2.5.2.1 The Key Parameters Relating a Photodetector to an Optical System

Some of the key parameters which relate a photo detector to the ultimate system where it is applied include its sensitivity, dynamic range, quantum efficiency, responsivity, internal gain, bandwidth and noise. The basic goal is to achieve the lowest possible noise at a given bandwidth or data rate.

The sensitivity of a photo detector is defined as the minimum mean optical power needed to achieve a given bit error rate, BER (CCITT standard requires a BER of 1e-10). The exact expressions for sensitivity vary according to the particular type of detector. Its dynamic range is the range of of input power levels over which the BER is acceptable [62]. The quantum efficiency, h, and the responsivity, R, are interlinked; the h is essentially the efficiency of converting an incident photon (optical power) into useful photo current, IPh, whereas the R is the amount of IPh obtained for a unit of incident optical power, Po. These last two parameters are wavelength dependent as discussed later in section 4.5. The internal gain of a detector refers to the generation of secondary electron-hole pairs created by the primary photo generated pair as in Avalanche Photo Diodes (APDs) or the electrical gain, b, of bipolar transistor based photo detectors. The bandwidth, B, refers to the range of frequencies over which the detector provides useful output at a pre-defined level (e.g. 3dB). Finally, the noise is a term which refers to spurious output in the absence of light; thus thermally generated dark current and surface and other leakage currents all add to this unwanted "signal" and contribute to the noise which essentially limits the sensitivity of the detector [61]. Terms often used to quantify this complex parameter include "signal to noise ratio" (SNR) and "noise equivalent power" (NEP) which is the incident optical power, at a given wavelength, necessary to obtain a IPh equivalent to the rms noise current within a unit bandwidth B.

There is often a fundamental trade-off between the sensitivity and the bandwidth of a detector. For example, in the case of a p-i-n detector with a FET pre-amplifier, a large resistor may be needed to extract a useful voltage to drive the gate of the FET. In doing so, however, the RC constant becomes large which limits the bandwidth of the receiver.

2.5.3 p-n, p-i-n and Avalanche Photo Diodes (APDs)

Photo diodes can be broadly categorised into two types: those without internal gain such as p- n and p-i-n diodes and those with such as APDs. The penetration depth of light before it is absorbed within a material increases with its wavelength (see section 6.3). Thus a wider depletion region is necessary for long wavelength operation. In the p-n junction, this is achieved by making the n-type material so lightly doped that it can be considered intrinsic; an n+ layer is added to reduce ohmic contact. This modified device is known as a p-i-n photo diode [61]. The intrinsic layer is wide enough to maximise absorption for a given wavelength and the low doping means it is fully depleted under normal reverse bias resulting in fast collection of photo generated electron hole pairs. Due to the spectral limitations of Si and thermal instabilities and large dark currents associated with Ge, p-i-n diodes have been designed and fabricated using InGaAs which are sensitive over 0.95 to 1.65 mm wavelength range [63] and have dark currents in the pA range at room temperature [64] . Substrate entry heterojunction p-i-n based on InGaAs p+ and i layers and InP n layers have been used to eliminate absorption in the top p+ layer. However, this design suffers from charge trapping in the InGaAs/n-InP heterointerface although it is not a severe limitation to its performance [65].

In Avalance Photo Diodes (APD), the structure of the basic p-n diode is further modified to create an extremely high electric field; the APD consists of a n-p-i-p+ type layer structure. In addition to the depletion n-p region where majority of absorption takes place, the high field region (i region) accelerates the primary photo generated pairs to acquire sufficient energy to excite new electron-hole pairs by impact ionization [61]. This is known as carrier multiplication; hence these devices have inherent gain. In order to minimise noise, the electric field at avalanche breakdown must be as low as possible. In Si, this has been achieved by using a reach through structure (RAPD) where the multiplication region is much wider than the n-p region. Much of the material problems associated with Si and Ge p-i-n diodes are also relevant to APDs and heterojunction devices have been realised using various compound semiconductor material systems including InGaAs/InP. However, the narrow bandgap of InGaAs gives rise to unacceptably high level of band-to-band defect tunneling currents which precede avalanche field. In common with the Si RAPD, this problem is significantly reduced by using a separate absorption and multiplication region in the SAM-APDs with the gain occurring at InP p-n junction where tunneling is much less [66]. As in InGaAs/InP p-i-n, the issue of charge trapping at the heterointerface discontinuity is also a limitation in these APDs. However, Campbell et al [67] have reported the use of a InGaAsP (with a bandgap located between InGaAs and InP) quaternary grading layer to smooth out the discontinuity and hence improve speed performance in separate absorption, grading and multiplication (SAGM) APDs. Noise arising from multiplication region in APDs has been addressed by Capasso et al [68] by incorporating a super lattice structure (SL) in AlGaAs/GaAs and Kagawa et al [69] in the InGaAs/InAlAs SL-SAM-APDs.

