysis suggest that the measured bandwidth is limited by RF loss in the conductive GaSb substrate rather than RC or transit-time limits alone, pointing to clear routes for further speed improvements (e.g., semi-insulating substrates). The grating-spacing dependence corroborates the photon-trapping (grating-coupler) mechanism and aligns with simulation trends.
This work demonstrates, for the first time, photon-trapping structures integrated with MWIR SACM APDs to break the traditional tradeoff between low dark current and high quantum efficiency. Using an ultrathin (200 nm) AlInAsSb absorber and metal grating photon trapping, the devices achieve EQE ~22% at normal incidence (and ~24% via edge coupling), extremely low dark current (~1 × 10^-6 A cm^-2 at 180 K), low excess noise (k ≈ 0.01), high gain (~700 at 240 K), ~7 GHz bandwidth and >200 GHz gain-bandwidth product. These metrics represent multi-fold improvements over previous 2 µm APDs and indicate substantial SNR gains relative to both earlier AlInAsSb and state-of-the-art HgCdTe APDs. Future improvements could include semi-insulating substrates (e.g., InP) to reduce RF loss and push bandwidths >30 GHz, AR coatings to raise EQE in edge-coupled devices, and extension of the photon-trapping thin-absorber concept to longer wavelength cut-off APDs.
• Measured EQE (~22%) is below simulations (~36–38%) due to fabrication nonidealities (metal thickness and profile, sidewall bevel, Ti adhesion layer oxidation). • Conductive GaSb substrates introduce RF loss that limits bandwidth below RC and simulated transit-time expectations; semi-insulating substrates are needed for higher speed. • Reported dark current and maximum gain metrics are at reduced temperatures (e.g., 180 K for dark current, 240 K for max gain); room-temperature dark current and highest-gain performance are not fully benchmarked. • Absorber not fully depleted immediately at punch-through, affecting bandwidth behavior at low gains. • Edge-coupled EQE lacks anti-reflection coating; potential performance is inferred from simulations rather than demonstrated. • Unity-gain determination requires excess-noise-based fitting; direct unity-gain point is not trivially observable from I–V alone.
Related Publications
Explore these studies to deepen your understanding of the subject.

