Physics and Technology Forefronts

Silicon Lasers

By Bahram Jalali

Figure 1. Difference of light emission mechanisms between direct and indirect bandgap materials.

Silicon, the wonder material of the 20^th century, appears to have one more trick up its sleeve. After dominating the digital electronic industry, the bedrock of the digital world, it is now being considered as a platform of choice for building optoelectronic devices. Silicon’s mature and large-scale manufacturing base is believed to be precisely what is needed to effect a much needed reduction in the cost of photonic devices that are currently being manufactured in compound semiconductors such as Gallium Arsenide and Indium Phosphide. Such a cost reduction can bring the power of optical networks to the desktop computer and to home entertainment systems. Imagine downloading digital movies in seconds! The technology, termed silicon photonics, would be compatible not only with complementary metal-oxide-semiconductor (CMOS)-based electronic integrated circuits, but also with micro-electro-mechanical systems (MEMS). It could enable a new generation of electro-opto-mechanical chips that perform the job of today’s complex systems at a fraction of the cost, size, and power dissipation.

For some time now, it has been known that when surrounded by its native oxide (SiO₂), otherwise known as ordinary glass, silicon becomes a low-loss optical waveguide at wavelengths of 1300 nm and 1550 nm, the regions of the spectrum where optical networks operate. Using such waveguides as basic building blocks, optical filters and switches have been developed, and by adding Germanium to the chip, photo-detectors have also been produced. In other words, most of the building blocks that make up an optical communication system are already available. If we also had optical amplifiers and lasers, the tool box would be complete.

Silicon light sources and amplifiers have remained elusive because of the material’s bandstructure. Silicon has an indirect bandstructure, which means that the upper and the lower electronic states (conduction and valence bands) do not occur at the same value of crystal momentum (Fig.1). Because visible or infrared photons have negligible momentum (compared to that of the electron), the de-excitation of an electron (recombination) needs to be mediated by emitting or absorbing a phonon in order to conserve momentum. Such 2^nd order radiative recombination events do not occur frequently, as characterized by a very long lifetime that is in the order of one second. The electron in the upper state has to sit and wait until a phonon with the right momentum shows up. While waiting, it becomes prone to non-radiative recombination with the energy being dissipated as heat. The experimentally measured lifetime in silicon is in the millisecond to microsecond range, depending on the impurity or defect concentration. This suggests that the desired radiative processes are insignificant compared to undesired nonradiative recombination. Even when using the highest purity silicon, devices have an electrical to optical conversion efficiency of only 10^-4 to 10^-3. The situation would be different if one could break free of the low radiative recombination rate of bulk silicon.

According to the Heisenberg Uncertainty Principle, when the electron is localized, its momentum becomes uncertain. This phenomenon may offer a solution to the indirect bandgap of silicon. An interesting case study is the semiconductor, GaP, which is used for light emitting devices despite its indirect bandgap. Here the momentum conservation requirement is relaxed when an electron is localized at a Nitrogen impurity site. A possible technique to create quantum confinement is the use of silicon nanocrystals that occur naturally in a Silicon Rich Oxide (SRO) thin film. When a SiO_x (x < 2) film is subjected to high temperature anneal, the excess silicon leaves the oxide matrix and forms nanometer size grains of crystalline silicon dispersed throughout the oxide. The nanocrystals are excited by pumping the material with a high intensity light beam. Several research groups, notably at the University of Trento in Italy and at the University of Rochester, NY, have reported optical amplification in these films. However, the observed characteristics cannot be explained on the basis of electron localization in the nanocrystals, and the observations are highly dependent on how the sample was prepared. Consequently, questions remain regarding the nature of the observed optical gain. Additionally, the emission wavelength is in the 800-900 nm range, i.e., outside of the important telecommunication bands.

Other attempts at exploiting quantum confinement in defect sites or in intentionally formed quantum structures are showing promise. Researchers at Brown University have reported evidence of lasing at cryogenic temperatures (≤70 K). A two dimensional array of nanometer size holes were etched into a thin film of silicon that resides on oxide layer. The sample was cleaved (forming mirrors) and pumped optically. The mechanism for light emission (1270 nm wavelength) is believed to be from defects on etched silicon surfaces. In anther report, a team consisting of researchers at the University of California, Irvine, and in Taiwan used a junction diode in which the dopants were confined to nanometer size regions. They reported compelling evidence of stimulated emission at room temperature. Optical amplification was not demonstrated, but the room temperature operation and the electrical pumping feature are indeed intriguing.

