Phones, laptops, supercomputers ... everything could change radically in a few years thanks to a new type of molecular circuit that eliminates the biggest problem in today's computers, achieving power and fault tolerancecomparable to that of the human brainwith minimal use of energy. .
The big problem that limits the power of any modern computer is its own architecture.Roughly, a machine has a central processor in charge of performing operations with data that comes from memory.This architecture - calledvon Neumann or Princeton- is inevitablylimitedby the communication channel between processor and memory.Every time a program has to execute an instruction, the processor has to access data in memory, process it, and send it back to memory.
This causes a bottleneck - calling von Neumann - which isphysically impossible to solveand can only be partially alleviated with different tricks, such as buffers built into the processor or process predictors that save a small part of the access to the main memory of the computer. system.
Now, in astudy publishedin the scientific journal Nature, an international group of engineers from Ireland, the United States, India and Singapore claim that their new circuit completely destroys the von Neumann bottleneck.It is a new type of architecture capable ofsimultaneouslystoring and processing information, as the human brain does.
How does it work
This circuit is called a 'memristor', a fusion of memory and transistor.It is not a new concept.It wasfirst proposed in 1971at the theoretical level but no one has been able to get it beyond the laboratory.All the experimental designs operated in a very limited temperature range and could only be made withextremely rareand expensiveelements, such as niobium and vanadium.
But this new design removes those limitations.It is an organic-metallic molecule of77 atoms.The molecule has the ability to consistently store data but also process it with a logical tree of instructions similar to the dendrites of neurons.
This logical tree exhibits the same neuroplastic properties of the brain: it can basically change configuration as needed, exposing it to different voltage levels.
An experimental prototype memristor manufactured by the US Department of Energy
Sreetosh Goswami, one of the authors of the study, says that the molecule exhibits "a flexibility and adaptability similar to the connectors of the human brain."Your memristor can be “instantly reconfigured for different computational functions simply by changing the applied voltages.Also, just as nerve cells store data, the same devicecan store informationfor future access and processing. "Goswamiwas the one who found the keyin the manufacture of the material to build this memristor, which consists of an iron molecule in its central part linked to organic molecules.
Their processing power is such that the team couldn't believe their data.According to Goswami, it is extraordinary: “our device was doing something like the brain does but in a different way.When you learn something new or are deciding something, thebrain reconfigures itselfand changes its physical 'wiring'.Similarly, we can logically reprogram or reconfigure our devices using different voltage pulses ”.
Unimaginable speed
According to the inventors of the new memristor, its architecture also allowsmultiple datato beprocessed simultaneouslyusing its instantly reconfigurable instruction tree.In fact, they claim that their memsistor has such processing power that one of these molecules can replace thousands of transistors in a traditional processor.
The memsistor can be connected in parallel with others to obtain a process speed previously unknown.The team also claims that because of the way it operates, it is resilient to failure.And since everything - processing and data storage - is carried out in the same nanoscopic circuit, theconsumption and time savingshave no equivalent in the current industry.
Thompson holding the magic molecule that makes up the memristor
Another of the study's authors - Damien Thompson, in the image above these lines -assuresthat they are “very excited about the possibilities because these devices show all the characteristics of brain computing.First, a gigantic number of tiny identical molecular network processors work in parallel [like neurons].Even more importantly, they areredundantand reconfigurable, which means that a device can solve problems even if its individual components do not work perfectly or in the same way at all times ”.
As always with all these kinds of cutting-edge developments, you'll have to wait and see how long it takes to get to market.But it is clear that, just when we are about to reach the limit of computing traditional architectures, we now have two interesting new avenues open:quantum computingtechnologyand this new type of circuitry that emulates the behavior of themost advanced computer.that we know: the human brain.