CMOS ICs contain a parasitic p-n-p-n structure (adjacent well/substrate diodes) that behaves like a thyristor. A heavy ion depositing enough charge along its track can forward-bias this structure into its low-impedance "on" state — single-event latchup (SEL) — which is a physical short, not a data error:
I_latch = (V_dd − V_hold) / R_latch
E(t) = ∫ I_latch · V_dd dt (Joule heating while latched)
destroyed if E(t) > E_damage before the circuit breaker trips
Once latched, current stays high (bounded only by R_latch, not by the parasitic gate) until power is removed — an unprotected die self-heats until it fails. A current-limiting protection circuit watches supply current and cuts power once it exceeds Itrip continuously for the response delay td, then reconnects power after a cool-down. This is exactly the race modelled here:
- Strike rate — Poisson-distributed ion hits across the array; higher Vdd also raises the trigger probability, matching real bias-dependence of the latchup window.
- Itrip — set it too high and small latch currents never trip protection at all; too low and normal transients would false-trigger (not modelled, but the trade-off is real).
- Trip delay td — a slower circuit lets more energy accumulate before cutoff; push it high enough and dies start cooking before the breaker reacts.
Real-world relevance: this is why spacecraft power rails use current-limiting FETs or fold-back regulators per board, not just fuses — a fuse is too slow, and this same physics is why radiation-hardened parts use epitaxial substrates and guard rings to raise the latchup trigger threshold in the first place.