Modern data processing is digital, based on two states, zero one or on / off, while analog computers have many possible states - it's like the difference between turning the world on and off and using a dimmer that allows many different degrees of light. Neuromorphic, or brain-inspired engineering, has been the subject of research for over 40 years, but is only now starting to come to the fore as we are starting to hit the limits in digital processing and we increasingly need fast image processing, for example for autonomous cars. "We have powerful computers, no doubt, but the problem is we have to store the memory in one place and process it in another," explains Saptarshi Das, a Penn State professor.
Flipping data between memory and logic consumes a lot of energy and slows down processing speed, and what's more, computer architecture requires a lot of space. And if memory processing and storage took place in one place, the bottleneck would be eliminated: - We create artificial neural networks that are to imitate the energy and area efficiency of the brain. The brain is so compact that it fits inside our head, while supercomputers cover the area of two or even three tennis courts, another researcher adds.
And just as synapses that connect neurons in the brain can be reconfigured, scientists want to approach the issue of reconfiguring artificial neural networks by adding an electric field to the graphene sheet - a layer of carbon atoms one atom thick. In their latest research, it was possible to obtain at least 16 possible memory states in this way, unlike two in a traditional memristor (one of the basic passive electronic elements - works as a single memory cell, it can be used to store one bit of information, and its resistance can be current controlled).
- We have shown that we can control a large number of memory states with high precision, using simple graphene field transistors - they conclude. The only question is whether the commercialization of such a solution is possible and profitable? According to scientists, yes, especially since many huge concerns dealing with semiconductors are actively interested in biomimetic processing.