Introduction
Microelectronics based on silicon had brought great changes to our society in the 20th century. Keeping pace with Moore’s law, the transistor gate lengths in complementary metal-oxide-silicon (CMOS) are now scaling down continuously. Unfortunately, as gate lengths approach their physical limit, circuit delay and electronic power dissipation will increase up to an unbearable level, so electronic interconnects will inevitably encounter its bottleneck [1]. Silicon photonics is promising to extend Moore’s law by adding photonic components on the chip to construct photonic interconnects, where silicon light source is the most challenging part, and has been studied for decades worldwide [2]. A practical silicon photonic link has been demonstrated recently after the realization of the practical silicon light source [3]. However, accompanied by the advantages of on-chip photonic link, one flaw is the size mismatch between photonic and electronic components in a chip. Thus, surface plasmon (SP) source with deeply sub-wavelength size is intensely studied as the next generation Si-based light source [4].
Silicon hybrid photonic lasers
Due to indirect band gap of silicon and extreme difficulty in developing other practical Si-based emitters, hybrid silicon light sources have been adopted in practice by bonding III-V semiconductors on silicon [5,6]. Most III-V semiconductors exhibit direct band gap as principal photonic materials. Especially, InGaAsP, AlGaInAs, and InGaNAs multiple quantum wells grown on indium phosphide (InP) substrates emit light with tunable wavelength from 1.3-1.6 μm, which is standard wavelength for silicon interconnects. There are three types of bonding techniques used frequently for combining InP material and silicon. One is direct bonding, which bonds InP on silicon without any insertion between them as illustrated in Fig.1. University of California in Santa Barbara and Intel cooperation jointly reported the first electrically driven InP/Si hybrid laser fabricated by direct bonding [5]. The light generated in InP part couples into the just below silicon waveguide. But, direct bonding method needs a rigorous clean room and atom-scale smooth surfaces for two components to be bonded. Other two bonding methods are indirect bonding known as polymer bonding and metallic bonding, where the insertion indicated in Fig.1 is polymer or metal. With the help of the adhesive insertion, indirect bonding does not require a chemically clean and atom-scale smooth surface compared to direct bonding and these post-bonding approaches have great flexibility that allows the integration of the prefabricated devices finished within separate processes. Ghent University reported a disk laser bonded on a silicon waveguide using thick benzocyclobutene (BCB) [7].
Fig.1 Hybrid integration by bonding III-V semiconductors on silicon. When there is no insertion between them, it is known as direct bonding; for indirect bonding, polymer or metal is frequently chosen as an insertion material. But, polymer bonding results in an electrically insulating layer; and metallic bonding in an optically insulating layer. A transparent and conducting insertion is an interesting alternative in some cases |
However, polymer bonding usually results in an electrically insulating layer between the two parts being bonded; likewise, a weakness of metal bonding is the strong light absorption of metal, blocking light from transmitting through the bonding. Thus, in some case, metal bonding method has to be adopted in some selective area, where the bonding metal in the light coupling area has to be removed [8], as illustrated in Fig. 2. In fact, an interesting alternative is selecting transparent and conducting material, such as Indium-Tin Oxide (ITO) as the adhesive insertion to overcome the weaknesses of the two traditional indirect bonding methods. The transmission spectra of ITO films in visible and near infrared regions (400-2000 nm) are quite flat, decreasing very slightly in near infrared region. Bonding can also achieve the integrations of III-V detectors, modulators and other photonic components onto the silicon chip for interconnect [9].
