TECHNOLOGY EXPLAINER
Semiconductor Lasers
— the light source of optical communications is decided by crystal, facet and heat
Every signal in optical communications starts at a semiconductor laser somewhere. In the data centre, InP-based DFB lasers and EMLs have handled the longer distances and GaAs-based VCSELs the shorter ones. Each is a block of crystal a few to a few hundred micrometres across, yet its performance is set by substrate dislocations, compositions made differently from place to place, an aperture formed by oxidation, and the facet the light leaves through. And much of the electricity never becomes light at all. It becomes heat.
- Semiconductor lasers in three points
- Wavelength is set by the material — InP and GaAs
- Edge emitters: DFB lasers and EMLs
- Our calculation: the bandwidth a 200G EML needs
- Surface emitters: the VCSEL and its oxide aperture
- A materials engineer's view (1): a VCSEL is set by oxidation and temperature
- External light sources and efficiency — 90% of the electricity becomes heat
- A materials engineer's view (2): reliability lives at facets and interfaces
- 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 targets in a standard, evaluation results published by a company (and whether for a development or production part), and values from papers.
1. Semiconductor lasers in three points
A semiconductor laser combines a crystal that emits light when current flows (the gain section) with a structure that sends the light back and forth and lets only certain wavelengths build up (the cavity). Smit and colleagues at TU Eindhoven describe a laser as a composite building block made of a gain section and a cavity, and note that the cavity can be formed by a ring, a Fabry-Perot arrangement (two mirrors) or a Bragg gratingSourced.
- DFB laser: an edge-emitting laser with a grating built in alongside the light-emitting layer, so that it lases at the single wavelength set by the grating period (our explanation)
- EML: a DFB laser with an electro-absorption (EA) modulator integrated on the same chip. Mitsubishi Electric explains that it monolithically integrates the LD and EA modulator, which have different structures, on an InP substrateSourced
- VCSEL: a surface emitter that sends light out perpendicular to the substrate. Michalzik and colleagues at the University of Ulm note that a VCSEL in the 850 to 980 nm band needs about 8 µm of epitaxial layers, of which the active layer is just a few quantum wells tens of nanometres thickSourced
Larger InP substrates and heterogeneous integration of InP onto silicon were covered in our explainer on photonic integrated circuits and indium phosphide, and the idea of an external light source placed outside the package (ELSFP) in our explainer on co-packaged optics. Modulation formats (PAM4 and so on) are covered in our explainer on optical modulation and modulation formats, and wavelength plans such as CWDM in our explainer on wavelength multiplexing and microrings. This article covers the structure, materials and reliability of the laser chip itself.
2. Wavelength is set by the material — InP and GaAs
The wavelength a laser emits is set first of all by the material of the light-emitting layer (its bandgap). Baier and colleagues at Fraunhofer HHI note that InP and the lattice-matched InGaAsP and InGaAlAs have a natural operating wavelength range of about 1300 to 1650 nm, which fits the optimum range for optical fibre wellSourced. The VCSELs Michalzik and colleagues discuss, by contrast, sit in the 850 to 980 nm band, with mirrors made of GaAs-AlAs multilayersSourced.
| Substrate (Sumitomo Electric standard specifications) | Growth method, dopant | Diameter | EPD (average) |
|---|---|---|---|
| GaAs for lasers (n-type) | VB, Si | 76 / 100 / 150 mm | ≤ 1×10² (76 mm), ≤ 5×10² (100 and 150 mm) cm⁻² |
| InP (n-type) | VB, S | 50 / 76 / 100 / 150 mm | ≤ 5×10² (50 mm), ≤ 5×10³ (76 and 100 mm), ≤ 1×10⁴ (150 mm) cm⁻² |
Sourced (Sumitomo Electric Industries, “GaAs substrates: substrates for lasers” [Reference 5] and “InP substrates: n-type substrates” [Reference 6]). EPD (etch pit density) is a proxy for dislocations. The GaAs laser substrates can also be ordered with orientations tilted 2° to 15° off (100).
When Coherent announced its 6-inch InP manufacturing capability (March 2024), it cited as background its “years of investment and operating experience in high-volume VCSEL array manufacturing for mobile handsets”Sourced. The picture is one in which experience of mass-producing GaAs VCSELs has become a foothold for moving InP to larger wafers (our commentary).
