IBM develops ‘brain like’ chip for energy-efficient AI
The prototype could lead to more energy-efficient AI following concerns surrounding emissions from warehouses packed with computers that are used to power AI systems.
The prototype semiconductor works by using its components in a similar way to connections in the human brain, using analogue memory resistors (memristors) that can store a range of numbers, unlike most digital chips which store information as 0s and 1s. These kinds of memristors could ‘remember’ their electric history in a similar way to synapses in a living person.
Cloud providers could use these naturally-inspired chips to reduce energy costs and carbon footprints, however it remains unclear as to when a commercially-viable chip of this design will be available.