Researchers have developed a new AI chip that increases the energy efficiency of common systems by six times. The chip uses novel material systems and mimics the processing methods of biological neural networks to dramatically reduce energy consumption.
With the ever-increasing use of AI systems, we will soon face a problem. Intelligent algorithms require a large amount of energy to efficiently perform calculations and output answers. Researchers around the world are therefore working on the next generation of computer chips to reduce energy consumption in the medium term.
Now a team has developed a new AI chip that increases the energy efficiency of common systems by six times. The chip uses a new type of material system and mimics the processing methods of biological neural networks. This should drastically reduce the energy consumption of AI technologies. Both the execution of computing operations and data storage and energy efficiency are said to have been significantly improved.
AI chip: Higher energy efficiency through more effective data transmission
The new chips use memristors, also known as memory resistors, which are made of so-called entropy-stabilized oxides (ESOs). This material system includes over half a dozen elements and offers finely tunable memory capacities.
Memristors are similar to biological neural networks in that they do not require an external memory source and do not waste energy on data transfer. This technology makes AI tasks more energy efficient than conventional central processors. Another advantage of the ESO-based chips is the processing of time-dependent information.
These occur in dynamic content, such as audio and video data. This works because the system can adapt the composition of the ESOs to different time formats. This allows it to process information in real time and recognize connections.
AI requires increasing amounts of globally available energy
The AI chips could have a significant impact on the development of future systems with artificial intelligence. The use of memristors reduces energy consumption while increasing efficiency. This also seems necessary. According to forecasts, AI could account for half a percent of global energy consumption in 2027. This is roughly comparable to the annual energy consumption of the Netherlands.
The advances could reduce the environmental impact of the increasing energy requirements of AI systems and increase their performance. Ultimately, intelligent algorithms are likely to continue to play a crucial role in the future and simplify many processes in our everyday lives.
Also interesting:
Source: https://www.basicthinking.de/blog/2024/07/07/ki-chip-energieeffizienz/
