Hardware flaws causing data corruption do exist, but most are silently rectified without public awareness, notes Alan Cox.
In response to your published letters asserting the reliability of computers themselves (Computers rarely go wrong, but computer systems often do, 17 January), the reality is not entirely optimistic. Although computer hardware is generally more reliable than most current software, it’s crucial to acknowledge the existence of data-corrupting flaws in hardware. Furthermore, the identification of such flaws tends to increase over time.
Instances like rowhammer gain notoriety as security vulnerabilities, while others, like the Pentium FDIV incident, become widely known through recalls. However, a majority of these flaws go unnoticed by the general public, quietly addressed or mitigated in microcode.
Research papers from Google (Cores That Don’t Count) and Facebook (Silent Data Corruptions at Scale), along with a review of published errata documents, indicate that assuming a computer’s reliability is a valid basis for a criminal conviction should be critically examined. Additional supporting evidence, such as logs, audit trails, and robust internal self-checking within the software, should be considered.