3. Edge emitters: DFB lasers and EMLs
An edge-emitting laser sends its light out of the facet (a cleaved crystal plane) at the end of an elongated chip. In Smit and colleagues' scheme, a Fabry-Perot laser uses the cleaved facets as mirrors, while a DFB laser uses a Bragg grating as its cavitySourced.
Why add a separate modulator?
Direct modulation (DML), switching the laser current itself on and off, is the simplest approach, but Baier and colleagues note that the modulation rate of directly modulated lasers is practically restricted to the 25/28 Gb/s level, whereas an EA modulator, which changes absorption through the quantum-confined Stark effect (QCSE), can go much fasterSourced. The laser is kept shining at constant power and only an optical “shutter” is opened and closed at high speed — that is the idea behind the EML (our commentary).
The laser and the modulator want different crystals
Baier and colleagues divide EML integration schemes into three: butt-joint coupling, the identical layer structure, and waveguide interconnectionSourced.
- Butt joint: the laser and the modulator can be optimised separately — for example InGaAsP quantum wells for the laser and InGaAlAs for the modulator. InGaAlAs sweeps carriers out quickly when switching from absorbing to transparent, which favours ultra-high-speed modulation, while InGaAsP laser structures are known for “uncritical reliability behaviour”. Fabrication is more demanding, though, and the interface can reflect laser light back. Several commercial manufacturers use this approachSourced
- Identical layer structure: simple and cheap to make. But because the band edge is the same throughout, the DFB lasing wavelength has to be placed (through the grating period) on the long-wavelength side of the gain spectrum, so more drive current is needed for the same optical outputSourced
The problem seen in our explainer on photonic integrated circuits and indium phosphide — making different crystal compositions in different places — turns up unchanged even in an EML, which has only two components (our commentary).
A company example: a 200 Gbps-per-wavelength EML
A 2025 Mitsubishi Electric Technical Report sets out the design and evaluation results of an EML running at 200 Gbps per wavelengthSourced. The company has announced that it has been in volume production since April 2024 with this “200 Gbps (112 Gbaud PAM4) EML chip”Sourced.
- Structure: a buried-heterostructure DFB-LD for high optical power, and an EA modulator with a high-mesa waveguide for high extinction ratio and fast operation. The two differently structured parts are monolithically integrated on an InP substrate in what the company calls a “hybrid waveguide structure”. A spot-size converter (SSC) is integrated beyond the modulatorSourced
- The trade-off: higher speed needs a shorter EA modulator to cut capacitance, but a shorter one has a lower extinction ratio. A high-mesa structure raises the fraction of light confined in the absorption layer, so even a short modulator can reach a high extinction ratioSourced
- Wavelengths: a line-up of eight wavelengths at 10 nm spacing from 1,271 to 1,341 nm (CWDM8)Sourced
| Item | Target specification (the report cites IEEE P802.3dj) | Measured (Mitsubishi Electric) |
|---|---|---|
| 3 dB cutoff frequency | ≥ 60 GHz | 60 to 76 GHz |
| OMA (optical modulation amplitude) | ≥ 6.4 dBm | 8.2 to 9.9 dBm |
| Dynamic extinction ratio | ≥ 3.5 dB | 4.8 to 6.0 dB |
| TDECQ | ≤ 3.4 dB | 1.3 to 2.5 dB |
Sourced (Mitsubishi Electric Technical Report Vol. 99 No. 3, 2025 [Reference 1], Table 1). Measured at a chip temperature of 55 °C and LD current of 100 mA, with RF probes on a submount. These are results for the chip alone, not transceiver performance. The target column gives the values the report attributes to IEEE P802.3dj.
4. Our calculation: the bandwidth a 200G EML needs
The Mitsubishi Electric report walks step by step from 200 Gbps to the bandwidth requiredSourced. FEC (forward error correction) overhead brings the effective bit rate to 226.875 Gbps; PAM4 carries two bits per symbol, so the baud rate is 113.4375 Gbaud; and the 3 dB cutoff frequency needed is half of that, about 60 GHz. Against the 35 GHz a 100 Gbps EML needed, the report says the response speed has to be roughly doubledSourced.
5. Surface emitters: the VCSEL and its oxide aperture
A VCSEL (vertical-cavity surface-emitting laser) sandwiches its layers between mirrors above and below (Bragg reflectors), sending light back and forth perpendicular to the substrate. According to Michalzik and colleagues, each mirror is a stack of alternating layers of different refractive index, each a quarter-wavelength in optical thickness, and the crystal is grown by MOCVD or MBESourced.
