A hardware erratum is a documented case where a chip, most often a processor, behaves differently from what its published specification says it should. Unlike a software bug, the defect is etched into physical silicon that has already shipped to customers, so it cannot simply be patched and redistributed. Errata range from harmless quirks to serious flaws like the Pentium FDIV division error or the F00F lockup, and managing them has become a routine, formalized part of how the chip industry operates.
CPU vendors publish these defects in dedicated documents. Intel issues a Specification Update for each processor family and individual erratum overviews such as its invalid-instruction erratum; AMD publishes a Revision Guide for each processor family. These documents enumerate each known issue, describe the conditions that trigger it, state which steppings (manufacturing revisions) are affected, and indicate whether and how it has been fixed. They exist precisely because operating-system writers, BIOS and firmware authors, and compiler teams need a reliable, vendor-maintained catalog of what the hardware actually does, not just what it was designed to do.
Because the silicon cannot be recalled cheaply in most cases, the workaround usually lives in software or in updatable microcode. Modern processors decode many instructions into lower-level micro-operations using a microcode store that can be patched at boot time, so some errata are fixed by loading a microcode update from the BIOS or operating system. Other errata are contained at a higher level: the F00F bug, for example, was mitigated by rearranging the operating system’s interrupt tables so a benign fault fired before the processor could wedge itself. When a defect is severe and unfixable, the only remedy is replacing the chips, as Intel ultimately did for the FDIV flaw.
Errata also carry a security dimension. A behavior that looks like an obscure corner case can become an attack surface once someone realizes an unprivileged program can trigger it, as the F00F lockup and later speculative-execution flaws showed. Tracking and disclosing errata is therefore not only a matter of correctness but of coordinated security response, and a stepping’s errata list is one of the first things a careful systems engineer consults.
The errata sheet is, in a sense, an admission built into the engineering culture of hardware: complex chips ship with known bugs, those bugs are catalogued openly, and the layers above the silicon, microcode, firmware, and software, are expected to absorb the difference between the specification and the reality.