The successful realization of light emission and amplification in optical fibers that are doped with Erbium has motivated efforts aimed at Erbium in silicon. However, silicon is not a good host for Erbium resulting in poor emission at room temperature. The reason is believed to be the back-transfer of energy from the excited Er ions to silicon and also the low concentration of Erbium that can be accommodated by silicon.

But optical fiber is made from SiO₂, so why not add Erbium to SRO? Similar to its fiber counter part, Erbium doped SRO exhibits light emission at the technologically important wavelength of 1550 nm. The material can be excited electrically, by sandwiching the film between a silicon and a metal film. A voltage applied to the so-called Metal-Oxide-Semiconductor (MOS) structure causes electrons to tunnel through the oxide, and in the process, to excite the Erbium atoms. Silicon nanocrystals also become excited and transfer their energy to nearby Erbium ions. Light emitting diodes with efficiencies of about 10%, as high as commercial Gallium Arsenide devices, have been reported by STMicroelectronics in Italy. Unfortunately, energetic electrons injected into the oxide (by tunnelling) cause premature device failure (a link between “hot” electrons and device failure is well established in the electronics industry). In addition, the maximum output power of silicon LEDs is currently far less than their Gallium Arsenide counterparts. Nonetheless, the observation of efficient electrical to optical conversion is a major development, and research on means to overcome the current limitations is underway.

Under funding from the Defense Advance Research Projects Agency (DARPA) in 2001, my group at UCLA proposed the use of Stimulated Raman Scattering (SRS) as a mean to realize amplification and lasing in silicon. So far, this has been the most successful approach and has yielded high gain amplifiers and efficient pulsed lasers. Raman back-scattering in silicon was used in the late 60’s and early 70’s as an analytical tool to study the vibrational properties of the material. More recently, SRS has been exploited in optical fibers to create amplifiers and lasers. However, several kilometers of fiber are typically required to create a useful device, suggesting that the approach is not applicable to silicon. Often overlooked was the fact that the gain coefficient for SRS in silicon is approximately 10³ to 10⁴ times higher than that in silica fiber. Additionally, owing to the large refractive index of silicon, silicon waveguides can confine the optical field to an area that is approximately 100 to 1000 times smaller than the modal area in a standard single-mode optical fiber, resulting in proportionally higher Raman gain. When combined, these facts make it possible to observe SRS over the interaction lengths encountered on a chip. This was validated through experimental demonstration of spontaneous and stimulated Raman scattering in silicon waveguides in 2002-2003 at UCLA. The first Si laser was demonstrated by us in 2004 and made use of the Raman effect under pulsed pumping. The laser showed a surprisingly high slope efficiency of 34% (as good as conventional lasers) and peak pulse power of over 2.5 Watts. The work was followed by Intel Corporation’s demonstration of a continuous-wave (CW) Raman laser in 2005. At present, CW operation is frustrated by low output power and efficiency that are caused by accumulation of electrons that are generated by two photon absorption. Unfortunately, a diode used to sweep out the electrons is only partially effective.

Fortuitously, the two photon absorption vanishes in the mid infrared. This along with silicon’s excellent thermal conductivity and high optical damage threshold makes the silicon Raman laser an ideal technology for power combining (taking several low power lasers and combining their beams to create a high power lasers) and wavelength shifting. A joint UCLA/Northrop Grumman team is investigating high power mid infrared sources by combining a silicon Raman laser with low power and shorter wavelength lasers. This technology is aimed at biochemical detection, with the potential to also be applicable to free space optical communication and eventually to defense against heat seeking missiles.

Bahram Jalali is Professor of Electrical Engineering at UCLA. He also serves as Chair of the UCLA Optoelectronic Circuits and Systems Laboratory, and is Director of the DARPA Consortium for Optical A/D System Technology.