Why Is the VCSEL the Workhorse Laser of Short-Reach Data Centers?

In short-reach data-center links (intra-rack, inter-rack, <100 m), the VCSEL (Vertical-Cavity Surface-Emitting Laser) is the workhorse of the SR family — 100G/400G/800G SR4/SR8 are almost all 850 nm VCSELs^[4]. It is also the core laser in iPhone Face ID, many automotive/robotic LiDARs, and optical mice.

The VCSEL differs fundamentally from the FPLD/DFB/EML lasers in the laser-classification article: emission direction. Edge-emitters radiate from the chip side (cleaved facet); the VCSEL radiates perpendicularly from the chip surface. That difference drives its cost, testability, and arrayability.

This article answers one question: why is the VCSEL the workhorse laser of short-reach data centers? The answer: the vertical cavity (two DBR mirrors) enables wafer-level testing and dense arrays, and combined with the 850 nm multimode-fiber and silicon-photodetector ecosystem, it becomes the most economical short-reach laser.

1. The vertical cavity: how it differs from an edge-emitter

A conventional edge-emitter has a “horizontal” cavity: the two cleaved facets are the mirrors, and light exits in the chip plane. The cavity is the full chip length (hundreds of μm), so the gain is large and the ~30% Fresnel reflectivity of the cleaved facets is enough.

The VCSEL turns the cavity vertical: the cavity is formed by two DBR (Distributed Bragg Reflector) mirrors parallel to the wafer surface, with the light-emitting active region sandwiched between them, and light exits perpendicular to the chip surface^[2].

A DBR mirror is a stack of alternating high/low-refractive-index thin layers, each λ/4 thick in the material — that thickness makes the reflections from each interface add in phase, forming a high-reflection band. Because the refractive-index contrast of semiconductor alloys is small, about 30 layer pairs are needed to reach >99% reflectivity; the bottom (high) mirror exceeds 99.9%, and the top (output coupler) mirror is 99.5–99.9%^[2]. Why such high reflectivity? The VCSEL cavity is only 1–2 μm long (an edge-emitter is hundreds of μm), so the short-cavity gain is small and only mirrors above 99% reflectivity can compensate and reach threshold. The top mirror “leaks” slightly more than the bottom one — that leakage is the output coupling that lets the laser light out; higher output coupling raises the threshold current, so the top mirror’s reflectivity is a trade-off between threshold and output power.

The DBR mirrors also serve as electrodes: the top mirror is p-doped, the bottom n-doped, forming the diode junction; current flows along the optical axis (top to bottom).

Between the mirrors is the active region, typically several quantum wells (nanometer-scale layers); carriers confined in the thin layers let the laser reach threshold at low current. VCSEL threshold currents are commonly below 1 mA^[2] — the combined result of the short cavity, quantum wells, and good optical confinement.

Fabrication uses mesa etching to carve a cylindrical pedestal in the chip surface, defining the emission aperture (typically a few μm in diameter) — the aperture sets the beam size and mode. Current confinement has two approaches: oxide confinement (air-bridge) — high-Al AlGaAs forms an oxide layer that confines current to the aperture, the current mainstream; and ion implantation — the older approach, implanting ions to damage the lattice outside the aperture^[1].

Vertical emission also brings an optical advantage: the beam is round and symmetric (edge-emitters are elliptical with astigmatism), with small divergence that a single lens can collimate, and a larger alignment tolerance when coupling into multimode fiber.

Figure 1 | The vertical-cavity structure of a VCSEL. Click for full size.

2. Why vertical emission is cheap and reliable

The VCSEL’s biggest advantage is not optical — it is manufacturing and test:

• Wafer-level testing: edge-emitters can only be tested after cleaving into individual chips, so a bad die wastes the whole process; the VCSEL emits from the surface, so every device can be tested directly on the wafer, bad parts are culled early, and cleaving/packaging cost is never spent on bad parts^[1]
• Low cost: no cleaving, no complex packaging alignment
• Arrayable: vertical emission lets 8–16 VCSELs be densely packed on one chip, matched to ribbon coupling into multimode fiber^[1]. 400GBASE-SR8 is exactly 8 × 50G VCSELs in parallel^[4]

Low threshold and low heat also mean high reliability and long life, suitable for consumer products such as phones and optical mice.

