recent progress in lasers on silicon

recent progress in lasers on silicon

(Parte 1 de 3)


The photonics market today is shared by several materials systems, including compound semiconductors (indium phosphide, InP, and gallium arsenide, GaAs), elementary semiconductors (silicon, Si, and germanium, Ge), silica and rare-earth-doped glasses (glass fi bre, for example) and polymers. Each system targets particular applications or components. Today, the use of Si photonics is dwarfed by compound semiconductors and Si microelectronics, mostly due to the problems associated with making Si a host material for effi cient light emission, and thus subsequently realizing a laser. Fift y years ago the birth of the laser started a scientifi c and technological revolution. Two years later, diode lasers were demonstrated in group i–v compound semiconductors, and this was around the same time that Si-based transistor radios achieved mass popularity. Since then many scientists and engineers have researched lasing on Si substrates1. Rapid advances in Si photonics over the past two decades have been driven not only by the need for more complex, higher functionality and lower cost photonics integrated circuits, but also by pin count and power limits for communications, as summarized in the International Technology Roadmap for Semiconductors (ITRS)2. Electronics giants such as Intel, IBM, Hewlett Packard, STMicroelectronics, IMEC and Alcatel-Th ales have teamed up with research institutes around the world with support from government, industry and academia to drive progress in Si photonics. Th e current momentum and potential for making a useful laser in or on Si are signifi cant.

Fundamentals At the time of the demonstration of the fi rst laser fi ft y years ago, the fundamental hurdle to realizing stimulated emission in Si was understood: optical transitions must obey the laws of conservation of energy and momentum, but these conditions are not satisfi ed simultaneously in crystalline Si. In direct bandgap materials (GaAs and InP, for example) radiative recombination occurs rapidly and effi ciently via a simple two-particle process, as shown by the simplifi ed band diagram in Fig. 1 (left ). Direct bandgap materials have a structure in which the lowest energy points of both the conduction and valence bands line up vertically in the wave vector axis; that is, they share the same crystal momentum. Th is is the principal reason why GaAs-, InP- and GaN-based materials have been the dominant material systems for semiconductor diode lasers since their fi rst demonstration in 1962.

Si, like Ge, is an indirect bandgap material, and is not naturally capable of accomplishing effi cient radiative recombination. Free electrons tend to reside in the X valley of the conduction band, which is not aligned with free holes in the valence band (Fig. 1, right). Th erefore if a recombination is to lead to emission of a photon, a third particle must be involved to carry away the excess momentum,

Recent progress in lasers on silicon

Di Liang* and John E. Bowers

Silicon lasers have long been a goal for semiconductor scientists, and a number of important breakthroughs in the past decade have focused attention on silicon as a photonic platform. Here we review the most recent progress in this fi eld, including lowthreshold silicon Raman lasers with racetrack ring resonator cavities, the fi rst germanium-on-silicon lasers operating at room temperature, and hybrid silicon microring and microdisk lasers. The fundamentals of carrier transition physics in crystalline silicon are discussed briefl y. The basics of several important approaches for creating lasers on silicon are explained, and the challenges and opportunities associated with these approaches are discussed.

which results in slow optical transition rates. A major non-radiative process is Auger recombination, in which an electron (or hole) is excited to a higher energy level by absorbing the released energy from an electron–hole recombination. Th e Auger recombination rate increases with injected free-carrier density and is inversely proportional to the bandgap. Free-carrier absorption (FCA) represents another major non-radiative process wherein the free electrons in the conduction band can jump to higher energy levels by absorbing photons. In high-level carrier injection devices (lasers and amplifi ers, for example) or heavily doped layers, free-carrier loss is orders of magnitudes higher than the material gain1. For both Auger recombination and FCA, the electrons pumped to higher energy levels release their energy through phonons, rather than by emitting photons. Th ey also have much shorter lifetimes (τnonrad) than those of radiative processes (τrad) in Si, resulting in an extremely poor internal quantum effi ciency ηi of light emission, which is defi ned as3 τnonrad τnonrad + τrad ηi = and is generally of the order of 10–6. Consequently, semiconductor laser research over the past fi ft y years has primarily focused on compound semiconductor substrates, but now there is intense interest in lasers on Si.

