TECHNOLOGY EXPLAINER
Photodetectors
— turning light back into electricity is a matter of thinness and dark current
A signal sent as light is finally turned back into electricity by a photodetector (photodiode). At optical fibre wavelengths, the light is received by InGaAs on an InP substrate or by Ge grown on silicon. For speed you want the absorption layer thin; for sensitivity you want it thick. And the dark current that flows even without light is set by crystal defects and interfaces.
- Photodetectors in three points
- Which wavelengths can be received is set by the bandgap
- The speed limit — the trade-off between capacitance and transit time
- A company example: a back-illuminated InGaAs PD for 200 Gbps
- A materials engineer's view (1): dark current is set by crystal and interface
- Ge photodetectors on silicon
- A materials engineer's view (2): putting thermal expansion mismatch to work
- APDs — amplification by avalanche
- What is still hard
- Glossary / References / Claim-to-source audit
Sourced = stated in published material or a peer-reviewed paper (link given)
Our calculation = a figure this article derived, with the assumptions spelled out
Not yet confirmed = a plan or research-stage result with no confirmed production record
Structural readings and materials-design interpretations are marked separately as Commentary.
Performance figures are kept apart as company technical material and evaluation results for development parts or values from papers.
1. Photodetectors in three points
A photodetector (photodiode, PD) shines light on a semiconductor pn (or pin) junction and collects the electrons and holes created by absorbed photons as a current. Hamamatsu Photonics' technical material describes the InGaAs PIN photodiode as a photovoltaic device with a PN junction, just like a Si photodiodeSourced.
- What it is made of: at optical fibre wavelengths, InGaAs lattice-matched to InP, or in silicon photonic circuits selectively grown GeSourced
- What makes it hard: speed (bandwidth), sensitivity (A/W) and dark current (current with no light) pull against one another. Hamamatsu Photonics explains that the cutoff frequency set by capacitance and the one set by transit time trade off against each otherSourced
- Detectors that amplify: there are also avalanche photodiodes (APDs), which multiply the current internallySourced
Where Ge detectors fit within a silicon photonic circuit (one of its four components) was covered in our explainer on silicon photonics, InP substrates and integration on InP in our explainer on photonic integrated circuits and indium phosphide, and the EML on the transmit side in our explainer on semiconductor lasers. The amplifier circuit behind the detector (TIA) and signal processing are covered in our explainers on optical interconnects and SerDes and on coherent optical communication and DSP. This article focuses on the crystal, layer structure and dark current of the detector.
2. Which wavelengths can be received is set by the bandgap
A semiconductor absorbs only light with more energy than its bandgap. Hamamatsu Photonics' technical material gives the relation between bandgap Eg (eV) and cutoff wavelength λc (µm) as λc = 1.24 ÷ Eg, and explains that InGaAs with a composition lattice-matched to InP has a bandgap of 0.73 eV at room temperature and a cutoff wavelength of about 1.7 µmSourced. The bandgap of InGaAs also shifts with the In-to-Ga ratio, and the company offers long-wavelength types with cutoffs of 1.9, 2.1 and 2.6 µm alongside the standard 1.7 µm typeSourced.
Silicon detectors barely absorb near-infrared communication wavelengths (see our explainer on silicon photonics), so silicon photonic circuits use Ge instead. Yang and colleagues (2010), writing on Ge/Si photodetectors, note that Ge is regarded as an ideal candidate for low-cost, high-performance infrared detectors because it has a large absorption coefficient for near-infrared light at 1,300 to 1,550 nm and is compatible with existing silicon processingSourced.
3. The speed limit — the trade-off between capacitance and transit time
Hamamatsu Photonics' technical material sets out, as equations, the factors that determine a detector's response speedSourced.