2.5.4 ITO/n-GaAs Schottky Photo Diodes

The use of a Schottky barrier photo diode has many advantages for very high speed applications. In common with a conventional p-i-n detector, the absorption layer thickness can be engineered to obtain the optimum compromise between external quantum efficiency and detector bandwidth. An advantage however, is that there is no slow component associated with minority carrier effects in the p+ region of a p-i-n photo diode [70]. Planar Pt/n-GaAs Schottky diodes with 100 GHz bandwidth have been reported by Wang et al [71]; the metal thickness was only 100Å to allow for optocoupling.

An inherent disadvantage of the Schottky photo diode, however, is the high series resistance and low efficiencies arising from the semi-transparent metal layer. This is apparent in the relatively low quantum efficiency of 19% and high series resistance of 190W obtained by Emeis et al in their p-InGaAs Schottky diodes (for operation at 1.3mm wavelength) with 50Å Ni semi-transparent metal contact [72].

Using a practically transparent and highly conductive layer of Indium Tin Oxide (ITO) to form the metal/semiconductor junction solves both these problems [73]. Figure 2.9 shows the band diagram of such a device.

Figure 2.9: Band diagram of an ITO/n-GaAs Schottky photo diode.

The absorption layer is usually lightly doped to maximise depletion and is situated directly underneath the metal contact. Light enters through the transparent ITO contact and creates photo generated electron-hole pairs. These are then swiftly separated by the built-in depletion field giving rise to a photo current. The speed of response of such a device depends on the transit time of photo-generated carriers across the depletion region, the junction capacitance and parasitic circuit element contribution. In a monolithic structure, device isolation, achieved by either proton bombardment or mesa etch or a combination of both, reduces these parasitics. Furthermore, a planar structure is suitable for monolithic integration with other circuit elements such as HBTs or HEMTs.

Further discussion on Schottky diodes and photo detectors are presented in section 6.1

2.5.5 Transparent-Gate High Electron Mobility Transistors (TG-HEMTs)

The use of GaAs Metal Semiconductor Field Effect Transistors (MESFETs) and AlGaAs/GaAs High Electron Mobility Transistors (HEMTs) as optoelectronic detectors have been recently reported [74,75]. These are both essentially field effect devices where current is transported laterally through a channel between two horizontal electrodes, source and drain, that can be modulated by a Schottky electrode placed at an intermediate lateral location [76]. The more advanced heterojunction HEMT device usually consists of epitaxially grown n-type AlGaAs layer on an undoped GaAs layer grown on a semi insulating GaAs layer. The channel is formed by mobile charge accumulation at the heterointerface since electrons from the wider bandgap AlGaAs "supply layer" transfer across into the undoped GaAs in order to occupy a lower energy state. Extremely high electron mobility results from diminished scattering due to ionised impurity as the coulombic field within the undoped GaAs crystal lattice is greatly reduced [77].

As in the ITO/n-GaAs Schottky photo diodes, radiation with energy greater than the bandgap of AlGaAs gives rise to a photo-generated electron-hole pair in the supply layer directly above the channel. This pair is separated by the built-in Schottky field giving rise to a photo current and corresponding photo voltage across the junction. In turn, this photo voltage effectively reduces the channel width and can be used for the desired optical control [78]. We have recently shown that using a transparent ITO gate enhances this control [79]. On the other hand, a photo conductive effect dominates if the energy of the incident radiation is between the bandgaps of the AlGaAs and GaAs; in this case, electron-hole pairs are generated in the GaAs channel and are separated by the lateral source-drain electric field thus adding to the drain current.

