Building better photodiodes

The photodiode has been serving mankind for many years. Since its invention in the 1950s, it has been used to read data off countless optical discs and bar code scanners, sense the onset of twilight for numerous streetlights around the globe and detect billions and billions of pulses of light that underpin optical communication. In recent times there has been a rapid build-out of Internet capacity. This has propelled a hike in the performance of photodiodes, which have made major strides in power-handling and detection speeds. In turn, this improved photodiode performance has helped engineers to fully exploit many opportunities in the burgeoning field of photonics. Working in the photonic domain, rather than its electronic counterpart, is tremendously beneficial to engineers: Bandwidth is higher, losses are smaller, cabling is lighter, and there is greater resistance to electromagnetic interference. As a result, photodiodes can find application in areas traditionally dominated by electronics, such as data transmission links and oscillators. What’s more, these diodes can play a role in terahertz signal generation, arbitrary waveform synthesis, photonic analogue-to-digital conversion and high-speed wireless links. Although the majority of fibre-optic links deployed around the world are digital, a growing number of applications are using analogue optical links, which convey electrical signals over an optical carrier in analogue form (see Figure 1). Examples include cable TV, local oscillator (LO) distribution for radio telescopes, beam-forming networks for phased array antennas and ‘antenna remoting’ for military radar. In addition, the Atacama large millimetre/sub-millimetre array in Chile – one of the largest radio telescopes in the world – uses analogue optic links to distribute the photonic LO over 100 GHz (see image below).

    Figure 1. A simplified diagram of analogue optic link These analogue links can be viewed as replacements for conventional electrical cables or waveguides, which are often impractical, due to their high loss and limited bandwidth. One reason why analogue optical links can realise very high speeds is that they are able to overcome the limitations of analogue-to-digital and digital-to-analogue conversion, which are found in digital transmission. With analogue optical links it is possible to transmit signals over vast distances by using a substantial amount of RF power between a centre station and remote locations, such as antenna feeds. This reduces the complexity of the instruments held at remote locations, and cuts their maintenance requirements. The performance of these links depends on the capability of the photodiode in providing optical-toelectrical conversion. A high gain, large bandwidth link demands a diode with high-power-handling capacity and high-speed operation. High linearity of the photodiode is valued too, because this minimises the signal distortion in the link and enables it to maintain a large, spurious-free dynamic range. High-performance photodiodes can also be deployed in oscillators, an indispensable component in virtually all modern electronic systems. Most of these oscillators are electronic, and they tend to rely on ‘high-Q’ quartz crystal resonators to achieve high spectral purity. But as frequency increases their performance drops off, due to either a weakening of the resonance or an introduction of phase noise, which results from frequency multiplication. It is possible to address these weaknesses with an optoelectronic oscillator (OEO) (see Figure 2 for an example of this approach). This class of oscillator produces an ultra-stable microwave signal at frequencies of up to tens of gigahertz by exploiting the low-loss properties of optical resonators. Approaches can involve long fibre delay lines, Fabry–Pérot cavities and whispering gallery mode resonators. Many OEOs are based on a transposed gain oscillator, while others use a dual-mode laser or an optical frequency division technique. One of the strengths of these OEOs is that they generate microwave signals in both the electrical and optical domains, a trait that makes them specifically suitable for integration with other photonic systems.

Figure 2.Schematic diagram of an OEO based on a transposed gain approach Ideally, high-RF-power-output photodiodes are used in OEOs, because this increases the signal-to-noise ratio and lowers the system phase noise. Further enhancements in performance are possible when the RF power output from the photodiode is large enough to eliminate the electronic amplifier and its corresponding noise from the loop. The introduction of extra phase noise during the photo-detection process can be addressed with low distortion photodiodes. Device design PIN photodiodes are widely used to meet the requirements for high optoelectronic conversion efficiency and large bandwidth. These devices contain an intrinsic absorber sandwiched between heavily doped n-type and p-type layers that give rise to a space charge region. When photons hit the device, they spawn electron-hole pairs, which are pulled apart by an applied bias voltage acting in partnership with an internal electric field established by ionized dopants. Electrons and holes flow in opposite directions, creating a photocurrent in the external circuit. Many PINs operate at 1.55 μm, the wavelength for longdistance optical communication. Producing this class of device involves epitaxial growth of a lattice-matched InGaAs layer on an InP substrate. A key decision is to select the optimal thickness for the depleted, ternary absorbing layer: Get it too thick and the diode is too slow, due to excessive transit times; but get it too thin and the device does not absorb enough incident light, leading to a signal that is too weak. Another key characteristic for the photodiode is its power handling capability. This is strongly influenced by the space-charge effect, and can be traced back to the spatial distribution of photo-generated carriers as they pass through the depletion layer. In the photodiode, the electric field generated by the free carriers – the spacecharge field – opposes that established by ionized dopants and the applied bias voltage. This leads to a total electric field in the depletion region that drops to almost zero at high current densities. When this happens, carrier transit time increases and RF power output falls, due to a combination of compression and saturation. Making matters worse, the voltage drop across the load resistor reduces the effective bias voltage, pushing the diode toward saturation (see Figure 3 for a summary of the major factors limitinghigh-power photodiode performance). In an ideal world the electrical output of the photodiode follows its optical input in a linear fashion. But in practice this is never the case – nearly every physical mechanism has some degree of nonlinearity. This is present in optical transmission, carrier generation and transport, and compounding this issue, the non-linear mechanism that dominates varies, depending on the power level, frequency range and bias voltage of the photodiode. So to produce a diode with good performance, an engineer has to weave a well-chosen path that trades conflicting requirements for high-linearity with the need for certain levels of performance in key areas.