- Capacitance (CR time constant): cutoff frequency fc = 1 ÷ (2π Ct RL). The terminal capacitance Ct is proportional to the active area and inversely proportional to the square root of the reverse voltage. For low capacitance, a small active area and a thick depletion layer are better
- Transit time (drift time): the time for carriers to cross a depletion layer of thickness W is trd = W ÷ vds, and the cutoff frequency it sets is fc(trd) = 0.44 ÷ trd. Here a thin depletion layer is better
- Carrier speed: in InGaAs, the hole drift velocity saturates at a field of about 10⁴ V/cm, at about 5×10⁶ cm/s
- Diffusion delay: charge created by light absorbed outside the junction can take several microseconds or more to diffuse to the electrode
Assumptions: vds = 5×10⁶ cm/s (from Hamamatsu Photonics), fc = 0.44 ÷ trdOur calculation
- W = 1 µm → trd = 20 ps → about 22 GHz
- W = 0.5 µm → 10 ps → about 44 GHz
- W = 0.3 µm → 6 ps → about 73 GHz
- W = 0.2 µm → 4 ps → about 110 GHz
Assumptions and limits: this simply applies to the whole depletion layer the equation and hole saturation velocity Hamamatsu Photonics gives in its explanation of APDs. Electron velocity, capacitance, electrodes and packaging are not included. It is a calculation to show the trend, not the bandwidth of any particular device.
4. A company example: a back-illuminated InGaAs PD for 200 Gbps
On 20 August 2024 Mitsubishi Electric announced that it would begin sample shipments of a “200 Gbps pin-PD chip for 800 Gbps / 1.6 Tbps optical fibre communications” on 1 OctoberSourced. The company said it had been in volume production of a 200 Gbps EML chip for transmission since April 2024, and that on the receive side there had been a problem of few products meeting the required performanceSourced. Its 2025 Technical Report describes the design in detailSourced.
- The problem: multi-lane optical transceivers want a large receiving area to ease optical alignment, but enlarging the receiving area of a surface-illuminated device enlarges the pn junction, raising capacitance and lowering speed. A conventional back-illuminated device (receiving area φ15 µm) struggles to exceed 50 GHz of bandwidth. Thinning the absorption layer for bandwidth means the sensitivity falls short of 0.60 A/W
- The solution: a back-illuminated design that brings light in from the substrate side, keeping the pn junction (formed by selective Zn diffusion, about 10 µm across) small while shaping the InP substrate itself into a convex lens to enlarge the effective receiving area. The anode electrode doubles as a metal mirror, reflecting light that has passed through the absorption layer so that it is absorbed on a second pass
- The crystal: on a semi-insulating Fe-doped InP substrate, an n-type contact layer, n-type InP layer, undoped InP layer, absorption layer, undoped InP window layer and p-type contact layer are grown epitaxially in that order. The layers outside the receiving area are etched away, and Fe-doped InP is regrown to bury them and protect the interfaces
| Item | Mitsubishi Electric evaluation results (development part) |
|---|---|
| Dark current | about 1 pA at the operating voltage of 2 V (room temperature, five devices) |
| Activation energy of dark current | about 0.66 eV (nearly equal to the bandgap of the InGaAs absorption layer → diffusion current dominates) |
| Responsivity | about 0.60 A/W |
| 3 dB bandwidth | about 65 GHz (2 V, 1,310 nm, 0 dBm, probed directly on the chip) |
| Responsivity tolerance | about φ10 µm without the lens → about φ40 µm with it |
Sourced (Mitsubishi Electric Technical Report Vol. 99 No. 3, 2025 [Reference 2]). These are evaluation results for the chip alone, not the receiver sensitivity of a transceiver.
Quantum efficiency η = responsivity (A/W) × 1.24 ÷ wavelength (µm)Our calculation
- 0.60 × 1.24 ÷ 1.31 ≈ 0.57 (about 57% of incoming photons become current)
- Optical alignment tolerance: φ10 µm → φ40 µm is 16 times in area ((40 ÷ 10)²)
Assumptions and limits: this treats the 0.60 A/W responsivity as a value at 1.31 µm. The report measured the two-dimensional distribution at 1,310 nm, but does not state the wavelength at which responsivity was measured, so this is an indicative calculation.
