Siliconization of Photonics

Intel IDF

Light Wave We have developed a silicon and germanium-based photo detector that can detect laser light signals at speeds of over 40 gigabits per second…it is the world;s best performing Silicon Germanium photo detector” remarked Mario Paniccia, an Intel Fellow and Director of the Photonics Technology Lab. That’s 40 gigabits per second with a dark current of less than 200 nanoamperes My ears perked up as a I was sitting in the main amphitheater inside San Francisco’s downtown Moscone Center. I knew Intel was investigating photonics applications, but was unaware of their most-recent achievements. Combining germanium with silicon has long been seen as an impossible design; the materials produced several significant physical shortcomings including dark current that made them unmanageable for practical use. I was taking in this technological precedence in the audience just behind the general press section at Intel’s Developer Forum (IDF).

IDF is an amalgamation of part technology expo, part applied science and part futurist dream. The three-day event heralds Intel’s advances and innovations in new devices and technologies to bolster company confidence from technology analysts, journalists and especially investors. At the same forum, Intel reinforced their processor roadmap which incorporates the concept of Terascale computing, which amounts to scaling dual and quad processors to multi-core systems upwards of 32, even 80 cores. Terascale computers would operate on terabytes of data all on a single machine. When combined with our recent breakthroughs in silicon photonics, these experimental chips address the three major requirements for terascale computing: teraOPS of performance, terabytes-per-second of memory bandwidth and terabits-per-second of I/O capacity, said Justin Rattner, Intel’s chief technology officer and senior fellow, in an earlier IDF. While any commercial application of these technologies is years away, it is an exciting first step in bringing terascale performance to PCs and servers[1]

Siliconization of Photonics

Intel Corporation is investing in a future of optical devices based on their well-developed Silicon fabrication techniques. Silicon has dominated the microelectronic industry as the ideal material to build electronic circuits. It is abundant, cheap, and readily oxidizes to produce an insulator, SiO2. As a semiconductor material it’s electrical properties can be highly controlled; Silicon can easily be doped to create wells to make tiny electronic MOSFET gates. Fabrication techniques exist to create almost any structure desired. State-of-the-art processes and technologies can now produce precise forty-five nanometer tolerances and complex quad-core processor designs exceeding two billion transistors. The idea is simple, leverage the billions of dollars spent for well-developed fabrication techniques from over forty years of development in the microelectronic industry and apply them to optical applications. However, there are many inherent problems using silicon with optical applications.

Silicon by nature is an indirect bandgap material; there is no practical mechanism for light emission as done in most semiconductor lasers. Many of the standard methods of modulation require materials that are simply incompatible with Silicon. Additionally many material properties like the Pockels effect, which is normally used for fast modulation, is not possible in Silicon. At frequencies of interest, Silicon is transparent which allows for the construction of waveguides, however this same property makes it impossible to make a receiver. Without overcoming and addressing these problems, silicon simply cannot not be used for optics.

Integration by Parts

Four main areas constitute the building blocks for a realizable on-chip solution: Light emission, on-chip modulator, Light guides, and photo-detection. In order to harness the microelectronic industry’s silicon fabrication abilities, all components are to be manufactured on Silicon and packaged into complete, single-chip solutions. The applications of such devices will provide for the computational bandwidth which will be required beyond the limits of current electrical-only devices. Several recent breakthroughs in on-chip signal modulation are opening avenues to make Silicon Photonics not only feasible, but competitive and equitable.

Light Emission

Traditionally gallium arsenide or indium phosphide have been used for light-emitting applications. Both of these materials have direct-bandgaps allowing electrons to move from higher energies to lower energies by emitting light. Electrons in indirect bandgap do not readily emit light. In order for electrons to relax in energy, they must interact with phonons or lattice vibrations of the crystal which produces heat instead of light. Silicon has an indirect bandgap and therefore cannot emit photons and achieve optical gain for lasing, however there is another process in which Silicon can exhibit lasing known as the Raman effect or Raman Scattering.

Raman scattering occurs when photons strike an atom and produce photons at a different frequency than the incident photon. These photons interact with other atoms which stimulates more photons producing amplification. It is a similar principle harnessed for optical amplification in silica by doping with Erbium. Silicon’s Raman scattering is more than 10,000 times stronger than silica which makes it a possible means of producing light[2]. In order for this effect to be utilized, losses in waveguides and material absorption need to be as small as possible.