In comparison with the Schottky photo diode, the responsivity of these devices is many times greater as a direct consequence of the in-built gain of the FET structure. An inherent limitation of these devices as optical detectors however arises from the fact that the combined vertical depth of the AlGaAs and GaAs channel layers, typically 0.1mm, are inadequate to ensure maximised photo absorption.

2.5.6 Heterojunction Photo Transistors (HPTs)

With the problems encountered with APDs in the 1970's for long wavelength use and the advent of high speed HBTs in addition to other fundamental benefits offered by these devices, renewed interest in (heterojunction) photo transistors (HPTs) as optical detectors has been aroused. The HPT offers the dual function of detection and amplification in a single device. In comparison with its rivals such as p-i-n/FET or APD/FET photo receiver combinations, this device is a particularly attractive alternative because of its relatively simple structure, ease of fabrication and integrated nature [80]. From an optical communication point of view, HPTs can be easily integrated monolithically in existing MMIC and OEIC processes resulting in high reliability and low parasitic noise [81].

Studies using HBTs with opaque emitter contacts show excellent suitability of these devices as photo detectors [82] and mixers in coherent photo receivers [83] in terms of optical performance where signal to noise ratios in excess of 30dB have been obtained. Optoelectronic mixing in HPTs has considerable potential for simplification of the optical/electrical interfaces in some configurations of these systems. Mixing in the HPT is easily realised since the base terminal can be pumped either optically or electrically.

In the photo transistor, radiation incident on the device passes through the wide gap emitter unattenuated and is absorbed in the base, base-collector depletion region and bulk collector. Under normal common emitter operation mode, the base-collector junction is reverse biased. Hence it can be likened to a p-i-n photo diode (the collector doping is relatively very low). The base-collector junction acts as the light gathering element or the absorption layer. Internal gain is achieved through normal transistor action whereby light absorbed affects the base current (effectively adding to the Icbo component) giving multiplication of photo current through the device. The optical gain, G, for a heterojunction photo transistor is given by [84,85]:

The quantum efficiency in (eqn. 2.45) is a function of the device parameters only and is independent of the current gain. Other aspects of the HPT are discussed later in Section 6.2.

2.5.7 Light Emitting Diodes (LED) and Vertical Cavity Surface Emitting Lasers (VCSEL)

A detailed theoretical discussion on the operation of LEDs and LASERs is clearly beyond the scope of this thesis. Yet, like detectors these transmitting devices constitute an inseparable part of the optical link. Solid state implementation of these devices essentially involve a p-n junction designed specifically for light emission rather than absorption. In the case of the LED, the p-n junction is formed using degenerate semiconductors to ensure forward current is dominated by the recombination process resulting in spontaneous emission of photons [86]. LEDs have been realised using a host of III-V compounds ranging from GaN for ultraviolet, GaAs for blue through to near infra-red and InGaAsP for emission beyond this spectral range. Unlike LEDs, laser operation is dominated by stimulated emission of light which results in highly monochromatic radiation. Laser modulation is easily achieved by modulating the forward current. Since the photon lifetimes are very short, high speed modulation can be realised. AlGaAs/GaAs lasers are used for 0.8 - 0.9 mm emissions while longer wavelengths of 1.3 - 1.55 mm are catered for by the InGaAsP/InP system.

Internal reflection and substrate absorption losses limit the external quantum efficiencies of LEDs. A significant portion of the drive current crosses the p-n junction directly beneath the contact, generating light that is obscured. Given that at the p-n junction the emitted light intensity is at its peak, this is particularly undesirable. The high resistances associated with a semi-transparent metal electrode lead to rapid decrease in junction current density due to its exponential dependence on the voltage; in turn this significantly reduces the emission from points at a lateral distance from the contact. Therefore, the use of a transparent contact not only allows the entire junction area to emit (which would otherwise be obscured) but it does so with uniformity. Aliyu et al have reported very low threshold voltage of 1.7V for 20 mA in their study using ITO contacts on AlGaInP visible LEDs [87].

