Human biology could be much more energy efficient than current computing.
Modern computers are a triumph of technology. A single computer chip contains thousands of millions of nanometric scale transistors that work extremely feasiblely and at a pace of millions of operations per second.
However, high speed and reliability are cost Significant energy consumption. Data centers and domestic computer appliances, such as computers and mobile phones, represent about 3% of global electricity demand, and AI use is likely to increase consumption further.
But what if we could redesign how computers work, so that they could perform computing tasks as quickly as today, spending much less energy?- Asks Heiner Linkeno .
In this case, Nature can offer us some potential solutions.
In 1961, the IBM scientist Rolf Landauer It has led to the question of knowing if we need to spend so much energy on computing tasks.
The expert created Landauer’s limit, which states that a single computational task – for example, set a bit, the smallest unit of computer information, to have a value of zero or one – must spend about 10–²¹ Joules (J) by Energia.
This is a very small amount, despite the many thousands of millions of tasks that computers perform. If we could use computers to these levels, the amount of electricity used in computing and residual heat management with cooling systems would not be concerned.
However, there is a if not. To perform a bits operation near Landauer’s limit, it has to be performed in an infinitely slow way. Computing in any finite period of time is expected to cost an additional amount that is proportional to the pace to which computing is made. In other words, The faster the computing, the more energy it is used.
More recently, this has been demonstrated by experiences created to simulate computational processes: energy dissipation begins to increase measurablely when performing more than one operation per second.
Processors that work at a clock speed of a billion cycles per second, which is typical in current semiconductors, use about 10–¹¹J by bit – about ten billion times more than Landauer’s limit.
The solution may conceive of computers in a fundamentally different way. Reason by which traditional computers work at a very fast pace is the fact that they operate in series – One operation at a time. If instead it was possible to use a large number of “computers” working in parallel, each could work much more slowly.
For example, a “hare” processor could be replaced, which performs a thousand million operations in a second, Millo Millions of “Turtle” processorseach taking a complete second to perform their task, with a much lower energy cost by operation.
A study in Nature In 2023 of which Heiner Linke was co -author showed that a computer could then work near the Landauer limit, using fewer energy orders than current computers.
Turtle power
Is it possible to have thousands of millions of independent “computers” to work in parallel? Asked the investigator.
Parallel processing to a smaller scale – explains – is already used today, for example, when about 10,000 graphic processing units or GPUS work at the same time to train artificial intelligence models.
However, this is not done to reduce speed and increase energy efficiency, but out of necessity. The limits of heat management make it impossible to further increase the computing power of a single processor, so processors are used in parallel.
An alternative computing system that is much closer what would be necessary to approach Landauer’s limit is known as Network -based biocomputation. It uses biological motor proteins, which are small machines that help perform mechanical tasks inside cells.
This system involves coding a computational task in a nanofabricated maze of channels with carefully designed intersections, which are usually made of patterns of silicon cookies. All possible paths through the maze are explored in parallel by a very large number of long, similar to wire molecules called biofilamentoswhich are fed by motor proteins.
Each filament has only a few nanometers in diameter and about a length micrometer (1000 nanometers). Each acts as an individual “computer”, encoding information through their space position in the maze.
This architecture is particularly suitable for solving the so -called combinatory problems. It is about problems with many possible solutionssuch as task programming, which are computationally very demanding for computers in series.
More efficient biocomputers
Experiences confirm that A biocomputer of this type requires between 1,000 and 10,000 times less energy per calculation than an electronic processor.
This is possible because the biological motor proteins themselves evolved in a way that no more energy than necessary to perform their task to the required pace. This is typically of a few hundred steps per second, one Million times slower than transistors.
Currently, only small biological computers have been built by researchers to prove the concept. To be competitive with electronic computers in terms of speed and computing, and explore a large number of possible solutions in parallel, Network -based biocomputation needs to be increased.
A detailed analysis shows that this should be possible with current semiconductor technology and could benefit from another great advantage of biomolecules in relation to electrons, namely their ability to transport individual information, for example in the form of an DNA label.
There are, however, numerous obstacles climbing these machines, including Learn to accurately control each of the bioofilaments, reduce their error rates and integrate them into current technology.
If this type of challenges can be exceeded in the coming years, resulting processors may solve certain types of difficult computational problems at an extremely reduced energy cost.
“Neomorphic computing”
Alternatively, it is an interesting exercise to compare the use of energy in the human brain. The brain is often acclaimed as being very energy efficient, using only a few watts – much less than AI models – for operations such as breathing or thinking.
However, they do not appear to be the basic physical elements of the brain that save energy. The shooting of a synapse, which can be compared to a single computational step, actually consumes approximately the same amount of energy a transistor needs by bit.
However, brain architecture is highly interconnected and works fundamentally differently from both electronic processors and network -based biofomputers. The so -called neuromorphic computing tries to emulate this aspect of brain operations, but using new types of computer hardware as opposed to biocomputation.
It would be very interesting to compare neuromorphic architectures with Landauer’s limit to see if the same types of knowledge of biocomputation could be transferred here in the future. If so, it may also be the key to a huge leap in front of computer efficiency in the coming years.
