Silicon photonics shines

Intel makes chips, but not all of them are the familiar semiconductor variety. The company also makes silicon photonic chips, part of a growing industry that uses the material to control light.

In a presentation at the SPIE Advanced Lithography and Patterning conference, Haisheng Rong, an Intel senior principal scientist, talked about the technology and its prospects. The most important application, he said, is optical communications, the technology that carries the world’s internet traffic by modulating laser beams. Other uses lie in biomedical and 3D sensing as well as quantum computing and communications.

For all these applications, it’s vital to lower costs while improving performance. Regarding the first of those needs, traditional photonics solutions typically have a few discrete components instead of the billions of transistors integrated on a chip.

“That makes the cost very high,” Rong said of this low level of integration.

Decades ago, researchers at Intel and elsewhere started investigating silicon as a solution to this problem. Silicon has some advantages and disadvantages when it comes to controlling light. On the plus side, it’s transparent at the infrared wavelengths used for long range optical communication. It also has a huge manufacturing base for fabrication of devices with microscopic features and thus photonic devices made out of silicon could be very compact.

On the other hand, silicon has several minuses. Because it is transparent in the infrared communication window, it can’t detect a light signal. What’s more, at the time that scientists started working on silicon photonics, they also couldn’t generate light using the material.

Rong noted, though, that silicon can be a waveguide, a device that directs light where desired. That capability meant silicon had promise as a way to control light. For instance, an early demonstration silicon photonic waveguide was a ring resonator. Light entering the circular structure could only exit if it was the right wavelength. So, this meant the resonator could be a gate, letting only certain wavelengths through.

That transmission happened with high efficiency and low loss after researchers overcame certain problems. Some were lithography and patterning related, such as minute unevenness in the sidewalls of the microscopic ring. These walls needed to be exceedingly smooth to ensure high transmission efficiency. As Rong noted in a Q&A session after his presentation, “The sidewall roughness is critical.”  

In his talk, Rong recapped solutions to the other problems. Incorporating the semiconductor germanium in the process made detectors possible. As for generating light, the compound semiconductor indium phosphide did that. Today, Intel integrates high performance indium phosphide lasers on its silicon photonics chips.

The laser performance is very good, Rong said. In fact, he noted that silicon lithography technology makes it possible to do laser arrays, groups of lasers sending out light at different regularly spaced wavelengths. That is something difficult to achieve with discrete components.

Putting the light source, waveguides, and detectors on the same chip increases the level of integration, lowering cost, and increasing performance. After years of development, Intel started shipping silicon photonic optical communication devices in 2016, with the data transmission rates rising steadily from the starting rate of 50 gigabits per second as new devices rolled out.

There are some key differences between silicon electronics and photonics, according to Rong.

There is, for one thing, no photonic equivalent of a basic building block like a transistor. Another difference is that photonic dimensions tend to be much larger but require tighter control in terms of percentages for some of the dimensions. Two ring resonators that are microns in diameter may differ in length by only a few nanometers, for instance. But that slight variation will mean the transmission wavelength windows of the resonators will differ enough to be a problem. That type of control is beyond what lithography and patterning can deliver today. So, in order to get devices to yield, Intel developed a way to tune resonators across a wafer to make them function more uniformly.

As for the future, one near-term goal is to move photonics chips as close to electronic ones as possible because this approach produces the best performance. A long-term goal is to take that idea as far as it can go by putting the photonics and electronics together on a single integrated device.

“We believe the long-term convergence of computing and communication will happen in silicon,” Rong said.

Hank Hogan is a science writer based in Reno, Nevada.