热力学计算是一个新兴领域，旨在利用热波动而非对抗它来进行计算。传统电脑工程师致力于消除热杂讯以确保计算准确，但热力学计算试图将这种随机的能量波动转化为计算资源，这可能彻底改变我们对计算的认知，并大幅降低设备的能耗与散热需求。

这个概念受到自然界热力学过程的启发，将系统状态的变化视为在能量景观中的轨迹。正如细胞内蛋白质（如乳糖酶）在热波动的帮助下折叠成最稳定的低能状态，热力学电脑能透过达到平衡状态或在非平衡的驱动下，利用这些波动寻找最佳解。

目前已有几家新创公司（如 Normal Computing 和 Extropic）在矽基电路上模拟并实作这项技术。他们展示了利用杂讯驱动的网路来执行矩阵反转或生成式人工智慧演算法，其能源效率远胜过传统的数位神经网路，这也为未来的低耗能计算及理解复杂生物系统提供了新方向。

Thermodynamic computing is an emerging field that aims to harness thermal fluctuations rather than fight against them. While traditional computer engineers work hard to eliminate thermal noise to ensure computational accuracy, thermodynamic computing seeks to turn these random energy fluctuations into a computational resource, potentially revolutionizing our understanding of computation and drastically reducing energy consumption and heat dissipation.

The concept is inspired by thermodynamic processes in nature, viewing the evolution of a system's state as a trajectory through an energy landscape. Just as proteins within cells, such as lactase, fold into their most stable low-energy states with the help of thermal fluctuations, thermodynamic computers can use these fluctuations to find optimal solutions by either reaching equilibrium or being driven out of equilibrium.

Several startups, including Normal Computing and Extropic, are already simulating and implementing this technology on silicon-based circuits. They have demonstrated that noise-driven networks can perform tasks like matrix inversion or generative AI algorithms with significantly higher energy efficiency than traditional digital neural networks, paving the way for low-power computing and a better understanding of complex biological systems.