Department of Electrical and Computer Engineering, University of California, Santa Barbara, California 93106, USA. *e-mail:

Ener gy X

Wave vector


Holes hv

InP Si

Direct recombination

Free-carrier absorption

Auger recombination

Phonon Indirect recombinationГ

Figure 1 | Energy band diagrams and major carrier transition processes in InP and silicon crystals. In a direct band structure (such as InP, left), electron–hole recombination almost always results in photon emission, whereas in an indirect band structure (such as Si, right), free-carrier absorption, Auger recombination and indirect recombination exist simultaneously, resulting in little photon emission.


Th e recent and widespread availability of nanotechnology has allowed the traditional phonon-selection rule in indirect bandgap materials to be relaxed by breaking the crystal-symmetry or by phonon localization through the creation of nanostructures in crystalline Si. Th e motivation is to achieve quantum confi nement of excitons in a nanometre-scale crystalline structure4. A number of groups have reported enhanced light-emitting effi ciency and optical gain in low-dimensional (that is, of the order of the de Broglie wavelength) Si at low temperatures. Th ey include porous Si5–8, Si nanocrystals9–12, Si-on-insulator (SOI) superlattices13 and photoniccrystal-like nanopatterns14, and Si nanopillars15,16. However, achieving room-temperature continuous-wave (CW) lasing based on these temperature-dominated processes remains a challenge3,17,18.

Despite being fundamentally limited by an indirect bandgap and low mobility, Si exhibits a number of important properties that make it a good substrate, if not necessarily a good gain medium for diode lasers. First, Si wafers are incredibly pure and have low defect density. Second, state-of-the-art 32 nm complementary metal–oxide–semiconductor (CMOS) technology is suffi ciently advanced to fabricate virtually all Si photonic components, which are mostly still in the micrometre regime. Both factors allow for Si waveguides with propagation losses that are typically one order of magnitude lower than compound semiconductor waveguides. Furthermore, Si has a high thermal conductivity, which is a very useful characteristic for an active device substrate. SiO2, the highquality native oxide of Si, serves as a protective layer and a naturally good optical waveguide cladding, owing to its large refractive index diff erence from Si (Δn ~ 2.1). Th is is one of the major advantages of Si over Ge and other semiconductors for use in integrated circuits. Further loss-reduction in Si waveguides by oxidation19 and hosting rare-earth doping in SiO2 brings additional benefi ts to passive Si lightwave circuits. Although low waveguide loss does not change the ultralow band-to-band radiative emission effi ciency in Si, it improves the effi ciency of Si lasers that rely on a nonlinear eff ect such as Raman scattering.

Silicon Raman lasers Th e Raman eff ect refers to the inelastic scattering of a photon by an optical phonon. When incident light is absorbed by an atom or molecule at a vibrational state, the system energy is raised to an intermediate higher state. In most cases, the energy quickly drops back to the original vibrational state by releasing a photon with the same frequency, which is known as Rayleigh scattering, and is analogous to elastic scattering. Yet it is also possible to observe very weak (approximately one in ten million photons) additional components with lower and higher frequencies than the incident light due to the absorption or emission of optical phonons, namely the Stokes and anti-Stokes transitions, respectively.

