Transistor-Like Modulator Helps 'Siliconize' Photonics

Scientists from Intel Corp. (Santa Clara, Calif.) have achieved a major advance using silicon manufacturing processes to create a novel "transistor-like" device that can encode data onto a light beam. The ability to build a fast photonic (fiber optic) modulator from standard silicon could lead to very low-cost, high-bandwidth fiber-optic connections among PCs, servers and other electronic devices, and eventually inside computers as well. "I like to say our quest is to siliconize photonics — to do what's being done today in photonics using standard CMOS," said Mario Paniccia, director of silicon photonics research at Intel.

As reported in the Feb. 12 issue of Nature, Intel researchers split a beam of light into two separate beams as it passed through silicon, then used a novel transistor-like device to hit one beam with an electric charge, inducing a phase shift. When the two beams of light are recombined, the phase shift induced between the two arms makes the light exiting the chip go on and off at >1 GHz, 50× faster than previously produced on silicon. This on and off pattern of light can be translated into the 1s and 0s needed to transmit data.

"We're actually using transistors to control the light, and that's the breakthrough," Paniccia explained. "We've created a modulator. I bring the light into the silicon chip with a waveguide, and I split it into two beams. Each of those beams goes through this structure, which phase shifts the light. If the phase shifter is off, then the two beams recombine — all this happens on a silicon chip — and the output is a digital 1. If I turn on the phase shift, and I use this effect by applying a voltage to this transistor-like device, and send light through it, I can phase shift it. And if I can align the two beams to be 180° out of phase, when they recombine I get a digital 0."

"This is a significant step toward building optical devices that move data around inside a computer at the speed of light," said Patrick Gelsinger, senior vice president and chief technology officer at Intel. "It is the kind of breakthrough that ripples across an industry over time, enabling other new devices and applications."

To date, the fabrication of commercial optical devices has favored expensive and exotic materials requiring complex manufacturing, thus limiting their use to such specialty markets as wide area networks and telecommunications. "What's special about this is, to date, no one has experimentally demonstrated a technology that allows you to do in silicon at speeds of about 20 MHz," Paniccia said. "You can use indium phosphide and lithium niobate and these other III-V materials, and you can get modulators that run in excess of 40 GHz, even 100 GHz. But those are III-V exotic materials. If you wanted to use silicon as an optical material, and you wanted to get functionality with it, 20 MHz really doesn't do much, when the industry is operating 2.5 to maybe 4 Gb/sec and then moving on to 10 Gb/sec. This breakthrough was not only that we developed a phase-shifting technology that's like a transistor, but we've developed a technology that's extremely fast. We've shown that the speed of the current generation was in excess of 10 GHz. We've also shown that we have a module that can encode data at 1 Gb/sec. That's an enormous leap in speed and performance. We also believe we can scale this technology to much faster speeds in the future."

This silicon-based optical modulator is a MOS capacitor waveguide phase shifter. It comprises an n-type doped crystalline silicon slab (the silicon layer of the SOI wafer) and a p-type doped polysilicon rib with a gate oxide sandwiched between. Carriers in the silicon change both the refractive index and optical absorption and subsequently the phase of the light. (Source: Intel)

Efforts to integrate photonics in silicon have suffered from two impediments, Paniccia noted. "One is that silicon is an indirect bandgap material, so fundamentally it doesn't lase. The second one is the modulation speed. We think we've lowered one of the barriers." Intel uses an external laser source, he explained, using couplers to guide the light in and out of the fiber.

The receive side requires photodetectors that work in silicon, to convert the light back to electrons. "Finally, you need ways to assemble all these, in what we call passive alignment. Today, 50-60% of the cost often is in assembly," Paniccia said. "If we do passive alignment, we can use silicon as the work batch and then put down components using silicon lithography and other self-aligning things and integrate these components to reduce the total cost of the devices."

The last component is intelligence, Paniccia noted. Although integration is not trivial, making the photonic devices CMOS-compatible is key to long-term development. "That's why this transistor-like statement is very powerful. You have the opportunity to integrate electronics on the same piece of silicon some day in the future."

For more information about wafer processing, go to www.semiconductor.net/wafer