For confining current to a small area, Michalzik and colleagues list mesa etching, making regions highly resistive by ion implantation, and selective lateral oxidation of Al-rich layers (Al₀.₉₈Ga₀.₀₂As or AlAs)Sourced. They compare them: proton implantation has been perfected as a technique for commercial VCSELs with good manufacturability and reliability, while selective oxidation, with its low optical loss in the cavity, has produced the best-performing devicesSourced.
- Oxidation conditions: an AlAs layer about 30 nm thick is oxidised from the side in a water-vapour atmosphere at 400 °C. The lateral oxidation rate is in the 1 µm per minute range, and mesa etching is requiredSourced
- Aperture size: the mesa is made a few tens of micrometres larger than the target active diameter, and mesa size and oxidation time let the active diameter be set anywhere from 1 µm to over 100 µmSourced
- Transverse modes: as a rule of thumb, devices lase in a single transverse mode up to active diameters of about 4 µm; larger ones also lase in higher-order modesSourced
6. A materials engineer's view (1): a VCSEL is set by oxidation and temperature
The most important dimension in a VCSEL — the diameter of the aperture that light and current pass through — is set not by lithography but by how far a semiconductor layer has oxidised in from the outside.
Our calculation: using Michalzik et al.'s figure of about 1 µm per minuteOur calculation,
- to leave a 6 µm aperture in a 30 µm mesa, oxidation must run (30 − 6) ÷ 2 = 12 µm in from each side → about 12 minutes
- if the oxidation rate is off by 10%, the oxidation depth is off by 1.2 µm, and the aperture diameter changes by about 2.4 µm (about 40% of 6 µm)
Assumptions and limits: the 30 µm mesa and 6 µm aperture are examples chosen by this article, and the oxidation rate is assumed constant. The real rate varies with layer composition, thickness and temperature, and Michalzik et al. give it only as an order of magnitude.
The reading is that materials and process variables — Al composition, layer thickness, and the uniformity of temperature and water vapour in the furnace — turn directly into device characteristics such as threshold current and transverse modes. And the oxidised layer stays right beside the active layer as an insulator whose volume and stress state have changed (our commentary).
Temperature pulls resonance and gain apart
Michalzik and colleagues explain that because a VCSEL's cavity is so short, its lasing wavelength is set by the cavity resonance rather than by the gain peak. For VCSELs in the 800 to 1000 nm range they give a shift of the resonance wavelength of about 0.07 nm/K (mainly from the change in refractive index, about a tenth from thermal expansion) and a shift of the quantum-well gain peak of about 0.32 nm/K (mainly from bandgap shrinkage)Sourced.
The temperature stability of a VCSEL's wavelength is set by a material constant — the temperature coefficient of refractive index of the Bragg reflectors and cavity — and moves on a different number from the temperature behaviour of the light-emitting layer. The difference between two material constants sets the operating temperature range, which makes this a device where the materials side has a great deal of room to shape the design (our commentary).
7. External light sources and efficiency — 90% of the electricity becomes heat
Silicon photonics cannot emit light, so the light source has to be supplied separately (see our explainer on silicon photonics). Shekhar and colleagues' 2024 Nature Communications paper describes as a practical solution hybrid integration that mounts, together with the silicon photonic chip, pre-tested sub-millimetre DFB lasers made for data centres in the millions at low cost and high yieldSourced. Coherent, too, lists “high-power CW lasers for silicon photonics applications” among the products it is qualifying on 6-inch InPSourced. The idea of ELSFP, which moves the light source outside the package, was covered in our explainer on co-packaged optics.
This is where efficiency starts to matter. Shekhar and colleagues write that most lasers in the C, L and O bands have a wall-plug efficiency (WPE), the conversion of electrical power into light, of only about 10%Sourced.
Assumption: WPE = 10% (taking Shekhar et al.'s “about 10%” as it stands)Our calculation
- Electrical power needed for 100 mW of light = 100 ÷ 0.10 = 1 W
- Of that, the part that becomes heat = 1 − 0.1 = 0.9 W
- Line up eight of them and the heat is about 7.2 W
Assumptions and limits: 100 mW and eight lasers are examples chosen by this article, not values for any particular product. Real WPE varies widely with device, temperature and output power.
