With novel optoelectronic chips and a new partnership with a top silicon-chip manufacturer, MIT spinout Ayar Labs aims to increase speed and reduce energy consumption in computing, starting with data centers. Backed by research at MIT and elsewhere, Ayar has developed chips that move data around with light but compute electronically. The unique design integrates speedy, efficient optical communications into traditional computer chips, replacing less efficient copper wires.

Ayar Labs’ optoelectronic chips move data around with light but compute electronically. (Image courtesy of Ayar Labs)

Ayar believes that the chips can reduce energy usage by about 95 percent in chip-to-chip communications and increase bandwidth tenfold over their copper-based counterparts. In large data centers, they believe the chips could cut total energy usage by 30 to 50 percent. The chips could also be used in supercomputers, which have similar efficiency issues and speed constraints as data centers. The technology could also improve optics in various fields, from autonomous vehicles and medical devices, to augmented reality.

The idea of the original research was to help data transmission keep up with Moore's Law. The number of transistors on a chip may double every two years, but according to the researchers, the amount of data pushed across those copper pins hasn't grown at the same rate.

The researchers chose light because an optical wire can transmit multiple data signals on different wavelengths, while copper wires are limited to one signal per wire. Optical chips can, therefore, transmit more information using significantly less space. Moreover, photonics produces very little waste heat compared to copper wires, which generate a large amount of waste heat, thus reducing efficiency in individual chips. This is an especially important issue in data centers, where copper wires run inside and between servers.

Computer chips send data between chips with different functions, such as logic chips and memory chips. With copper-based communications, however, the chips can't send and receive enough data to take advantage of their increasing processing power. That's caused a bottleneck, where chips must wait long durations to send and receive data. More than half the time in data centers, for instance, circuits are waiting for data to come and go.

The collaboration integrated optical components onto silicon chips fabricated using the traditional CMOS semiconductor manufacturing process that churns out chips for pennies. CMOS doesn't lend itself well to optics, so industry veterans assumed you'd have to make major changes to get it to work. To avoid making changes to the CMOS process, the researchers focused on a new class of miniaturized optical components, including photodetectors, light modulators, waveguides, and optical filters that encode data on different wavelengths of light, and then transmit and decode it. They essentially “hacked” the traditional method for silicon chip design, using layers intended for electronics to build optical devices, and enabling chip designs to include optics more tightly configured than ever inside a chip's structure. In 2015, chips, were successfully manufactured that contained 850 optical components and 70 million transistors and performed as well as traditional chips. This year, Ayar's first prototypes should reach U.S. data centers, with a planned 2019 commercial release.

In addition to solving the chip input-output problem, Ayar is also excited about what its new technology means for the field of optics in general. Optical sensors, for instance, are used in self-driving or semiautonomous vehicles and expensive medical equipment. Lowering manufacturing costs, while increasing computational power, could make those technologies much less expensive and more accessible.

For more information, contact Sara Remus at Sara Remus at This email address is being protected from spambots. You need JavaScript enabled to view it., 617-253-2709 .

Photonics & Imaging Technology Magazine

This article first appeared in the November, 2018 issue of Photonics & Imaging Technology Magazine.

Read more articles from this issue here.

Read more articles from the archives here.