New device could make processors run 1,000 times faster without additional waste heat — scientists say it could reduce data center energy demands

Researchers in Japan have created a device that promises to boost computer processing speeds, without generating massive amounts of additional heat.

Two of the limiting factors in high-performance computing, especially for the processors used in data centers, are the costly energy inputs required and the massive amount of waste heat generated. Generally, the faster a processor performs, the more heat it generates.

This principle applies to the largest and smallest machines; most people are familiar with the sound of fans whirring to cool down components when a computer is performing a particularly complex function. Cloud data centers, meanwhile, might have tens of thousands of servers, each generating massive amounts of heat from their processors.

But a new device, called a "non-volatile switching element," is capable of rapid processing without the problematic heat generation that's typically associated with fast processing, scientists have discovered.

The new device could process a bit — the smallest unit of information, represented as a "1" or a "0" — in just 40 picoseconds, or 40 trillionths of a second. For comparison, conventional chips struggle to process a bit in less than a nanosecond, or a billionth of a second.

In the new study, published May 14 in the journal Science, the scientists demonstrated that ultralow-power switching in the picosecond range was possible.

Tapping into the power of light

The researchers built this nonvolatile switching element device from ultrathin layers of tantalum (Ta) and Mn₃Sn atop a silica base. They chose tantalum, a refractory metal that can store and release electricity, and Mn3Sn because it is antiferromagnetic, meaning it has stable magnetic properties and is resistant to interference from external magnetic fields.

Then, they used an ultrafast pulse generator to control rapid pulses of light ‪—‬ as quick as 60 picoseconds per pulse ‪—‬ within the normal communication wavelength band. Each pulse of light passed through a high-speed photodetector called a uni-traveling-carrier photodiode (UTD-PD).

When the nonvolatile switching element device received pulses from the UTD-PD, the spins of the electrons in the material changed and the scientists recorded a minuscule magnetic force.

In the laboratory trials, the nonvolatile switching element operated consistently and reliably, despite performing over a billion switches, thereby proving the device's inherent stability. What's more, the process didn't require a continuous flow of electricity for the magnetic information to be maintained.

Most importantly, the processing generated minimal additional heat compared with that generated by a conventional computing processor. The nonvolatile switching element device could therefore bypass the challenge of high-speed processing by operating in a way that did not generate massive amounts of heat.

Minimizing waste heat

Waste heat is currently a major barrier to scaling up data centers' processing power, the scientists noted in the study ‪—‬ and this device could remove that limitation. Due to the low power requirements and low thermal generation, the nonvolatile switching element could dramatically reduce the power demands of processors.

However, manufacturing enough of these devices to make a difference may pose further challenges. Tantalum is a rare metal that is already in high demand, so there may be supply issues to overcome. The device would also need to be tested outside laboratory conditions, where external environmental factors could hinder the results.

Following the successful laboratory demonstration, a prototype chip could be ready by 2030, the scientists said in the study.

The researchers think a further reduction in the thickness of the Mn₃Sn layer will reduce power consumption even more. The next challenge, they added, will be to develop a commercially viable bulk manufacturing process capable of building the device at scale.