Baier and colleagues note that EMLs are still commonly operated under thermoelectric (Peltier) cooler (TEC) control, typically at a heat-sink temperature of 40 to 50 °CSourced. Mitsubishi Electric's evaluation condition, too, is a chip temperature of 55 °CSourced. A laser is a component that gives off heat while being held at a constant temperature, and where that heat goes is the job of materials outside the chip — the submount, the bonding material, the package (our commentary).
8. A materials engineer's view (2): reliability lives at facets and interfaces
The reliability of lasers for optical communications is checked against a common industry yardstick. Telcordia's GR-468 (Issue 2, September 2004) sets out generic reliability assurance requirements (qualification testing, accelerated ageing tests, lot-to-lot controls and so on) for optoelectronic devices used in telecommunications, including lasers, LEDs, photodetectors and modulatorsSourced.
On 24 March 2025 OpenLight announced that it had completed GR-468 qualification for lasers, EA modulators and photodetectors with InP integrated on silicon, built on Tower Semiconductor's PH18DA processSourced. According to the company, qualification requires 2,000 hours of long-term reliability testing including HTOL (high-temperature operating life), damp heat storage, temperature cycling, temperature and humidity bias, and electrostatic discharge (human body model), and two lots exceeded 5,000 hours and another exceeded 15,000 hours with zero failuresSourced (as published by the company).
In the same announcement OpenLight claims that because its lasers are integrated into silicon without a laser facet, there is no risk of catastrophic optical mirror damage (COMD), a common failure mode in traditional lasersSourced (the company's claim). It also says the lasers are made with a single epitaxial growth with no regrowth, grating or blocking junctions, and that over 99% of the InP epitaxy is transferred to the silicon wafer without defectsSourced (the company's claim).
On VCSELs, Michalzik and colleagues explain that unlike edge emitters, they can be run up to their maximum output power without concern, because power densities stay in the lower kW/cm² range and cannot induce optical damage to the semiconductor material or the laser facetSourced.
Put these two statements side by side and a common thread appears (our framing).
- Facets: in an edge emitter, light passes at high density through a cleaved crystal plane. The condition of that surface and the light density are among the factors that set lifetime
- Interfaces: regrowth and butt joints create interfaces inside the crystal. OpenLight citing “no regrowth” as an advantage is the other side of the same coin
- Composition: Baier and colleagues write that InGaAsP laser structures are known for their “uncritical reliability behaviour”, so the choice of composition is itself a reliability variableSourced
Shekhar and colleagues note that heterogeneously integrated lasers need care over thermal isolation of the light-emitting region by the buried oxide (BOX) and mismatch in thermal expansion coefficients, and that adding redundant lasers can improve the failure rate (FIT)Sourced. Surfaces the light passes through, interfaces inside the crystal, and the escape route for heat — almost everything about laser reliability can be described in the language of materials and bonding (our commentary).
9. What is still hard
(1) Low efficiency, and heat
Shekhar and colleagues describe the WPE of about 10% as a metric that needs more focused researchSourced. Which technologies will improve it, and by how much, is not settled within the scope of this article's researchNot yet confirmed.
(2) At the next speed, balancing bandwidth and extinction ratio gets harder still
As the Mitsubishi Electric report shows, even at 200 Gbps the trade-off between modulator length (capacitance) and extinction ratio had to be overcome through structureSourced. How far EMLs can go in the 400 Gbps-per-wavelength generation could not be confirmed in the primary sources this article consultedNot yet confirmed.
(3) Much of the reliability data is self-published by companies
OpenLight's zero-failure hours and its “over 99%” transfer are figures published by the company, not the results of third-party verification. The full text of GR-468 is sold commercially, and this article limits itself to what can be confirmed on the official summary page. Quantitative data on laser degradation mechanisms (degradation of facets or crystal defects) could not be confirmed in the primary sources within the scope of this article, and are not given.
- Lasers for the optical fibre band (about 1300 to 1650 nm) are InP-based; VCSELs at 850 to 980 nm are GaAs-basedSourced
- On the same 100 mm substrate, the upper EPD limit for InP is 10 times that of GaAs (for lasers) (Sumitomo Electric specifications)Our calculation
- In an EML the laser and the modulator want different crystals — butt joints that pair InGaAsP with InGaAlAs are in commercial useSourced
- A VCSEL's aperture is set by oxidation depth, at around 1 µm per minute. A 10% error changes a 6 µm aperture by about 40%Our calculation
- Laser WPE is about 10%, so every 100 mW of light brings about 0.9 W of heatOur calculation
- Reliability comes down to facets, interfaces and the escape route for heat — problems of materials and bonding (our commentary)
10. Glossary
- DFB laser
- Distributed feedback laser. A grating built into the laser makes it lase at a single wavelength.