5. A materials engineer's view (1): dark current is set by crystal and interface
The most interesting part of the Mitsubishi Electric report from a materials standpoint is how it reads the temperature dependence of the dark currentSourced.
- The activation energy obtained from the temperature dependence of dark current is about 0.66 eV, nearly equal to the bandgap of the InGaAs absorption layer
- That means the dark current is dominated by diffusion current in the InGaAs absorption layer, with generation-recombination and leakage currents small
- This suggests little dark current flowing via defects in the epitaxial layers or on the sides of the junction
- The company attributes this to selective Zn diffusion and a semiconductor-buried structure using Fe-doped InP, which suppress the formation of interface states and reduce leakage along the sides of the pn junction
Hamamatsu Photonics' technical material likewise divides APD dark current into surface leakage current flowing along the interface between the PN junction and the surface passivation film, and the like, and recombination current, tunnelling current, diffusion current and the like generated inside the semiconductorSourced.
Translated into materials terms (our framing):
- Diffusion current: the floor set by the material's bandgap. Unavoidable
- Generation-recombination current: created by defects (deep levels) in the crystal. Set by epitaxial quality
- Surface and sidewall leakage: set where the junction meets the surface, and by the interface with the passivation film
An activation energy close to the bandgap is therefore also a “certificate of quality”: the latter two — crystal defects and interfaces — have been suppressed enough that the device has reached the limit of the material's physics. The choice to fill the etched sidewalls with the same InP crystal rather than an insulating film reflects a design philosophy that avoids interfaces (our commentary).
Assumptions: dark current ∝ exp(−Ea ÷ kT), Ea = 0.66 eV (Mitsubishi Electric's estimate), k = 8.617×10⁻⁵ eV/KOur calculation
- 25 °C (298 K) → 70 °C (343 K): exp[0.66 ÷ k × (1/298 − 1/343)] ≈ about 29 times
- 25 °C (298 K) → 85 °C (358 K): likewise ≈ about 74 times
- For a device at 1 pA at room temperature, that is about 74 pA at 85 °C
Assumptions and limits: a simplification assuming diffusion current dominates and the activation energy stays constant. 70 °C and 85 °C are temperatures chosen by this article, not guaranteed product temperatures. Hamamatsu Photonics also notes that dark current rises exponentially with temperature, with a coefficient that depends on the device's bandgap and reverse voltageSourced.
6. Ge photodetectors on silicon
In silicon photonic circuits, detectors are made of Ge. imec explained in 2018 that its silicon photonics prototyping platform uses 200 mm SOI wafers and integrates Ge photodetectors and other devices through selective-area growth of Ge and standard CMOS metallisation, and that its modulators and photodetectors have electro-optic bandwidths above 50 GHzSourced. Shekhar and colleagues' 2024 Nature Communications paper states that Ge photodetectors are already offered by many commercial silicon photonics foundriesSourced.
Yang and colleagues (2010) note that the lattice constant of Ge is 4% larger than that of Si, which can produce a high density of threading dislocations in relaxed Ge layers, leading to high dark current and difficult process integrationSourced. The paper shows by simulation that lowering the threading dislocation density in the Ge layer greatly reduces the dark currentSourced.
7. A materials engineer's view (2): putting thermal expansion mismatch to work
Yang and colleagues describe one more materials effectSourced.
- Experiments have confirmed that Ge grown epitaxially on Si carries tensile strain
- The strain arises from the difference in thermal expansion between Ge and Si as the layer cools from the high growth temperature
- It depends on growth temperature and is expected to reach at most 0.34% when cooled from the melting point of Ge (937 °C)
- Tensile strain narrows the direct bandgap of Ge and has been proposed as a way to extend detection to longer wavelengths (the L band)
Thermal expansion mismatch is normally a nuisance that causes warpage and delamination. Here it works as a function that extends the wavelengths the detector can receive. The same Ge-on-Si interface produces dislocations (a cause of dark current) through the lattice mismatch and useful strain through the thermal expansion mismatch — a bad effect and a good effect living together at one heterointerface (our commentary).
