A Shaped Magnetic Pulse Could Slash the Energy of Storing a Bit

Flip a single bit from a 0 to a 1 and you pay a tax. Not a metaphorical one: a real, physical dab of energy, dissipated as heat, every time a scrap of memory changes its mind. Multiply that dab by the traffic of a modern data centre, by every search and every image a large language model dreams up, and the tax bill starts to look like one of the defining engineering problems of the century. The scale is no longer hypothetical: the International Energy Agency reckons the world’s data centres drew around 485 terawatt-hours in 2025 and will roughly double that to 950 by 2030, when they would account for about 3 percent of global electricity, with AI-focused facilities growing fastest of all. So here is a question worth asking. How cheaply can a bit be flipped, in principle?

A team at the University of Edinburgh has a fresh answer, and it hinges on the shape of a magnetic pulse. Not its strength, which is where most of the effort has traditionally gone, but its shape in time.

The idea comes from a corner of mathematics called optimal control theory (OCT), the same toolkit that works out the most fuel-efficient path for a spacecraft or the leanest route through a supply chain. Point it at a magnet and it asks a narrow question: of all the ways you could push these spins from pointing up to pointing down, which one wastes the least energy on the way? The answer, it turns out, is not the brute-force shove that conventional memory relies on. It is a carefully sculpted pulse, chirped and twisting, that stays perpendicular to the magnetization the whole way round and front-loads its work into the opening stretch of the flip.

What does that buy you? Rather a lot, if the sums hold up.

The group, led by Elton Santos and with Mohammad Badarneh and PeiYu Cai doing the heavy computational lifting, modelled the trick in three atom-thin magnets: two iron-based compounds and a material called CrSBr. In their simulations the sculpted pulse reversed a magnet in the picosecond range, a few trillionths of a second, while spending up to two orders of magnitude less energy than the conventional field-driven approach. For one material, the model did the same job just as quickly with a magnetic field roughly ten times weaker than usual.

“Every digital operation has an energy cost, and that cost becomes increasingly important as AI and data-intensive technologies continue to expand,” says Santos.

Timing the Spins Like a Swing

A magnetic bit is a little forest of spins, all pointing the same way; writing to it means tipping them over to point the other way. The old way is to slam an opposing field across them and wait for them to topple, a bit like pushing a swing by leaning on it. Wasteful, and slow. The optimal pulse instead nudges the spins in time with their own natural precession, the way you would time your pushes to a swing already moving, so almost none of the energy is squandered fighting the material’s own tendency to spring back. Push the damping low enough, in the magnets that allow it, and the model drops into the femtojoule range, a scale where a single write costs less than the twitch of a biological synapse. Get there and you are within touching distance of the Landauer limit, the hard thermodynamic floor on what erasing a bit can ever cost. That floor is not just a chalkboard abstraction: it sits at roughly three zeptojoules, about 2.9 × 10⁻²¹ joules, at room temperature, and after going untested for half a century it was finally confirmed in 2012 using a single microscopic bead in a double-well trap. Physicists have since measured it directly in nanoscale magnetic bits, the very technology this study is trying to make cheaper, and found the dissipated heat consistent with Landauer’s bound.

Where the Saving Actually Hides

There is a catch, and to their credit the researchers don’t bury it. The optimal pulse is not always a smaller pulse. For one benchmark the peak field it demands is actually larger than what some conventional devices use; what OCT minimizes is the total heat dumped over the whole flip, not the height of the field at any instant. The energy saving is real in the model, but it’s a savings of a particular kind, and the university’s talk of ultra-low energy storage glosses over that distinction.

Two more caveats. This is theory and simulation, not a chip you can hold; the team sketches a plausible experimental setup but has not built one. And most of the numbers come from runs at absolute zero, with only a handful of finite-temperature checks, each averaged over 100 independent simulations to tame the thermal noise.

For all that distance between a simulation and a fabrication line, the gap the method is chasing is enormous, which is part of why it’s worth chasing. Today’s silicon transistors dissipate on the order of a femtojoule per switch, roughly a million times above the room-temperature Landauer floor, so the theoretical headroom for doing better is vast even if only a sliver of it is ever reachable in practice.

Still, the reason to pay attention is that the method doesn’t seem fussy about its medium. The same math that shapes a magnetic pulse can, the authors argue, be turned loose on electrical currents or ultrafast laser pulses, the two technologies most likely to sit inside future memory. “Although we first developed the theory using magnetic field pulses, the mathematics is far more versatile than that,” Santos says. That portability is the ambitious part: not a single gadget, but a recipe for wringing waste out of whichever switching mechanism eventually wins.

Whether any of it survives contact with a real fabrication line is the open question, and it is a big one. But if even a fraction of the projected savings carries over from the simulation to silicon, the humble act of flipping a bit could get an awful lot cheaper.

Frequently Asked Questions