The February 2023 issue of IEEE Spectrum is here!

Close bar

Forget Moore’s Law—Chipmakers Are More Worried About Heat and Power Issues

As transistors become smaller, problems with heat dissipation and power consumption are driving the industry toward specialization

2 min read
Abstract illustration of semiconductor
Illustration: iStock

Power consumption and heat generation: these hurdles impede progress toward faster, cheaper chips, and are worrying semiconductor industry veterans far more than the slowing of Moore’s Law. That was the takeaway from several discussions about current and future chip technologies held in Silicon Valley this week.

John Hennessy—president emeritus of Stanford University, Google chairman, and MIPS Computer Systems founder—says Moore’s Law “was an ambition, a goal. It wasn’t a law; it was something to shoot for.”

 “It is definitely slowing down,” he says, “but to say it’s dead is premature.”

That slowing, at this point, isn’t his biggest concern. The real problem, Hennessy says, is the failure of Dennard scaling, an observation that as transistors get smaller and circuits become faster, a chip’s power consumption stays the same.

“Who would have thought,” he says, “that microprocessors would have to slow down clock speeds or turn off cores to keep from burning up?” Hennessy spoke as part of a panel at a Churchill Club forum held Monday in Menlo Park.

The power consumption of microprocessors was also a hot topic at Arm TechCon 2019 on Tuesday, with Sha Rabii, Facebook’s head of silicon and technology engineering, indicating that the energy used by microprocessors and the heat that chips give off is a major roadblock on the road to augmented reality glasses.

How to tackle the problem? The key, several industry veterans suggested, may be specialization.

“Either we keep going on [an] evolutionary path, with faster CPUs and everything happening in software, or we look at it as a systems problem and think about what we would do differently” if the industry were designing from the ground up, said Navin Chaddha, managing director of the Mayfield Fund, speaking at the Churchill Club event. “I believe the world is moving to specialization,” Chaddha says, instead of focusing on doubling raw processing power every 18 to 24 months.

The recent rush of startups producing processors designed to do deep learning—such as Cerebras Systems, Mythic, and Syntiant—are examples of this kind of thinking. But there may be limits to just how much specialization can help.  

Arm Holdings CEO Simon Segars would seem to agree with Chaddha. Segars kicked off Arm TechCon by announcing that the company will, for the first time, allow developers to insert custom instructions into the core of Arm chips, allowing more efficient processing of algorithms. This capability will permit at least some degree of specialization, which is a big departure for a company that has always focused on standardized products. 

The Conversation (0)

Two Startups Are Bringing Fiber to the Processor

Avicena’s blue microLEDs are the dark horse in a race with Ayar Labs’ laser-based system

5 min read
Diffuse blue light shines from a patterned surface through a ring. A blue cable leads away from it.

Avicena’s microLED chiplets could one day link all the CPUs in a computer cluster together.

Avicena

If a CPU in Seoul sends a byte of data to a processor in Prague, the information covers most of the distance as light, zipping along with no resistance. But put both those processors on the same motherboard, and they’ll need to communicate over energy-sapping copper, which slow the communication speeds possible within computers. Two Silicon Valley startups, Avicena and Ayar Labs, are doing something about that longstanding limit. If they succeed in their attempts to finally bring optical fiber all the way to the processor, it might not just accelerate computing—it might also remake it.

Both companies are developing fiber-connected chiplets, small chips meant to share a high-bandwidth connection with CPUs and other data-hungry silicon in a shared package. They are each ramping up production in 2023, though it may be a couple of years before we see a computer on the market with either product.

Keep Reading ↓Show less