Structural architecture: how the two laser families differ at the device level

VCSELs and edge-emitting lasers differ in a single foundational dimension — the direction in which light leaves the device — and that difference cascades into nearly every engineering trade-off that follows. VCSELs emit light perpendicular to the wafer surface, sandwiching a thin active region between two distributed Bragg reflector (DBR) mirror stacks. Edge-emitting lasers (EELs) emit light parallel to the wafer substrate from cleaved or etched facets.

This geometry determines testability — and therefore cost. Because VCSELs emit upward through the wafer surface, entire wafers can be probed and tested before dicing, allowing thousands of devices to be characterised simultaneously. Edge-emitting lasers cannot be tested on-wafer: the active facets are only exposed after cleaving, requiring individual device preparation and testing. According to PatSnap’s patent intelligence analysis, Mellanox Technologies explicitly identified this on-wafer testability as a decisive cost differentiator in its 2018 VCSEL modulation speed patents.

The longer cavity inherent to EEL geometry provides higher single-pass gain. This enables higher output optical power and operation across a broader range of wavelengths — particularly in the near-infrared and telecom bands (1310 nm, 1550 nm) where GaAs-based VCSELs have historically struggled to compete. Apple’s 2023 patent on an integrated edge-generated vertical-emitting laser directly acknowledges that edge-emitting lasers can operate reliably in near-infrared wavelength ranges where VCSELs may not perform acceptably, and that EELs can deliver greater optical power or reliability for sensing applications such as proximity detection.

Beam shape is another consequential structural difference. VCSELs produce a circular output beam, making them efficient couplers to multimode optical fibre and — with appropriate optics — to silicon photonic waveguides. EELs produce an elliptical beam from their rectangular facets, requiring aspheric or cylindrical optics for efficient fibre coupling. For 2D array integration, VCSELs have a clear advantage: native two-dimensional arrays are straightforward to fabricate on-wafer. EEL arrays are naturally one-dimensional, and two-dimensional configurations require complex packaging.

Modulation bandwidth: where VCSELs plateau and how engineers push past it

The ultimate modulation bandwidth of a standard 850 nm oxide-confined VCSEL is approximately 24–25 GHz — a ceiling set not by the mirrors or the gain material alone, but by the interplay of parasitic capacitance, photon lifetime, and current-induced self-heating. Research from Ioffe Institute established this limit quantitatively, finding that even reducing photon lifetime from 4 ps to 1 ps by adjusting mirror loss could not overcome the excess damping imposed by self-heating, with the optimum oxide aperture sitting near 4–6 µm.

Three engineering levers are used to push beyond this plateau. First, aperture optimisation: smaller oxide apertures reduce parasitic capacitance but increase resistance and thermal density. Mellanox Technologies’ patents from 2020 and 2024 specifically target aperture-ratio engineering to minimise parasitic capacitance for 50 Gb/s and above operation. Second, photon lifetime tuning: the Bimberg Chinese-German Center for Green Photonics identified optimisation of photon lifetime alongside a novel multi-hole oxidation aperture geometry as dual pathways to simultaneously reduce power consumption and increase bandwidth in short-wavelength VCSELs. Third, coupled-cavity designs: a hexagonal transverse-coupled-cavity VCSEL demonstrated a 3-dB roll-off modulation bandwidth of 45 GHz — five times greater than a conventional VCSEL on the same epiwafer — by harnessing the Vernier effect to increase aperture while maintaining single-mode operation.

“A hexagonal transverse-coupled-cavity VCSEL achieved a 3-dB modulation bandwidth of 45 GHz — five times greater than a conventional VCSEL fabricated on the same epiwafer — by harnessing the Vernier effect to increase aperture while maintaining single-mode operation.”

For longer-wavelength VCSELs, topology governs performance differently. Research from ITMO University on 1550 nm wafer-fused VCSELs showed that the double-mesa size directly controls parasitic capacitance and thereby the parasitic cutoff frequency. The smallest topology — S-type — achieved above 13 GHz modulation bandwidth by reducing parasitic capacitance of the reverse-biased p+n-junction region. This topology-performance relationship is specific to wafer-fused InP-based designs and does not directly translate to GaAs-based 850 nm devices, according to IEEE-published photonics research on heterogeneous laser integration.

