1. Theory

  • Definition: A uni-traveling-carrier photodiode (UTC-PD) absorbs the light in a neutral, narrow-gap p-type layer, and only the electrons travel through the depleted wide-gap collector, because the photogenerated holes are majority carriers in the p layer and are collected at the p contact within the dielectric relaxation time.
  • Band structure: The absorption region is surrounded by a diffusion-blocking layer on the p side and by a wide-gap collector on the n side; the holes, which move with a much lower velocity, are quickly captured by the p contact, while the electrons are swept across the collector in a strong drift current.
  • InGaAs/InP stack: A modified UTC (MUTC) has a graded p-doped absorber (80 nm undepleted with doping steps of , and , plus 180 nm partially depleted), a 250 nm InP drift layer, a 30 nm n-InP cliff layer of , and InGaAsP grading layers at the heterointerfaces.
  • Near-ballistic variant: A near-ballistic UTC-PD (NBUTC-PD) has a 150 nm graded InGaAs absorber, a 280 nm n-InP collector and a 90 nm p-InAlAs charge layer, with 20 nm graded layers, and a built-in field of 2.1 kV/cm to 84 kV/cm.
  • Other wavelengths: InGaAsP absorbers on InP for 1064 nm, an InGaAs/GaAsSb multiple-quantum-well absorber for 2 μm, and an InAs/GaSb superlattice on GaSb for the mid-infrared (cutoff 5.6 μm at 300 K).
  • Why it is used: At high optical power the space charge in a PIN diode lowers the field and slows the response; the electron-only transport reduces this effect and gives a large saturation current at high bandwidth, with bit rates of several hundred gigahertz in the book.
  • Trade-off: The speed is limited by the electron diffusion through the absorber (), so a thin absorber (80 nm to 220 nm) is used and the efficiency is recovered by illuminating through the waveguide, where .
  • Measured devices: A waveguide MUTC reached more than 220 GHz at V; a near-ballistic UTC-PD gave more than ; a 1064 nm device gave 80 mA at compression.

2. Working Principle

  • Absorption: A photon with creates an electron-hole pair in the p-type absorber; the extra hole relaxes in the dielectric relaxation time and the photoelectron becomes a minority carrier.
  • Electron transport: The electron crosses the neutral absorber by diffusion and by the quasi-electric field of the graded doping, , and is blocked from the p contact by the wider-gap barrier layer, so that it leaves towards the collector.
  • Collection: The reverse bias depletes the collector, and the electrons drift across it at the peak or overshoot velocity ( in one InP collector with 20 kV/cm to 40 kV/cm), which gives the transit limit .
  • Field control: The cliff or charge layer sets the field in the collector (20 kV/cm to 50 kV/cm for a cliff doping of ) and counters the screening of the field by the electrons; a higher doping lowers the barrier but, at , the field becomes too low for fast transport.
  • Saturation: When the electron density in the collector cancels the field at the absorber side, , the response saturates, and the saturation current rises with the bias (8.2 mA to 10.6 mA from 1.5 V to 2.5 V in a device).
  • Bandwidth rises with current: The bandwidth increases with the photocurrent in several devices, through the self-induced field in the absorber or through the velocity overshoot in the collector, and decreases again when the field is too high for the overshoot.
  • Band offsets: A conduction-band step between the absorber and the drift layer blocks the electrons at a low bias, which gave a bandwidth of 120 MHz at zero bias and 6.58 GHz at V in an InAs/GaSb device.
  • Readout: The photocurrent is delivered to a 50 Ω load or a coplanar line, and the overall bandwidth also includes of the junction and the parasitic capacitance (about 4.5 fF in the best waveguide device).

3. Applications

  • Millimetre-wave and terahertz photonic links: A MUTC photodiode module covered 110 GHz to 170 GHz with a roll-off of about and was used for a 160 Gb/s fibre-terahertz communication system.
  • High-power analog and radio-over-fibre links: The large saturation current at high bandwidth, with a current-bandwidth product above , supports photonic generation of microwave power through heterodyne beating.
  • Coherent receivers: A UTC structure with a double collector is proposed for 100 Gb/s coherent systems, where the photodiodes operate at high optical power.
  • Modulator drivers in integrated circuits: A high-bandwidth, high-saturation-power diode provides the photocurrent swing that drives an optical modulator in a transceiver, as in the book.
  • Mid-infrared detection: UTC structures on GaSb detect to 5.6 μm, with a bandwidth of 6.58 GHz in a 20 μm device.
NameFormulaRelated FormulaeVariables/Constants
Electron transit time through the neutral absorber








Built-in field from a graded doping profile




Collector transit, RC limit and overall bandwidth







Response of a partially depleted (modified UTC) absorber

Efficiency, responsivity and waveguide length





Space-charge-limited photocurrent density



RF output power and the current-bandwidth product





Important Questions

Ques: Why does electron-only transport raise the saturation current compared with a PIN diode?