There is a hidden advantage too: wavelength stability. The VCSEL wavelength temperature coefficient is <0.1 nm/°C, more stable than edge-emitters, so short-reach applications need no complex temperature control^[2].

3. Wavelengths and material systems

The VCSEL wavelength is set by the material system:

• GaAs/AlGaAs system covers 750–980 nm (850 nm most common): the lattice constants of GaAs and AlGaAs vary little with composition, so many lattice-matched epitaxial layers can be grown — the key that makes the DBR mirror feasible^[2]
• InP system (InAlGaAsP on InP) reaches 1300–2000 nm, used for sensing and free space^[2]
• 940 nm: sunlight background is stronger at 850 nm; 940 nm reduces background-light interference, so outdoor LiDAR/3D sensing prefers it^[1]
• 1060 nm: an industry trend for longer-reach MMF (not a mainstream standard), with slightly better loss and dispersion in multimode fiber than 850 nm, letting the same OM4 reach farther

850 nm is the data-center mainstream: the SR family (100GBASE-SR4/SR10, 400GBASE-SR8) all use 850 nm VCSELs with multimode fiber (OM3/OM4) to 100 m^[4]. OM3’s bandwidth is about 500 MHz·km and OM4’s about 4700 MHz·km — the higher the bandwidth, the farther the same rate reaches (OM4 pushes 400G to 150 m)^[3]. 850 nm also sits in the high-sensitivity region of silicon photodetectors, keeping both transmitter and receiver cheap; 940 nm is its co-mainstream for LiDAR/3D sensing (silicon detectors still perform well at 940 nm).

4. The limit: why the VCSEL stays short-reach

The VCSEL’s strengths (low cost, arrayability) are also its limits:

• Multimode fiber: the MMF “bandwidth × distance” product is lower than single-mode, capping reach (OM4 about 150 m at 400G)^[3]
• Mode noise: VCSELs can exhibit longitudinal/transverse mode hopping, which accumulates into noise over long reaches — together with the bandwidth × distance product it forms the short-reach constraint
• Wavelength: 850 nm loss in single-mode fiber is about 2.5 dB/km (versus 0.2 dB/km at 1550 nm), unsuitable for long reach^[3]

So the VCSEL and the DFB/EML are complementary, not competing: the VCSEL rules 850 nm multimode short reach; the DFB/EML rule 1310/1550 nm single-mode mid and long reach. At the same 850 nm, the VCSEL costs far less than a DFB — wafer-level testing and no cleaving are the main reasons. Why doesn’t the VCSEL do 1310/1550 nm single-mode? The GaAs-system DBR loses efficiency above 1310 nm, and InP-system VCSEL fabrication is harder — single-mode mid/long reach remains the DFB/EML domain^[2].

As data-center rates move toward 800G/1.6T, VCSEL per-lane rates are rising (25G→50G→100G PAM4) — 100G PAM4 raises the modulation-bandwidth requirement, which is the current technical challenge. But the 850 nm multimode distance ceiling (bandwidth × distance product) is a physical limit — longer reach still goes to 1310/1550 nm single-mode DFB/EML.

Back to the opening question: the VCSEL is the short-reach data-center workhorse because the vertical cavity (two DBR mirrors) enables wafer-level testing and dense arrays, and combined with the 850 nm multimode-fiber and silicon-photodetector ecosystem it becomes the most economical short-reach laser. Scope: 850 nm MMF under 100–150 m; 940 nm for outdoor sensing. Unresolved limits: the 850 nm multimode bandwidth × distance ceiling and the 100G PAM4 VCSEL transition are still evolving.