[attachment=3775]
Author info sheet
Author:
Institution Name: Institution Address:
Topic Name & code: Title of the paper:
1)M.Vamsi Krishna 2)A.Anitha
St.ANN'S COLLEGE OF ENGG & TECHNOLOGY
Chirala
Prakasam Dist Andhra Pradesh
Photonics (EC-04) Silicon Photonic
ABSTRACT
We introduce our approach to opto-electronic integration, silicon photonics, and outline the key functions required for an opto-electronic integration platform generation, control, and detection of light. Recent research results for silicon-based optical components are discussed including a tunable external cavity laser, a 2.5 GHz optical modulator and a silicon-germanium waveguide based photodetector. Lastly, optical packaging challenges and potential next-generation designs are presented.
INTRODUCTION
Silicon photonics at can be defined as the utilization of silicon-based materials for the generation (electrical-to-optical conversion), guidance, control, and detection (optical-to-electrical conversion) of light to communicate information over distance. The most advanced extension of this concept is to have a comprehensive set of optical and electronic functions available to the designer as monolithically integrated building blocks upon a single silicon substrate.
Silicon Electronics in Optical Communications
In the electrical domain, silicon integrated circuits have been widely adopted in all layers of the network, including physical media drivers, media access controls, and for complex network intelligence functions. In principal, monolithic integration of electronics and optics is possible, can reduce unwanted electrical parasitics, and can allow for a reduction in overall size. There are numerous theoretical and practical obstacles to achieving full monolithic opto¬electronic integration. We are currently pursuing two parallel approaches to opto-electronic integration in silicon. The first is to achieve a high level of photonic integration with the goal of maximizing the level of optical functionality and optical performance. The second is to look for specific cases where close integration of an optical component and an electronic circuit can improve overall system performance. One such case would be to integrate a SiGe photodetector with a Complementary Metal-Oxide-Semiconductor (CMOS) transimpedance amplifier.
Integration: The Challenge and the Value
An optical transceiver module (sometimes called a transponder) is to be found at the terminations of virtually all optical communications links and the general functionality and architecture of the transceivers in service are similar regardless of performance requirements, communications protocols, and end-user applications. One vision of the future of optical communications links is one in which the drive for lower costs for higher performance (smaller size, lower power, higher data rate, greater transmit distance, expanded functionality, and expanded flexibility) will occur through an increased complexity in the optical domain. Some examples would be multiple wavelengths in one fiber from one ingress point, adaptive or reconfigurable optical components capable of recovering signal integrity under changing external conditions, all-optical packet switching, all-optical signal regeneration, and the use of shared optical media in so-called Passive Optical Networks (PONs) and optical Code Division Multiple Access networks (optical-CDMA). Such increased complexity in the optical domain will require increasingly sophisticated electronic control solutions, and at high data rates, there will be pressure to more closely integrate optical and electronic components. Monolithic integration of a suite of optical and electronic capabilities in one substrate is the natural progression of an integrated photonics vision.
The most likely insertion points for integrated photonics will be in places where an extreme amount of data (aggregate bandwidth) is required in a very small space. Two such applications would be microprocessor data busses (i.e., from microprocessor to memory or between multiple processors in a server) and in the backplane of server racks. Paradoxically, these applications violate a widely held axiom that says that optics is the best choice for long¬distance transmission.The copper traces and copper cables own the application space for shorter distances. The key assumption that allows us to violate this axiom is that the aggregate bandwidth will be too large for a cost-effective copper-based solution. Present-day estimates indicate that a copper-based point-to-point serial page link will become prohibitively expensive above 20 Gbps.
SILICON LIGHT SOURCE
While a silicon laser is still out of reach, work is being done worldwide on silicon light emitters that emit both visible and infrared radiation. A silicon emitter is the missing piece for monolithic integration as it would enable all optical elements and drive electronics to be fabricated on a common substrate. Because we are using silicon waveguides to guide light, the emitter must be in the infrared region of the wavelength spectrum (> 1.1 urn) where optical absorption loss is low. We first summarize the different paths researchers are investigating to achieve electrically pumped light emission, known as Electro Luminescence (EL), from silicon. Until reliable and efficient silicon emitter can be produced, hybrid integration must be considered (i.e., using a non-silicon-based light source coupled to silicon waveguides). In such a hybrid integrated approach, we show how a simple gain element (III-V gain chip) coupled to a silicon-based Bragg filter can be used to form an ECL. Proof of principal of this tunable, single-mode laser is discussed.