In order to produce a silicon laser based on the Raman effect, the scattering requires an external photon source to pump the material. Because Raman amplification is small relative to other lasing amplifications, it requires a high-pump intensity. As these high-energy photons collide with atoms inside a silicon crystal, gain within silicon can be achieved[3][4]. However, the increased power of these pump photons also ejects electrons from the silicon atoms. These free carries act as recombination sites for photons known as two-photon absorption. As free carries take on photons which completely destroy the lasing process. This effect can be overcome by allowing electrons time to recombine by pulsing the laser, but turning the lasing on and off produces the detrimental effect of chirp. Chirp ultimately produce erroneous data at the output because different frequencies produced by the chirp propagate at different speeds[5].

Continuous lasing in Silicon was considered unfeasible, until techniques were explored to remove unwanted carriers. Integration of a diode-like feature can remove electrons that normally accumulate in the amplifier. Panniccia’s team at Intel created a diode by implanting electrodes on either end of the active region which effectively sweep away electrons and allow Silicon to undergo continuous lasing. The gain cavity is a Silicon-on-insulator design which uses a thin layer of SiO2 sandwiched between two layers of silicon. The light is confined in the SiO2 which has higher index of refraction. SOI designs were originally developed to improve power for portable electronics. Other materials must provide the light for lasing.

One approach to provide the necessary pumping energy is to use indium phosphide. A hybrid laser was developed by Professor John Bowers of the University of California Santa Barbara in tandem with Intel. The approach is to bond InP to a Si waveguide through tight control of the manufacturing process. Oxygen plasma is used to fuse the two materials together with an oxide layer 25 atoms thick[6]. By putting a contact on the InP, current can be made to flow producing light which couples with the waveguide. This process frees alignment issues by fabricating the laser directly with the silicon. The latest breakthrough using a similar process have succeed in making a 39.2GHz laser with 206mA on the gain section and 0.4 V forwards bias on the transistor-like absorber. This approach allows for high-speed synchronization applications by generating and distributing low phase noise clock signals without any RF drive circuitry[7].

The latest designs incorporate III-V materials into fully functional mode-locked lasers. Mode locked lasers have high extinction ratios, low jitter, low chirp and can be easily modulated. Bower’s group have achieved 40GHz with low jitter and extinction ratios that rival III-V based semiconductor Mode-Locked Lasers[8]. When these designs are combined with special topologies, adjusting the frequency of the mode-locked laser becomes as simple as adjusting the geometry. One particular promising topology couples around onto itself in a racetrack shape[9]. Adjusting the cavity length of the track selects the frequency of oscillation of the mode-coupled laser. The result is a repeatable design that has direct control of laser characteristics through fabrication.

On-chip Modulator

High-speed modulation is usually achieved by Pockels effect in which the refractive index of a material can be modified in proportion to the applied electric field. Among the materials that exhibit this effect are semiconductors such as gallium arsenide and indium phosphide, but not Silicon. Without this linear birefringence, modulation is limited to other mechanisms such as thermal or introduction of free carriers. Modulation in silicon using this free-carrier dispersion effect have only shown around 20MHz[10] with theoretical maximum bandwidths of 1GHz[11]—far too low for even the slowest optical systems. Normally direct modulation is achieved by the means of lithium niobate, which has a strong piezo-electric effect. With lithium niobate, an applied electric field changes the speed at which light propagates through it. Silicon does not exhibit this property, instead, carriers must be injected into the device to induce a phase shift—however, the maximum speed is limited by the ability to move charge carriers.

Complete light modulation occurs by splitting and phase-shifting one branch with respect to the other. By phase-shifting relative to one another and recombining, a full phase shift results in a amplitude modulation of zero as the intensities of light effectively cancel one another out. In February of 2004, Intel announced the first Gigahertz silicon modulator which integrated a scalable, transistor-like device that was able to quickly move charge carriers and transmit up to 10Gb/s. The total length of the entire modulator tested by Intel was 15 mm including the input and output waveguides and splitters. An applied voltage induced charges near the gate dielectric of the capacitor which changed the phase of any light passing through it. High-speed circuitry was used to drive the transistor-like device making it capable of inserting electrons and holes on opposite sides of the oxide layer. In this design the driving circuitry was simply wirebonded to pads and used 3.3v; the total on-chip loss was 10dB[12]. At the time, it was predicted that improvements to the circuitry and optimizing doping profiles would allow for 1v CMOS driving circuitry with as little as 2 dB on-chip loss[13].

