Shmoo Plot

A shmoo plot is a two-dimensional pass/fail map produced during chip testing — typically voltage on one axis and frequency (or timing margin) on the other — that shows exactly which combinations of operating conditions a specific die actually works under, rather than a single pass/fail number. It is the test-engineering equivalent of the [process window](process-window): the same concept of a bounded region of acceptable operation, but measured electrically on a finished, packaged part at [wafer probe](wafer-probe) or final test, instead of optically during lithography.

$$\text{Pass}(V, f) = \begin{cases} 1 & \text{die operates correctly at voltage } V \text{ and frequency } f \\ 0 & \text{die fails (timing violation, functional error, or excess leakage)} \end{cases}$$

Sweeping $V$ and $f$ across a grid and plotting the result turns a yes/no spec check into a shape — and the shape itself is the real engineering signal, carrying information no single-point test could ever reveal.

Shmoo Plot — Voltage vs. Frequency Pass/Fail Map each cell is one measured operating point on one actual die — the shape is the test result supply voltage → clock frequency → PASS region nominal spec point guardband margin too slow to switch too fast — setup time violated at any voltage a wide, well-centered pass region on real silicon is the electrical proof that the design's PPA margin actually exists

Why this single plot carries so much more information than a go/no-go production test:

  • The width of the pass region at the qualified operating voltage is the real timing guardband. A design verified in simulation to meet timing at a given voltage and frequency still has to prove it on actual silicon — process variation, on-die temperature, and supply noise all shift the real boundary, and the shmoo plot is literally a picture of how much margin exists between the qualified spec point and the edge of failure, tying directly back to the [PPA](ppa) tradeoff that set that spec point in the first place.
  • Its shape diagnoses the failure mechanism, not just its presence. A shmoo that fails only at low voltage and high frequency points to a setup-time (speed) limitation; one that fails at high voltage too points to a different mechanism (often oxide reliability or leakage-driven thermal failure) — a single pass/fail test at one corner would never reveal this, but the full 2D map does, which is why failure analysis teams pull a shmoo plot before anything else when a part fails in the field.
  • It is how speed binning actually gets decided. Rather than guessing which parts can be sold as a higher-frequency SKU, a manufacturer runs the shmoo sweep on sampled production silicon and sets bin boundaries directly from where the real pass regions cluster — the shmoo plot is the underlying data, the bin spec sheet is just a summary of it.
  • It is run across the full temperature and process corner range before any single unit ships. A part isn't qualified from one shmoo at room temperature — it's qualified from a family of shmoo plots across hot/cold and slow/fast/typical silicon corners, and the design has to keep a nonzero pass region in the worst of all of them simultaneously, which is the electrical analogue of a [process window](process-window) that has to hold across every process corner a fab will ever ship.
$$\text{Guardband} = V_{\text{pass edge}} - V_{\text{qualified spec}}$$

is the number that ultimately gets reported up to a product team: a comfortable, positive guardband measured directly from the shmoo plot is what lets an engineering team sign off on a part with confidence that ordinary unit-to-unit variation in [first-pass silicon](first-pass-silicon) won't push any shipped die outside its advertised operating range.

Take shmoo plot further

Ask the copilot about this term, or have our engineers assess it against your process.

Ask CFSGPT about this → Request a technical assessment → Open the CFS app