Our calculation: a strain of 0.34% corresponds to an extension of 3.4 µm per millimetre of lengthOur calculation. It looks small, but the premise of the paper's argument is that it is enough to move the band structure.
8. APDs — amplification by avalanche
According to Hamamatsu Photonics' technical material, an APD (avalanche photodiode) multiplies the photocurrent internally through avalanche multiplication, in which carriers accelerated by a strong reverse voltage collide with the crystal lattice and create new electron-hole pairsSourced.
- Division of roles (SAM structure): InGaAs has a small bandgap, so a large reverse voltage produces large dark current. The structure therefore separates an InGaAs layer that absorbs light from an InP layer that multipliesSourced
- A material constant sets the noise: the ratio k of the ionisation rates of electrons and holes is a physical constant intrinsic to the semiconductor material, and in InP holes ionise more readily (k > 1). Holes generated in the InGaAs are therefore sent into the InP layer to be multiplied. Multiplication comes with statistical fluctuation (excess noise); approximating the excess noise factor as F ≈ Mx, the company's example gives x ≈ 0.7Sourced
- Temperature: as temperature rises, lattice vibration intensifies and carriers are more likely to collide with the lattice before reaching the energy needed for ionisation, so the gain at a given voltage falls. The voltage has to be adjusted or the temperature held constantSourced
- There is an optimum: raising the gain raises both signal and noise, so there is a gain at which S/N is maximisedSourced
Taking F ≈ Mx with x = 0.7 (Hamamatsu Photonics' example)Our calculation,
- M = 10 → F ≈ 100.7 ≈ 5.0
- M = 20 → F ≈ 200.7 ≈ 8.1
Doubling the gain doubles the signal and raises the excess noise factor by about 1.6 times. That is why more gain is not always better. Assumption: the approximation is applied as it stands; these are not values for any particular product.
On APDs in silicon photonics, Shekhar and colleagues write that APDs combining bandwidth, sensitivity and low noise at low voltage help improve S/N on the receive side, but that compared with Ge photodetectors, APDs generally fall short in bandwidth, linearity and power handling, and need an optimal bias as well as stabilisation against temperature and voltage driftSourced. They also caution that raising the gain alone, while the intrinsic responsivity stays low, does not improve performanceSourced.
9. What is still hard
(1) The tug-of-war between thinness and sensitivity goes on
Mitsubishi Electric solved the problem that a thin absorption layer falls short of 0.60 A/W by sending light back and forth with a metal mirror and turning the substrate into a lensSourced. How far surface-illuminated detectors can keep up in generations with even higher per-wavelength rates could not be confirmed in the primary sources this article consultedNot yet confirmed.
(2) Dislocations and dark current in Ge-on-Si
The roughly 4% lattice mismatch between Ge and Si remains a cause of dislocations and dark currentSourced. Production values for dark current and dislocation density in each foundry's Ge photodetectors could not be confirmed in the primary sources within the scope of this article, and are not given.
(3) Reliability
OpenLight announced in March 2025 that it had completed Telcordia GR-468 qualification for active components with InP integrated on silicon, including photodetectors (PDs)Sourced (as published by the company). Mitsubishi Electric says that selective Zn diffusion and the semiconductor-buried structure can be expected to give high reliabilitySourced, but no long-term reliability test results are given within the scope of the report.
- The wavelengths a detector can receive are set by its bandgap; InGaAs lattice-matched to InP (0.73 eV) reaches about 1.7 µmSourced
- Speed is set by thinness and smallness. Looking at transit time alone, 1 µm gives about 22 GHz and 0.3 µm about 73 GHzOur calculation
- Mitsubishi Electric's 200G PD won sensitivity from a thin absorption layer by turning the substrate into a lens and the electrode into a mirror. Quantum efficiency is about 57%Our calculation
- A dark-current activation energy close to the bandgap is evidence that crystal defects and interface leakage have been suppressedSourced
- In Ge-on-Si, the lattice mismatch produces dislocations while the thermal expansion mismatch produces useful tensile strainSourced
10. Glossary
- Photodiode (PD)
- A semiconductor device that absorbs light and passes a current.
