Can the power of light unlock the next advances in artificial intelligence computing?
MIT spinoff Lightelligence is seeking to answer that question with a new kind of AI hardware that is the “world’s first realization of deep learning neural network computing in a photonic integrated circuit,” the company said in a press release.
The Cambridge-based startup announced on Friday that it has raised a $10 million seed financing round backed by “top VC investors and leading industry technologists” to support development. The company said the round was led by Baidu Ventures, which is the venture capital arm of Chinese tech giant Baidu, and a group of U.S. semiconductor executives.
The large seed round for Lightelligence comes as funding for AI hardware startups is red hot, having raised more than $1.5 billion last year, which is more than double from two years ago, according to CB Insights research cited in a recent New York Times article.
Lightelligence is acutely aware of the growing opportunities in the AI hardware space. In its press release announcing the seed round, the startup cited Nvidia’s skyrocketing stock price thanks to companies using its graphical processing units for AI computing. It also references Intel’s six AI hardware acquisitions from the past year.
The company said what separates itself from the other AI hardware startups is that instead of using an electronic circuit as the basis for its computer chip, it uses a photonic circuit. By using the power of light, the company said it enables” ultra low latency, high throughput and extremely high power efficiency.”
The technology is based on years of research in nanophotonics, deep learning and optical computing by the company’s founding team at MIT, which includes CEO Yichen Shen, who was named to Forbes 30 Under 30 for 2018. One of the company’s directors is Marin Soljacic, founder of wireless charging company WiTricity and an MIT professor who runs the school’s Photonics and Modern Electro-Magnetics Group.
Lightelligence said its technology has “superior performance” over competitors in many areas. For cloud computing, the chip acts as a co-processor for CPUs to “accelerate deep learning training and inference.” The company is also looking at applications for edge computing.
“Edge computing for us means specialized application where our customers have a specific algorithm to employ,” Paul Xie, one of the company’s co-founders, told BostInno in an email. “We can then design our chip to maximize certain aspects of the performance, whether it’s speed, power or latency. Examples of such application include cameras, blockchain hardware and high-frequency trading hardware.”
The company’s website lists 10 job openings, with titles ranging from “deep learning architect” to “photonic design engineer” and “computer chip architect.”