Soft Errors in Memory and Logic

Memories and logic are the most upset-prone parts of a system; neutron testing quantifies soft-error rates and validates error correction.

Where the bits flip

Dense memory arrays and fast logic hold enormous numbers of small storage nodes, each with a modest critical charge. That makes them the parts most likely to register a single-event upset. A soft error does not damage the hardware, but a corrupted bit in the wrong place can crash a function, so the rate must be known and managed.

Multi-bit upsets

As cells pack closer, one particle can flip several adjacent bits at once, a multi-bit upset. This challenges simple error-correcting codes, which are designed to fix single-bit errors, and drives interleaving and stronger codes. Neutron testing at 14 MeV reveals how often multi-bit events occur in a given part.

• SEU: single-bit upset, correctable by simple ECC
• MBU: multiple bits from one strike, harder to correct
• SEFI: the device stops functioning until reset

Validating protection

A campaign measures raw upset rates and then confirms that the system's error correction and scrubbing keep the effective error rate within budget. Kronos delivers this validation as part of its neutron service.

Scrubbing and system budgets

Memory protection is not only about detecting an error but about correcting it before a second error accumulates in the same word. Periodic scrubbing, reading, correcting, and rewriting memory, keeps single-bit errors from piling into uncorrectable multi-bit ones. A neutron campaign measures the raw upset and multi-bit rates that set how often scrubbing must run, turning a physics measurement directly into a system design parameter.

This is a public overview only. It contains no classified information, no operational detail, and no weapons-design content.