in

Chips that recover energy when working

Michael Frank has been working in a technological discipline that seems like science fiction for more than three decades. One that apparently pursues something impossible. And yet, it may not. The goal?

Than using a chip manages to recover energy.

Chips have always done the opposite. They consume energy. A lot. More and more, especially now that we are immersed In the era of glottone artificial intelligence.

And yet Frank is immersed in the development of the call “reversible computing“As they explain In IEEE SpectrumLast summer Frank left his work as a scientific engineer in Sandia National Laboratories to join a very special startup called Vaire Computing.

It is there where he intends to become what he has already estimated in his studies: according to his investigations, reversible computing could Allow energy efficiency 4,000 times higher to that of alternative options.

In Vaire Frank and his team are working in a chip prototype that will be manufactured in the first quarter of 2025 and that is very ambitious. Even so, for the first time a chip is expected to recover energy when used in an arithmetic circuit.

Then, in 2027, they hope to have a supply processor designed for artificial intelligence inference. The advanced version of said chip, which will theoretically be 4,000 times more efficient than the current ones, It will take 10 or 15 years to arrivetheir creators warn.

What is reversible computing

In 1961 Rolf Landauer, an IBM engineer, discovered that deleting a bit of information on a computer had an energy cost and produced heat. For him there was a solution: try to compute tasks without erasing information.

Xor
Xor

A traditional xor door is not reversible: you cannot recover entries just knowing the exit. If an additional output is added, a copy of one of the inputs is reversible. Thus, the two outputs can be used to “disassemble” the door Xor and recover the entrances and, with them, the energy used in the calculation. Source: IEEE Spectrum.

Landauer made clear that it seemed unchanged, because store all the information You would fill the memory With unnecessary data.

More than a decade later Charles H. Bennett, physicist and information theoretical in IBM, discovered a solution. Instead of storing intermediate results in memory, you could reverse the computing process, “disbelieve it” to store only the final inputs and outputs.

“Almost 100% of the energy used by a chip ends up being dissipated as heat, so we are basically evil. But in a reversible chip you never dissipate energy. Do not allow energy to become heat, and recycles internally.”

The idea was taking shape in academic environments, and in the 90 Frank ended up being one of those who immersed themselves in their promises. The problem is that when developing that concept they saw that although energy was recovered in the circuit, it was lost in the external power supply.

Vaire Computing was founded by researcher Hannah Earley and entrepreneur Rodfo Rosini. Earley explained In TechCrunch How “almost 100% of the energy used by a chip ends up being dissipated as heat, so we are basically missing it. But In a reversible chip you never dissipate energy. Do not allow energy to become heat, and recycles internally. “

In Varie they have apparently propose na solution with traditional CMOS transistors. To do this, they use the so -called “resonator”, a system that manages to recover energy by changing a bit from 0 to 1 so as not to turn it into heat and store most of it into the transistor itself.

That idea has already been explored in the past, but in Vaire they seem to have managed to integrate that resonator into the chip. They will start with a little ambitious chip and then work on those chips for inference to which they will be theoretically what will be what will be theoretically really demonstrate How far can you get with reversible computing, also well explained in this small essay published in the Newsletter Exponential View.

Of course, It is one thing to promise those efficiencies and another very different to achieve them. The proposal is very interesting and of course could raise a revolution that accelerated the development of much more powerful chips and that were not so voracious in terms of energy consumption. It remains to be seen if they will achieve it. Hopefully yes.

Image | Wu Yi

In Xataka | The 2 Nm chips race will start in 2025. And it will be the most fierce of all

What do you think?

Leave a Reply

Your email address will not be published. Required fields are marked *

GIPHY App Key not set. Please check settings

China is advancing at breakneck speed in nuclear fusion. It already has something ready that until now only the Netherlands had

Barcelona vs. Valencia, La Liga: where to watch it live in the United States