If you’re going to break the law, make it Moore’s Law

A startup is using photonics to build a faster chip—and it’s just one of several approaches to ushering in a post-Moore’s Law world.

In the news

It might seem risky to proclaim the imminent demise of Moore’s Law. In 1965, Intel founder Gordon Moore (who passed away last year) made one of science’s all-time-great predictions when he said the number of transistors on a computer chip would double every year. Moore later revised that to every two years, and the seemingly outlandish forecast has held true for nearly six decades.

Doomsayers have been claiming Moore’s Law was no longer achievable since at least the mid-1980s—but Nvidia’s new AI chip features more than 200 billion transistors. That’s up from 64 transistors per chip in the year Moore first made his prediction. Nevertheless, the naysayers are growing louder and more numerous. Is this the end for Moore’s Law?

While expert opinions vary, let’s agree it’s not feeling as peppy as once it did.

Enter Lightmatter, formed by MIT alums using photonic technologies to reinvent how chips communicate and calculate. The startup’s timing is excellent; the coming age of artificial intelligence will place unprecedented demand on computers’ processing capabilities, not to mention the requisite energy.

Where traditional chips rely on electrons alone for both processing and data transport, Lightmatter is adding photons—light itself—to the mix. The startup has two products. One focuses on communication among chips, while the other specializes in AI calculations.

The idea is that while electricity has myriad strengths (notably in the field of memory), light outperforms it in other areas, such as the matrix manipulations required in deep-learning models and other types of AI.

“With photonics, you can perform multiple calculations at the same time because the data is coming in on different colors of light,” says Lightmatter co-founder Nicholas Harris. He offers an example: One color could feature a photo of a cat; another a photo of a dog; and a third a photo of a tree. “And you could have all three of those operations going through the same optical computing unit … at the same time. That drives up operations per area and reuses the hardware that's there, driving up energy efficiency.”

The Cognizant take

Mr. Moore, meet Mr. Heisenberg. While many have whiffed in predicting an end to Moore’s Law, there’s good reason to say its end is in sight. As chips continue to be miniaturized, experts say the Heisenberg uncertainty principle—which postulates limits on computational accuracy—will eventually come into play, halting Moore’s established trajectory. More than a decade ago, one paper foresaw 2036 as the year this limit would come into effect—and that prediction is looking solid to many in the scientific community.

With this deadline looming, several approaches to a post-Moore future are under development. In addition to Lightmatter’s photonics-based approach, quantum computing shows promise. Quantum effects, including superposition and entanglement, may be able to overcome the miniaturization woes traditional chips are encountering.

Silicon alternatives, field-programmable gate arrays (a hardware solution popular in large data centers) and spintronics are among the other approaches undergoing experimentation. Meanwhile, at least one startup claims it’s extending the lifespan of Moore’s Law through the use of nanotechnology.

One major issue Lightmatter’s founders are tuned into is energy efficiency. “By 2040, some predict that around 80% of all energy usage on the planet will be devoted to data centers and computing,” Harris says, “and AI is going to be a huge fraction of that.” Lightmatter’s chips, he says, address this in two ways: They make use of existing hardware and increase the number of operations per area.

The digital world as we know it was made possible by the astonishing progress described by Moore’s Law. While that prediction may be running out of steam in the strictest sense, innovation in the fields of photonics, quantum and nanotechnology may yet have the final say.