RESEARCH TO WATCH
Photos
Key takeaways:
Conventional transistors are expected to reach their maximum energy efficiency by 2037, yet demand for computing power continues to grow.
New devices made of combined ferroelectric and ferromagnetic materials (magnetoelectric multiferroics) could radically improve the speed and energy efficiency of computer processors and also add memory capability.
The DARPA-funded research is led by University of Michigan Engineering and includes partners from the University of Wisconsin-Madison.
An alternative to the transistor could make computers 100 to 1,000 times faster and more energy efficient, according to a team of researchers from University of Michigan Engineering and the University of Wisconsin, funded with up to $6M from the Defense Advanced Research Project Agency Fast and Curious program.
The new device—which the researchers call an M3BOLT—could go beyond the theoretical energy-efficiency limits of today's computers. Modern transistors are expected to reach that limit by 2037, yet the boom in AI and edge computing will continue to raise the demand for computing power. The M3BOLT could help meet the demand by using magnetic and sound waves instead of moving electrons.
"The transistor is arguably the most consequential device of the modern era, so the objective is not to claim that we will replace it everywhere. But major advances often begin by questioning an assumption that has become almost invisible," said Roozbeh Tabrizian , U-M associate professor of electrical and computer engineering who is leading the design of the M3BOLT device.
"What are the fundamental limits to computing speed and efficiency, when we aren't tied to the physics of any one device? We are exploring what becomes possible with an alternative," Tabrizian said.
By fundamentally changing the physics that underpin computers, the researchers hope to also integrate more advanced computing into remote and portable systems.
"The ability to harness and process information could scale tremendously," said John Heron , U-M associate professor of materials science and engineering and the principal investigator of the project. "Autonomous or mobile devices could carry better computers without huge power supplies."
Today, improvements in computer energy efficiency primarily come from packing chip components closer together or shrinking transistor sizes. When electrons travel shorter distances, less power is converted to, and wasted as, heat. Smaller transistors also require lower voltages to switch on and off, but when transistors reach a few nanometers in size, quantum tunneling allows electrons to leak through when they are switched off, wasting energy.
"We are trying to change the fundamental physics of the device, which alone can be 1,000 times more efficient. Then, if you add in new architectures, you could improve the efficiency even further," said Wei Lu , the James R. Mellor Professor of Engineering at U-M.
He leads the design of M3BOLT circuits.
"Right now we are building devices for extremely low-power applications. But imagine that someday we can build supercomputers with them. You might not need a nuclear power plant to run a large data center," said Lu, professor of electrical and computer engineering and materials science and engineering.
Using magnetic and sound waves as bits
The researchers hope to improve chip efficiency using a layered system of two materials developed by Heron's lab . The bottom material is a crystal called lead magnesium niobate-lead titanate, shortened to PMN-PT based on the chemical symbols of the elements in the material. On top is a film of magnetic iron-gallium alloy.
PMN-PT is ferroelectric—its charged elements naturally face a specific direction. When put inside an electric field, the orientation of the charge can flip, and it remains locked in place even when the electric field is turned off.
The shifting of PMN-PT atoms during electric field switching causes the magnetic poles in the iron-gallium alloy to wobble in a wave-like pattern, called a magnon. The magnetic waves move through the iron-gallium film with very little resistance, so less energy is lost as heat than in conventional materials. And because they can move at terahertz frequencies , they could allow computers to process data hundreds of times faster than today's consumer gigahertz processors.
The magnetic waves can be stopped by flipping the orientations of the magnetic poles within the iron gallium. Then the poles no longer wobble, and the waves become simple sound waves. These two states, with magnetic or nonmagnetic waves, could serve as 1 and 0 states, instead of the two voltage levels in conventional transistors.
The orientation of the magnetic poles is also retained in the iron-gallium film when the device is powered off, allowing the same material to also act as computing memory. When materials can simultaneously handle memory and computing, less data needs to be transferred between different parts of the computer, which saves energy and space.
Researchers from the University of Wisconsin will contribute to the study under the leadership of Jiamian Hu , associate professor of materials science and engineering. They will provide physics-based computer simulations that guide and accelerate design by predicting how the device's architecture affects performance, as well as the impact of defects inside and at the interface of the materials.
Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Defense Advanced Research Projects Agency.