Energy efficiency is a distinguishing VCSEL strength at the device level. Small oxide-aperture VCSELs operating at low bias currents achieve both energy efficiency and lower relative intensity noise (RIN), meeting the requirements of the 32G Fibre Channel standard, as confirmed by King Abdul-Aziz University research on 980 nm temperature-stable VCSELs. At the system level, however, energy efficiency comparisons depend heavily on the modulation format, driver ASIC architecture, and thermal management — all of which are active areas of patent activity from Mellanox, Trumpf Photonic Components, and AMS Sensors.

Where edge-emitting lasers hold the advantage

Edge-emitting lasers retain clear superiority in three domains: telecom-band wavelengths, high output power, and large-scale WDM multichannel integration. The longer cavity inherent to EEL geometry provides higher single-pass gain, enabling operation across the full telecom band — including 1310 nm and 1550 nm — where GaAs-based VCSELs cannot compete without wafer fusion or other heterogeneous integration techniques.

Waveguide engineering is central to EEL performance optimisation. Research from Qilu University of Technology reviewed the Coupled Large Optical Cavity (CLOC) approach, which exploits resonant optical coupling between waveguides to select high-order modes, as a cost-efficient solution for improving diode laser efficiency. Microsoft Technology Licensing pursued a multi-stripe EEL architecture with independently excitable lower-power and higher-power optical cavities on the same substrate, enabling dynamic range control — a function not easily replicated in standard VCSEL designs.

At the system integration scale, EEL-based multichannel arrays have demonstrated aggregate bandwidths that dwarf what individual VCSEL modules can achieve. A hybrid integrated light source developed by PETRA demonstrated over 10 Tbit/s aggregate bandwidth with 1,000 channels, exploiting a spot-size converter with SiOx slab layer for wide fabrication margin. Power uniformity across such large arrays reached a minimum standard deviation of 0.49 dB across 200 output ports — a level of consistency that is critical for WDM coherent systems where channel power imbalance directly limits reach and spectral efficiency. Standards bodies including ITU specify tight power uniformity requirements for WDM transmission systems, making this a practically significant metric.

For sensing applications at near-infrared wavelengths, Apple’s 2023 patent on an integrated edge-generated vertical-emitting laser describes a hybrid architecture that combines EEL emission direction with vertical output coupling — specifically to access wavelengths and power levels where VCSELs may not perform acceptably. This represents a clear use-case boundary: for wavelengths and power levels where VCSELs are insufficient, EELs remain the preferred choice, as noted by WIPO patent filings in the photonic sensing category.

Datacenter deployment: application domains and integration implementations

VCSELs dominate short-reach intra-datacenter interconnects operating over multimode fibre at 850 nm — a position established by cost, energy efficiency, and array integration advantages, and reinforced by a decade of system-level demonstrations. Mellanox demonstrated 80 Gb/s PAM-4 transmission over 500 m using a single-mode VCSEL module, and 64 Gb/s per lane in a full transceiver link experiment for 200 GbE short-reach intra-datacenter optical interconnects, employing single-mode, single-polarization VCSELs with a dedicated driver chip and linear receiver.

At the form-factor level, Hitachi’s VCSEL-based active optical cable demonstrated 25.78 Gbit/s × 4-channel error-free transmission over 100 m OM3 multimode fibre in a form factor 55% smaller than standard QSFP28, with error-free operation maintained at 70°C case temperature through optimised heat-dissipation structures. Thermal management is a recurring integration challenge: the Finisar integrated optical transceiver patent describes a VCSEL array on a laser diode substrate with a dual heat-sink architecture that thermally isolates the laser driving and photodiode driving circuitry — a design specifically addressing the thermal crosstalk that arises when high-speed driver electronics and optical emitters share a compact package.

For physics experiment environments requiring radiation hardness, 850 nm VCSEL-based transmitters operating at 25 Gbps were prototyped with the LOCld65 ASIC driver in the MTx+ module, demonstrating that VCSEL systems can meet specialised ruggedised deployment requirements beyond standard datacenter conditions. This breadth of VCSEL deployment — from hyperscale datacenters to high-energy physics experiments — reflects the technology’s maturity and manufacturing scalability, consistent with data published by ITU on global optical fibre deployment trends.

