Characterization Vehicle
A characterization vehicle (CV) is a test chip built purely to measure a process, not to ship as a product, packed with simple structures — ring oscillators, transistor arrays, via chains, SRAM bitcells — that extract the electrical numbers a design team needs before anyone commits a real product design to that process.
A CV solves the problem that a process design kit can't be built from a datasheet or a simulation alone — the numbers a design team relies on have to come from silicon that's actually been measured. Foundries publish transistor models, parasitic extraction rules, and timing libraries for every process they offer, and every one of those numbers traces back to real measurements taken on simple, purpose-built test structures rather than guessed from theory — a characterization vehicle is where that measurement actually happens, run early enough in a process's life that the resulting models are ready before the first real product tapes out on that node.
The structures are deliberately simple, which is the opposite instinct of normal chip design, and that's exactly the point. A real product chip tangles thousands of different effects together in one piece of logic, which makes it nearly impossible to isolate what any single transistor or interconnect structure is actually doing electrically — a ring oscillator built from identical inverters measures gate delay and nothing else, a via chain measures only interconnect resistance and defect rate, and an SRAM bitcell array measures only the cell's minimum operating voltage and stability margin, because keeping each structure deliberately narrow is what makes the resulting measurement trustworthy.
| Test structure | What it measures | Feeds into |
|---|---|---|
| Transistor arrays (W/L sweep) | Threshold voltage, drive current, leakage | SPICE transistor models |
| Ring oscillators | Gate delay, frequency vs. voltage and temperature | Timing libraries, speed binning |
| SRAM bitcell array | Minimum operating voltage, read/write margin | Memory compiler, Vmin specs |
| Via and interconnect chains | Resistance, open/short defect rate | Parasitic extraction rules |
| Process monitors | Line width, film thickness, overlay | Design rule limits |
A CV runs early in a process's life specifically so the resulting models are ready before anyone needs them for a real design. Foundries typically push a characterization vehicle through a new or significantly updated process well ahead of the first customer product tapeout, because building an accurate process design kit — the transistor models, timing libraries, and design rules a design team actually works against — takes real measured silicon data, and that data has to exist before a customer can trust their simulation results on that node. A process that changes meaningfully partway through its life, through a yield improvement or a device tweak, often gets a follow-up CV run to re-characterize and update the PDK rather than leaving design teams working against stale numbers.
Read the characterization vehicle through a measurement-first lens: a process design kit isn't a theoretical prediction of how a process will behave, it's a direct report of how that process actually behaved on real silicon, and a CV is the deliberately simple, deliberately isolated test chip that makes that report possible before a single real product bets its design on the answer.