The heart of an optical communication transmitter is the laser diode (LD) — without it, there is no light carrying data. The laser’s type sets the link’s rate, reach, and cost. But no single laser does everything: on the same 1310 nm single-mode fiber link, a 10G link and a 100G-per-lane link use different lasers.
This article answers one question: why does optical communication need three kinds of lasers — FPLD, DFB, and EML?
The main answer has two parts: cavity structure sets spectral purity (multimode vs. single-mode), and modulation method sets speed (direct current modulation vs. an integrated electro-absorption modulator). The three form a ladder of cost × speed × spectral purity, each matched to different links. The same 1310 nm wavelength appears in both 10G and 800G modules — what differs is the laser type (DFB vs. EML).
1. Cavity Structure Sets the Spectrum
The FPLD (Fabry–Pérot Laser Diode) has the simplest structure: the two ends of the p-i-n chip are cleaved — not sawed, but scored and snapped along the single crystal’s natural crystal plane, leaving an atomically flat facet that serves directly as a mirror. The Fresnel reflection at the semiconductor–air interface forms the two cavity mirrors — InP has a refractive index of about 3.2, giving a single-surface reflectivity of about 27% (computed from ((3.2−1)/(3.2+1))²). No extra mirror fabrication is needed, which makes the FPLD the cheapest laser diode.
The cavity is a Fabry–Pérot interferometer, and its spectrum has two problems:
- Multimode: several longitudinal modes (each one a distinct wavelength) oscillate at once, with a typical spacing of about 100 GHz[1]
- Mode hopping: as temperature changes, lasing jumps between adjacent longitudinal modes, making the wavelength–temperature curve discontinuous[1]. The mechanism: temperature shifts the gain peak, and the longitudinal mode closest to it takes over oscillation
The wavelength temperature coefficient is about 0.3 nm/K, dominated by the temperature dependence of the bandgap energy[1]. Positioning: the FPLD fits links that tolerate multimode light — short-reach low-rate links, CATV, sensing[1].
The DFB (Distributed Feedback laser) etches a Bragg grating into the active region: the periodic structure reflects only one wavelength (Bragg condition λB = 2·neff·Λ), providing wavelength-selective distributed feedback[2]. With a λ/4 phase shift in the middle of the grating, the DFB selects a single longitudinal mode stably: only the mode matching the Bragg wavelength receives enough feedback from the grating; all other modes are suppressed[2]
- Spectrum: single-mode, linewidth from a few MHz (telecom class) up to 100 MHz (unoptimized low-cost devices)[2]
- Wavelength drift: about 0.06 nm/K[5] — about one fifth of the FPLD
The DFB wavelength range covers 0.8–2.8 μm: 1550 nm dominates long haul (EDFA amplification), 1310 nm serves shorter distances[2].
Figure 1 | Feedback and modulation architectures. An EML integrates a DFB source with an EAM rather than introducing another laser-cavity type. Click for full size.
The consequence is direct: wavelength-division multiplexing (WDM) requires a stable wavelength per channel. DWDM packs dozens of channels into one fiber with a 100 GHz channel spacing (ITU grid) — about 0.8 nm at 1550 nm (λ²·Δν/c); a drifting channel encroaches on its neighbor and causes crosstalk. The FPLD’s drift and mode hopping disqualify it from WDM; the DFB is the standard DWDM channel source (a TEC holds the chip temperature and locks the wavelength to the ITU grid).
2. Modulation Method Sets the Speed
Direct modulation writes data by changing the drive current: carrier density change → gain change (intensity modulation) + refractive-index change (frequency modulation, i.e. chirp). The coupling between the two is set by the linewidth enhancement factor (α). Below 10G the chirp penalty is acceptable; at higher speeds chirp broadens the spectrum, and fiber dispersion converts the frequency sweep into pulse broadening, limiting rate × distance. The speed ceiling of direct modulation is not the laser’s own oscillation speed but the chirp penalty imposed by the fiber.
The EML (Electro-absorption Modulated Laser) places a DFB laser and an electro-absorption modulator (EAM) on the same InP chip[3]. How it works:
- The DFB section emits continuous wave (CW) light, its wavelength locked by the grating
- The EAM section changes its absorption with a voltage of only a few volts[3]: the Franz–Keldysh effect shifts the absorption edge, and the quantum-confined Stark effect (QCSE) in quantum wells delivers a high extinction ratio[3]
Intensity modulation is decoupled from carrier dynamics: chirp is low, and by adjusting the bias voltage it can even go negative, compensating fiber dispersion and extending reach[3].
