现代社会中信息存储和处理的重要性日益增强，如何提高信息处理速度并降低能耗是当前面对的关键问题，而信息擦除过程在信息处理中占据重要地位并内在的影响其能量消耗。目前针对信息擦除热力学的讨论集中关注于单个存储单元中擦除方案的优化。然而，一方面受限于非平衡擦除过程中速度与效率的trade-off关系，内在限制了优化潜力；另一方面未涉及实际存储介质（如磁介质）中存储单元粗粒化编码的情况。我们分析发现：采用简单多数定律编码的存储单元可以在加快信息擦除速度的同时提升热效率，为非平衡擦除过程进一步优化提供了新途径。在本项目中，我们将系统的研究不同粗粒化编码方式下存储单元中信息擦除的热力学，尝试通过粗粒化编码进一步提高擦除过程的性能和热效率，并将结果应用于具体信息功能模型和过程，探索粗粒化编码在复杂信息处理过程中的应用，为存储介质的设计、微观热机和分子马达的研究以及机器学习过程的认识和改进等提供新思路。

As information storage and processing are of increasing importance for modern society, it is the crux that how to improve the speed and reduce the energy consumption of information processing simultaneously. According to Landauer’s theory, information erasure plays a crucial role in information processing and greatly affects the energy consumption. Therefore, it is essential to investigate the thermodynamics of non-equilibrium information erasure in memory units. Up to now, the discussions on the thermodynamics of information erasure mainly focus on the optimized erasure protocol of a single memory unit, remaining two topics barely touched: Firstly, as the non-equilibrium information erasure process is restricted by the trade-off relation between erasure speed and thermodynamic efficiency, the further optimization of erasure performance is quite limited; Secondly, the thermodynamics of memory units under coarse-grained decoding, which is common in actual storage mediums such as the magnetic storage medium, is seldom discussed. We explored that, by employing memory units under simple majority-rule decoding, the information erasure process could be accelerated while the thermodynamic efficiency is improved. This finding provides new methods for the further optimization of non-equilibrium information erasure. In the present project, we focus on the thermodynamics of non-equilibrium information erasure under different manners of coarse-grained decoding. We try to further improve the erasure performance and thermodynamic efficiency of memory units by employing coarse-grained decoding. The results will be applied to functional information processing models and processes such as bipartite model, feedback control and neural network learning to explore the applications of coarse-grained decoding in the thermodynamics of complexed information processing. We believe that these results will benefit the practical applications such as the design of storage medium, researches of micro engines and molecular motors as well as the improvement of machine learning.