- Fabry-Perot (FP) laser
- A laser that uses the cleaved facets at both ends as mirrors. Tends to lase at several wavelengths.
- EML
- Electro-absorption modulated laser. A DFB laser and EA modulator on one chip.
- EA modulator
- A modulator that switches light on and off by changing the strength of absorption with an electric field.
- QCSE
- Quantum-confined Stark effect. The shift of the absorption edge when a field is applied to a quantum well.
- DML
- Directly modulated laser. The signal is imposed by varying the laser current itself.
- VCSEL
- Vertical-cavity surface-emitting laser. Emits light perpendicular to the substrate.
- Bragg reflector (DBR)
- A multilayer mirror of alternating layers of different refractive index, each a quarter-wavelength thick.
- Selective oxidation (oxide confinement)
- Oxidising only an Al-rich layer from the side to turn it into an insulator, confining current and light to the centre.
- Transverse mode
- The shape of the light's distribution across the beam. A single transverse mode is easier to handle.
- Buried heterostructure
- A structure in which both sides of the active layer are filled with another crystal to confine current and light.
- High mesa
- A structure with both sides of the waveguide deeply etched. Confines light strongly.
- SSC
- Spot-size converter. Expands the small on-chip light spot to a size that matches the fibre.
- CWDM8
- Eight wavelengths at 10 nm spacing from 1,271 to 1,341 nm (the example in the Mitsubishi Electric report).
- PAM4
- A modulation format that sends two bits at a time using four intensity levels.
- TDECQ
- A metric for the quality of the transmitted waveform. Smaller is better.
- OMA
- Optical modulation amplitude. The difference in optical power between a “1” and a “0”.
- WPE
- Wall-plug efficiency. The fraction of electrical input power that becomes light.
- TEC
- Thermoelectric cooler, a Peltier-based temperature controller that holds the laser at constant temperature.
- COMD
- Catastrophic optical mirror damage. A failure mode in which intense light damages the facet.
- GR-468
- Telcordia's generic reliability assurance requirements for optoelectronic devices used in telecommunications.
- HTOL
- High-temperature operating life test. Accelerates degradation by running devices continuously at high temperature.
11. References (primary sources)
- 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
- M. Baier, N. Grote, M. Moehrle et al. (Fraunhofer HHI) “Integrated transmitter devices on InP exploiting electro-absorption modulation”, PhotoniX 1:4 (2020, DOI: 10.1186/s43074-020-0003-4) — photonix.springeropen.com
- M. Smit et al. (TU Eindhoven and others) “An introduction to InP-based generic integration technology” (revised version of the 2014 paper, published by JePPIX as chapter 1 of a book, PDF) — jeppix.eu
- R. Michalzik, K. J. Ebeling (University of Ulm) “Operating Principles of VCSELs” (book chapter made available by the university, PDF) — uni-ulm.de
- Sumitomo Electric Industries “GaAs substrates: substrates for lasers”, standard specifications (in Japanese) — sei.co.jp
- Sumitomo Electric Industries “InP substrates: n-type substrates”, standard specifications (in Japanese) — sei.co.jp
- 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
- Coherent “World's First 6-inch InP Scalable Wafer Fabs…”, 25 March 2024 — coherent.com
- Telcordia (Ericsson) “GR-468: Generic Reliability Assurance Requirements for Optoelectronic Devices Used in Telecommunications Equipment”, document summary page — telecom-info.njdepot.ericsson.net
- OpenLight “OpenLight Achieves Successful Completion of Telcordia GR-468 Qualification for Silicon Photonics Components”, 24 March 2025 — openlightphotonics.com
- 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
12. Claim-to-source audit
| Claim in the text | Basis | Label |
|---|---|---|
| That the 200 Gbps EML monolithically integrates a buried-heterostructure DFB-LD and a high-mesa EA modulator on an InP substrate in a “hybrid waveguide structure”, with an integrated SSC; 226.875 Gbps with FEC, 113.4375 Gbaud with PAM4, a required 3 dB cutoff of about 60 GHz against 35 GHz for a 100 Gbps EML, and the need to roughly double response speed; the trade-off that a shorter modulator lowers capacitance but also extinction ratio, and the use of a high-mesa structure to raise confinement in the absorption layer; eight wavelengths at 10 nm spacing from 1,271 to 1,341 nm; the Table 1 targets (citing IEEE P802.3dj) and results (60 to 76 GHz, 8.2 to 9.9 dBm, 4.8 to 6.0 dB, 1.3 to 2.5 dB); and the evaluation conditions (55 °C, 100 mA, RF probes on a submount) | Mitsubishi Electric Technical Report Vol. 99 No. 3 (2025), Reference 1 https://www.giho.mitsubishielectric.co.jp/giho/pdf/2025/2503104.pdf | Sourced |