- pin photodiode
- A structure with a lightly doped i layer between the p and n layers, where the light is absorbed.
- InGaAs
- An alloy of indium, gallium and arsenic. At the composition lattice-matched to InP it receives near-infrared light.
- Cutoff wavelength
- The limit beyond which longer wavelengths can no longer be absorbed. λc = 1.24 ÷ Eg.
- Responsivity (A/W)
- The current per watt of light. Set by quantum efficiency and wavelength.
- Quantum efficiency
- The fraction of incoming photons collected as electrons in the current.
- Dark current
- The current that flows when no light is falling on the device. A source of noise.
- Activation energy
- An energy obtained from the temperature dependence of dark current. A clue to which kind of current dominates.
- Depletion layer
- The region near the junction free of carriers. Charge created there is swept out quickly by the field.
- Drift time (transit time)
- The time carriers take to cross the depletion layer.
- CR time constant
- The product of device capacitance and load resistance. A large value slows the response.
- Back-illuminated
- A structure in which light enters from the side opposite the junction (the substrate side).
- Selective Zn diffusion
- Diffusing zinc only into chosen areas to form p-type regions.
- Semiconductor-buried structure
- A structure in which etched sidewalls are filled back with crystal such as Fe-doped InP.
- Selective growth
- Growing crystal only in chosen places on a wafer. Used for Ge photodetectors.
- Threading dislocation
- A line defect that runs from the interface up through the film to its surface in growth on a foreign substrate.
- Tensile strain
- A state in which the crystal is stretched. In Ge it narrows the bandgap.
- APD
- Avalanche photodiode. A detector that multiplies carriers internally.
- SAM structure
- An APD structure with separate absorption and multiplication layers.
- Excess noise factor
- How much the fluctuation of multiplication adds to noise. Approximated as F ≈ Mx.
- Ionisation rate ratio k
- The ratio of the rates at which electrons and holes create electron-hole pairs by collision. A constant intrinsic to the material.
11. References (primary sources)
- Hamamatsu Photonics “Technical note: compound semiconductor photosensors” (PDF, in Japanese) — hamamatsu.com
- Mitsubishi Electric S. Okuda et al., “200 Gbps EML and PD for 800 Gbps / 1.6 Tbps transmission in data centres”, Mitsubishi Electric Technical Report (Mitsubishi Denki Giho) Vol. 99 No. 3 (2025) (PDF, in Japanese) — giho.mitsubishielectric.co.jp
- imec “Silicon photonic interposers for 400Gb/s and beyond optical interconnects”, imec magazine, October 2018 — imec-int.com
- S. Shekhar, W. Bogaerts, L. Chrostowski, J. E. Bowers et al. “Roadmapping the next generation of silicon photonics”, Nature Communications 15, 751 (2024, DOI: 10.1038/s41467-024-44750-0) — nature.com
- J.-H. Yang, Y. Wei, X.-Y. Cai, J.-Z. Ran (Lanzhou University) “The effects of threading dislocations and tensile strain in Ge/Si photodetector”, Microelectronics International 27(2), 113–116 (2010, DOI: 10.1108/13565361011034803) (PDF) — bdt.semi.ac.cn
- Mitsubishi Electric “Sample shipments begin of the ‘200 Gbps pin-PD chip for 800 Gbps / 1.6 Tbps optical fibre communications’”, 20 August 2024 (in Japanese) — mitsubishielectric.co.jp
- OpenLight “OpenLight Achieves Successful Completion of Telcordia GR-468 Qualification for Silicon Photonics Components”, 24 March 2025 — openlightphotonics.com
12. Claim-to-source audit
| Claim in the text | Basis | Label |
|---|---|---|
