Chipkill

Chipkill is a memory protection technology developed by IBM that extends traditional ECC memory to protect against DRAM chip failures. By using a higher-density coding scheme that spreads data and parity across multiple DRAM chips within a memory rank, Chipkill enables the system to detect and correct errors that would be unrecoverable by standard ECC in the event of a single chip failure. The goal is to improve memory availability in servers and other mission-critical systems.

Implementation relies on a memory controller capable of mapping data across several chips and applying a more powerful error-correcting code. Instead of a single 64-bit data word protected by an 8-bit ECC, Chipkill distributes the same data across several chips with additional parity information, allowing reconstruction if one chip goes faulty. The exact layout and protection level vary by platform, but common designs allow correction of single-chip faults and detection of multi-bit errors within a rank.

Chipkill originated with IBM in the 1990s and became associated with IBM’s eServer and pSeries lines; the term is often used to describe similar high-availability memory protection features in other vendors’ systems. While powerful, Chipkill requires hardware support and compatible memory modules; it does not eliminate all memory failures but can reduce the chance of system downtime due to memory faults.