Analogue computers use less energy than digital ones
metamorworks/Getty Images
Analogue computing has emerged as a potential solution to the energy consumption challenges faced by data centres due to the increasing demand for artificial intelligence models. While digital computers process data using binary digits, analogue computers operate using continuously varying quantities such as electrical resistance.
Although analogue computers have traditionally been less accurate than digital ones, recent advancements have shown promise. Researchers at Peking University, led by Zhong Sun, have developed analogue chips that can efficiently solve matrix equations, which are essential for tasks like data transmission, scientific simulations, and AI model training.
The innovative approach involves a pair of chips working together – one providing rapid but low-precision solutions, while the other refines the results to achieve high accuracy. The collaboration results in a significant reduction in error rates, matching the precision of standard digital calculations.
While the current chips are capable of handling small matrices, scaling up to tackle the complex problems in large AI models would require more extensive circuits. Despite this limitation, analogue chips offer superior speed and energy efficiency compared to digital counterparts.
One notable advantage of analogue computing is that larger matrices do not significantly impact processing time, unlike digital systems which face exponential challenges with increasing matrix size. This efficiency could potentially surpass the throughput of high-end GPUs while consuming significantly less energy.
Although the theoretical benefits of analogue computing are promising, practical implementation for real-world applications may require hybrid solutions integrating analogue and digital technologies. James Millen from King’s College London emphasizes the potential of analogue computing in accelerating matrix calculations and reducing the energy demands of AI-driven processes.
As the world grapples with the energy implications of AI proliferation, the development of efficient computing solutions like analogue chips could pave the way for a more sustainable and high-performance computing landscape.
