The arrival of brain-inspired computing: meet the AI chip that learns how to learn
US-based researchers have developed an AI chip that can, in essence, “rewire itself” like a human brain as it learns information.
A team led by engineers from Purdue University in Indiana has showcased circuit components which can change their function based on electronic pulses. The chip components can switch from being a resistor, a memory capacity, or an artificial neuron. This development could lead to the creation of “dynamic neural networks in hardware that can rewire themselves as they learn—just like the brain”, as reported by Singularity Hub.
The devices are made from perovskite nickelate, a material with electrical properties that can be altered by adding hydrogen ions to its structure. Researchers have noted that it is possible to create patterns of conductivity, or for the chip to shift from one configuration to another with ease.
The technology could lead to “brain-inspired” computing, while also providing a chance to create electrical components from the same material, which would simplify the chip making process. Researchers also claim that they are looking into how to create large-scale chips.
The world’s biggest AI chip was revealed in April 2021, featuring 2.6 trillion transistors. Large-scale chips are useful for reducing cost by linking chips together, while removing the need for data to be transferred between different AI chips in a server.