Hardware entropy is a coupled system
好的,我现在需要帮用户总结一篇英文文章的内容,控制在100字以内。首先,我得仔细阅读用户提供的文章内容,理解其主要观点和关键信息。 文章开头提到作者记录了Mac Mini M4上的58个硬件熵源的原始字节,使用了他开发的开源框架OpenEntropy。这些熵源包括时钟抖动、DRAM定时、PLL振荡器、中断调度和USB QRNG等。接下来,作者使用text-embedding-3-large对这些字节窗口进行了嵌入,并进行了比较。 然后,作者发现有19个来源来自10个不同的硬件类别,落在同一个聚类中。在去除主要共享因素后,这些来源仍然保持聚类。在压力测试下,像dram_row_buffer和page_fault_timing这样的来源会一起移动,而USB量子RNG几乎不动。 作者还提到他没有测试SHA-256/ChaCha20条件是否去除相关性(可能去除),并且这只是一台机器的结果。此外,像NIST SP 800-22这样的单流测试不会发现这种跨源的相关性。 现在我需要将这些信息浓缩到100字以内。首先确定关键点:记录了58个硬件熵源的数据,使用OpenEntropy框架;通过嵌入分析发现19个来源聚类;压力测试显示某些源相关性;单流测试无法检测跨源相关性。 接下来,组织语言确保简洁明了。可能的结构是:记录数据、分析方法、发现结果、结论或限制。 最后检查字数是否符合要求,并确保没有使用“文章内容总结”等开头用语。 </think> 研究者从Mac Mini M4的58个硬件熵源记录原始字节数据,并通过自建开源框架OpenEntropy进行分析。结果显示19个来源聚类且具相关性,在压力测试下部分源行为一致而USB量子RNG稳定。单流测试无法检测跨源相关性。 2026-3-18 03:55:32 Author: www.reddit.com(查看原文) 阅读量:7 收藏

I recorded raw bytes from 58 hardware entropy sources on a Mac Mini M4 (clock jitter, DRAM timing, PLL oscillators, interrupt scheduling, a USB QRNG, etc.) using an open-source framework I built called OpenEntropy. Then I embedded the byte windows with text-embedding-3-large and compared them.

19 sources from 10 different hardware categories landed in the same cluster. After removing the dominant shared factor, they stayed clustered. Under stress tests, sources like dram_row_buffer and page_fault_timing moved together while the USB quantum RNG barely moved at all.

I haven't tested whether SHA-256/ChaCha20 conditioning removes the correlations (it probably does), and this is one machine. But per-stream tests like NIST SP 800-22 wouldn't catch this since they don't look across sources.


文章来源: https://www.reddit.com/r/netsec/comments/1rwtf1l/hardware_entropy_is_a_coupled_system/
如有侵权请联系:admin#unsafe.sh