If a scattering medium is irradiated with pump and signal beams simultaneously, the pump beam excites the constituent molecules or atoms to a higher vibrational level, while the signal beam, which has a frequency resonant at the Stokes transition, triggers the generation of another Raman Stokes photon. Th us, amplifi cation can be achieved through stimulation of the Stokes transition. Th is technique is known as stimulated Raman scattering, and has enabled the realization of Raman glass fi bre amplifi ers with gain bandwidths of over 100 nm. Th e Raman gain coeffi cient in Si is around fi ve orders of magnitude larger than that in amorphous glass fi bres because of the well-organized single-crystal structure20. However, Si waveguide loss is also several orders of magnitude higher than in glass fi bre, making fabrication of a low-loss Si waveguide one of the keys to realizing net Raman gain in Si. Furthermore, the tight optical confi nement in an SOI waveguide leads to an ultrasmall waveguide eff ective area, which in turn lowers the pump power threshold for stimulated Raman scattering. A pump with energy well below the Si bandgap is typically used to avoid elevating the electrons up to the conduction band and also to suppress FCA — both of which prevent lasing in Si. Initial studies demonstrated up to 0.25 dB of stimulated Raman gain for a Stokes signal at 1,542.3 nm for SOI waveguides, using a 1,427 nm pump laser with a CW power of 1.6 W (ref. 21). Such high pump powers, however, induce another optical loss mechanism — two-photon absorption (TPA). TPA is a nonlinear loss mechanism in which two photons combine their energies to boost an electron in the valence band to the conduction band. Free carriers further induce FCA and dump more optical power inside the cavity. TPA increases with the number of photons in a waveguide, and therefore becomes a limiting factor when using high optical pump powers. Th e fi rst demonstration of a pulsed Si Raman laser22 overcame TPA by using a long delay together with a short optical pulse, thus allowing the carriers generated during TPA to recombine prior to the next pass of the optical pulse. Following demonstrations used a p-i-n (p-type/intrinsic/n-type layers) structure in the waveguide to sweep free carriers away under

Pump power (mW)Laser output (mW) Wavelength (nm)

80 dB p-region n-regionV bias

Directional coupler Iinc

Ring cavity

Laser output

Pump Bus waveguide zIp(0)Ip(L)

Si substrate

Al contact

Buried oxide n-regionp-regionSi rib waveguides

Al contact SiO2 passivation

1 μm

25V 10V 5V 0V

R elativ e spectr al po wer ( dB ) b cd

Figure 2 | Low-threshold Si Raman racetrack ring laser. a, Schematic of a device with a p-i-n junction design. I(0) and I(L) are the pump power at the starting point and after a round trip in the cavity, respectively. The light propagation direction is given by z. b, SEM cross-section of a directional coupler and p-i-n junction region. c, Laser output power against coupled input pump power, showing a higher output power achieved at a higher reverse bias on p-i-n junction for a 3 cm cavity. The error bars here are derived from diff erent measurement traces. d, High-resolution spectrum showing a lowthreshold Si Raman racetrack ring laser with a side-mode suppression ratio of over 70 dB. Figure reproduced from ref. 25, © 2007 NPG.


NATURE PHOTONICS | VOL 4 | AUGUST 2010 | 513 a reverse bias23, as this reduced the free-carrier lifetime to minimize TPA-induced FCA. An alternative method involved reducing the volume-to-surface ratio of the waveguide to increase the surface recombination rate of the carriers. Th e fi rst successful demonstration of a CW Si Raman laser followed soon aft er24, with a lasing threshold at an eff ective pump power of ~182 mW for a reverse bias of 25 V.

Th e threshold pump power was recently reduced using a high

Q-factor racetrack ring resonator cavity and an optimized p-i-n diode structure25. Th e cavity resonance eff ect enhances the light fi eld inside cavity. Figure 2a is a top-view schematic of the racetrack ring cavity, with the p-i-n regions highlighted. A large bend radius of 400 μm helps to minimize waveguide bending losses, even though high-index-contrast SOI waveguides can typically support low-loss light propagation in a more compact bending structure. To utilize the pump power effi ciently and achieve a low lasing threshold, the directional coupler in the 1.6-cm-long bus waveguide is designed so that it is close to the critical coupling for the pump wavelength (1,550 nm) but has low coupling for the Stoke signal wavelength (1,686 nm). Th e narrow gap in the evanescent coupler was fi lled with boron phosphorus Si glass to eliminate any air voids that form, which helps to reduce losses. A thin layer of SiO2 buff er layer on top of the Si surface is deposited before coating with boron phosphorus

Si glass to prevent phosphor and boron from diff using into the Si during the thermal refl ow step and inducing FCA later on25.

