Advances in microprocessors have transferred the computation bottleneck away from CPUs to better communications between components. That trend is driving the advance into optical interconnection of components, now moving from systems to boards to chip packages to chips themselves.
A related issue with input-output (I/O)-intensive applications such as server farms is the energy consumption required to transport bits of data around. Using photonics technology for I/O components can both improve performance and reduce energy consumption.
But to be commercially viable these photonic I/O devices must leverage the vast existing silicon infrastructure and integrate with silicon as much as possible. That means these components can’t directly follow the spectacular successes of the optical fiber systems that run the Internet, cautions Lionel Kimerling, MIT professor of materials science and engineering and director of the MIT Microphotonics Center.
“We don’t look at this the way we still look at fiber, which is to stuff as much bandwidth as we can onto the fiber and send it as far as we can,” Kimerling says. Instead, developers working to integrate optics tightly with silicon electronics must address not only bandwidth but packaging and cost issues.
Kimerling and other MIT researchers have developed a wealth of optical interconnect technologies. They've also been working closely with industry since the 1990s, “finding where the pinch points are and what needs to be done to solve them,” he says. And with funding from the National Institute for Standards and Technology, the Microphotonics Center joined with the International Electronics Manufacturing Initiative to create the Photonics System Manufacturing Consortium, which aims to develop a viable roadmap for manufacturing integrated photonics on silicon.
To read more, click here.