Silicon photodiode threatens III-V detectors

For some years now, Intel has been looking for a way to “siliconize” photonics. The chip giant wants to build optical devices on a silicon substrate to drive the manufacturing process to higher volumes and lower cost. Now the company says it has made a breakthrough in one of the key components that would be required — a high-gain silicon-based optical detector.

Intel has added germanium to silicon to create an avalanche photodiode (APD) that is better at detecting high-speed, low-intensity signals than existing devices. The results were published online in the journal Nature Photonics on December 7.

“This is the first time that a silicon photonics device has better performance than any recorded performance from an equivalent device in III-V materials, specifically indium phosphide,” claims Mario Paniccia, Intel fellow and director of the company s photonics technology lab.

That s a big deal because the performance of silicon-based components is usually worse than their counterparts in other materials. Paniccia said the target performance in the company s silicon photonics work is usually 90 percent of the performance for an order of magnitude in cost reduction. The APD is an exception to this rule.

Most optical components are made from III-V semiconductor materials, such as GaAs and InP, because these materials are good at creating and detecting light. Silicon is not good at these functions &mdash hence the addition of germanium to the device to allow light to be “caught” as it impinges on the device.

However, although silicon is not the optimal photonic material, it has very good electronic properties and this can be exploited in an APD to create gain.

In a standard PIN detector, light arriving at the absorption region is converted into an electron-hole pair. An applied voltage separates the electron-hole pair, which move towards the electrodes to generate current. The APD operates on the same principle, but with one key difference — the multiplication region. When the electron is pulled into this region it creates additional electron-hole pairs, and these in turn create further electron-hole pairs. Now a single photon incident on the device can produce tens or even hundreds of electrons.

The performance of the APD is measured as its gain-bandwidth product. This parameter, which is the gain of the device multiplied by the speed at which it operates, is fixed for a particular type of material. For an indium phosphide APD the figure is typically around 120 GHz.

In the Nature Photonics paper, the researchers report devices operating at a wavelength of 1300 nm and a data rate of 10 Gbit/s, which had a gain of “over 30”. The best result was a “world record” gain-bandwidth product of 340 GHz.

Intel s silicon photonics work is mostly aimed at short-reach connections, between or even inside electronic chips, but the silicon APD would probably find its first applications in mainstream telecoms, where it could provide a low-cost replacement for detectors in long-haul links. Other applications such as sensing, imaging, quantum cryptography or bio-chips have also been mooted.

The enhancement in gain-bandwidth product can be useful in a couple of ways. For starters, the extra gain could be used to extend the transmission distance, because the device can detect weak signals that have lost intensity as they travel down an optical fibre. Alternatively, it could be used to provide leeway in system design, allowing cheaper, low-power lasers to be used on the transmit end of the link, which could be particularly useful for applications like fibre-to-the-home.

The third option is to use the device to provide more moderate gain at higher speeds. Intel suggests that the silicon APD could help to lower the cost of 40 Gbit/s systems, although it hasn t yet built a device that can operate that fast.

Paniccia emphases the fact that this is a research result, although he doesn t see any particular obstacles to commercialization. “Now it s about optimizing those devices for performance, packaging the devices and getting the manufacturing process qualified,” he says.

“We have work to do in terms of reducing the dark current and improving the sensitivity, which is based on the dark current,” Paniccia adds. The dark current, which flows even when no light is present, is caused by the slight lattice mismatch between the silicon and germanium layers of the device.

The prototype APDs were produced on a commercial CMOS production line alongside memory chips being made by Intel spin-out Numonyx, where much of Intel s silicon photonics work is now carried out. Experts at the University of Virginia and the University of California, Santa Barbara, provided consultation and assisted with testing, Intel said.

This result is the latest in a string of silicon photonics advances from the chip giant. Back in 2004 Intel pushed the performance of silicon modulators, which at the time only ran at 20 MHz, out to 10 Gbit/s. Two years later it boosted the data rate to 40 Gbit/s, and then last summer demonstrated an array of eight modulators on a single chip, giving an aggregate capacity of 200 Gbit/s.

In 2006 Intel demonstrated of a hybrid silicon laser, which integrated a tiny piece of InP onto a silicon-based structure to provide light emission.

According to Paniccia, Intel now believes that the hybrid silicon laser is the way forward for silicon photonics, rather than the all-silicon laser based on the Raman effect that Intel demonstrated back in 2005.

Of course, Intel is not alone in its pursuit of silicon photonics. Other companies in the field include CyOptics, Lightwire and Luxtera. In fact, Luxtera reported a monolithic detector based on silicon in March 2007. Like Intel, Luxtera used germanium on silicon, but the device had a lower sensitivity than Intel s silicon APD.