Figure 3. Factors that limit the performance of high-power photodiodes While addressing these issues, the photodiode designer must not neglect thermal management. Multiplying the photocurrent by the bias voltage reveals that a highpower photodiode needs to dissipate nearly 1 Watt from an active area of less than 10-4 cm2. The junction temperature of the diode – which is governed by factors such as heat conductance of semiconductor layers, the photodiode geometry and heat sink design – can exceed 200 °C (see Figure 4). Even higher diode temperatures are possible when the bias voltage is cranked up to improve photodiode saturation. This further increases the importance of good thermal management.  

Figure 4 (a). The layout image and (b) thermal image of 34-μm backsideilluminated MUTC photodiode obtained by thermal reflectance imaging New variants During the last decade engineers have increased the performance of the photodiode by modifying its design. Variants introduced include the uni-traveling carrier (UTC) photodiode, the partially-depleted-absorber photodiode, and the dual-depletion region photodiode. Our group at the Department of Electrical and Computer Engineering at the University of Virginia believes that the UTC photodiode merits special attention: It delivers superb performance and has potential for further improvements. This class of photodiode features a quasi-neutral absorber and a transparent depletion region. When incident photons create electron-hole pairs in the un-depleted absorption layer, electrons behave as minority carriers and holes as majority carriers. Via a combination of diffusion and drift, electrons are transported to the depleted, high-field collection layer, where they then drift toward the ncontact at their high saturation velocity. In contrast, holes, thanks to the quasi-neutrality of the absorption layer, respond very fast – within the dielectric relaxation time that is determined by their collective motion. This means that there is a fundamental difference between a UTC photodiode and its PIN counterpart: In the former structure, electrons and holes contribute to the response current, with the low-velocity hole transport dictating the device’s speed. An additional strength of the UTC design is that it alleviates the space-charge effect, thanks to a more balanced electron and hole distribution profile. Electrons are able to maintain their high velocity at relatively low electric fields, enabling the UTC photodiode to achieve high speed and high saturation output photocurrent, even at a low bias. Through modifications to the UTC structure, we have taken the device to a new level of performance in certain areas, such as output power. Our modified UTC (MUTC) features several elements that contribute to the device’s outstanding performance (see Figure 5). This includes an intrinsic InGaAs layer inserted between the p-type InGaAs absorber and the InP drift layer. Additional design flexibility follows from this – the device can then be optimized for both high responsivity and high speed.

Figure 5. Cross section of the MUTC photodiode. The numbers in brackets indicate the order of fabrication steps Our quasi-neutral InGaAs absorber consists of four step-graded p-type layers. Grading creates a quasielectric field that drives electron transport in the absorber region. To increase the saturation current, we use charge compensation. To this end, we slightly dope the depletion region to pre-distort the electric field in such a way that it is initially higher where the spacecharge effect is most severe at high current. Another feature that pre-distorts the electric field is the cliff layer – a very thin layer of n-type InP sandwiched between the InGaAs absorber and the InP drift layer. The cliff layer increases the electric field in the intrinsic InGaAs absorber and speeds the passage of electrons through the InGaAs/InP hetero-junction interface. Demonstrated superiority Measurements on our 40-μm diameter MUTC photodiode, using back-illumination and a 5 V reverse bias, show that this diode has a 3-dB bandwidth of 24 GHz and 0.69 A/W responsivity at 1.55 μm. At the 3-dB bandwidth frequency, the 1-dB saturation current was 146 mA. One option for further increasing RF power is to divide the optical signal between a parallel array of photodiodes, before combining their outputs in a coherent manner. This approach delivers an RF power improvement of 5-6 dB over a single photodiode, according to our measurements of home-built photodiode arrays based on a traveling-wave and Wilkinson power combiner. Further improvements to MUTC photodiode performance are possible with better thermal management of the device. To realise this, we have flip-chip bonded our diode to an AlN substrate, which has much higher thermal conductivity than InP. The optical path remains in InP, while electrical and thermal accesses to the photodiode are provided on the AlN substrate (see Figure 6).

Figure 6. (a) Cross-sectional view of the flip-chip bonded photodiode. (b) SEM picture of the bonded chip on probe pads Using the same-sized photodiode as before, the flip-chip technology enabled a hike in the 1-dB saturation current to more than 180 mA at 11 V reverse bias. The corresponding RF output power reached a record-high of 28.8 dBm (0.75 W) at 15 GHz (see Figure 7 to compare the performance of our MUTC photodiode to that of other high-power photodiodes detailed in recent publications.)

Figure 7. Summary of recent results: Output RF power of a single photodiode versus measurement frequency Many researchers are also focusing efforts on improving device linearity. Our approach is to incorporate a heavily doped p-type absorber in the UTC photodiode design (HD-MUTC). This modification improves linearity by reducing the influence of voltage on photodiode capacitance and responsivity. We were able to realise a third-order intercept point (OIP3) of 55 dBm for our HDMUTC at low frequencies; at 20 GHz it retained a high value of 47.5 dBm. Our efforts highlight the rapid development of photodiode technology, which continues to contribute to the advancement of photonics. Improvements are being made in many different areas, including optimisation of device geometry and epitaxial structure, attempts to enable more effective optical/electrical coupling, better heat management, and integration with a silicon platform. It is clear that in future more and more applications will utilise high performance photodiodes while taking advantage of the benefits of photonics technology.