At high power the carriers in the depleted region set up a space charge that lowers the field, and the response saturates when this charge cancels the field at the injection side, . In a PIN diode the slow holes (about ) stay in the depleted absorber for a long time and dominate the charge, whereas in the UTC-PD the holes disappear into the p contact and the fast electrons ( to ) carry the current, so that the charge density is two to seven times smaller (computed). For nm and V the limit is , the same order as the of a measured device (computed). Reference: L. A. Coldren, S. W. Corzine and M. L. Mašanović, Diode Lasers and Photonic Integrated Circuits, 2nd ed., Wiley, 2012, Sec. 8.3.1, Fig. 8.24 and Example 8.8; Ultra-fast Waveguide MUTC Photodiodes over 220 GHz, arXiv:2402.07491.

Ques: Why can a UTC-PD absorber be slower than a PIN absorber of the same thickness, and how is this overcome?

In a purely diffusive absorber grows with the square of the thickness, whereas the PIN hole time grows only linearly. For μm, and the PIN diode reaches 60 GHz and an absorber wholly p-type only 37 GHz (the patent figures). The remedies are a thin absorber, a graded doping that adds the field , and a partially depleted absorber: with μm and μm the weighted time gives about 115 GHz (computed), the patent value being 116 GHz. The UTC structure therefore gains by thin absorbers and by electron-only space charge, not from the diffusion itself. Reference: US Patent 6,909,161 B2, Photodiode (Google Patents); L. A. Coldren, S. W. Corzine and M. L. Mašanović, Diode Lasers and Photonic Integrated Circuits, 2nd ed., Wiley, 2012, Sec. 8.3.1, Fig. 8.24 and Example 8.8.

Ques: Does the bandwidth rise with photocurrent because of velocity overshoot in the collector or because of a self-induced field in the absorber?

The near-ballistic UTC-PD paper attributes the rise to velocity overshoot: above about 2 V the collector field exceeds the critical field of about 10 kV/cm, and a larger photocurrent lowers the net field and raises the electron velocity, not the diffusion in the absorber. The 1064 nm paper and a modelling paper instead describe a field induced by the photogenerated carriers that assists the electrons in the undepleted absorber. With and a 150 nm absorber has ps, of which the diffusion is only 0.38 ps (computed), so for thin absorbers the collector effect is the more likely, and for the thicker or graded 220 nm to 450 nm absorbers both can contribute. Reference: J.-W. Shi et al., near-ballistic uni-traveling-carrier photodiode, IEEE Photon. Technol. Lett. (2005) (staff.ee.ncu.edu.tw/jwshi, shi2005.pdf); InGaAsP/InP uni-travelling-carrier photodiode at 1064 nm wavelength, arXiv:1906.06983; Springer, Opt. Quantum Electron., doi:10.1007/s11082-017-1205-2 (equivalent-circuit model of the UTC-PD, abstract only).

Ques: Why is a cliff or charge layer placed in the collector?

The cliff layer is an n-type layer of 30 nm to 50 nm next to the absorber that shapes the field in the absorption and collection regions and suppresses the space-charge effects. A doping of kept 20 kV/cm to 50 kV/cm across the 250 nm drift layer, and the simulated range of to showed that a higher doping lowers the conduction-band barrier and raises the field, which relieves the space charge at high current, but at the field is too low for fast transport. The saturation current of the device rose from 8.2 mA to 10.6 mA when the bias went from 1.5 V to 2.5 V, while at V the field was too high for the overshoot. Reference: Ultra-fast Waveguide MUTC Photodiodes over 220 GHz, arXiv:2402.07491; InGaAsP/InP uni-travelling-carrier photodiode at 1064 nm wavelength, arXiv:1906.06983.

Ques: Why is the efficiency-bandwidth trade-off fundamental, and how does the waveguide geometry address it?

