DFB laser array

A DFB laser array is a single InP photonic chip containing multiple DFB lasers, each designed for a different center wavelength, with shared electrical and optical infrastructure. The array provides multi-wavelength laser sources for WDM transmitters and tunable-laser implementations without requiring multiple discrete laser modules.

Standard architectures.

Architecture	Channel count	Wavelength setting	Application
Multi-channel DFB array + on-chip combiner	4 – 16	Lithographically fixed	Multi-channel WDM transmitter (one chip per direction)
Selectable DFB array (one-of-N)	8 – 16	Lithographically fixed; current selects	Tunable laser via channel selection
DFB array + Tunable EAM grouping	4 – 16	Lithographic + thermal fine tune	High-density WDM with per-channel modulation
Wavelength-locked DFB bank	8 – 96	Lithographic + thermal lock	ITU-grid transmitters in line systems

Wavelength setting. Each DFB in the array has its grating period set during chip fabrication, giving a fixed center wavelength. Typical channel spacing: 100 GHz or 200 GHz (matching ITU CWDM and DWDM grids), or 25 / 50 GHz for denser arrays. Thermal fine-tuning of each individual laser allows $\pm 0.5$ nm wavelength adjustment per laser, sufficient to lock to the exact ITU grid wavelength after fabrication tolerances are accounted for.

On-chip multiplexer. The outputs of the array are combined into a single waveguide by either:

• MMI tree combiner: cascaded 2:1 multiplexers; each stage adds 3 dB intrinsic loss for an N-channel combiner → 10 log₂(N) dB total
• AWG combiner (arrayed waveguide grating): wavelength-selective combiner with theoretically 0 dB intrinsic loss; only the channel at the correct wavelength is transmitted; out-of-grid wavelengths blocked
• Star coupler: low-loss broadband splitter; 1:N intrinsic loss

For multi-channel transmitter applications (all lasers ON simultaneously), the AWG combiner is preferred since it avoids the 10 log₂(N) dB loss penalty of MMI combiners.

For one-of-N tunable laser applications (only one laser ON at a time), MMI combiners are simpler and acceptable since the per-channel loss is fixed at $\sim 6 - 10$ dB and the unused lasers are off.

Standard implementations.

4-channel multi-wavelength transmitter for 400G LR4 / FR4:

• Four DFB lasers at 1271, 1291, 1311, 1331 nm (CWDM4 grid)
• Each laser directly-modulated at 25 – 100 Gbps
• MMI-tree or AWG combiner integrated on chip
• All four channels exit through a single optical fiber
• Used in QSFP-DD-FR4 transceivers

8-channel tunable laser for coherent CWDM:

• Eight selectable DFBs spanning $\sim$ 70 nm (e.g., 1262 to 1330 nm in 10 nm steps)
• One-of-N selection by drive current
• 10 GHz fine-tuning per channel via temperature
• Used in early-generation coherent transponders

Sampled-grating DBR array (SG-DBR):

• Each grating-period combination produces a different reflection comb
• Vernier tuning selects 1 of $\sim 10^2$ possible wavelength combinations
• Continuous tuning over 40 – 50 nm via fine-tuning of constituent currents
• Standard in coherent transponders 2010 – 2020

Performance specifications for production multi-wavelength DFB arrays:

Parameter	Typical value
Channel count	4 – 16
Wavelength accuracy (post-fab + thermal tuning)	$\pm 50$ pm to ITU grid
Per-channel output power	10 – 100 mW
Per-channel SMSR	$> 35$ dB
Channel-to-channel power balance	$< 1.5$ dB after MMI combiner
Total active chip area	1 – 5 mm² per channel

Yield and cost. Each laser must pass independently. Yield drops geometrically with channel count: a 4-channel array at 95% per-laser yield = 81% array yield; 16-channel = 44% array yield. This is the dominant cost-scaling problem for high-channel-count arrays and motivates active research in redundancy (more lasers than channels needed, select working subset) and improved per-laser yield.

Comparison: discrete lasers + external multiplexer. A multi-channel transmitter can also be implemented with N separate packaged DFB modules followed by an external WDM multiplexer. This approach:

• Higher unit cost (per-laser packaging)
• Lower yield risk (one laser failure doesn't kill the entire transmitter)
• More flexibility (replace a single failing laser without re-spinning the whole transmitter)
• Larger physical footprint

For lower channel counts (≤ 4), discrete-laser implementations dominate due to lower yield risk. For higher channel counts (8+), monolithic DFB arrays are typically cheaper at production scale.

References: Yariv & Yeh, Photonics: Optical Electronics in Modern Communications, 6th ed., Ch. 14 on integrated laser arrays; Murakami et al., Multi-channel DFB laser array for 100 GbE applications, IEEE JSTQE 2013.