Silicon Turns One Wavelength into Many

Breck Hitz

There are many challenges to be overcome before photonic connections can replace electronic ones in computers and other equipment. Nonetheless, that replacement soon will become imperative because miniaturization is driving electronic interconnects to their limits in terms of power dissipation, latency and bandwidth. One of the bigger challenges is designing tiny photonic interconnects that take advantage of the technology’s full bandwidth potential.

Figure 1. Several years ago, Princeton University scientists generated light in multiple WDM channels by broadening the output of a mode-locked laser in dispersion-compensated fiber and dividing that broad spectrum into individual channels with a WDM splitter. EDFA = Erbium-doped fiber amplifier.

A photonic interconnect can carry vast amounts of information by spreading it across multiple spectral channels, a technique known as wavelength division multiplexing (WDM). In practice, the problem is generating all the spectral channels in a manner that is consistent with the size and power restrictions inside modern computers.

Figure 2. Recently, UCLA scientists shrank the Princeton concept to fit on a single chip. They broadened the pulses from a mode-locked laser in a 2-cm-long silicon waveguide and divided the broad spectrum into individual channels with microdisk resonators. Reprinted with permission of Applied Physics Letters.

The most straightforward approach is to have a separate laser diode for each channel, but that would require far too much space and consume far too much power. In addition, lasers cannot operate efficiently at the elevated temperatures encountered on computer chips.

Several years ago, scientists at Princeton University in New Jersey demonstrated a different approach. They were addressing the problem of generating WDM channels in a telecom environment, where space and power are more plentiful than with computer interconnects. They passed the pulses of a mode-locked fiber laser through a length of dispersion-compensated fiber, where self-phase modulation broadened the pulses’ spectrum into a continuum tens of nanometers wide (Figure 1). Then they carved that broad continuum into multiple channels with a WDM splitter.

Figure 3. The scientists fabricated the chip illustrated in Figure 2 with the microdisk resonators beneath the silicon waveguides. As shown on the left, the spectrum was carved into three channels, not four as shown in Figure 2. Each microdisk resonator (right) was a slightly different size from the others, so each coupled a different wavelength out of the straight waveguide. Reprinted with permission of Applied Physics Letters.

With its long lengths of fiber and its bulk optical filters, the Princeton approach never was intended for computer interconnects. But recently, scientists at the University of California, Los Angeles, took this concept and cleverly compressed it to fit on a single monolithic chip that may one day be useful in computer interconnects.

The UCLA investigators passed the pulses of a mode-locked laser through a 2-cm-long silicon waveguide, where self-phase modulation broadened them into a spectral continuum. Microdisk resonators then carved the continuum into individual WDM channels (Figure 2).

Figure 4. The 2-cm-long silicon waveguide broadened the pulses by a factor of ∼2.5. The scientists believe that greater broadening could be achieved in shorter waveguides by shrinking the waveguide’s cross section to boost the intensity of the power in the waveguide. Reprinted with permission of Applied Physics Letters.

In a glass fiber, self-phase modulation is the result of only one effect, the classic Kerr nonlinearity. In silicon it’s more complicated because both the Kerr nonlinearity and the nonlinear refraction resulting from the presence of free carriers — electrons and holes created by two-photon absorption — contribute to self-phase modulation. The scientists calculated both effects, and in the end, their experimental results were in reasonable agreement with the calculations.

Figure 5. The normalized outputs of the three microdisks show distinct peaks at 1548.5 nm (port 1), 1549.1 nm (port 2) and 1547.9 nm (port 3). Other spectral features are discussed in the text. Reprinted with permission of Applied Physics Letters.

They fabricated the devices illustrated in Figure 2 with a technique they call separation by implantation of oxygen (SIMOX) three-dimensional sculpting (see “Photonics Goes Underground in Silicon,” Photonics Spectra, October 2005, p. 18). This technique allows 3-D placement of photonic and electronic components on a single monolithic silicon chip. The scientists placed the microdisk resonators on a layer beneath the silicon waveguides (Figure 3).

The 2-cm waveguide increased the spectral width of the incoming pulses by a factor of 2.5 at the –20-dB level (Figure 4). A different wavelength emerged from each of the three microdisk resonators (Figure 5). In these spectra, the secondary peaks were separated from the primary peaks by the free spectral range of the disks, about 2.6 nm. The scientists believe that the small peak at 1548.3 nm from port 3 was a higher-order mode of that microdisk.

Applied Physics Letters, Aug. 6, 2007, Vol. 91, 061111.