The bandwidth falls as while the absorbed fraction grows with , so a 150 nm absorber absorbs only 7% (book value ) to 14% (, computed) of a normally incident beam. In a waveguide detector the light travels along the absorber, and the length gives 92 μm for 90% with (the book example), but the capacitance grows with the length (1 fF/μm). The to waveguide MUTC devices reached internal responsivities of 0.567 A/W to 1.099 A/W, but external values of 0.161 A/W to 0.312 A/W because of a coupling and propagation loss of (a factor 0.28, computed), so the coupling is the remaining weakness. Reference: L. A. Coldren, S. W. Corzine and M. L. Mašanović, Diode Lasers and Photonic Integrated Circuits, 2nd ed., Wiley, 2012, Sec. 8.3.1, Fig. 8.24 and Example 8.8; Ultra-fast Waveguide MUTC Photodiodes over 220 GHz, arXiv:2402.07491.

Ques: Which bandwidth figures can be trusted: 280 GHz to 290 GHz estimated, or the 110 GHz and 220 GHz that were measured?

The near-ballistic devices were measured flat to about 110 GHz, and the 280 GHz to 290 GHz values were extracted from an equivalent circuit with a 25 Ω load, with a transit-time limit quoted above 400 GHz. The waveguide MUTC devices gave more than 220 GHz for the and areas, a limit of the measurement set-up, while the device gave 165 GHz, limited by the RC time of its larger capacitance. The extrapolated values depend on the circuit model and on the 25 Ω load, so the measured figures (110 GHz, more than 220 GHz) are firm, and the extracted ones are likely to be reached only if the parasitics are as modelled. The D-band module value of 147 GHz is a simulation. Reference: J.-W. Shi et al., near-ballistic UTC-PD with a flip-chip package and a microlens, Shi group paper (staff.ee.ncu.edu.tw/jwshi, shi2010.pdf; year and journal taken from the file name only); Ultra-fast Waveguide MUTC Photodiodes over 220 GHz, arXiv:2402.07491; D-band MUTC Photodiode Module for Ultra-Wideband 160 Gbps Photonics-Assisted Fiber-THz Integrated Communication System, arXiv:2411.17297.

Ques: Why does the ideal power agree with the measured output of the 1064 nm devices?

The formula gives the power of a fully modulated sinusoidal photocurrent in a matched load, and the sources quote the output at the frequency and the saturation current at compression. For 80 mA and 50 Ω the ideal power is , which is after the roll-off, against measured at 13.0 GHz; for 69 mA it is and , against at 17.8 GHz (computed). At 215 GHz a 10.6 mA device would give , but was measured, which is consistent with the roll-off of a 165 GHz device, though the source does not give the loss. The 2 μm paper writes the ideal power with instead of , which is the same for . Reference: InGaAsP/InP uni-travelling-carrier photodiode at 1064 nm wavelength, arXiv:1906.06983; Ultra-fast Waveguide MUTC Photodiodes over 220 GHz, arXiv:2402.07491; Two-micron wavelength high speed photodiode with InGaAs/GaAsSb type-II multiple quantum wells absorber, arXiv:1905.07258.

Ques: Why does a conduction-band step between the absorber and the collector slow a UTC-PD at low bias?

An unintended conduction-band discontinuity at the interface of the absorber and the drift layer forms a barrier for the electrons that have crossed the absorber, which at a low bias traps or blocks them; thermal emission helps at high temperature, and at a high reverse bias the depletion field overcomes the barrier. In the InAs/GaSb superlattice device the bandwidth was about 120 MHz at 0 V, 5.1 GHz at V and 6.58 GHz at V, and the bandwidth was transit-time limited since the calculated RC limits (27.85 GHz for 20 μm) were higher than the measured ones. Graded InGaAsP layers are grown at the InGaAs/InP interfaces of the InGaAs/InP devices to smooth the band discontinuities, and the waveguide MUTC reached its bandwidth at only V. Reference: High Speed Mid-Wave Infrared Uni-traveling Carrier Photodetector, arXiv:2001.10663; Ultra-fast Waveguide MUTC Photodiodes over 220 GHz, arXiv:2402.07491; InGaAsP/InP uni-travelling-carrier photodiode at 1064 nm wavelength, arXiv:1906.06983.

Related pages in this vault: PIN Photodiode, Waveguide Photodiode, PN Photodiode, Interband Cascade Infrared Photodetector (ICIP), Phototransistor (HPT and Photo-FET), Photodetectors (PIN and Avalanche Photodiodes), Indium Phosphide Quantum-Well Electroabsorption Modulator, Germanium and Silicon-Germanium Electroabsorption Modulator, Lithium Niobate Mach-Zehnder Modulator.