Light Guides

The frequency of choice for optical communication has long been 1550nm due to the low losses in silica optical fibers at that frequency. Silicon is transparent in the near infrared spectrum which makes it a natural light guide. Waveguides structures use the same SOI approach as lasers by fabricating oxide channels within silicon. They can be made using standard lithography techniques by etching to make the channels. To date, sizing of waveguides has been matched to the component of interest. Whenever optical fibers were coupled into a system, tapered channels allowed for easy connection. Tapers behave like funnels to bring light in from fibers or bring it back out. Coupling loss between a tapered end and fiber was measured to be 4dB[14]. The surfaces where light enters and exists must be highly polished as imperfections and surface roughness highly degrade transmission. A successful system with lasing requires amplification must exceed the absorption of these waveguides.

Optical Detectors

For the same reason that Silicon is a good waveguide makes it an abysmal optical detector; optical transparency makes silicon useful for guiding light, but impossible for detection. Detection can only be achieved by the absorption of light at the specified signal frequency—Silicon alone is unable to do this. Germanium is a material that can absorb light in the 1300 to 1600nm range and it is CMOS compatible. Compared to other, more exotic materials Germanium tends to be cheaper and more efficient as a siliconized receiver.

Mario J. Paniccia’s group at Intel’s Santa Clara Photonics lab have been exploring Germanium-on-silicon detectors. By placing a thin strip of germanium on the top of a silicon waveguide, light gets evanescently coupled into germanium from the waveguide. A vertical p-i-n junction can be formed by placing germanium on top of p-type silicon waveguide with a n+ germanium contact on top. One of the main advantages with this design is that light interacts with the germanium along the length of the waveguide as opposed to interacting solely at surface. The longitudinal geometry improves sensitivity and enables higher frequencies of operation[15]. With a -2 volt bias, Paniccia’s group have shown about a a 30GHz bandwidth with a maximum speed of 31 Gigabits per second—sufficient receiver speeds for optical communication[16].

Depositing Germanium is not difficult; getting the two materials to properly bind without defects is challenging. Mismatches in the crystal lattices of germanium and silicon cause defects at the interface due to the additional strain caused by the larger germanium lattice spacings. Dark current occurs when electrons and holes are generated within the depletion region of the device and get swept away by the high electric field. The additional charge generation is directly related to crystallographic defects within the depletion region. These defects can be removed by improving the process of deposition. By optimizing thermal growth conditions, the impact of the defects can be minimized in the edges where they would impede performance. Controlling the deposition process allows for better match between layers of the germanium and the silicon. How to exactly depositing uniform layers on top of silicon to minimize strain is now an Intel trade secret.

The latest design can detect rates of 40 gigabits per second. The detector is said to more efficient and with the capability to recover a cleaner signal[17]. Paniccia’s group have produced all-optical clock recovery units that utilize a ring structure. The exact structure is defined by lithographic techniques which makes integration and matching to any desired bit rate possible. Pulses up to 30.37 GHz with very low jitter were achieved using this design[18].

“This is a significant step toward building optical devices that move data around inside a computer at the speed of light. It is the kind of breakthrough that ripples across an industry over time, enabling other new devices and applications. It could help make the Internet run faster, build much faster high-performance computers and enable high-bandwidth applications like ultra-high-definition displays or vision recognition systems” explained Pat Gelsinger, senior vice president and chief technology officer[19].


Silicon optics combine the repeatability and low-cost of Silicon fabrication processes with the power and speed of optical communications. As high-speed, high-bandwidth demands continue to push microelectronics to faster and more data-intensive designs, companies are looking to alternatives of the copper interconnect on and between chips. Applications of siliconized optics have the potential to open entirely new areas in technology: Network communication by including optical-electrical conversion embedded into fiber links themselves, silicon lasing for biomedical devices, and especially addressing the multi-core pipeline data deficiency. The ability to produce, modulate, guide, and receive light have been proven functional on silicon-based designs. The immediate concern is the integration of the individual components into an all-in-one package. As Tao Yin, An Intel Research Lab Engineer, remarked, “Most of the work has to do with combining the packaging all three components—the hybrid silicon laser, the modulator, and the photo detector—onto a single silicon chip. Once mass production of the optical communications chips commences, it will become possible to use them in order to integrate computers and optical devices and transmit terabits of aggregate data per second. Intel hopes to enable tera-scale computing in the the not too distant future[20].”


