Laser research could hold key to unlocking exponential speed increase in computers
In the never-ending race to create faster, more powerful microchips, the tech industry is increasingly running up against a challenging bottleneck. Microchips use electricity to communicate with each other, but electrical signals are relatively slow and inefficient. Trying to make computers faster by increasing chip speeds would take so much electrical power the chips would fail—or even melt.
Mark Wistey, associate professor in the Department of Physics at Texas State University, envisions a way to sidestep this problem, by having microchips use lasers to communicate. Lasers can be used to efficiently transmit tremendous amounts of data very quickly, at a fraction of the heat and energy cost associated with wires and electricity. But laser communication has never been cost effective at such a small scale and that presents unique challenges.
"Right now, your computer has a processor and memory, and we want to be able to get data between those chips using light instead of electricity because light is faster and carries more 'channels' or colors. You can move data at a much higher speed using less power. So it's a win-win situation," Wistey said. "Our goal is to find materials we can grow on silicon that would be good light emitters to make lasers. We cannot make lasers out of silicon itself, since silicon cannot emit light, so we have to find some other material.
"The elements germanium and carbon are terrible emitters of light, so you would think a mixture of the two would be terrible," he said. "The surprising thing is, an alloy of germanium and carbon will be an excellent light emitter, according to our theoretical modeling. Our goal now is to prove this experimentally."
The alloy itself has proven difficult to grow in the laboratory. Carbon molecules aggressively bond together, resulting in clumps of graphite that render the alloy useless. When the alloy fails, instead of a material that looks like a diamond lattice, it comes out looking more like soot.
"Our goal is to find ways to grow the material that give us what's called high substitutionality," Wistey explained. "Basically, we take out a germanium atom and put a single carbon atom in its place, with no defects. We've had good luck with that. We've been able to show we can get carbon into germanium with no defects that we can measure.
"Our next goal is to get enough carbon in so that we can actually make a laser out of it," he said. "We'd like to get data transferred between chips by lasers and even within the chip. If we could get data transmitted within the chip using light instead of electricity, that could be a big win for energy and speed."
The energy savings could be substantial. Silicon chips are tiny, and individually use only a small amount of power. But those same chips are ubiquitous in modern society, found in smart phones, appliances, cameras and automobiles. With the advent of cloud computing, massive data centers have sprung up around the world, consuming vast amounts of energy.
"Estimates are that somewhere between 1 and 2 percent of all U.S. electricity goes to data centers. And that's not even counting the computers on your desk or on your lap," Wistey said. "Those would also be more efficient this way. The first place these lasers-on-silicon would show up would be in high performance computers, like in data centers. And then they would trickle down to the rest of the market, like our laptops and phones.
"If you buy a computer, they get advertised as a multi-core processor. What that means is that there are four or eight computers within the same chip. Each of those eight have to talk over the same wire to the memory, which is a different set of chips. They're competing for bandwidth for data communication, and Intel is seriously talking about thousand-core CPUs now," he said. "Now if you put a thousand of them on a single chip, 900 are just going to be sitting idle, twiddling their thumbs because they're waiting for their turn at the data channel. If we can make that data channel much, much faster, then we can make the computers continue to become faster. As the number of cores in a computer goes up, the need for faster communication goes up even faster. We also now know that artificial intelligence is mostly about connections. Your brain has as many neurons as there are transistors on a big chip, but it has 1,000 times more connections and that's what makes the difference. Having lasers right on the chip would let up make these beautifully long, elegant paths that we can't do with conventional electronics.
"Right now, it's a materials challenge," Wistey said. "We're at the stage of learning how to grow these materials well and in a way that's manufacturable and affordable."