Diagram shows components -- Sensay A, Antilles, and Sensay B -- arranged parallel to each other within an outer structure. The dimensions of the outer structure are labeled. It is 117mm long by 25 mm wide by 25 mm deep.

Miami faculty, small company team up to develop smaller, faster, more efficient computing device

We’ve all seen the message tacked onto the end of emails: “Please consider the environment before printing.” For those who do, indeed, consider the environment, digital often seems the better choice. Not printing that email saves a tree. Buying the digital version of a movie bypasses plastic waste. Holding a videoconference avoids the carbon emissions associated with travel to a face-to-face meeting.

But while having many of our digital possessions tucked away in the cloud may mean they leave virtually no footprint on our personal environments, they nevertheless leave a sizable footprint on the global environment. That’s because “the cloud” is actually millions of networked servers housed in huge data centers. According to an article in Yale Environment 360, “The biggest [data centers], covering a million square feet or more, consume as much power as a city of a million people. In total, they eat up more than 2 percent of the world’s electricity and emit roughly as much CO2 as the airline industry.”

Obviously, there’s no question of turning back; for environmental better or worse, digital is here to stay. So, where the analog world may have beat a path to the door of the inventor of a better mousetrap, the online world may beat a path to the door of the inventor of a better data center. That could end up being a team of researchers from Miami University and their industrial partner, Look Dynamics.

Powering artificial intelligence

The Miami researchers – Dave Hartup, Gokhan Sahin, and Chi-Hao Cheng in the Department of Electrical and Computer Engineering; John Femiani in the Department of Computer and Software Engineering; and Anthony Rapp in the Department of Physics – are working with photonic processing company Look Dynamics on a project funded by the U.S. Department of Defense’s Defense Advanced Research Projects Agency (DARPA). The project aims to create computing hardware that is not only smaller and more energy-efficient, but also faster, enabling higher performance hardware for artificial intelligence (AI) systems.

According to Hartup, AI, and specifically deep learning, are “hot topics” in engineering because of their use in technologies such as autonomous vehicles, advanced medical imaging, and remote sensing. But generating the powerful algorithms behind that AI requires computers that consume large amounts of energy and space. These issues of sustainability (all those data centers!) and portability limit the application of AI to applications where power and space are readily available.

In collaboration with Look Dynamics, Hartup, Sahin, Cheng, Femiani, and Rapp – along with undergraduate students Owen Hichens and Janelle Ghanem – are helping to overcome these limitations by creating hardware that functions in a completely different way from conventional computers.

A prototype of the novel AI optical processing hardware being developed by Miami University researchers and Look Dynamics.

Replacing electrons with photons

Conventional computers and devices that are controlled by conventional computers – like smart TVs, gaming consoles, and microwaves – are sometimes called “electronics” because they function by moving electrons along circuits. The flow of electrons is controlled by computer chip components called transistors. To process large amounts of information, computer chips contain many transistors, but adding too many slows down processing speeds. And using more transistors results in higher power consumption and generates more heat, which must then be dissipated by fans, which require even more power. So far, scientific advances have enabled a steady increase in the number of transistors on each computer chip, but there’s consensus among electrical engineers that a hard limit is on the horizon.

What the Miami team and Look Dynamics are working on is optical computing hardware. Instead of electrons, optical computing devices rely on photons, particles that make up light. Because photons are transmitted in free space, they are unconstrained by the need for circuits and transistors. As a result, optical systems are able to achieve a high degree of what electrical engineers and computer scientists call “parallelism,” efficiently performing many calculations and carrying out many processes simultaneously.

“The hardware we’re working on can implement AI algorithms 1,000 times faster with 1,000 times less power,” Hartup says, “and it’s 500 to 1,000 times smaller than conventional hardware.”

That’s exactly what’s needed to expand the use of AI to new applications where power and space are limited. New contexts require new AI algorithms, and the more efficiently those algorithms can be implemented, the more quickly technologies can be brought to market. Smaller algorithmic computing devices enable more portable, wearable, or seamlessly integrated technologies.

Enabling new AI applications

Hartup says portable technologies are of particular interest to project sponsor DARPA. Many of the things that AI is really good at enabling, like image recognition and the detection and tracking of moving objects, have obvious relevance to defense. That relevance is sometimes lost if the technology can’t be applied in the field.

“If you’re talking about something like advanced AI algorithms for image processing, you’re not going to carry around a rack of electronics capable of doing that,” Hartup says. “It’s too big and heavy. But with an optical system, it’s small enough and light enough to carry around.”

In the context of data centers, optical computers’ small size means improved sustainability. Swapping out conventional systems with smaller, faster optical ones could allow the physical footprint of data centers to be maintained or reduced, even as the proliferation of AI-enabled technologies ratchets up demand for computing capacity. And because optical computers use less electricity, data centers’ carbon footprints could shrink as well.

For all the complex technology involved, what the Miami-Look Dynamics team is doing boils down to something very simple: applying new design – optics – to make an existing, useful thing – a computer – even more useful. Metaphorically speaking, they’re building a better mousetrap, and DARPA has been the first to take what will surely become a well beaten path to their door.

Images courtesy of Dave Hartup.