13 February 2004—The road to faster computing is paved with optical interconnects that can speed signals from computer to computer or from circuit to circuit much faster than the metal wires used today. Currently, the essential optical components that convert data from electrical to optical bits and back again, mainly used in networks to communicate over distances from meters to kilometers, are made of exotic and expensive materials—too pricey to put into PCs, workstations, and servers.

But a new silicon device that can modulate light at a rate of 1 GHz is the first step in developing a set of optical components that can take advantage of silicon's vast infrastructure to build complete optical systems cheaply. The device, an optical modulator that encodes data 50 times faster than any previous silicon component, was developed by researchers at Intel Corp., Santa Clara, Calif., and announced in yesterday's issue of the journal Nature.

"It has removed one of the biggest barriers to making optical devices out of silicon," says Vic Krutul, senior manager of Intel's Photonics Technology Strategy. He adds that the company hopes to start putting these silicon devices into products in the second half of this decade. The modulator is just one component of a transceiver, which also requires a laser light source, drivers for the outgoing optical signal, and amplifiers and photodetectors to convert an incoming optical signal back into an electrical signal.

The modulator has two basic elements. The first is a silicon waveguide. "Silicon is transparent at the infrared wavelengths used today for optical communication," explains Mario Paniccia, director of silicon photonics research at Intel. "By using silicon-processing techniques, we can sculpt optical waveguides into the silicon surface that confine and direct the light just like an optical fiber."

The second element is a transistor-like device that can shift the phase of light that passes through it. Just before the point in the light path where the light wave is to be modulated, the waveguide divides into two branches. One branch contains the phase-shifting capacitor, which is made up of a silicon oxide layer sandwiched between the crystalline silicon chip and a polycrystalline silicon cap. Metal contacts at the top of the device apply the electrical signal that is to be converted to an optical signal.

After the light passes through the capacitor, the light waves from the two branches merge. If they are in phase, the recombined wave is essentially the same as the initial one, except for some attenuation. But if the two waves are out of phase by 180 degrees—that is, if the peaks of one wave line up with the valleys of the other—they interfere with one another and the amplitude of the combined wave goes to zero.

The modulator's basic principle of operation goes back to the mid-1980s, when researchers predicted that the phase of light can shift when it passes through regions with large charge densities, explains Paniccia. In the phase-shifting device, voltage applied to the polycrystalline upper layer creates a densely charged area on each side of the oxide by pulling charge carriers in each silicon layer toward the central oxide, as in a capacitor.

When the light passes through this region, its phase shifts. The degree of the shift is proportional to the amount of charge that accumulates near the oxide, which is in turn proportional to the voltage applied. It is also proportional to the length of this charged region. The modulator announced yesterday shifts the phase half a wavelength, enough to induce interference, by applying about 5 volts to the capacitor.

Intel engineers told Spectrum that they have already improved on the device reported in the Nature paper by putting a capacitor in each branch and biasing each at 2.5 V. That way, says Paniccia, you don't have to drive the device over such a large voltage range, making it easier to operate at gigahertz frequencies. The 2.5-V dc bias is just enough to shift the phase of each wave by 90 degrees, or a quarter of its wavelength. Since the phases of the waves in the two branches undergo the same shift, there is no destructive interference when the two waves recombine.

To modulate the light, the voltage on one of the two phase shifters is raised enough to shift the phase by another 90 degrees and the voltage on the other is reduced to zero, so there is no phase shift in the second branch. In this state, the net phase difference is 180 degrees, or half a wavelength, and the two waves cancel each other out when they recombine.

Salvatore Coffa, research director of soft computing, silicon optics, and postsilicon technologies for STMicroelectronics NV, in Geneva, Switzerland, told IEEE Spectrum that getting a silicon optical modulator to run at gigahertz frequencies is an important result. "It demonstrates that good old silicon can do almost everything." Coffa recently achieved a breakthrough by making an efficient silicon light-emitting diode [see "Light from Silicon," January 2004].

The use of polycrystalline silicon in the upper layer of the Intel device to separate it from the metal contacts was very clever, adds Coffa. "One of the problems in integrated modulators, even those that are not silicon-based, is that you need metal contacts to apply the signal, but you can't put them right on top of the active area because they have a very large optical absorption." Although the polycrystalline silicon layer eliminates the attenuation of the optical signal by the metal contacts, it is not the ideal solution, because the material attenuates the light more than the crystalline silicon does. But, in the Nature paper, the Intel researchers report that they are hoping to improve transmission by growing crystalline silicon on top of the oxide using a technique called epitaxial lateral overgrowth. Making the oxide thinner and the capacitor shorter will also reduce the attenuation, says Paniccia. "Our goal," he says, "is to continue to improve the performance from this 1-GHz benchmark to much higher frequencies in the future." In addition to faster, more efficient modulators, Intel will be working on the other pieces of the silicon optical communication puzzle needed to realize the dream of freeing chip-to-chip communications from its copper wire cage.

For background on optical interconnects, see "Linking With Light," August 2002.