TPA-induced FCA nonlinear optical loss can also reduced by optimizing the p-i-n reverse-biased diode. Balancing the trade-off between a good metal/Si contact and induced free-carrier absorption loss ensures that the diff usion of electrons and holes under reverse bias produces a uniform fi eld across the optical mode, which allows effi cient carrier removal without signifi cantly increasing the linear optical loss. Figure 2b shows a scanning electron microscopy (SEM) cross-sectional image of the directional coupling region and incorporated p-i-n diode structure. Th e average optical loss of this particular racetrack ring was measured to be 0.20 ± 0.05 dB cm–1, which indicates a negligible contribution of the p-i-n diode to the linear propagation loss when the p- and n-region separation is greater than 6 μm (ref. 25). An extremely short free-carrier lifetime of <0.4 ns was obtained for this device, resulting in a substantial reduction in the lasing threshold. Under a reverse bias of 25 V, the laser had a threshold of 20 mW and a maximum output power of 50 mW (Fig. 2c). Th ese represent fi ve- and tenfold improvements, respectively, over the fi rst CW Si Raman lasers24. As the bias voltage is lowered, the laser output begins to saturate earlier, owing to the relatively longer eff ective carrier lifetime. However, the lasing threshold changes only slightly because the TPA-induced FCA is much weaker at lower pump powers around the threshold. Silicon Raman lasers benefi t signifi cantly from high spectral purity, which results from the absence of a linewidth enhancement (a common eff ect in diode lasers). For example, linewidths of <100 kHz and side-mode suppression ratios of over 70 dB (see Fig. 2d) are well beyond the best performance of present diode lasers25.

Epitaxial lasers on silicon Compared with Si, GaAs and InP have lattice mismatches of 4.1% and 8.1%, respectively, and thermal expansion coeffi cient mismatches of 120.4% and 76.9%, respectively. Th ese result in a threading or misfi t dislocation density of 108–1010 cm–2 when either compound is grown on a Si substrate26. Numerous approaches, including special surface treatment27, strained superlattices28,29, low-temperature buff ers30 and growth on patterned substrates31 have been used to reduce the dislocation density to around 105–106 cm–2, but this is still around two orders of magnitude higher than in InP- or GaAs-based epitaxial wafers for room-temperature CW lasers. Recent advanced epitaxial techniques with SiGe32,3 and

GaSb34 buff er layers have enabled the realization of GaAs-based CW diode lasers on Si substrates at room-temperature. However, their reliability still remains a big issue for any future practical application. An exciting approach is the epitaxial growth of compound semiconductors lattice-matched to Si, such as GaNAsP35,36.

Another exciting approach is Ge-on-Si (or SiGe-on-Si) epitaxial growth. Key photonic components from this material system, including p-i-n37 and avalanche photodetectors38,39 and modulators40,41, have demonstrated performances comparable or even better than their i–v counterparts in certain aspects. Pure Ge has a signifi cant mismatch with Si in terms of its lattice constant and thermal expansion coeffi cient. Germanium has an indirect band structure, but the energy gap (0.8 eV) from the top of the valence band to the momentum-aligned Γ valley is close to the actual bandgap (0.6 eV), which increases the chance of radiative recombination between the Γ valley and the valence band. Th e larger thermal expansion coeffi cient of Ge naturally leaves thermal tensile strain in Ge aft er growth on a Si substrate, and a moderate tensile strain of 0.2–0.25% is able to reduce the energy diff erence between the Γ and L valleys to 115 meV (refs 42,43). In addition, strain raises the light-hole band, which increases optical gain for high injection42. Free electrons, incorporated through heavy n-doping, quickly fi l up the L valey to a level equal to that of the Γ valey, which increases the probability that those free carriers will begin to occupy the Γ valley for radiative recombination. Th ese techniques have enabled room-temperature direct-bandgap electroluminescence43,4 and CW room-temperature optically pumped operation of Ge-on-Si lasers45.

Th e fi rst Ge-on-Si laser operating at room-temperature was fabricated by selectively growing 1.6 μm × 0.5 μm Ge waveguides epitaxially on Si (ref. 45). A thermally induced tensile strain of 0.24%, together with a phosphorous doping level of 1 × 1019 cm–3, allowed enhanced light emission from direct gap of 0.76 eV. A cross-sectional SEM picture of the Ge waveguide is shown in the inset of Fig. 3. Th e

Emission intensit y (a.u.)

Wavelength (nm)

(Parte 1 de 3)