| That InP and InGaAsP / InGaAlAs have a natural operating wavelength range of about 1300 to 1650 nm; that directly modulated lasers are practically restricted to the 25/28 Gb/s level while EA modulators using QCSE can go much faster; the EML integration schemes (butt joint, identical layer, waveguide interconnection); that butt joints allow InGaAsP for the laser and InGaAlAs for the modulator, that InGaAlAs favours carrier sweep-out, that InGaAsP is known for “uncritical reliability behaviour”, that fabrication is demanding and the interface may reflect light, and that several commercial manufacturers use the approach; that the identical layer structure is simple but places the DFB wavelength on the long-wavelength side and raises drive current; and that EMLs are commonly operated under TEC control at a heat-sink temperature of about 40 to 50 °C | Baier et al., PhotoniX 1:4 (2020), Reference 2 https://photonix.springeropen.com/articles/10.1186/s43074-020-0003-4 | Sourced |
| That a laser is a composite building block of a gain section and a cavity, and that the cavity can be a ring, an FP arrangement (cleaved facets) or a Bragg grating | Smit et al. (book chapter 1 published by JePPIX), Reference 3 https://www.jeppix.eu/wp-content/uploads/2025/03/Chapter-1-Introduction.pdf | Sourced |
| That VCSELs in the 850 to 980 nm band need about 8 µm of epitaxial layers, with an active layer of a few quantum wells tens of nanometres thick, grown by MOCVD or MBE; that the mirrors are alternating layers of quarter-wavelength optical thickness (GaAs-AlAs as an example); the methods of current confinement (mesa, ion implantation, selective oxidation of Al₀.₉₈Ga₀.₀₂As or AlAs), with proton implantation commercially proven for manufacturability and reliability and selective oxidation giving low loss and the best performance; oxidation of an AlAs layer about 30 nm thick in water vapour at 400 °C at about 1 µm per minute laterally, active diameters from 1 µm to over 100 µm set by mesa size and oxidation time, and single transverse mode up to about 4 µm; that the lasing wavelength is set by the cavity resonance, at about 0.07 nm/K in the 800 to 1000 nm range (mainly refractive index, about a tenth thermal expansion), with the gain peak at about 0.32 nm/K (bandgap shrinkage), and calculations for different gain offsets; and that, unlike edge emitters, VCSELs can be run to maximum output because power densities stay in the lower kW/cm² range and cannot cause optical damage | Michalzik and Ebeling, “Operating Principles of VCSELs” (book chapter made available by the University of Ulm), Reference 4 https://www.uni-ulm.de/fileadmin/website_uni_ulm/iui.inst.140/Diverse/vcsel-chapter-ram_kje-d.pdf | Sourced |
| For GaAs laser substrates (n-type, VB, Si): diameters of 76 / 100 / 150 mm, EPD ≤ 1×10² (76 mm) and ≤ 5×10² (100 and 150 mm) cm⁻², and orientations 2° to 15° off (100) | Sumitomo Electric Industries, GaAs laser substrate standard specifications, Reference 5 https://sei.co.jp/sc/products/gaas/laser.html | Sourced |
| The diameters and EPD of n-type InP substrates (VB, S) (≤ 5×10² / ≤ 5×10³ / ≤ 5×10³ / ≤ 1×10⁴ cm⁻²) | Sumitomo Electric Industries, n-type InP substrate standard specifications, Reference 6 https://sei.co.jp/sc/products/inp/n.html | Sourced |
| That hybrid integration mounting pre-tested sub-millimetre DFB lasers, made in the millions, with silicon photonic chips is a practical solution; that most C, L and O band lasers have a WPE of about 10%, a metric needing more research; and that heterogeneously integrated lasers need care over thermal isolation by the BOX and CTE mismatch, and that redundant lasers can improve FIT | Shekhar et al., Nature Communications 15, 751 (2024), Reference 7 https://www.nature.com/articles/s41467-024-44750-0 | Sourced |
| That experience of high-volume VCSEL array manufacturing for mobile handsets was cited as background to 6-inch InP; and that the products being qualified on 6 inch include “high-power CW lasers for silicon photonics applications” | Coherent press release (25 March 2024), Reference 8 https://www.coherent.com/news/press-releases/worlds-first-6-inch-inp-scalable-wafer-fabs-paving-the-way-for-the-next-generation-of-lasers-for-ai-transceivers-and-6g-wireless-networks | Sourced |
