Germanium Photodetector

# Germanium Photodetectors in Silicon Photonics

Silicon is transparent at the infrared wavelengths used for optical communication (1310 nm, 1550 nm, the C-band), so silicon itself cannot absorb photons there and generate electrical current — it is useless as a photodetector material. Germanium, alloyed with silicon or deposited as a thin layer on the silicon substrate, has a much smaller bandgap (0.66 eV vs. silicon's 1.12 eV) and becomes absorbing at 1550 nm. A germanium photodetector (Ge PD) is a reversed-biased p-n junction in germanium that sits directly atop a silicon waveguide, couples the light evanescently into the germanium absorber, generates electron-hole pairs, and collects them as current — the silicon photonics answer to "how do we detect infrared light on the same chip as our modulators and waveguides?"

The integration of Ge photodetectors *on* silicon is mature enough that today's silicon photonics transceiver chips integrate 4–16 photodetectors per channel (one for the data signal, others for automatic gain control, wavelength locking feedback, and monitor outputs), all sharing the same substrate and process flow. This tight co-integration is why silicon photonics can achieve the power efficiency and cost per bit that makes it attractive to datacenters — a monolithic chip trades a few extra photolithography steps and germanium deposition for the ability to put photodetection, modulation, wavelength selection, and electrical signal conditioning all on one die.

The relationship between Ge photodetectors and the rest of the silicon photonics stack is complementary to that of silicon modulators: modulators convert electrical signals (from the transimpedance amplifier, the digital driver) into optical waves using the free-carrier plasma dispersion effect (id 971); photodetectors do the reverse, converting optical waves back into electrical current. Both use the same silicon waveguide platform, both are fabricated in the same process flow, and both have the same fundamental constraint — speed vs. efficiency. A modulator that is very efficient at converting electrical to optical power does so slowly; one that is fast often wastes more electrical power. Similarly, a photodetector with high responsivity (converts a lot of optical power into current) does so with higher capacitance and slower rise time; a fast detector has lower responsivity and needs more amplification downstream.

The transimpedance amplifier (TIA), sitting right next to the photodetector, is the necessary counterpart: its job is to convert the tiny photo-current from the Ge PD (often in the picoamp-to-nanoamp range before amplification, just above the dark current floor) into a clean voltage signal ready for a limiting amplifier and clock-recovery circuit. The TIA's transimpedance (the ratio V_out / I_in, measured in ohms) sets the gain and also the bandwidth — a high transimpedance gives more signal but slower response, a lower transimpedance is faster but needs a more sensitive subsequent stage. This trade-off is built into the receiver architecture: the Ge photodetector decides how much light is converted to current, the TIA decides how clean and fast that signal becomes, and the downstream receiver signal processing (automatic gain control, equalization) adapts to whatever combination of speed and sensitivity the frontend actually achieves.

The physical integration of Ge on Si is the practical core of silicon photonics' cost advantage: germanium deposition, junction formation, and metal contact patterning add just a handful of extra mask steps to a silicon logic or analog process, because germanium and silicon are lattice-matched and form a stable heterostructure. A company already manufacturing silicon chips has, in principle, the process ready to go. This is why so many of the early silicon photonics companies were founded by researchers from semiconductor logic houses (Intel, IBM) rather than traditional optics companies — the technology plugged directly into their existing fabs and process expertise.

Ge photodetectors are not the only way to detect light on a silicon chip — InGaAs heterojunctions, avalanche photodiodes, and quantum-dot photodetectors all work and have advantages in specific regimes (sensitivity, speed, integration density). But germanium remains the most cost-effective and process-compatible option for mainstream silicon photonics because it leverages the existing silicon fabrication ecosystem without requiring materials or process steps fundamentally alien to the world's largest semiconductor industry. The silicon photonics transceiver as a commercial product exists, in large part, because someone figured out how to make germanium work seamlessly on top of silicon, converting infrared photons into the electrical signals that silicon electronics already know how to process.