IBM has created a chip that works like a brain and could make artificial intelligence (AI) use less energy. This could help with the problem of AI systems using too much power in computer warehouses.
The chip might make AI chips for smartphones use less energy. This is because it uses parts that work like connections in the human brain.
The main advantage is that this chip works in a similar way to the brain, which is really good at doing things while using very little power. This could mean that big and difficult tasks could be done in places with not a lot of power, like cars, phones, and cameras. It could also help companies that provide computer services use less energy and be kinder to the environment.
The new chip uses things called memristors (memory resistors) that can hold different amounts of information, unlike regular digital chips that only use 0s and 1s. This is more like how the brain works, with its connections called synapses.
These memristors can “remember” their history of electricity, kind of like how synapses work in real brains. When you put a lot of memristors together, they can make a network that’s similar to a brain. While this is exciting, it’s important to remember that making a computer using memristors is hard, and there are problems like how much it costs to make and difficulties in manufacturing that need to be fixed before lots of people can use it.
The new chip is more energy efficient because of these parts, but it also has some regular digital parts that make it easier to use in current AI systems.
Lots of phones today have AI chips to help with things like pictures. For example, iPhones have a “neural engine” chip. In the future, IBM hopes chips like these could make phones and cars work better, with longer-lasting batteries and new things they can do.
In the end, chips like this one from IBM could save a lot of energy if they replace the chips in computers that power powerful AI systems. They could also help save water that’s used to cool down big computer centers that need lots of energy. A big center can use as much electricity as a medium-sized town.
James Davenport, who teaches about computers at the University of Bath, thinks what IBM found is interesting, but he wants people to know that this chip isn’t a simple solution. It’s a first step towards solving the problem of making AI systems use less energy.