類腦計算機會成為「競爭殺手」嗎？

The demand for electricity from data centres, artificial intelligence (AI), and cryptocurrency is growing rapidly. A recent report suggests that by 2026, the energy consumption by these sectors could double compared to 2022 levels. This would mean that the energy usage by these sectors could be equivalent to Japan's annual energy needs. In response to this, companies like Nvidia are focusing on developing more energy-efficient hardware. However, there is a growing interest in exploring a fundamentally different type of architecture for computers - one that mimics the human brain and is more energy-efficient. This approach, known as neuromorphic computing, involves electronic devices imitating neurons and synapses, resembling the electrical network of the brain. While this concept is not new and has been under development since the 1980s, the increasing energy demands of AI are driving efforts to bring this technology into practical use. Although current systems primarily serve as research tools, major players like Intel and IBM are eyeing commercial applications. SpiNNcloud Systems, a spinout of the Dresden University of Technology, has announced plans to sell neuromorphic supercomputers for the first time, with a focus on energy-efficient AI applications. Another potential application is in edge computing, where data is processed in real-time on connected devices with power constraints, such as autonomous vehicles and wearable technology. However, there are significant technical challenges, particularly in developing the necessary software for these chips to function effectively. Additionally, the cost of creating these new chips, whether using silicon or other materials, is a notable concern. Despite these challenges, companies like Intel and IBM have made substantial progress in developing large-scale neuromorphic research systems, indicating the potential for practical applications in the future.