VCSELs are also being extended toward coherent applications. IHP Solutions presented a silicon photonic coherent transceiver in which a 1550 nm VCSEL serves as the transmitter in an external modulation configuration, achieving stable optical injection locking with direct phase modulation — claimed as the first such demonstration using two vertical-emitting sources. This extends the VCSEL application space into territory previously reserved for distributed feedback (DFB) edge-emitting lasers.

EEL-based approaches retain a stronghold in multichannel WDM interconnects and chip-to-chip links requiring telecom wavelengths. Bandgap Technology Corporation’s foundational 1993 patents established the architectural concept of three-dimensional OEIC stacks using VCSEL arrays, receivers, and logic monolithically integrated — a concept that has since been extended by EEL-based multichannel sources for ultra-high-bandwidth applications. The patent landscape shows a clear division: VCSEL innovation concentrates around bandwidth optimisation, driver ASIC co-design, and thermal management for intra-datacenter use; EEL innovation concentrates around waveguide engineering, multichannel WDM integration, and miniaturisation toward chip-scale active volumes.

Silicon photonics convergence and the road to 800 GbE and beyond

Both VCSEL and edge-emitting laser technologies are converging toward silicon photonics platforms — a trajectory driven by the need to reduce interconnect power density and increase integration density for co-packaged optics in next-generation switching systems. The convergence pathway differs between the two device families, but the destination is the same: photonic integrated circuits where laser sources are heterogeneously integrated with silicon waveguides, modulators, and photodetectors on a single substrate.

For VCSELs, University of Toronto demonstrated hybrid integration of an O-band VCSEL onto a silicon photonic chip via a grating coupler that simultaneously maintains single polarization emission and achieves -5 dB coupling efficiency. IHP Solutions’ silicon photonic coherent transceiver uses a 1550 nm VCSEL in an external modulation configuration with optical injection locking — a configuration that separates the laser source from the modulation function, enabling high-speed coherent signalling without requiring the VCSEL itself to achieve coherent-grade linewidth. Research from institutions including NIST has characterised the linewidth requirements for coherent optical systems, establishing the performance targets that VCSEL-silicon photonic integration must meet.

For edge-emitting lasers, the miniaturisation trajectory toward chip-scale integration is represented by NTT’s LEAP laser — an InP-based photonic crystal structure with an active volume of only 2.5 × 0.3 × 0.15 µm³ that achieves 25 Gbit/s at 10 fJ/bit. This represents an extreme reduction of edge-emitting cavity principles toward dimensions compatible with on-chip integration alongside silicon photonic components. The PETRA multichannel hybrid integrated light source used a spot-size converter with SiOx slab layer to couple EEL outputs into integrated waveguides with wide fabrication margin — a practical engineering solution for the beam shape mismatch between elliptical EEL outputs and silicon waveguide modes.

The immediate pressure driving this convergence is the 800 GbE and 1.6 TbE Ethernet standards. National Taiwan University’s 2022 review explicitly states that 850 nm VCSELs currently face constraints in meeting future 800 GbE and 1.6 TbE Ethernet standards, requiring new device designs and modulation formats. This is not a distant challenge: hyperscale datacenter operators are already deploying 400 GbE infrastructure, and the engineering decisions made in the current patent filing cycle will determine which laser architecture — or combination of architectures — powers the next generation of AI training cluster interconnects. The PatSnap analytics platform tracks these filing trends across all major assignees in real time.

The patent landscape reflects this pressure. Mellanox (now NVIDIA) holds multiple patents targeting 50 Gb/s and above VCSEL operation via aperture-ratio engineering. Trumpf Photonic Components holds active EP patents on VCSELs with multiple active layer structures connected via tunnel junctions for improved gain-switching behaviour. AMS Sensors holds active EP patents on high-speed VCSEL devices for short-haul digital communications. On the EEL side, Microsoft Technology Licensing’s multi-stripe architecture and NTT’s LEAP laser represent the frontier of edge-emitting miniaturisation. The competitive intensity of patent filings in both categories — reviewed across over 40 patents and publications in this analysis — signals that neither technology has reached a definitive endpoint.