The key term in integrated circuits is: integrated† The ability of manufacturing facilities to integrate things sets limits on what processes are available and what materials can be used safely. As soon as you propose a different material or process, the whole chain is broken, and anyone who suggests such a thing should expect people to question your suitability for your current position. “Compatibility” is why you won’t find any laser-powered integrated circuits in your laptop.
However, the ability to make lasers with materials compatible with integrated circuits may have received a boost, with a demonstration of glowing (but not yet lasing) silicon.
Looking from the bright side
Optics and lasers are the backbone of high-speed data transfer. You do not use copper wires to transport data at 1Tb/s. Instead, you use glass and some finely tuned and very expensive laser diodes. But laser diodes are made using processes and materials incompatible with those used to make integrated circuits. So while it’s possible to create an optical connection between a RAM module and a CPU, for example, you have to somehow glue the optics in the right place on the silicon chip. Research labs are happy to sacrifice PhD students to such ventures, but PhD bots don’t scale well, are low-maintenance and their deployment leads to dark looks.
A better solution would be to make silicon emit light, but he doesn’t like that. The reason is not that complicated, but it does take a few words to explain.
Light is created by charged particles such as electrons that give up energy to emit a photon. In a semiconductor, electrons can’t just have an old energy — they must have an energy allowed by the structure of the material. The available electron states in semiconductors are limited, which can lead to problems. For example, an electron may have a pretty old chunk of energy, but not be able to emit it because all the lower energy states are already occupied by electrons.
The situation is even darker than that, because states are defined by more than just their energy; their properties include momentum, angular momentum and more. So there can be a low energy state for an electron with energy left, but the electron cannot enter that state directly because it has the wrong momentum.
This is exactly the problem faced by electrons in silicon. An electron can be excited and must emit a near infrared photon to get rid of its excess energy. But all available low-energy states require the electron to simultaneously emit a photon and change its momentum (usually by bouncing back on a silicon atom). Since that combination is highly unlikely, the electron just bounces around and loses both energy and momentum through collisions.
Materials where electrons cannot easily pass energy barriers are called indirect bandgap semiconductors. Very little light is emitted in these semiconductors.
Silicon isn’t the only material like this. The researchers were inspired by the behavior of its closest relative, germanium, which is also unlit. If germanium’s crystal structure (the way the germanium atoms are arranged) changes, the electrons can emit photons without having to change momentum. In other words, the bandgap changes from indirect to direct based on the structure of the material.
To make silicon glow, the researchers turned to alloys. The nice thing about semiconductors is that their optical and electronic properties tend to change smoothly when you deposit them. If you add germanium to silicon, the resulting alloy will exhibit some of the characteristics of germanium. However, some properties can be continuously adjusted, others cannot. A bandgap is direct or indirect, not a mixture. So, how much germanium does it take to make silicon glow?
The answer turns out to be about 65 percent.
To arrive at this number, researchers grew silicon-germanium alloy nanowires. The nanowires allowed them to choose a template material that forced the silicon-germanium alloy to form the correct crystalline structure to form a bandgap. It also allowed the researchers to compare crystalline structures of different quality.
The researchers’ experiments put quality in the spotlight: the nanowires had to be almost perfect to get a reasonable amount of light from them. All crystals contain flaws (holes or dislocations) and high quality crystals have fewer. But once the researchers perfected making the nanowires, the nanowires produced would emit a surprising amount of light.
The amount of light is low compared to mature laser diode technology, but still bright compared to what was previously available. The shiny silicon seems to reveal a viable route to place all the necessary elements for optical communication on a silicon chip. This means that communication with peripherals such as memory, hard drives and more can be faster. It also means things like clock synchronization on the chip are easier to organize. But for all of that to happen, the researchers’ faint glowing wire has to work its way into the fab’s list of acceptable materials.
Nature, 2020, DOI: 10.1038/s41586-020-2150-y (About DOIs)