EC-25.2 Characterising a Device: What Is Measured, and With What

What this is and why it exists

A model is only as good as the measurements it was fitted to.

Device measurement is its own craft, and it is harder than it looks. The currents of interest run from picoamperes to amperes. The device is a few micrometres across. The probe, the cable and the connector are part of the circuit.

This topic is what is measured, on what structure and with what instrument. It is also how a parameter is extracted from a curve rather than read off it.

The vocabulary

• Characterisation — measuring a device to determine its behaviour and its model.
• Threshold voltage — the gate voltage at which a transistor begins to conduct.
• Transconductance — how much the output current changes per volt of input.
• Leakage — current flowing when the device is meant to be off.
• Extraction — computing a model parameter by fitting a measured curve.
• Parameter analyser — an instrument for measuring current over many decades.
• Guarding — a driven shield that removes cable leakage from a measurement.
• Test structure — a device placed on the wafer purely to be measured.
• Scribe line — the space between dies where those structures live.

The mental model

A small set of measured curves determines most of a device model. Current against gate voltage. Current against drain voltage. Capacitance against bias. Leakage against temperature. Each has a standard extraction method, and each method rests on assumptions.

That is the point worth internalising. A threshold voltage is defined by a procedure, not by nature. Three common extraction methods give three different answers on the same device, all of them defensible. Quoting a parameter without naming its extraction method is close to meaningless, and comparing two laboratories' figures without checking the method is worse.

The instruments matter more here than almost anywhere else. Measuring picoamperes requires a guard: a driven conductor around the signal path, held at the same potential. No voltage then appears across the cable's insulation, so no leakage flows. That is the practical difference between a benchtop meter and a parameter analyser, and it is why one costs a hundred times the other.

Test structures live in the scribe line between dies. Dedicated transistors, resistors, contact chains and capacitors are measured on every wafer to monitor the process. They are the earliest signal that something has drifted, appearing long before any finished part is tested.

Probing has its own difficulties. The probe must contact well without destroying the pad, and its contact resistance appears in series with everything measured. Repeating a measurement on the same site is the standard way of proving that you measured the device rather than the setup.

Finally, temperature. Almost every parameter of interest changes with it, and different mechanisms often move in opposite directions. A characterisation at one temperature cannot support a specification over a range, however carefully it was done.

What you should now be able to explain or do

• List the measurements that determine a transistor model.
• Explain why a threshold voltage depends on the extraction method used.
• Say what guarding does and why picoampere measurement needs it.
• Describe what test structures are for and where they live.
• Prove that a measurement reflects the device rather than the contact.
• Say why characterisation must span the temperature range.

Check yourself