| That the InGaAs PIN photodiode is a photovoltaic device with a PN junction, like a Si photodiode; λc = 1.24 ÷ Eg, with InP-lattice-matched InGaAs at Eg 0.73 eV at room temperature and a cutoff of about 1.7 µm; that the bandgap varies with composition, with long-wavelength types at 1.9, 2.1 and 2.6 µm besides the 1.7 µm standard; that Ct is proportional to active area and inversely proportional to the square root of reverse voltage, the CR cutoff 1/(2πCtRL), and that a small active area and thick depletion layer favour CR; trd = W/vds and fc(trd) = 0.44/trd, InGaAs hole drift velocity saturating at about 10⁴ V/cm at about 5×10⁶ cm/s, and the trade-off between the two; that charge absorbed outside the junction can take several microseconds or more to diffuse; that dark current rises exponentially with temperature with a coefficient depending on bandgap and reverse voltage; and the APD principle (avalanche multiplication), the reason for the SAM structure (InGaAs absorption layer and InP avalanche layer), the components of dark current (surface leakage and the like / recombination, tunnelling, diffusion and the like), that the ionisation rate ratio k is material-specific with k > 1 in InP, that holes drift into the InP layer, F ≈ M^x with x ≈ 0.7 as an example, the gain that maximises S/N, and why gain falls as temperature rises | Hamamatsu Photonics technical note, “Compound semiconductor photosensors”, Reference 1 https://www.hamamatsu.com/content/dam/hamamatsu-photonics/sites/documents/99_SALES_LIBRARY/ssd/compound_kird9004j.pdf | Sourced |
| That multi-lane use calls for a larger receiving area but a larger junction raises capacitance, that a conventional back-illuminated device (φ15 µm) struggles to exceed 50 GHz, and that thinning the absorption layer leaves responsivity short of 0.60 A/W; the back-illuminated design, the substrate shaped into a convex lens, and the anode electrode doubling as a metal mirror; the layer structure on an Fe-doped InP substrate, buried regrowth of Fe-doped InP, selective Zn diffusion (about 10 µm across), and suppression of interface states and leakage; dark current of about 1 pA at 2 V, an activation energy of about 0.66 eV close to the InGaAs bandgap indicating diffusion-dominated current, and the implication of little leakage via defects and junction sidewalls; responsivity tolerance from φ10 to φ40 µm, responsivity of about 0.60 A/W, and 3 dB bandwidth of about 65 GHz (with measurement conditions); and the statement that high reliability can be expected | Mitsubishi Electric Technical Report Vol. 99 No. 3 (2025), Reference 2 https://www.giho.mitsubishielectric.co.jp/giho/pdf/2025/2503104.pdf | Sourced |
| That imec's silicon photonics prototyping platform uses 200 mm SOI and integrates Ge photodetectors through selective Ge growth and standard CMOS metallisation, and that its modulators and photodetectors have bandwidths above 50 GHz (2018) | imec magazine (October 2018), Reference 3 https://www.imec-int.com/en/imec-magazine/imec-magazine-october-2018/silicon-photonic-interposers-for-400gb-s-and-beyond-optical-interconnects | Sourced |
| That Ge photodetectors are offered by many commercial silicon photonics foundries; that APDs combining bandwidth, sensitivity and low noise at low voltage help S/N, that APDs generally fall short of Ge photodetectors in bandwidth, linearity and power handling and need stabilised bias and drift, and that raising gain while intrinsic responsivity stays low does not improve performance | Shekhar et al., Nature Communications 15, 751 (2024), Reference 4 https://www.nature.com/articles/s41467-024-44750-0 | Sourced |
| That Ge has a large absorption coefficient at 1,300 to 1,550 nm and is compatible with silicon processing; that the Ge lattice constant is 4% larger than Si, which can produce a high density of threading dislocations, leading to high dark current and difficult integration; the simulation showing that lower threading dislocation density lowers dark current; and that tensile strain in Ge on Si arises from thermal expansion mismatch on cooling, depends on growth temperature, is expected to reach at most 0.34% when cooled from the Ge melting point (937 °C), and narrows the direct bandgap to extend detection to longer wavelengths (the L band) | Yang et al., Microelectronics International 27(2), 113 (2010), Reference 5 https://bdt.semi.ac.cn/download/0.481743961058849.pdf | Sourced |
