Increasing demand for data-intense computing applications—such as artificial intelligence, large language models and high-performance computing—has created a need for computing infrastructure that can handle large workloads with high energy efficiency. Advances in silicon-based complementary metal–oxide–semiconductor technology have led to more efficient field-effect transistors, but these devices are fundamentally limited by thermionic injection. As a result, on–off switching efficiency cannot be improved beyond 60 mV of drive voltage per decade of current. Operation of electronics at cryogenic temperatures, such as 77 K, can overcome this limit and provide performance improvements. Here we explore the development of computing at cryogenic temperatures. We examine the changes in electrical transistor and material properties observed at low temperatures, and highlight the need for further studies on cryogenic noise, reliability, variability and thermal management. We also consider the potential performance improvements at the device and circuit level of such technology.

对数据密集型计算应用程序（如人工智能、大型语言模型和高性能计算）的需求不断增长，这催生了对能够以高能效处理大型工作负载的计算基础设施的需求。硅基互补金属氧化物半导体技术的进步导致了更高效的场效应晶体管，但这些器件从根本上受到热离子注入的限制。因此，每十倍频程电流的驱动电压超过 60 mV 时，无法提高开关开关效率。在低温（如 77 K）下运行电子设备可以克服这一限制并提高性能。在这里，我们探讨了低温计算的发展。我们研究了在低温下观察到的电晶体管和材料特性的变化，并强调了进一步研究低温噪声、可靠性、可变性和热管理的必要性。我们还考虑了此类技术在器件和电路级别的潜在性能改进。