Microchip Based on Phase Change Modules Makes Language Models More Energy Efficient
Phase Change Instead of Digital Bits: IBM Researchers Have Developed a New Type of Analog Microchips That Could Reduce the Enormous Power Consumption of Artificial Intelligence. The Chip Processes and Stores Information Through the Phase Change of a Material. In Initial Tests, a AI Language Model Running on These Analog Chips Achieved an Energy Efficiency of 12.4 Trillion Operations per Watt – That’s 14 Times More Energy-Efficient Than Common AI Systems, as Reported by the Team in ‚Nature‘.
Source: scinexx.de/Nature, IBM Research (Nadja Podbregar) 25.08.2023