The difficulty in making a silicon light emitter arises from silicon's indirect band-gap. This indirect bandgap results in radiative (light emitting) decay being less likely compared to other non-radiative (e.g., Auger recombination) routes, and thus in a less-efficient corresponding light emission. Forming a laser or even a light emitter from silicon is therefore difficult, although not impossible, and research worldwide has shown light emission from silicon and silicon-based materials by a wide variety of different methods; see, for example, these range from photo-luminescence in textured bulk silicon, to fabrication of nano-scale or porous silicon to doping with exotic ions, to Raman emission.
To achieve infrared light emission from silicon, the silicon must be doped with a suitable material, such as p-FeSi2, or Erbium. Erbium-doped silicon waveguides have shown infrared light emission these kind of doped bulk silicon devices suffer from a major problem although emission can be relatively strong below 100 K, the emission intensity falls rapidly when the device is heated to room temperature. This greatly limits the application of these devices.
A different approach to enhance the efficiency of light emission in silicon is to reduce the other non-radiative mechanisms for electron hole recombination. This can be done by restricting carrier diffusion to the non-radiative recombination centers in the lattice. This increases the probability for radiative transitions and hence increases light emission efficiency. Silicon nano-crystals suspended in silicon-rich oxide restrict carrier movement while still allowing electrical pumping. Other means to obtain carrier confinement and efficient emission of infrared wavelengths include using Ge/Si quantum dots or crystalline defects. e.g., ytterbium or terbium allows emission at 0.980 and 0.540 urn in resonant cavity silicon LEDs. For these devices to be used in practical applications, however, their lifetimes and reliability still need to be optimized.Another limitation for all forward-biased silicon light emitters is their low direct modulation speeds (~1 MHz). This means that realistically this kind of silicon emitter will require an external modulator for high-speed communication links. Reverse biasing has the potential for achieving higher direct modulation speeds (~200 MHZ), but at the moment this comes at the expense of light emission efficiency.
Device Architecture
The work towards a silicon-based emitter is ongoing but still far from mature. Until an efficient, reliable silicon-based light source is available, a photonic integrated system will need to use a conventional III-V material light emitter. This section describes how a silicon waveguide-based Bragg grating can be used in an external cavity to alter the lasing properties of a III-V gain chip to produce a useful source for optical communications. The strong thermo-optic effect in silicon can be used to tune the lasing wavelength by heating the silicon grating. The driving force behind this is to produce an inexpensive narrow line-width source suitable for optical communications.
The Bragg grating is fabricated by etching a set of 1.2x2.3 um, 3.4 urn deep, trenches into a 4 um thick Silicon-on-Insulator (SOI) wafer. One thousand of these trenches are laid out in a
Paly-sUiconpillars in A waveguide
line along the waveguide with a range of periods around 2.445 urn (although these were laid out as rectangular trenches, due to litho resolution, they were rounded after processing). These trenches are then filled with poly-silicon and annealed to reduce the loss due to the poly-silicon. The poly-silicon is then chemically/mechanically polished to obtain a planar surface and the 3.5 um wide, 0.9 um deep rib is patterned using standard lithography and etching. The last step in the fabrication is to deposit a final, 0.5 um thick, and low-temperature layer of oxide to provide the necessary upper cladding for the rib waveguides. A schematic of the Bragg grating is shown in Figure 1.
Figure 1: Schematic of polycrystalline-
The novel property of this Bragg grating is that it only reflects a narrow, 0.5 nm wide range of wavelengths back through the waveguide with a reflectivity of 70%. An example of a reflection spectrum from a 1500-trench grating filter is shown in Figure 2. As a separate component these Bragg filters can be used in optical communication networks as channel filters for wavelength division multiplexed systems.
Figure 3: Schematic of the external cavity laser
The ECL is formed by butt coupling a Single Angled Facet (SAF) gain chip to a waveguide containing the polycrystalline/crystalline silicon Bragg grating. The laser cavity is formed between the Bragg grating as one end mirror, and a 90% high-reflection coating of the gain chip as the other mirror. The 8° angled facet between the two chips decreased the effective reflectivity of that facet to ~10-3. Combining an angled facet with a 1% anti-reflection coating resulted in an effective facet reflectivity of ~10-5. The output of the laser was taken from the 90% high-reflectivity coated side of the laser diode with a conical polished (140°) lensed single-mode fiber [Figure 3]. The purpose of this lensed optical fiber was to increase the coupling between the laser and the optical fiber.
