Silicon photonics offers the potential of combining the speed and compactness of photonics with the functionality and CMOS fabrication of electronics, a happy combination that may result in vastly more capable computers, communications and consumer items. One of the essential technologies needed to implement silicon photonics is light modulation, but that is anything but straightforward.Conventional light modulators, such as the ones in today’s fiber optic telecommunications systems, rely on the classic electro-optic (Pockels) effect in lithium niobate and similar materials. But that approach won’t work in silicon photonics because silicon exhibits a negligible electro-optic effect. Instead, scientists rely on the index change induced by the injection of free carriers — electrons and holes — when they force an electric current through a silicon waveguide. If the waveguide is one arm of a Mach-Zehnder interferometer, the amplitude of the transmission through the interferometer is modulated by the current that passes through the waveguide.Free carriers can be generated quickly by the current, but the modulator’s speed is severely limited by how fast the carriers deteriorate when the current is switched off. The solution to that problem was demonstrated several years ago when scientists fabricated a p-i-n junction around the waveguide. A reverse bias across the junction swept the carriers out of the waveguide very quickly, so that the modulator speed was ultimately limited by the RC time constant.Even that limitation was recently diminished by scientists at Intel, who fabricated a silicon Mach-Zehnder modulator with a traveling-wave drive scheme that allows the electrical and optical signals to copropagate along the waveguide (see “Silicon Photonic Modulator Transmits Data at 30 Gb/s,” Photonics Spectra, April 2007, page 106). That design has allowed silicon modulators to operate at rates as high as 40 Gb/s, which is, for now at least, the world’s record.But silicon Mach-Zehnder modulators are large, on the order of millimeters in length, which precludes packing hundreds of them onto a single chip, as is desirable in many applications. An alternative has been micrometer-size ring resonators fabricated adjacent to silicon waveguides. The waveguide’s transmission can be modulated by tuning the microring into resonance with the light passing through the waveguide. But the inherent drawback of this approach is the fact that the microring is resonant, which renders it susceptible to changes in temperature and to other environmental variables.Now William M.J. Green and his colleagues at the IBM Thomas J. Watson Research Center in Yorktown Heights, N.Y., have seemingly married the best of both worlds by demonstrating a Mach-Zehnder modulator whose arms need be only a few hundred micrometers in length and whose total area is only several times more than that of a microring. The interferometer’s tiny silicon arms — rib waveguides only 550 nm wide and 220 nm high — are the secret of its success. The arms’ small cross-sectional area, ∼0.12 μm2 as compared with ∼1 μm2 for a conventional silicon Mach-Zehnder, leads to high carrier density as well as to high overlap between the carriers and the optical mode in the waveguide.Figure 1. The scanning electron microscope image on the left shows one arm of the Mach-Zehnder and the p-i-n junction surrounding it. The heavily doped p- and n-sections are hatched, and the nickel-silicide contacts are shown in false-color gold. The extension of the nickel silicide beyond the edges of the doped regions is a fabrication error that imposed unnecessary loss on light propagating through the rib waveguide and that reduced the modulator’s extinction ratio. Despite the narrowness of the 35-nm-thick silicon slab providing a path between the doped material and the waveguide, the junction showed a surprisingly low resistance of only 49 Ω. The IBM scientists fabricated an 80-μm loop in one arm of their Mach-Zehnder (as shown on the right), resulting in a free-spectral range of 7.3 nm. An equal-arm interferometer would have a larger, more desirable bandwidth, but the existence of spectral interference fringes in this experimental device facilitated in differentiating between thermal and current-induced index changes and in measuring the modulator’s extinction ratio. Images reprinted with permission of Optics Express.Electrically, the IBM scientists measured a forward resistance across the p-i-n junction surrounding one arm of their Mach-Zehnder to be 49 Ω (Figure 1). This value is significantly smaller than the multi-kΩ resistances typically measured in tiny silicon rib waveguides and leads to lower power consumption during operation. Optically, they illuminated the interferometer with a broadband LED and measured a transmission loss of about 12 dB at 1550 nm. They believe this loss was due principally to small fabrication errors described in the caption to Figure 1 and could be significantly reduced without changing the device’s electrical characteristics. Likewise, presumably fixable fabrication problems resulted in an extinction ratio through the interferometer of only 6 to 10 dB.Figure 2. The scientists drove the p-i-n junction with a 9-MHz sinusoidal signal (a) and observed the resulting transmission through the interferometer (b). The half-wave voltage was the voltage required to turn the optical signal around; i.e., the voltage difference applied between the two points indicated by arrows in the bottom trace.But the modulator’s efficiency — given by the product of the half-wave voltage and the length of the Mach-Zehnder arm — was two orders of magnitude better than the typical values in larger Mach-Zehnders. The scientists measured the half-wave voltage dynamically by driving the modulator with a 9-MHz, 1.8-V signal and observing the voltage necessary to turn the transmission around (Figure 2). The product, 0.36 V-mm in a 200-μm-long arm, indicates an index change of 4 × 10–3, or a charge density of 1.5 × 1018 cm–3. This efficiency, a direct result of the high carrier density in the waveguide and of the overlap between those carriers and the optical mode, enables the device to be as much as 50 times shorter in length than a conventional silicon Mach-Zehnder modulator.Figure 3. The modulator generated good eye diagrams at modulation rates as high as 10 Gb/s when driven with a pseudo-random bit sequence. LMZM is the length of the arm in the Mach-Zehnder interferometer. The modulator generated good eye diagrams at modulation rates as high as 10 Gb/s (Figure 3). Higher rates are probably limited, at least in this design of the modulator, by the ∼16-GHz RC time constant. There is no obvious reason, however, why an Intel-like traveling wave drive could not be employed in a future design.Power consumption is another important parameter in modulator performance, and the IBM scientists maintain that this value is consistently underestimated by the commonly used approach of quantifying only the DC power consumption. With a forward resistance of 49 Ω and drawing a DC current of 2.17 mA at 5 Gb/s, their modulator’s DC power consumption was 230 μW. But because the average amplitude of the radio-frequency (RF) voltage applied to the 49-Ω load was ∼1.4 V, the RF drive power was vastly larger, at 41 mW. This latter value, the RF consumption, is usually overlooked, they say, but is in fact the realistic number that should be used as the basis for link power budgeting. They believe that the total consumption by their tiny Mach-Zehnder modulator, effectively only 5 pJ per bit at 10 Gb/s, is far smaller than that consumed by conventional modulators.Optics Express, Dec. 10, 2007, pp. 17106-17113.