Recently, Chinese scientists have successfully developed a novel computing architecture that significantly enhances computational efficiency and processing power. Breaking away from the limitations of the traditional von Neumann architecture, this new design adopts a Computing-in-Memory (CIM) approach, tightly integrating data storage and processing units to drastically reduce energy consumption and latency caused by data movement. The research team, comprising members from Tsinghua University and the Chinese Academy of Sciences, published their findings in the prestigious international journal Nature Electronics.Leveraging emerging non-volatile memory devices such as memristors, the architecture enables parallel computation and intelligent scheduling directly at the hardware level—making it especially well-suited for high-demand tasks like artificial intelligence and big data analytics. Experimental results show that, compared to current GPUs and dedicated AI chips, this new architecture achieves over 10 times better energy efficiency and 3 to 5 times faster computation speeds.This breakthrough not only offers a potential solution to China’s longstanding challenges in advanced semiconductor technology but also presents a new pathway for global computing innovation. In the future, the technology could be widely applied in autonomous driving, smart cities, scientific computing, and other fields, accelerating the deployment of next-generation high-performance, low-power computing systems.
近日，中国科学家成功研发出一种新型计算架构，显著提升了计算效率与算力性能。该架构突破了传统冯·诺依曼体系的限制，采用存算一体（Computing-in-Memory）设计理念，将数据存储与处理单元高度集成，大幅减少数据搬运带来的能耗与延迟。研究团队来自清华大学与中科院等机构，其成果已在国际顶级期刊《自然·电子学》上发表。新架构利用新型非易失性存储器件，如忆阻器（Memristor），在硬件层面实现并行计算与智能调度，特别适用于人工智能、大数据分析等高负载任务。实验表明，相较现有GPU和专用AI芯片，该架构在能效比上提升超过10倍，同时运算速度提高3至5倍。这一突破不仅有望缓解我国在高端芯片领域的“卡脖子”问题，也为全球计算技术发展提供了新路径。未来，该技术可广泛应用于自动驾驶、智慧城市、科学计算等领域，推动下一代高性能、低功耗计算系统的落地。
原创文章，作者：admin，如若转载，请注明出处：https://avine.cn/11960.html