Redundancy Repair
Redundancy repair is the practice of building spare rows, columns, or circuit blocks into a memory chip so that defective cells found during test can be electrically swapped out for working spares, turning a die that would otherwise fail into one that ships.
Redundancy repair solves the problem that a single stuck bit anywhere in a multi-billion-transistor memory array would otherwise scrap the entire die. A modern DRAM or NAND die packs billions of identical memory cells onto one piece of silicon, and even a mature process still produces a handful of defective cells per die purely from random particle and pattern-dependent defects — without a way to route around those few bad cells, every one of those dies would be thrown away even though the overwhelming majority of the array works perfectly.
That repair capacity is deliberately built into the die before the first wafer is ever tested, because there's no way to add spares after the fact. Memory designers at DRAM and NAND makers like Samsung, SK hynix, and Micron reserve a small percentage of extra rows and columns, typically in the low single digits of the array, laid out identically to the main array so any one of them can stand in for a failing row or column once a defect map comes back from wafer sort — the repair itself happens through a laser that permanently blows a fuse to redirect addressing, or increasingly through an eFuse or antifuse circuit that does the same rewiring electrically without a separate laser step. Automated test equipment from vendors like Advantest and Teradyne runs the defect scan and feeds it into redundancy-analysis software that calculates the repair solution before any fuse gets blown.
| Repair mechanism | How it's triggered | Common use |
|---|---|---|
| Laser fuse | Focused laser beam blows a physical fuse link | Legacy DRAM, high-volume commodity memory |
| eFuse | Electrical current blows an on-chip fuse | Modern DRAM, SoC redundancy and configuration |
| Antifuse | Electrical pulse creates a permanent short | High-reliability and automotive-grade parts |
| Built-in self-repair (BISR) | On-die logic runs repair automatically at power-on test | SRAM caches inside logic SoCs |
Redundancy repair is what turns raw defect density into the yield number a fab actually reports, which is why it shows up directly in cost-of-ownership math. A die with built-in redundancy can absorb several defective cells and still ship as a fully functional part, so the real yield of a memory product reflects not just how clean the process is but how much spare capacity was designed in and how well the repair algorithm uses it — two dies hit by the same defect density can have very different outcomes depending entirely on whether enough redundancy was available to repair around the damage.
The tradeoff is area: every spare row and column sits on the die doing nothing until it's needed, so more redundancy means a larger, more expensive die for the same usable capacity. Memory designers have to size the redundancy budget against the process's known defect density before the design is even taped out, aiming for enough spares to recover the yield losses they expect without wasting so much area that the extra die cost outweighs the yield gained — getting that balance right is one of the core yield-engineering decisions behind every memory product.
Read redundancy repair through a yield-recovery lens rather than a reliability lens: it isn't primarily there to catch chips that fail in the field, it's there to decouple a memory die's final yield from its raw defect count by rewiring around whatever wafer sort finds — spare rows and columns reserved up front, a defect map generated at test, and a permanent fuse-based repair in between, converting dies that would otherwise be scrap into shippable parts at the cost of some die area that, on a good day, never gets used at all.