A prototype AI chip from IBM may finally be the solution to the resource-hungry gigantic data centers that power current-gen AI models and algorithms. IBM has developed an AI chip that functions like the human brain and is energy efficient.
Several concerns are centered around the emissions of data centers that power AI systems. IBM’s prototype chip has the potential to create AI chips that are more power-efficient and less battery-draining, which will be particularly useful for smartphones.
The chip’s efficiency is attributed to its components, which operate similarly to connections in the human brain. According to scientist Thanos Vasilopoulos at IBM’s Zurich research lab, the human brain achieves impressive performance while consuming minimal power.
This increased energy efficiency could enable more complex tasks like cars, mobile phones, and cameras in low-power environments. Additionally, cloud service providers could leverage these chips to reduce energy costs and carbon footprint.
Unlike conventional digital chips that store information as binary 0s and 1s, this new chip employs analog memristors (memory resistors) that can hold a range of values. This analog behavior resembles the way synapses work in the human brain. Prof Ferrante Neri from the University of Surrey explains that memristors fall within nature-inspired computing, emulating brain function.
Memristors can “remember” their electric history, akin to biological synapses. When interconnected, these memristors can form networks resembling biological brains. The potential for chips using this technology is promising, as they could usher in brain-like chips shortly. However, challenges lie ahead, such as material costs and manufacturing complexities.
The new chip incorporates analog and digital elements, making it compatible with existing AI systems. Many modern phones include AI chips for tasks like photo processing. IBM envisions the new chip enhancing the efficiency of devices like phones and cars, leading to extended battery life and new applications. In the long run, these chips could save substantial energy by replacing the chips in data centers that power AI systems and even reduce the water needed for cooling these power-intensive facilities.
James Davenport, Professor of IT at the University of Bath, recognizes the potential of IBM’s findings but cautions that the chip is not a straightforward solution. Instead, it represents a possible initial step toward addressing these challenges.
