The photodetector and its receiver front-end convert light back into electrical bits, and recent fabricated demonstrations, including a single-chip 1.024 Tb/s CMOS receiver at 71 fJ/b and a 206 GHz photodiode, show the receiver is now where much of the remaining energy and bandwidth headroom in optical links sits.
Summary
An optical receiver front-end is the input half of any optical link. A photodetector, usually a reverse-biased photodiode, converts incoming photons into a photocurrent; a transimpedance amplifier (TIA) turns that small current into a voltage swing large enough for a decision circuit or ADC. Everything that matters about a link’s power budget converges here: the detector’s responsivity in amps per watt, its 3 dB bandwidth, the parasitic capacitance between detector and amplifier, and the amplifier’s input-referred noise, which sets the minimum received optical power (sensitivity) for a given bit error rate.
The design tension is structural. Photodiode quantum efficiency and bandwidth trade against each other, because a thicker absorption region catches more light but slows carrier transit; work on modified uni-travelling-carrier (MUTC) photodiodes attacks this directly, with an InP waveguide-integrated device reaching a 206 GHz 3 dB bandwidth at 0.81 A/W external responsivity and a bandwidth-efficiency product above 130 GHz. The second tension is integration: detector and amplifier fabricated in different processes accumulate parasitics at the interface, which is the explicit motivation for 3D heterogeneous integration by micro-transfer printing of a 0.06 mm2 BiCMOS chiplet onto a silicon photonic IC, giving 224 Gb/s PAM-4 at minus 5.2 dBm OMA sensitivity and 0.51 pJ/b.
The parameters that decide the field are therefore: energy per bit, bandwidth density, sensitivity, and whether the front-end can be built in a commercial foundry flow. Two 2026 results set the current markers. A monolithic analogue front-end in the GlobalFoundries Fotonix platform achieves 28.9 GHz bandwidth and 61.7 dBohm gain at 9.22 mW and 0.08 pJ/bit, with under 737 nA RMS input-referred noise current and PAM-4 eyes measured to 64 GBaud. A 32-channel WDM receiver monolithically integrated in 45nm CMOS runs 1.024 Tb/s aggregate at 71 fJ/b including photonic tuning and control power, with 4 Tb/s/mm2 bandwidth density and BER below 1e-12 with no equalisation, error correction or DSP.
Outside data-centre links, the same component class is under pressure from three other directions: sub-THz wireless, where the photodiode is the photonic mixer generating RF power; mid-infrared sensing and free-space optics, where fast room-temperature detectors are scarce; and quantum systems, where the receiver choice determines what information survives measurement at all.
Evidence base
- A 32-channel WDM receiver monolithically integrated in 45nm CMOS reached 1.024 Tb/s aggregate at 32 Gb/s per channel, 71 fJ/b including electronics and photonic tuning, 4 Tb/s/mm2 bandwidth density and BER below 1e-12 with no equalisation, FEC or DSP, reported 12 January 2026.
- A 3D heterogeneously integrated receiver made by micro-transfer printing a 0.06 mm2 BiCMOS chiplet onto a silicon photonic IC achieved 224 Gb/s PAM-4 at minus 5.2 dBm OMA sensitivity, BER 2.4e-4 and 0.51 pJ/b, reported 28 November 2025.
- A monolithic analogue front-end in the GlobalFoundries Fotonix platform measured 28.9 GHz bandwidth, 61.7 dBohm gain, 9.22 mW from 1.2 V, under 737 nA RMS input-referred noise and 0.08 pJ/bit, with PAM-4 eyes to 64 GBaud, reported 7 May 2026.
- An InP waveguide-integrated modified uni-travelling-carrier photodiode achieved a 206 GHz 3 dB bandwidth, 0.81 A/W external responsivity, bandwidth-efficiency product above 130 GHz and above minus 5 dBm RF power at WR-5.1 output, reported 28 May 2026.
- An InAs/InAsSbP nBp barrier detector for 3.0 to 3.7 microns achieved 2.4 GHz 3 dB and 8.0 GHz 20 dB bandwidth at room temperature from a 121 micron diameter device, described as best in class, while noting commercial availability of such detectors remains scarce, reported 17 June 2026.
- A mid-infrared upconversion detector translated 3 micron photons to the visible at 80% conversion efficiency, giving 37% overall detection efficiency, NEP of 1.8e-17 W/Hz^1/2 and photon-number resolving to 9 with 0.14% noise probability per pulse, reported 4 June 2026.
- Optical I/O for superconducting qubits using fibre-integrated photodiode arrays for control delivery showed only a 0.19% single-qubit gate fidelity reduction versus microwave operation, reported 24 December 2025.
Open questions
- What are the yield, throughput and long-term reliability of micro-transfer printing an electronic chiplet onto a silicon photonic IC at production volume, given the 0.06 mm2 BiCMOS demonstration is described as a first of its kind?
- Does the 71 fJ/b figure for the 32-channel WDM receiver hold across temperature, process corners and ageing once ring and MZI tuning must track a real data-centre thermal environment?
- Is above minus 5 dBm RF power from the 206 GHz MUTC photodiode sufficient for practical sub-THz wireless links, or does it still require amplification that erases the efficiency gain?
- Will any fast room-temperature 3.0 to 3.7 micron detector reach commercial availability, given the stated scarcity, and at what price and package?
Assessment drafted 2026-08-31 from up to 17 KB sources using the technology-scorecard framework; scores are a draft read pending review.