Fig.2 For light coupling, the metal crossing light coupling area has to be removed, that is, selective area metallic bonding, the expense of which is the alignment-caused low yield. But this bonding method separates III-V process from CMOS process, and so there is no process-compatible problem |
Surface plasmonic light source
A first 50 Gbps silicon photonic link was demonstrated in 2010 by Intel cooperation integrated with four hybrid silicon lasers [3]. From many researchers’ viewpoints, such photonic link is very practical and potent, but not a permanent solution because of size mismatch between hybrid lasers and electronic components on the chip. The deeply sub-wavelength size SP emitter has the potential to be the next generation light source for highly compact silicon photonic interconnect [1]. Analogous to other light sources, an ordinary SP diode produces/excites SPs and an SP laser diode produces coherent SPs, but both of them operate on the interactions (energy coupling) between the optical emitters and the metal. An SP laser is termed surface plasmon amplification by stimulated emission of radiation (SPASER), a nanoplasmonic counterpart of light amplification by stimulated emission of radiation (LASER), and was first suggested by Bergman and Stockman in 2003 [10]. In 2008, the first lasing SPASER was demonstrated [11], and then in 2009, two types of nano SPASERs were demonstrated [12,13]. These SPASERs were all optically pumped. Recently, quite a few SP diodes including Si-based one have been reported [14-17]. A conceptual structure for a (laser) SP diode is shown in Fig. 3(a), which consists of an active (gain) layer and a noble metal layer at least. Additionally, we insert an index matching layer on each side of the active layer for higher efficiency SP generation. When the active layer is organic semiconductor, the structure is known as an organic light emitting diode (OLED), which is naturally an SP diode due to the very thin thickness of the organic layer [15]. Our group observed a strongly polarized edge-emission from an ordinary OLED with a stacked Sm/Ag cathode. The polarization ratio of transverse magnetic (TM) mode and transverse electric (TE) mode is close to 300 [16]. The polarization results from the scattering of SPs at the device boundary. Such Si-based OLED is potentially an electrically excited SP source in silicon plasmonics. The operation principal for an SP (laser) diode is depicted in Fig. 3(b).
The (laser) SP diode operates on the emitter coupling its radiation to the SP mode (energy transfer) [18]. In order to estimate the key performance parameters, such as the energy conversion efficiency of the SP diode and the threshold condition of the SP laser, we can conveniently calculate the emitter’s power dissipation as a function of in-plane wave vector BoldItalic//, that is, power dissipation spectrum [19-22]. For a multilayer structure, each guided mode will produce a peak in the power dissipation spectrum. Combining the separate analysis of the mode refractive index, we are able to assign each peak to a specific mode (or modes). In computations, the emitter is considered to be a classical electric dipole and the SP diode structure is modeled as a micro- (or nano)-cavity with multilayer. The field produced by the emitters in the active layer is reflected in the metal cavity [20,21].
Fig.3 (a) Conceptual diagram of an electric excitation SP source, consisting of an active layer and a noble metal layer at least. A pair of index matching layers is added for generating higher fraction of SPs. Coupling between a SP source and a SP waveguide is also depicted, with a very high coupling efficiency. The total thickness between the two metal layers should be sufficiently thin, so it supports SP modes only; (b) operation principal for an SP (laser) diode. Radiation or injection excites a transition into electron-hole (e-h) pairs (vertical blue arrow). The e-h pairs recombine nonradiatively and transfer energy to the plasmon excitation of the metal nano-cavity through resonant coupled transitions (red line). (b) is redrawn after Ref. [18]. For a SPASER, the active (gain) layer is in population inversion state, and the SP mode gain distributed from the gain media must exceed the SP mode loss, which is usually a very large value. |
As an instance, Fig. 4 plots a typical power dissipation spectrum of an emitter in the structure shown in Fig. 3(a), which is simplified to a sandwiched structure of Au (40 nm)/active layer (120 nm)/Au (40 nm) as inset in the figure. The differential dissipated power b of the emitter is normalized by that in free space, b0, and the in-plane wave vector BoldItalic// normalized by the wave vector BoldItalicorg in the active layer (here, the index of the active layer is set to 1.6, a typical value for organic materials, and the wavelength used here is 620 nm) and defined BoldItalic///BoldItalicorg to be u. As indicated, the power emitted with u<0.58 labeled by “leaky mode” may leak into air; for the peak at u≈0.697, the emitted field strongly couples to the SP modes at the mental/air interface; for u≈1.408, the peak corresponds to the SP mode at the metal/dielectric interface, as seen in Fig. 4.
By integrating the power under the peaks of SP modes in the curves shown in Fig. 4, the fraction of the total power coupled to the two SP modes can be estimated, which corresponds to the energy conversion efficiency of the SP diode. If the active (gain) media distributes its optical gain according to the above computed fractions to a certain SP mode (mode gain) and the mode gain exceeds the mode loss, an SP laser in this mode can be realized [22].
Conclusions
Bonding InP on silicon provides a platform for the silicon optoelectronic integration, but silicon photonic link needs higher compact light source, so, the deeply sub-wavelength size SP emitter may be a promising alternative.