autoMBIST 0.4.0 adds repair-yield analysis to its open memory built-in self-test (MBIST) flow. A designer can inject cell, row, column, and clustered defects into a memory model, run the same repair logic intended for the chip, and compare the result with an exact solver. That turns the number of spare rows and columns from a rule-of-thumb choice into something a team can test before layout is finished.
The useful detail is that autoMBIST measures two different things. yield-sim asks what an ideal repair algorithm could fix, while yield-sweep runs the generated hardware logic against faults the test controller actually observed. If those curves separate, the team knows whether it needs more spare memory or better diagnosis logic. The project includes 43 functional fault types and more than 3,000 automated tests, so this is a research instrument with a visible test base rather than a spreadsheet around a toy model.
The numbers still depend on the defect distribution a team supplies, and a modeled distribution isn't foundry yield data. Even so, an open run that records its assumptions is much easier to review than a repair budget hidden inside a proprietary flow. Memory designers should run the sweep before physical design and carry the chosen defect model into the design review, where manufacturing and design teams can challenge the same evidence.