Performing some calculations with light is 10 times quicker and uses a quarter as much energy.
Biologists are testing a new way to efficiently sift through mountains of genomic data: lasers.
As Moore’s Law collapses and we reach the limits of what normal processors can do, unconventional techniques—often dismissed in the past as too complex—start looking like feasible ways to bite off niche computing problems and gain speed advantages. A UK startup called Optalysys thinks optical computing, which has failed to live up to its promise for years, could be used to spot similarities in large data sets like genomes.
Unlike regular chips, the processor Optalysys has developed uses laser beams to perform mathematical functions known as Fourier transforms. By encoding data onto light beams that can be made to interfere with each other, it executes some calculations in one shot that a regular computer would do in many stages. The results then feed back into regular chips. Nick New, CEO of Optalysys, says the technique is particularly well suited to tasks that identify regions of similarity in data.
The company’s first product puts the optical hardware required to perform such calculations on a PCI express card that fits into regular desktops or servers. Optalysys says the device performs Fourier transform calculations 10 times faster than an Nvidia P6000 graphics card, commonly employed for high-performance computation, using just 25 percent of the power.
The company will soon know whether those figures stand up in real use. This week, researchers at genomics institutes—including the Earlham Institute and the University of Manchester in the UK, as well as Oregon State University in the US—began beta-testing the hardware to search for similarities in DNA.
Optalysys created software that uses its hardware to replicate a tool biologists currently use to find short snippets of DNA inside huge genomes—a process New likens to “searching for a needle in a haystack.”
The firm hopes its speed advantage will be borne out for the researchers, who will use the system via the cloud for three months. The startup then plans to release a commercial version of the hardware and DNA software in the summer. New won’t reveal how much its first units will cost, but he says the optical hardware required to build them is falling rapidly in price because it’s used in consumer products like projectors.
To be clear, such hardware is unlikely to become a workhorse chip in your laptop. But it could help people perform certain tasks faster. Optalysys says it’s now focusing much of its attention on how its hardware could speed some of the calculations performed in AI algorithms for image recognition.