Our ability to pack ever-smaller transistors on a chip has enabled today’s age of ubiquitous computing. But in accordance with some experts, this approach is finally reaching its limits He explains the tip of Moore’s Law and a related principle referred to as Dennard’s scaling.

These developments couldn’t come at a worse time. Demand for computing power has skyrocketed lately, driven largely by the rise of artificial intelligence, and shows no signs of slowing down.

Now Lightmatter, an organization founded by three MIT graduates, is constant the remarkable progress of computing by rethinking the lifeline of the chip. Instead of relying exclusively on electricity, the corporate also uses light for data processing and transport. The company’s first two products, a chip that makes a speciality of artificial intelligence operations and a compound that facilitates data transfer between chips, use each photons and electrons to drive more efficient operations.

“The two problems we solve are: ‘How do chips talk?’ and ‘How do you do these (AI) calculations?'” says Nicholas Harris PhD ’17, co-founder and CEO of Lightmatter. “With our first two products, Envise and Passage, we’re addressing each questions.”

In a nod to the scale of the issue and the demand for AI, Lightmatter raised just over $300 million in 2023 at a valuation of $1.2 billion. Now the corporate is demonstrating its technology at among the world’s largest technology corporations in hopes of reducing the large energy demands of knowledge centers and AI models.

“We will enable platforms based on our interconnect technology that consist of a whole lot of 1000’s of next-generation computing units,” says Harris. “Without the technology we’re developing, this simply wouldn’t be possible.”

From idea to $100,000

Before joining MIT, Harris worked at semiconductor company Micron Technology, where he studied the basic devices behind integrated chips. This experience made him realize that the normal approach to improving computer performance – packing more transistors on each chip – was reaching its limits.

“I saw the information processing roadmap slowing down and I desired to determine the way to move forward with it,” Harris says. “What approaches can augment computers? Quantum computing and photonics were two of those paths.”

Harris got here to MIT to work on photonic quantum computing as a part of his doctoral research with Dirk Englund, an associate professor within the Department of Electrical Engineering and Computer Science. As a part of this work, he built integrated silicon-based photonic chips that might send and process information using light as a substitute of electricity.

The work led to dozens of patents and greater than 80 research papers in prestigious journals corresponding to . But one other technology also caught Harris’ attention at MIT.

“I remember walking down the hallway and seeing students pouring out of those auditorium-sized classrooms and watching live videos of lectures to see professors teaching deep learning,” Harris recalls, referring to the Artificial intelligence technology. “Everyone on campus knew that deep learning was going to be a giant deal, so I began learning more about it and we realized that the systems I used to be constructing for photonic quantum computing could actually be used for deep learning. “

Harris had planned to turn into a professor after earning his doctorate, but realized that he could attract more cash and innovate more quickly through a start-up, so he teamed up with Darius Bunandar PhD ’18, who also studied in Englund’s lab, and Thomas Graham MBA together ’18. The co-founders successfully launched into the startup world by winning the 2017 MIT $100K Entrepreneurship Competition.

See the sunshine

Lightmatter’s Envise chip takes the a part of computing that electrons do well, corresponding to memory, and combines it with what light does well, corresponding to performing the huge matrix multiplications of deep learning models.

“Photonics lets you do multiple calculations at the identical time because the information is available in for various colours of sunshine,” Harris explains. “You could have a photograph of a dog in a single color. In a unique color you can have a photograph of a cat. In a unique color, possibly a tree, and you can run all three of those processes concurrently through the identical optical processing unit, this matrix accelerator. This increases operations per area and reuses existing hardware, increasing energy efficiency.”

Passage leverages the latency and bandwidth benefits of sunshine to attach processors in the same option to how fiber optic cables use light to send data over long distances. It also allows chips the scale of entire wafers to operate as a single processor. Sending information between chips is central to running the huge server farms that power cloud computing and power AI systems like ChatGPT.

Both products are designed to make computers more energy efficient, which Harris says is required to maintain up with increasing demand without causing an enormous increase in power consumption.

“Some predict that by 2040, around 80 percent of all energy consumption on the planet might be spent on data centers and computing, and AI will account for a big portion of that,” Harris says. “When you take a look at using computers to coach these large AI models, it’s estimated that they may eat a whole lot of megawatts. Their electricity consumption is on the dimensions of cities.”

Lightmatter is currently working with chip manufacturers and cloud service providers for mass deployment. Harris points out that because the corporate’s devices are powered by silicon, they might be manufactured in existing semiconductor manufacturing facilities without major process changes.

The ambitious plans are intended to open up a brand new path for computer science that may have a big impact on the environment and the economy.

“We will proceed to check all of the parts of a pc to determine how light can speed them up, make them more energy efficient and faster, and we’ll proceed to switch those parts,” Harris says. “Right now we’re focused on connecting with Passage and processing data with Envise. But over time we’ll develop the following generation of computers, and it can all be about light.”

This article was originally published at news.mit.edu