Figure 4: The line-width of the ECL is 118 MHz at 1.539 [Jin
Technical Results
With the SAF gain chip butt coupled to the Bragg grating, the ECL runs single mode with a line width of 118 MHz, as shown in Figure 4. The optimized output power of the ECL, when the gain chip was driven at 250 mA, as measured out of the single-mode fiber was 450 uW. The output power is limited by a number of factors: the coupling of the gain chip to the waveguide, the 90% HR coating, and the coupling of the gain chip to the fiber. The 90% output coating can be optimized for increased output power at the expense of the laser threshold, while the coupling can be enhanced by tapering the silicon waveguide so that its optical mode better matches the mode of the gain chip.
By altering the period of the grating, different wavelengths can be fed back into the gain chip. This allows the lasing wavelength of the ECL to be changed. Figure 5 shows the spectra of the ECL for four different grating pitches. The wavelength range here is limited by the selection of gratings fabricated rather than the gain spectra of the SAF gain chip. The side-mode suppression ratio (ratio of peak emission to background) of the ECL can be seen to be over 40
dB.The refractive index of silicon can be altered by using the thermo-optic effect. This was done by placing the silicon die on a thermo-electric cooler and monitoring its temperature with a thermo¬couple. As can be seen in Figure 6, heating the 2.440 um Bragg chip from 27° C to 71° C resulted in a tuning of the ECL from 1.5395 um to 1.5455 um. A plot of lasing wavelength vs. die temperature is shown in Figure 7 for three different period gratings. The tuning is linear with die temperature with a rate of 12.6 nm/100° C.
Architecture
Here we present an experimental demonstration of a silicon optical intensity modulator with a
modulation bandwidth of 2.5 GHz at optical wavelengths of around 1.55 um. This modulation
frequency is two orders of magnitude higher than has been demonstrated by any silicon
waveguide modulator to date. The high-speed modulation is achieved by using a novel phase
shifter design based on a metal-oxide-semiconductor (MOS) capacitor embedded in a
passive silicon waveguide Mach-Zehnder Interferometer (MZI). Figure 8 is a schematic
representation of one MZI modulator discussed in this paper. Light wave coupled into the MZI
is split equally into the two arms, each of which may contain an active section which converts
an applied voltage into a small modification in the propagation velocity of light in the
waveguide. Over the length of the active section(s), the velocity differences result in a phase
phaseshiiter difference in the two waves. Depending on the relative phase of
the two waves after passing through the arms, the recombined 0„¢ wave will experience an intensity modulation
Phase shifter
Figure 8: Schematic of a Mach-Zehnder interferometer modulator with two phase shifter sections
The novel component, as well as the essence, of our silicon MZI modulator is the MOS capacitor phase shifter. Figure 9 is a schematic of its cross-sectional view. It comprises a01.4um n-type doped crystalline silicon slab (the silicon layer of the SOI wafer) and a p-type doped poly-silicon rib with a 120 A gate oxides sandwiched between them. The poly-silicon
rib and the gate oxide widths are both 02.5 um, and the total poly-silicon thickness at the centre of the waveguide is 00.9 um. In order to minimize the metal contact loss, we designed a wide (010.5 um) top poly-silicon layer on top of the oxide layers on both sides of the poly-silicon rib. Aluminum contacts are deposited on top of this poly-silicon layer as shown in Figure 9. Modeling and testing confirm that the waveguide phase shifter is a single-mode device at wavelengths around 1.55 um. In accumulation, the n-type silicon in the MOS capacitor phase shifter is grounded and a positive drive voltage, VD, is applied to the p-type poly-silicon causing a thin charge layer to accumulate on both sides of the gate oxide.
Figure 9: Schematic diagram showing the cross-sectional view of a MOS capacitor waveguide phase shifter in SOI. Optical mode propagates along the z
direction.
by
The voltage-induced charge density change ANe (for electrons) and ANh (for holes) is related to the drive voltage
(1)
where e0 and er are the vacuum permittivity and low-frequency relative permittivity of the oxide, e is the electron charge, tox is the gate oxide thickness, t is the effective charge layer thickness, and VFB is the flat band voltage of the MOS capacitor. Due to the free carrier plasma dispersion effect, the accumulated charges induce a refractive index change in the silicon. At a wavelength of 1.55 um, the index changes caused by electrons and holes, which were obtained from experimental absorption spectra through Kramers-Kronig analysis, are given by
Ane= 8.8 x 10-Ane= 8.8 x 10"22ANe (2) Anh= 8.5 x 10"18(ANh)08 (3)
Where electron and hole density changes are in units of cm-3. The change in refractive index results in a phase shift AO in the optical mode given by
(4)
Where L is the length of the phase shifter, A is the wavelength of light in free space, and Aneff is the effective index change in the waveguide, which is the difference between the effective indices of the waveguide phase shifter before and after charge accumulation. Because charge transport in the MOS capacitor is governed by majority carriers, device bandwidth is not limited by the relatively slow carrier recombination processes of pin diode devices. As a result, this capacitor-based design has allowed us to demonstrate bandwidth that is
Figure 11 (a): On-chip modulation voltage (VRMS) and photo-receiver output of an MZI containing a single 2.5 mm phase shifter (b): Phase shifter normalized response showing an intrinsic bandwidth of approximately 2.5 GHz. The device was biased into accumulation with
unprecedented in a silicon-based modulator.
