Efuse
An eFuse is a microscopic on-chip fuse that a chip can permanently and electrically blow exactly once, turning a reversible design decision into a one-way switch burned directly into the silicon.
An eFuse solves the problem that some decisions about a chip need to be locked in after manufacturing, not just set once at design time. A chip's mask set is fixed for every unit that comes off that design, but individual dies still need their own identity, their own feature set, or their own repair map written in after they're built — an eFuse gives each die a small block of bits that firmware or test equipment can set exactly once, permanently, without needing a battery or any kind of rewritable memory to hold the value.
That one-way behavior is exactly what makes eFuses useful for the handful of jobs on a chip that must never be undone. Chipmakers use fuse arrays to lock in things like a unique per-die serial number or cryptographic key, to permanently disable a defective core or cache way discovered during test, to set the SKU and clock speed a die was binned into so it can't be unlocked into a higher-end part, and to store the redundancy-repair map that reroutes addressing around bad memory cells — in every one of these cases, the whole point is that the setting has to survive power loss and resist tampering, which a rewritable register can't guarantee the way an irreversibly broken physical link can.
| Use case | What gets fused | Why it has to be one-way |
|---|---|---|
| Die identity | Unique serial number, crypto keys | Can't be cloned or rewritten after the fact |
| SKU / speed binning | Which product tier and clock speed a die shipped as | Prevents unlocking a cheaper part into a premium one |
| Core / cache disable | Which defective blocks were found bad at test | A disabled block must stay disabled for the part's life |
| Redundancy repair | Which spare row or column replaces a bad one | The repair map can't drift or reset |
| Debug lockout | Whether JTAG/debug access is still open | Production parts must not be re-openable for security |
Programming an eFuse takes a real, deliberate jolt of current, which is exactly why it doesn't happen by accident. A typical eFuse link needs on the order of several milliamps pushed through it for microseconds to tens of microseconds to heat the link past its melting or agglomeration point, a programming event that firmware or test hardware triggers on purpose through a dedicated fuse-box circuit — ordinary chip operation never gets close to that current density on a fuse link, which is part of why eFuses hold their state reliably for the life of the part instead of drifting or flipping under normal use.
eFuses only became practical once they could be built from the same metal or polysilicon layers already in a standard logic process, which is why they spread so quickly through the industry. Earlier one-time-programmable elements often needed a special laser-trim step or extra mask layers, adding cost and a separate test insertion just for fusing — an eFuse instead reuses interconnect material that's already part of a standard CMOS flow, so adding a fuse array costs little beyond the fuse-box driver circuit, which is why eFuses now show up by the thousands on everything from AMD and Intel CPUs to Apple SoCs, used for everything from feature locking to the redundancy repair covered separately.
Read the eFuse through a one-way-door lens: everything else on a chip can be reset, reconfigured, or rewritten, but the handful of facts that have to stay fixed for the life of the part — identity, bin, disabled blocks, the repair map — get burned into fuse links that a deliberate programming current breaks permanently, trading a tiny bit of die area for a guarantee that nothing afterward, intentional or accidental, can undo.