  1. Intel Develops Terascale Research Chips, September 2006.

  2. Paniccia, Mario. Koehl, Sean. The Silicon Solution. IEEE Spectrum. 19 November 2008.

  3. H. Rong, A. Liu, R. Jones, O. Cohen, D. Hak, R. Nicolaescu, A. Fang, and M. PanicciaAn All-Silicon Raman Laser, Nature, 20 Jan 2005.

  4. M. Paniccia, A. Liu, H. Rong, R. Jones, O. Cohen and D. Hak, Optical Amplification and Lasing by Stimulated Raman Scattering in Silicon Waveguides, Journal of Lightwave Technology, Mar 2006.

  5. Rong H. et. al. Nature 433, 725-728 (2005).

  6. H. Park, A. W. Fang, S. Kodama, and J.E. Bowers, “Hybrid Silicon Evanescent Laser Fabricated with a Silicon Waveguide and III-V offset Quantum Wells,” Optical Express, 13, 9460-9464, November 2005.

  7. B. R. Koch, A. W. Fang, H.-H. Chang, H. Park, Y.-H. Kuo, R. Jones, O. Cohen, O. Raday, M. J. Paniccia, J. E. Bowers, A 40 GHz Mode Locked Silicon Evanescent Laser, 4th International Conference on Group IV Photonics, Tokyo Japan, wb1 (invited), 19 Sep 2007.

  8. B. R. Koch, A. W. Fang, O. Cohen, and J. E. Bowers, Mode-locked silicon evanescent lasers (PDF), Optics Express, Vol. 15, No. 18, pp. 11225-11233, 21 Aug 2007.

  9. A. W. Fang, B. R. Koch, K. Gan, H. Park, R. Jones, O. Cohen, M. J. Paniccia, D. Blumenthal, J. E. Bowers, A racetrack mode-locked silicon evanescent laser, Optics Express, Vol. 16, No. 2, pp. 1393-1398, 17 Jan 2008.

  10. G. T. Reed, The Optical Age of Silicon, Nature, 12 Feb 2004.

  11. C. E. Png, G. T. Reed, R. M. H. Atta, G. J. Enseil, A. G. R. Evans, Proc. SPIE 4997, 190-197 (2003).

  12. L. Liao, D. Samara-Rubio, M. Morse, A. Liu, D. Hodge, D. Rubin, U. Keil, and T. Franck, High speed Silicon Mach-Zehnder Modulator, Optics Express, 18 Apr 2005.

  13. D. Samara-Rubio, U. Keil, L. Liao, T. Franck, A. Liu, D. Hodge, D. Rubin, and R. Cohen, Customized Drive Electronics to Extend Silicon Optical Modulators to 4 Gb/s, Journal of Lightwave Technologies, Dec 2005.

  14. H. Rong, R. Jones, A. Liu, O. Cohen, Dani Hak, Alexander Fang, and Mario PanicciaA Continuous-Wave Raman Silicon Laser, Nature, 17 Feb 2005.

  15. T. Yin, R. Cohen, M. M. Morse, G. Sarid, Y. Chetrit, D. Rubin, and M. J. Paniccia, 31 GHz Ge n-i-p waveguide photodetectors on Silicon-on-Insulator substrate, Optics Express, Vol. 15, Issue 21, pp. 13965-13971, 9 Oct 2007.

  16. Hitz, Breck, Integrated Germanium-on-Silicon Detector Opens the Eye at 40 Gb/s, Optics Express, Oct 17, 2007, pp. 13965-13965-13971.

  17. K. Greene, Intel Completes Photonics Trifecta. Technology Review, MIT, 10 October 2007.

  18. B. R. Koch, A. W. Fang, H. N. Poulsen, H. Park, D. J. Blumenthal, and J. E. Bowers, R. Jones and M. J. Paniccia, and O. Cohen, All-Optical Clock Recovery with Retiming and Reshaping Using a Silicon Evanescent Mode-Locked Ring Laser, Optical Fiber Communications Conference (OFC) 2008, Paper OMN1, Feb 2008.

  19. Intel Unveils Silicon Photonics Breakthrough: High-Speed Silicon Modulation, Technology@Intel Magazine, p1. February/March 2004.

  20. O. Reshef, Intel’s High Speed Optical Trinity, The Future of Things, 24 October 2007.

Comments are closed.



  • slide
  • slide
  • slide
  • slide