news & views

OPTOELECTRONICS Fast silicon photodiodes There is typically a compromise between speed and efficiency when designing silicon photodiodes. Now, researchers have exploited microstructuring to achieve fast and thin devices that are also efficient.

Michael B. Johnston

ow much Internet traffic did you a Intensity (z) b Intensity (z) generate today? Perhaps more than kphoton kphoton you realize given the increasing I I H 0 0 popularity of streaming audio or video content, ‘cloud’ data storage and social media. It is estimated that approximately 21 1 zettabyte (10 bytes) of Internet traffic n n 1 was transmitted globally last year , which Ii Ii is the equivalent of about 360 MB per e– – – + e e– e h + day per person in the world. Much of the i i h + + h long-distance high-volume Internet traffic h is transmitted via near-infrared (NIR) light through optical-fibre waveguides. At the p p end of the optical fibre, the optical signal is turned into an electrical signal, typically for use in silicon-based integrated circuits. z z However, at present, most receivers for long-distance optical-fibre communications systems are based on photodiodes made Figure 1 | Schematic of a silicon p–i–n photodiode. The diagram shows propagation of light with from other semiconductors such as intensity I0 and wavevector kphoton incident normal to the surface of a p–i–n photodiode. Absorption – + InxGa1–xAs or Ge, which are challenging of photons in the high-electric-field i region of the device leads to electron (e ) and hole (h ) pair and costly to integrate with silicon CMOS production that rapidly contributes to the photocurrent. Ii is the drop in intensity of light associated electronics on a single chip. with absorption in the i region. a, For propagation straight through a conventional thin device only Now writing in Nature Photonics, Ii/I0 of photons are absorbed in the i region, with a significant fraction of photons passing though the Gao and co-workers report a silicon active region. b, If, however, light can be scattered into a lateral mode, a much higher proportion of photodiode that utilizes photonic photons can be absorbed in the active region, as the lateral extent of a fast p–i–n photodiode is many microstructuring to develop a photodiode times the thickness of the i region. For conceptual clarity, changes in refractive index between air and that is both fast and responsive2. As their the layers are ignored. technology is compatible with current silicon CMOS technology, this work could lead to cheaper and better integrated fibre- hitting the device per second. The more photogenerated electrons and holes that optic communications systems. closely the electrical signal can follow determines the speed of the diode. In the Fibre optics is an excellent technology changes in the intensity of light, the faster case of a p–i–n diode, speed is typically for transmitting large volumes of the photodiode. This speed or response limited by diffusion of photogenerated information. The good news is that we time of a diode is important as the faster electrons and holes to the high-electric- are, at present, only exploiting a small the photodiode, the more closely spaced in field ‘depletion region’ of the device; the fraction of the available bandwidth of time the 1’s and 0’s (that is, on’s and off’s) of ‘drift’ transit time of electrons and holes light (~200 THz for NIR light) in our a digital signal can be, and hence the faster under the influence of the electric field in communications systems. However, information can be transmitted. the depletion region; and the capacitance of to increase the density of information So what limits the speed of a the device4. transfer we need to continually develop photodiode? Consider the operation To minimize the effect of slow diffusion, technologies to modulate the light at of a conventional p–i–n photodiode most light should be absorbed in the high- higher frequencies, control dispersion that consists of a sandwich of thin electric-field i layer, rather than in the n and in the optical fibres and be able to detect p-doped, intrinsic (i) and n-doped layers p layers. In addition, to minimize the drift increasingly shorter pulses of light3. of semiconductors. Light is absorbed time and achieve low-voltage operation the The purpose of a photodiode is to according to Beer’s law through the p, i and i region should be thin. Efficient and high- turn an optical signal into an electrical n layers. The transit of photons through the speed photodiodes can be realized with signal. The electrical current generated device and the generation of electron–hole direct-bandgap semiconductors, such as by an ideal photodiode illuminated with pairs occurs on a femtosecond timescale InGaAs, which have very high absorption light of a particular wavelength is directly and this does not limit the photodiode coefficients for NIR light. However, silicon proportional to the number of photons speed. Instead, it is the collection of the is an indirect-bandgap semiconductor

268 NATURE PHOTONICS | VOL 11 | MAY 2017 | www.nature.com/naturephotonics ©2017 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved. ©2017 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved.

news & views

and has a very low absorption coefficient However, these devices typically require temperature, thereby limiting efficient for NIR light. For example, NIR light higher externally applied voltages than the photodiode operation to wavelengths (λ = 850 nm) needs to travel through just devices described by Gao and colleagues. <1,100 nm. Although 800–900 nm

