The Next Leap

Some interesting new technologies are about to go on display. Chipmakers and systems companies have been working on quantum computing, photonics, and specialized AI processors, for the past several years, and those efforts are beginning to gain momentum.

The goal is no longer a doubling of performance and power. It’s now orders of magnitude improvement, and next week’s Hot Chips conference is an important showcase. Building on advances in partitioning, packaging, and highly integrated algorithms and accelerators, what’s new this year are commercialized efforts to harness qubits and light.

Until now, systems companies, big chipmakers, and startups have been relying on increased transistor density to boost speed. While that trend will continue — more slowly in two dimensions than in three — there also are developments in speeding up the processing on each compute element. That’s a different way of looking at the problem, and proponents believe it can greatly increase performance using less power.

This is particularly important for cloud providers, which find themselves in need of a way to differentiate themselves from a potential encroachment by edge computing. The efforts underway, if successful, would go a long way toward that differentiation.

Stepping stones are beginning to appear in the quantum world. Researchers at the University of Sydney, for example, have developed an algorithm that can be used to characterize noise in quantum computers, which is one of the reasons these devices need to operate deep in the Kelvin scale. The researchers say this is a first step toward noise metrology. At Yale University, researchers also have developed a way to stabilize qubits during operation, which is one of the key reasons why qubits are inaccurate.

Both private and government funding are flowing into quantum research. IBM and Google both have big investments in quantum computing. So do startups like D-Wave. And this week, AWS announced that it would help provide development tools and simulators, along with hardware tools, to researchers looking to get started with its Amazon Braket quantum computing service. The company also has a deal with Caltech to develop a quantum computing center.

On the AI front, two big hardware challenges are getting enough processing power into one or more chips, and building enough flexibility into these systems to be useful as algorithms continue to change. The first challenge may require a combination of compute technologies, from NPUs to other specialized accelerators, and integrating all of these elements plus some new ones is not trivial. For example, photonics also is beginning to show up as more than just a communications tool. Lightmatter, a startup, is working on a design to use light for computing, not just for communication.

The second challenge is flexibility, which is essential because everything is changing so fast. This is why embedded FPGAs are starting to gain momentum in the market. Along the same lines, Baidu is working on ways to use AI processors for diversified workloads.

And finally, there is much work on distributed compute architectures, in-memory computing, and much faster interconnects throughout the compute world. These are all in active development across the industry.

Taken as a whole, there is a push for multiple orders of magnitude improvements in performance, and potentially at significantly lower power. After years of research and limited movement, the compute world seems to have enough critical mass to make some significant leaps forward, even if it takes a few more years. Nothing is guaranteed, of course, and problems always crop up that no one expected. Still, the whole tech industry is making progress both in charted and uncharted territory, and the possibilities are very interesting.

Related
Upcoming Virtual Events in the Semiconductor Industry
Find the Hot Chips Conference, the Advanced Semiconductor Manufacturing Conference, Technology Symposium & Open Innovation Platform® Ecosystem Forum and more chip industry events here
Winners And Losers At The Edge
No company owns this market yet — and won’t for a very long time.
Making Silicon Photonics Chips More Reliable
More data and new applications are making this technology increasingly attractive, but it’s still not mainstream.