A year or two ago, nanotechnology was grabbing headlines. Today, it's silicon photonics, with its promise of chip-level integrated optical inter-connects. In the same way that the silicon chip moved beyond computer microprocessors to appear in everything from automobile keys to children's toys, silicon photonics may ultimately be deployed across a variety of applications ranging from biomedical devices to optocouplers.
The value proposition of silicon photonics is cost. "If you could make [telecom components] with silicon, then you could essentially benefit from the economies of silicon and enable very low-cost, affordable high-speed optic networking," says silicon photonics pioneer Bahram Jalili of the University of California, Los Angeles. Silicon photonics could leverage the already well-established CMOS technology, taking place in the same fabs. Or even on the same wafers, which is the ultimate goal.
That wouldn't happen immediately, of course. The thinking goes that the technology could first penetrate the metro/access telecom market, then filter down. "Silicon will never be as good performance as any III-V material," notes Mario Paniccia, director of the Photonics Technology Lab at Intel (Santa Clara, CA), "but if you get 90% of the way for 10 times lower cost, then you can go to the customer and say hey, do you really need that performance?"
Silicon Steps Up
Silicon functions quite well for passive optical elements such as waveguides and array-waveguide gratings. As an indirect bandgap material, however, silicon was long perceived as useless for active devices. That perception began to change in 2001, with the help of the Raman effect. Raman amplification has been leveraged in optical communications since the late 1990s. Photons excite vibrations (optical phonons) in the crystal lattice of silica optical fiber, triggering the emission of light at a different wavelength than the pump beam. In optical fiber, the effect is small enough to require kilometers of fiber for appreciable gain; the effect is four times as strong in silicon, making it practical for chip-level devices.
By optically pumping a silicon waveguide to trigger Raman amplification, and adding cavity mirrors, Jalili's group demonstrated pulsed-mode lasing in 2004. Continuous-wave (CW) performance was hampered by two-photon absorption in the waveguide, which acted to spontaneously generate electrons that saturated the gain. Adding a p/n structure provided the answer; a voltage applied across the junction draws electrons to the n region, mitigating the effect. Paniccia and the Intel group used this approach to produce a CW laser (see illustration).
Fig. 1 The silicon laser includes a PIN structure (the p-doped Si, intrinsic Si, and n-doped Si). A voltage applied across the PIN structure sweeps away the electrons generated by two-photon absorption, dramatically reducing loss. Illustration based on figure courtesy of Intel.
The Intel group has also produced a phase-shifted modulator operating at gigahertz rates. Meanwhile, David A. B. Miller and colleagues at Stanford University (Stanford, CA) have reported a silicon-germanium modulator based on the quantum-confined Stark effect. Although CMOS has traditionally been silicon based, in recent years, germanium has been introduced into the fab, making the Stanford device a candidate for the silicon photonics toolbox.
Realizing chip-level interconnects is far from trivial, of course. One challenge is real estate. Moore's Law has the industry worshiping transistor number density. A technologyno matter how promisingthat takes up space designated for transistors is going to be a hard sell. Jalili's group has begun three-dimensionally integrating silicon photonics, burying the photonic layer below the wafer surface so that the microprocessor can be built on top. This sort of architecture would allow chip-level photonic interconnects without compromising the electronic circuit; moreover, the photonic structure could be built and tested first, reducing the chance that an expensive IC might be scrapped because of faulty photonics.
To produce the structures, the group implants oxygen into the silicon at a specified depth, and heats it to form silicon dioxide. This is a standard method for producing silicon-on-insulator structures, and wouldn't add significant cost or time to chip fabrication, says Jalili. So far, the group has produced waveguides, wavelength filters, and routers, and has successfully fabricated a functional transistor on top of the subterranean structure. According to Jalili, Sony is currently working on the technology as part of their joint venture with IBM to develop next-generation microprocessors for Sony's video-game units.
Silicon photonics for intrachip or board-level interconnects faces challenges, however. "You're trying to get on a train that's already going 90 miles an hour," says Tom Hausken, director at Strategies Unlimited (Mountain View, CA). "The microprocessor business is a finely tuned business." In the tradition of disruptive technologies, silicon photonics may find its big opportunity in a completely different sector, such as optocouplers, biomedical devices, or sensors. "Silicon photonics is much more likely to have a chance in a relatively [undemanding] application where you can provide a small advantage and gain some market share," adds Hausken, estimating the optocoupler market alone at $1 billion.
"I think silicon photonics has the ability to impact wireless, bio, sensing, and chemical analysis," says Paniccia. "If you look at any of these disruptive technologies, it's not the high end, it's about when technology gets to a point where it's just good enough."
"What is most dear to my heart in silicon photonics is its potential as a mid-IR material," says Jalili. Diode lasers dominate the near-IR spectral region (0.8 µm to 1.5 µm), but the mid-IR region (2 µm to 30 µm) is wide open, he notes. "There is a pressing need. There aren't that many compelling technologies out there."
He points to medicine as a significant opportunity. A silicon laser operating at 2.9 µm, for example, could replace large-frame, water-cooled erbium-doped yttrium aluminum garnet (YAG) systems for laser skin resurfacing. Jalili also cites chemical and biochemical sensing, including lab-on-a-chip and microfluidic designs. "These are topics and applications that are rather pressing when it comes to homeland security," he observes. "The whole field of sensors and lab on a chipthat could very well be the first mass-production application of silicon photonics."
Kristin Lewotsky is a freelance technology writer based in Amherst, NH.