Silicon, the material of choice for electronic circuits, is not ideally suited for electro-optical applications. Its indirect bandgap creates severe problems when using it as a photonic source and also presents disadvantages when using it as a detector. Moreover, its weak electro-optical properties make it difficult to design efficient modulators with the material. Nonetheless, as technology evolves to the integration of electronic and photonic devices, laboratories worldwide are investigating ways of overcoming these problems. Recently, silicon lasers, modulators, amplifiers and other photonic devices have been demonstrated.Modulators are a crucial photonic building block and, until now, the low electro-optic coefficients of silicon have required that silicon modulators be on a relatively large (i.e., millimeter) scale. But a research group at Cornell University in Ithaca, N.Y., has demonstrated a micron-scale silicon device capable of high modulation depths at gigahertz speeds.Figure 1. The silicon modulator was based on a microring resonator. In the modulator's "off" condition, the ring was resonant with the wavelength propagating in the straight waveguide, and most of the light in the waveguide was coupled into the ring. To turn the modulator "on," a voltage was applied across the PIN junction, changing the ring's refractive index and shifting its resonance away from the wavelength in the waveguide. Images ©2005 Nature Publishing Group.The modulator is based on a microring structure (Figure 1), designed to be resonant at the wavelength propagating in the adjacent waveguide; that is, the circumference of the ring is an integral multiple of the propagating wavelength. Under this resonant condition, light is readily coupled from the waveguide into the ring, where it is eventually absorbed and converted to heat. The transmission through the waveguide drops to nearly zero.To destroy the microring's resonance -- and increase the waveguide's transmission from near-zero to near-unity -- it is necessary to change the ring's refractive index. The Cornell researchers did this by building a PIN junction across the ring and injecting electrons and holes into it. Figure 2. A scanning electron micrograph of the microring resonator shows the dimensions of the ring and the gap between the waveguide and the ring (inset).The 12-µm-diameter microring was a silicon rib waveguide, separated from the adjacent waveguide by a 200-nm gap (Figure 2). The researchers coupled CW light from a tunable laser into the waveguide and observed the transmitted spectra as they changed the voltage applied to the PIN junction (Figure 3). For voltages less than ~0.7 V, very few electrons and holes were injected into the ring, and it remained resonant at the probe wavelength of 1573.9 nm. And the transmission through the adjacent waveguide at the probe wavelength remained near zero. As the voltage rose and free carriers were injected into the ring, its resonance shifted to shorter wavelengths. And as the density of electrons and holes increased, free-carrier absorption of the photons traveling around the ring likewise increased, and the depth of the resonant loss notch decreased. (Note that it is the normalized transmission that is plotted in Figure 3.) Figure 3. The microring's resonance shifted to shorter wavelengths as the voltage applied across the PIN junction increased. And the normalized transmission at the probe wavelength (1573.9 nm) through the modulator increased from ~3 to ~100 percent as the resonance shifted away from that wavelength (inset).But as the voltage applied to the PIN junction rose, the transmission through the waveguide at the probe wavelength also increased (inset, Figure 3). A change of only 0.3 V changed the normalized transmission through the waveguide from ~3 to ~100 percent. Greater voltages had no effect on the transmission because they merely pushed the ring's resonance farther from the 1573.9-nm probe wavelength. (The ring's free spectral range was 15 nm, significantly greater than the range over which the probe laser was tuned.)Electrical switching speeds in PIN junctions typically are limited to ~10 ns, because it takes that long to inject the free carriers when a positive bias is applied. However, when the microring modulator was operated at higher voltages, the optical transmission could shift from minimum to maximum significantly faster than the PIN junction could reach electrical steady state. That's because only a small concentration of free carriers -- far fewer than are required for electrical switching -- was necessary to shift the ring's resonance away from the probe wavelength.Figure 4. The top traces show the electrical signal applied to the microring modulator, and the bottom ones the normalized optical transmission. The traces on the left show nonreturn-to-zero modulation at 0.4 Gb/s. Those on the right show return-to-zero modulation at 1.5 Gb/s.To measure the modulator's dynamic response, the researchers drove the modulator with electrical pulses from an amplified pulse-pattern generator (Figure 4). The switching speed was limited by the fact that the PIN junction was formed around only part of the ring. When the junction was forward-biased, free carriers diffused into the section of the ring that was not inside the junction, where the reverse bias could not efficiently extract them.They drove the modulator with both a return-to-zero signal and a nonreturn-to-zero signal. Interestingly, the return-to-zero format had an inherently faster switching speed because the forward-bias time was relatively shorter, and fewer free carriers had time to drift into the nonjunction part of the ring.