Compared with an external MZM (silicon photonics or TFLN), the EML’s advantage is integration: laser and modulator on one chip, with low coupling loss and small size; the MZM has a higher bandwidth ceiling but needs a separate chip and optical coupling.
Figure 2 | Separate electrical drives and a connected waveguide in an EML. Layers, spectra, and power envelopes are simplified and not to scale; bias and facet coatings depend on the device. Click for full size.
This is the basis for 100G per lane (53.125 GBd PAM4)[7] and 800G (8 lanes × 100G) being EML-based[6]. Take 800GBASE-DR8 as an example: the transmitter holds 8 EMLs, each lane at 100G PAM4[6]. The price: more complex fabrication (the EAM section needs selective epitaxial regrowth) and a TEC for temperature control (the EAM absorption edge is temperature-sensitive).
3. Which Laser for Which Link
| FPLD | DFB | EML | |
|---|---|---|---|
| Cavity | Two end mirrors (Fresnel) | Grating distributed feedback | DFB + EAM on one chip |
| Modes | Multimode (spacing ~100 GHz) | Single-mode | Single-mode |
| Linewidth | Multimode spectrum | A few MHz – 100 MHz | A few MHz (DFB section) |
| Wavelength drift | ~0.3 nm/K | ~0.06 nm/K | ~0.06 nm/K |
| Modulation | Direct current | Direct current | EAM voltage |
| Cost | Lowest | Medium | Highest |
| Typical use | Short reach, CATV, sensing | 10G/25G, DWDM, coherent CW | 400G/800G data center |
Application mapping:
| Link scenario | Laser choice |
|---|---|
| Intra-datacenter multimode (<100 m) | VCSEL 850 nm (a fourth class, vertical cavity; covered separately) |
| 10G/25G single-mode | DFB 1310 nm, directly modulated (1310 nm is the zero-dispersion window[4]) |
| 400G/800G data center | EML 1310 nm (800GBASE-DR8 = 8 lanes × 100G PAM4[6]) |
| Long haul / coherent | DFB 1550 nm CW + external modulator (1550 nm is the low-loss window[4]) |
Selection logic: spectral purity × modulation speed × cost — the link’s requirements decide which rung of the ladder you stand on.
Common misconceptions:
- “A DFB is just a more expensive FPLD” — the grating changes the feedback from two broadband mirrors to wavelength-selective distributed feedback; single-mode operation is a qualitative change, not an incremental one
- “Faster links always need an EML” — directly modulated DFB is enough for 10G/25G; the chirp penalty only binds at high speed
- “Silicon photonics will replace InP lasers” — silicon is indirect bandgap and cannot emit light; silicon photonics modules still need an InP light source
Self-check: after reading, you should be able to answer — ① Why can’t an FPLD be used in DWDM? ② Why does an EML have lower chirp than a directly modulated DFB? ③ Why do silicon photonics modules still need an InP light source?
4. Limits and the Next Generation
- EML bandwidth is approaching its limit: 200G per lane (1.6T) is being standardized by IEEE 802.3dj; thin-film lithium niobate (TFLN) MZMs and silicon photonics are the main alternative routes[6]
- Silicon does not emit light (indirect bandgap): silicon photonics modules still need an InP EML/DFB as the external light source, coupled into the silicon waveguide
- DFBs are sensitive to optical feedback: back-reflection can trigger coherence collapse, so modules usually integrate an optical isolator[2]
- Coherent is moving from long haul into data centers (DCI), and demand for DFB CW sources keeps growing
Back to the opening question: optical communication needs three kinds of lasers because no single cavity structure satisfies spectral purity, modulation speed, and cost at once. The FPLD is cheap but multimode with large drift, so it serves only links that tolerate multimode light. The DFB is stable and single-mode — the backbone of WDM and coherent — but direct modulation caps its speed. The EML integrates a high-speed modulator and takes over 400G/800G, at the price of the most complex fabrication. For example, a 10G SFP+ module uses a directly modulated DFB while an 800G OSFP module uses 8 EMLs — same 1310 nm, different lasers. Scope: this ladder applies to the InP edge-emitter system (1310/1550 nm); 850 nm multimode short reach belongs to VCSELs. Unresolved limits: the EML bandwidth ceiling and the transition to 200G per lane, plus the cost of TFLN and silicon-photonics integration, are still evolving.