Vertical Cavity Surface Emitting Lasers have recently aroused interest in a number of fields including fiber optic communication and optical computing. As the name suggests, VCSELs use an orthogonal cavity for light amplification rather than the common lateral Fabry-Perot approach that is associated with an edge emitting laser. High reflection is made possible by epitaxial growth of Bragg reflectors. As in LEDs, the optical path is obscured by the route of the injection current. We have demonstrated that ITO can be used to realise InGaAs/GaAs VCSELs for the first time resulting in very low threshold currents of 20 mA at room temperature [88].

2.5.8 Principal Receiver Configurations

Optical receivers can be classed into three basic categories: direct detection; coherent or heterodyne detection; and optically pre-amplified and tunable direct detection.

In direct detection communication systems, where the photo detector responds merely to the intensity of the incident optical signal, good sensitivity or large signal to noise ratios are very important. Since the noise is dominated by thermal sources in the preamplifier, the sensitivity can be improved either by applying gain or reducing the noise [62]. To this end, the most popular approach has been in favour of the p-i-n/FET combination [89] rather than single APDs. Other common direct detection based implementations involve using p-i-n/HBT [90] and Schottky/HEMT [91] photo receivers.

Coherent detection, or mixing, involves amplifying the incoming signal by multiplying it with a local oscillator. This has two advantages over direct detection: (a) the receiver sensitivity can approach the Shot noise limit of the signal and (b) much greater wavelength selectivity allows many more channels to be carried at different wavelengths by the same fiber. Heterodyne receivers have also been implemented in a wide range of device combinations: p-i-n/FET [92], p-i-n/HEMT [93]. However, in their comparison of a wide range of heterodyne receivers, Urey et al [83] reported that in terms of the available intermediate frequency (IF) signal/noise ratio, the best performance was obtained using HPTs as mixers.

Optical pre-amplification refers to a relatively new technology which has wide ranging implications. Majority of today's developmental work is devoted to the erbium doped fiber amplifier (EDFA) [55]. These utilise a silica fiber with a doped core to provide a medium which affords gain when optically pumped at an appropriate wavelength - presently 980 nm or 1480 nm because of the availability of solid state lasers which couple sufficient power at these wavelengths. The signal is optically amplified en route to the detector end of the link; this reduces component count, increases reliability and opens up full bandwidth of the fiber windows between switching centres. The sensitivity of pre-amplified receivers exceeds that of coherent systems at high bit rates representing a major cost advantage in its favour since it essentially uses a direct detection design; tuning is achieved by selecting a suitable filter.

2.5.9 A Brief Comparison Between Various Types of Detectors

A summary comparison with some of the key performance parameters for various types of optical detector devices is shown in Table 2.3. These represent typical figures quoted in the literature for advanced III-V detectors and not necessarily the best data which is now available. Following the discussion on individual devices in previous sections, the purpose of this table is to present a "at a glance" figures of merit for these devices. For a more careful comparison, one needs to take into account several other factors such as material systems, the layer structures and the wavelength of the optical radiation amongst others; one such performance comparison between HPTs, p-i-n/FETs and APD/FETs has been made by Tabatabaie-Alavi et al [80]. In some cases, where no data is reported, an estimate is presented (e.g. bandwidth estimated from quoted impulse response for the device) or left empty where this is not possible.

It can be noticed from the above comparison that p-i-n devices are designed to provide high speed whereas APDs are essentially high gain devices. Since from a system point of view, the product of the gain and the bandwidth is important, there will a region of overlap where either device may be equally suitable. Another noticeable feature of this table is the inherent trade- off between the quantum efficiency, h, and the bandwidth of the detector; the two ITO/n- GaAs Schottky detectors [70,73] can be used to illustrate this point at a first order comparison: a slightly (10%) narrower absorption region enhances speed at the cost of lowering the h. HPTs have also come into contention with advancements in device technology and material growth techniques; these devices combine high speed and high gain and can be used as mixers. Finally, it is clearly seen that the use of ITO as both transparent Schottky and emitter ohmic contacts improves the h without fundamentally reducing the bandwidth.