Nowadays, artificial intelligence is used in more and more areas of everyday life. Undoubtedly, it is a very useful technology that makes our lives easier. The problem is, the large amounts of energy required to operate AI-based systems. This is why technology companies and researchers around the world are actively trying to develop neural networks that consume less energy.
For this purpose, IBM has developed a new silicon‑based chip that contains all the basic features of a neural network and is 100 times more energy‑efficient than existing chips.
Neural networks, which are inspired by the way the human brain works, are usually implemented in software and embedded in a device.
IBM researchers have created micro-electrical artificial synapses on their chips that mimic the synapses that connect individual neurons in the brain. The paper, published in the journal Nature, cites neuroscience as the inspiration for the work.
This method addresses several critical issues, including low accuracy, which has prevented previous attempts to build artificial neural networks in silicon. The researchers examined the chip's functionality, assigning it two simple tasks: recognizing images related to handwriting and classifying colors.
The results were astonishing, as the system was absolutely accurate and consumed only 1 percent of the energy compared to others.
However, IBM has a long way ahead, as the chip design is quite bulky. It has five transistors and three other components compared to a single transistor in existing standardized chips.
Nevertheless, this development is of great significance as it could pave the way for the creation of advanced computers with AI embedded in their cores.