| That GR-468 is Issue 2 of September 2004 and covers generic reliability assurance requirements (qualification testing, accelerated ageing tests, lot-to-lot controls and so on) for telecommunications optoelectronic devices including lasers, LEDs, photodetectors and modulators | Telcordia GR-468 document summary page, Reference 9 https://telecom-info.njdepot.ericsson.net/site-cgi/ido/docs.cgi?ID=SEARCH&DOCUMENT=GR-468 | Sourced |
| That on 24 March 2025 GR-468 qualification was completed for PH18DA-based lasers, EAMs and PDs with InP integrated on silicon; the 2,000 hours of testing (HTOL, damp heat storage, temperature cycling, temperature and humidity bias, ESD-HBM), with two lots exceeding 5,000 hours and one exceeding 15,000 hours with zero failures; the claim that facet-free integration removes the risk of COMD; and the claims of a single growth with no regrowth, grating or blocking junctions, and of over 99% of the InP epitaxy transferred without defects | OpenLight press release (24 March 2025). The figures and advantages are as published and claimed by the company, Reference 10 https://openlightphotonics.com/newsroom/openlight-achieves-successful-completion-of-telcordia-gr-468-qualification-for-silicon-photonics-components | Sourced |
| That the 200 Gbps (112 Gbaud PAM4) EML chip has been in volume production since April 2024 | Mitsubishi Electric news release (20 August 2024), Reference 11 https://www.mitsubishielectric.co.jp/ja/pr/2024/0820/ | Sourced |
| The factor of 10 between EPD limits (5×10³ ÷ 5×10²); 226.875 ÷ 200 = +13.4% and 60 ÷ 35 ≈ 1.7 times; the oxidation example (12 µm per side from a 30 µm mesa to a 6 µm aperture, about 12 minutes, and a change of about 2.4 µm, or about 40%, in diameter for a 10% rate error); a 3.5 nm resonance shift, 16 nm gain shift and 12.5 nm difference for a 50 K rise; and 1 W of electrical power and 0.9 W of heat for 100 mW of light at 10% WPE, and 7.2 W for eight lasers | Our calculation. The mesa and aperture diameters, 50 K, 100 mW and eight lasers are examples chosen by this article. Coefficients are assumed constant | Our calculation |
| The outlook for improving WPE; how far EMLs extend to the 400 Gbps-per-wavelength generation; and quantitative data on laser degradation mechanisms | 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 |
| The explanation that the DFB grating period selects the wavelength, and the figure's labelling of FP lasers as tending to lase at several wavelengths; describing the EML as a “shutter”; the picture of GaAs VCSEL volume experience as a foothold for larger InP wafers; the reading that a VCSEL's aperture is set by oxidation depth, so materials variables turn into device characteristics; the reading that the difference between two temperature coefficients sets the operating temperature range; the point that the escape route for heat is a job for materials outside the chip; organising reliability into facets, interfaces, composition and the escape route for heat; and how the figures are drawn | Our own framing and commentary based on the published material. Not views expressed by the companies or institutions | Commentary |
| That Figs. 1 to 5 are explanatory drawings rather than real cross-sections or measured data, and that the hero image and Fig. 6 are AI-generated images | Our note | Commentary |
Last updated 26 September 2026. Sources are limited to primary material (company technical reports, official announcements and product specifications, peer-reviewed papers and reviews, book chapters made available by universities, and official summaries of standards documents). 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 the chip alone, and OpenLight's reliability data are figures published by the company. The outlook for improving WPE, EMLs for the 400 Gbps-per-wavelength generation, and quantitative data on laser degradation mechanisms are not stated here because no published primary source could be confirmed. All figures are explanatory concept graphics. Figs. 1 to 5 are vector drawings, and the hero image and Fig. 6 are AI-generated images; none of them shows a real cross-section photograph, micrograph or physical product.