| That on 20 August 2024 the company announced sample shipments of a 200 Gbps pin-PD chip from 1 October; that the 200 Gbps EML chip has been in volume production since April 2024; the problem of few receive-side products meeting performance requirements; 800 Gbps with four chips and 1.6 Tbps with eight; and the back-illuminated, integrated convex-lens structure compatible with flip-chip mounting | Mitsubishi Electric news release (20 August 2024), Reference 6 https://www.mitsubishielectric.co.jp/ja/pr/2024/0820/ | Sourced |
| That OpenLight announced completion of GR-468 qualification for lasers, EAMs and PDs with InP integrated on silicon (as published by the company) | OpenLight press release (24 March 2025), Reference 7 https://openlightphotonics.com/newsroom/openlight-achieves-successful-completion-of-telcordia-gr-468-qualification-for-silicon-photonics-components | Sourced |
| The check 1.24 ÷ 0.73 ≈ 1.70 µm; transit times and bandwidths for thicknesses of 1 / 0.5 / 0.3 / 0.2 µm (about 22 / 44 / 73 / 110 GHz); quantum efficiency 0.60 × 1.24 ÷ 1.31 ≈ 0.57, and the area ratio of 16 for the responsivity tolerance; the dark current multiplication at an activation energy of 0.66 eV (about 29 times at 70 °C and about 74 times at 85 °C); 0.34% strain as 3.4 µm per mm; and F ≈ M^0.7 giving about 5.0 at M = 10 and about 8.1 at M = 20 | Our calculation. Thicknesses, temperatures and gains are examples chosen by this article. Includes simplifications such as applying the saturation velocity across the whole depletion layer, a constant activation energy, and treating the responsivity as a value at 1.31 µm | Our calculation |
| How far surface-illuminated detectors extend to faster generations; production values for dark current and dislocation density in each foundry's Ge photodetectors; and long-term reliability test results for Mitsubishi Electric's PD | Not stated because no primary source could be confirmed as of this article's research (September 2026). This article's own judgement | Not yet confirmed |
| Organising dark current into diffusion, generation-recombination and surface / sidewall leakage, and reading an activation energy close to the bandgap as a certificate of quality; reading the choice to refill sidewalls with crystal as a design philosophy that avoids interfaces; reading thermal-mismatch strain as a function, and the framing that bad and good effects coexist at one heterointerface; and how the figures are drawn (such as bringing light in sideways from the waveguide) | Our own framing and commentary based on the published material. Not views expressed by the companies or institutions | Commentary |
| That Figs. 1 to 4, 6 and 7 are explanatory drawings rather than real cross-sections or measured data, and that the hero image and Fig. 5 are AI-generated images | Our note | Commentary |
Last updated 26 September 2026. Sources are limited to primary material (company technical material, technical reports and official announcements, official articles from research institutes, and peer-reviewed papers). Because the article includes numerical conversions and materials-design readings, those are marked as “Our calculation” or “Commentary” and kept separate from sourced fact. Mitsubishi Electric's values are evaluation results for a development chip, not transceiver performance. How surface-illuminated detectors extend to faster generations, production values for foundry Ge photodetectors, and long-term reliability test results are not stated here because no published primary source could be confirmed. All figures are explanatory concept graphics. Figs. 1 to 4, 6 and 7 are vector drawings, and the hero image and Fig. 5 are AI-generated images; none of them shows a real cross-section photograph, micrograph or physical product.