a 3 V DC bias
To create a large-signal modulation, we used the MZI that has two 10 mm phase shifters. The MOS capacitor is again biased into accumulation with 3VDC. With an applied single-ended voltage swing of 1.6 V (3.2 V differential swing), the phase shifters should provide sufficient phase shift for the modulator to exhibit an extinction ratio of 5.8 dB when it is biased at quadrature.
SI-BASED PHOTODETECTORS
The final active optical component that would need to be integrated onto an all-silicon optical platform is the photodetector. Silicon photodetectors have already found wide acceptance for
visible light (0.400-0.700 um) applications because of their near perfect efficiency at those wavelengths. However, most communication-grade semiconductor lasers are operating in the near infrared wavelength (usually 0.850, 1.310, and 1.550 um), a region where silicon is a poor detector. In order to improve the performance of silicon-based detectors, the most common approach is to introduce germanium to reduce the bandgap and extend the maximum detectable wavelength. The effect on the absorption coefficient and depth of penetration clearly shown in Figure 13. Note that the data in Figure 13 represent unstrained bulk material with no voltage applied. By introducing strain or electrical bias, it is possible to shift the curves slightly to a higher wavelength due to a reduction in the effective bandgap. This could be critical for detection at 1.550 um, where a pure Ge film with the appropriate strain or bias could potentially be shifted to reduce the penetration depth to acceptable values.
Figure 13: Absorption coefficient and penetration depth of various bulk materials as a function of wavelength. The green lines mark the important wavelengths for telecommunications of 1.310 and 1.550 [im.
Two critical benchmarks for a photodetector are directly related to the absorption coefficient or penetration depth of the light responsivity, and bandwidth. The responsivity is the ratio of collected photocurrent to the optical power incident on the detector. Responsivities for commercial III-V photodetectors are typically close to 0.8 A/W. unless the detector is poorly designed; the responsivity should clearly increase as the absorption coefficient increases. The bandwidth of a photodetector can be limited by the transit time required for the photocarriers to travel to the contacts or the RC time constant. If the light penetrates 10 um into the material, for example, some photocarriers might have to travel 10 um back to the surface to be collected by a top contact. Good detector design eliminates the lethargic diffusion current by using very thin films that can be fully depleted to prevent the generation of diffusion current or effectively reducing the diffusion length of minority carriers. The collection of the much faster drift current is then optimized by keeping the depletion width as thin as possible as determined by the penetration depth. If the penetration depth can be kept to below 2 um, the transit time alone could support a bandwidth of 10 Gb/s. The way around this problem is to illuminate the device from the side. By doing this, the transit time can be kept low while the effective length of the detector is increased from a few micrometers to as long as a few millimeters. This is the approach used for waveguide-based photodetectors. Another advantage of the waveguide detector is its planar nature, which lends itself to integration with other optical devices.
We are using the same SOI platform as the modulator work to make SiGe waveguide-based photodetectors. A cross-section of this structure is shown in Figure 14 where the SiGe layer is directly on top of a silicon rib waveguide. Our initial detectors used 18 Si0.5Ge0.5 multiple quantum wells as the absorbing material, with a well thickness of 4 nm separated by 25 nm of silicon Figure 15. They were made on 2.5 and 4 um thick SOI wafers with waveguide widths varying between 2.5 and 15 um, and silicon rib etch depths sufficient to achieve multimode
operation. The responsivity was as high as approximately 0.1 A/W at 1.319 um for some devices. We believe that this can be increased to 0.5 A/W through a combination of increasing the number of quantum wells, and changing the placement of the SiGe in the waveguide, among others. Further improvements to responsivity would entail increasing the germanium concentration in the quantum wells, forcing them to even thinner structures to prevent relaxation. This reduced area counters the higher absorption coefficient, pushing any performance gain into the region of diminishing returns. The bandwidth of the devices was limited to below 500 MHz due to a large offset in the valence band that hindered the transport of holes. This can be fixed by altering the film composition, and modeling predicts that data rates approaching 10 Gb/s could be possible. The advantage of this device structure is that it is fully strained, meaning that few, if any, defects are formed in the active SiGe material. These defects are known to increase the dark current of the device that reduce the Signal to Noise Ratio (SNR). Higher optical power would then be needed to compensate and achieve acceptable Bit Error Rates (BER). Our best devices had a dark current of less than 1 uA (<1 nA/um2) at 3 V, which is acceptable for most applications. For comparison, InGaAs pin photodetectors typically have dark currents close to 1 nA.