0.14 μm of In0.53Ga0.47As for half the photons An alternate technology is metal– was originally used for optical-fibre to be absorbed compared with 13 μm for semiconductor–metal (MSM) photodiodes, communications, it is uncommon now silicon. Thus, to create a conventional which consist of interdigitated Schottky for long-distance links owing to high photodiode from silicon, the i region needs contacts on a silicon layer5. This MSM attenuation of optical fibres at these to be 100 times thicker, leading to a long technology is highly compatible with wavelengths (~3 dB km–1 at 850 nm) and drift transit time and hence much lower CMOS technology, and is relatively high less-mature fibre amplifier technologies in speed of operation. speed owing to low capacitance, however, this band as compared with low-dispersion So, is there a way that the speed of the sensitivity is much lower than that of 1,300 nm (O band) and low-loss 1,550 nm silicon diodes can be increased without APD or p–i–n photodiodes. (C band) wavelength regions. The C band compromising their efficiency? Gao So why bother with silicon if direct- is most commonly used owing to excellent and co-workers have come up with a gap semiconductors such as InGaAs erbium-based fibre amplifiers and low photonic solution to this problem. By can already provide us with high-speed, attenuation (~0.2 dB km–1). However, micropatterning a silicon p–i–n photodiode efficient photodiodes? The issue is with for shorter communication links, such they were able to redirect normal incident compatibility with integrated circuits as ‘last mile’ internet links, device-to- light laterally along the plane of the that are dominated by silicon-based device connections or intranets, where photodiode. This thereby allowed them technologies, such as CMOS. Silicon cheap single-chip solutions are needed, to maintain a thin intrinsic region in photodiodes integrate well with other sensitive and fast CMOS-integrated silicon their photodiode with a high ‘vertical’ silicon-based components, even on single photodiodes look most promising. ❐ electric field and short drift-time (creating chips. On the other hand, Ge or InGaAs fast photodiodes) while also keeping photodetectors are usually built as discrete Michael B. Johnston is in the Department of Physics, the absorption, and hence efficiency, of components, or are combined with CMOS University of Oxford, Clarendon Laboratory, Parks their devices high. Thus, they avoid the integrated circuits by wafer-bonding or Road, Oxford OX1 3PU, UK. compromise between speed and sensitivity flip-chip techniques that add expensive e-mail: [email protected] by exploiting the larger lateral dimensions additional steps in the device fabrication of the photodiode to increase the effective process. However, alternative methods of References absorption of their device (Fig. 1). direct growth of Ge on Si (ref. 6) or III–V 1. White Paper on The Zettabyte Era: Trends and Analysis (Cisco, 2016). There are a number of other nanowires on Si (ref. 7) could lead to more 2. Gao, Y. et al. Nat. Photon. 11, 301–308 (2017). 3. Caplan, D. O. J. Opt. Fiber Commun. Rep. 4, 255–362 (2007). methods to achieve silicon-integrated integrated competing technologies. 4. Sze, S. M. & Ng, K. K. Physics of Semiconductor Devices 3rd edn fast photodetectors. Silicon avalanche Probably the biggest drawback of silicon (Wiley, 2016). photodetectors (APDs) utilize a photodiode technologies for optical-fibre 5. Eng, P. C., Song, S. & Ping, B. Nanophotonics 4, 277–302 (2015). 6. Michel, J., Liu, J. & Kimerling, L. C. Nat. Photon. 4, 527–534 (2010). multiplication effect to achieve very high communications is the bandgap of silicon. 7. Koblmüller, G. & Abstreiter, G. Phys. Status Solidi RRL speed and up to single-photon sensitivity3. Silicon’s bandgap energy is 1.12 eV at room 8, 11–30 (2014).

METAMATERIALS A low-energy Cherenkov glow Hyperbolic metamaterials are shown to enable the emission of Cherenkov radiation from low-energy charged particles travelling at slow speeds. The achievement could lead to new forms of light sources and detectors. Mário Silveirinha

t is well known that when fast charged energetic gamma rays1. This breakthrough accelerators. Charged particles also emit particles travel inside a material with earned him the Nobel Prize in Physics in radiation when they travel in a vacuum a speed exceeding the medium’s phase 1958 alongside Ilya Frank and Igor Tamm, near a grating, an effect discovered by I 2 velocity for light, they spontaneously emit who helped explain the phenomenon. Smith and Purcell . Furthermore, neutral electromagnetic radiation at the cost of a Often, the Cherenkov effect is associated polarizable atoms may spontaneously reduction in their kinetic energy. The effect with the blue glow of the water radiate away part of their kinetic energy is known as Cherenkov radiation and owes surrounding the core of nuclear reactors when they travel near a smooth planar its name to Pavel Cherenkov — a Russian created by highly energetic beta particles metal surface3. However, all of the above physicist who in the mid-1930s was a released by the radioactive decay of nuclei. cited phenomena are generally relevant student working under Sergey Vavilov In fact, traditionally, the Cherenkov effect only when the particles travel with ultra- at the Lebedev Institute in Moscow and has various applications in the context relativistic speeds. was studying the luminescence emitted of high-energy physics, for example, to Now, writing in Nature Photonics, from liquids under the incidence of highly detect cosmic ray events or in particle Fang Liu and co-workers from

NATURE PHOTONICS | VOL 11 | MAY 2017 | www.nature.com/naturephotonics 269 ©2017 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved. ©2017 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved.