20瓦就能运行下一代AI？科学家瞄上了神经形态计算

Core Viewpoint - Scientists are attempting to create a neuromorphic computer that mimics the human brain, potentially revolutionizing AI by significantly reducing energy consumption while enhancing processing speed [2][4][6]. Group 1: Current AI Challenges - The rapid development of large language models has led to an "energy crisis" in AI, with projected electricity costs for running these models reaching $25 trillion by 2027, surpassing the annual GDP of the United States [3][4]. - In contrast, the human brain operates on approximately 20 watts daily, comparable to a household LED bulb, prompting researchers to explore more efficient AI models [4]. Group 2: Neuromorphic Computing - Neuromorphic computing aims to replicate the structure and function of the human brain, utilizing energy-efficient electronic and photonic networks to integrate memory, processing, and learning [6][8]. - Key features of neuromorphic systems include: 1. Event-driven communication that activates circuits only when necessary, reducing power consumption [9]. 2. In-memory computing to minimize data transfer delays [10]. 3. Adaptability, allowing systems to learn and evolve over time without centralized updates [10]. 4. Scalability, enabling the architecture to accommodate complex networks without significantly increasing resource demands [10]. Group 3: Technological Advancements - Current neuromorphic computers possess over 1 billion neurons and 100 billion synapses, indicating the potential for brain-level complexity [15]. - Major tech companies like IBM and Intel are at the forefront of this technological revolution, with products like IBM's TrueNorth chip and Intel's Loihi chip designed to simulate brain activity [18]. - The global neuromorphic computing market is expected to grow exponentially, reaching $1.81 billion by 2025, with a compound annual growth rate of 25.7% [19].