Integration Issues with Germanium
The amount of germanium required for efficient photodetection is dependent on the wavelength. If detection at 1.310 or 1.550 um is desired, then very high (>40%) germanium concentrations are needed. This is much higher than that found in SiGe Heterojunction Bipolar Transistors (HBTs) or strained silicon, and as a result new integration issues have to be dealt with in the fab, including strain and stability.
Since most useful strained Si1-xGex films are metastable with respect to defect formation, exposing the wafer to high temperatures after growth can be problematic. Certainly, long times at temperatures above the growth temperature (550-650° C) should be avoided. Higher temperatures might be possible for short times, such as in rapid thermal annealing, but this is conditional on the film quality. Amorphorus, poly-crystalline, or relaxed single crystal films will not have this temperature limitation. Chemical stability is also an issue for films with high germanium concentration. Since germanium does not form a stable oxide like silicon does when exposed to oxidizing chemicals, the SiGe films tend to be susceptible to corrosion during wet cleans or Chemo-Mechanical Polishing (CMP). We have developed alternate processing modules to accommodate for the difference and maintain the integrity of the SiGe films.
OPTICAL COUPLING AND PACKAGING
One of the most difficult challenges facing high-index contrast optical systems is efficiently coupling light into and out of the chip. Particularly difficult is the coupling of light from a standard optical fiber or external light source to a silicon waveguide. Overcoming these challenges requires the development of processes and structures in addition to the core device.
Tapers
A single-mode fiber core (n = 1.5) usually has a diameter of 8 um with a symmetric mode while a silicon waveguide (n = 3.45) is typically only a few micrometers in width with an asymmetric mode. To overcome these large differences in effective index, core size, and symmetry, one frequently used method is to employ a waveguide taper. Tapers allow for a reduction in coupling loss through an adiabatic modal transformation and can also be used to increase the alignment tolerance of other optical devices, such as III-V lasers
Fiber Attach
In order to integrate the optical devices, discussed in this paper, into optical networks, they must be integrated with fibers. As discussed in the previous section, the small waveguide dimensions and high index contrast of the silicon system lead to a fundamental difference in the optical mode profile between the waveguide and fiber. The integration of waveguide tapers at the waveguide/fiber interface can solve this problem.
Current fiber attach techniques are "active," relying upon the closed loop optimization of fiber position in order to ensure low loss coupling. This technique is time consuming however and hence costly. Passive alignment techniques for fiber attachment remove the need for closed-loop optimization by creating highly precise lithographically defined structures on the silicon
surface in order to align the fiber to the waveguide aperture.
Figure 19: Scanning Electron Micrograph of several U-grooves, two of which are populated with optical fibers and aligned to silicon waveguides
Active alignment techniques are typically capable of placement tolerances better than 1 um. The accuracy required of a passive alignment technique will depend upon the mode field overlap of the fiber and waveguide modes, which can be controlled by the waveguides and tapers.
CONCLUSION
Although research in the area of planar optics in silicon has been underway for several decades, recent efforts at Intel Corporation have provided better understanding of the capabilities of such devices as silicon modulators, ECLs and SiGe detectors. Incorporating silicon in an ECL opens a path towards hybrid silicon photonic integration, or even a Silicon Optical Bench(SiOB) platform for silicon photonics. Silicon modulators operating at 2.5 GHz have demonstrated two orders of magnitude improvement over other known si-based modulators, with theoretical modeling indicating performance capabilities beyond 10 GHz. And initial results from SiGe photodetectors have shown the feasibility of monolithically integrated waveguide detectors. Through further research and demonstration of novel silicon photonics devices, we hope to continue bringing the vision of integrated silicon photonics into focus as a viable future for commercial opto-electronics.
REFERENCES
[i] Silicon Photonics: An Introduction, John Wiley and Sons
[ii] J Zhao, A Wang, PJ Reece, and M GAL, "Efficient silicon light-emitting diodes"
