Analog IC · Lesson 9

Matching and layout

Arrange unit cells; common centroid, dummies and Pelgrom mismatch.

📘 Chapter introduction(click to show)

In this chapter you will learn to

  • ✓Distinguish random mismatch from systematic mismatch.
  • ✓Use Pelgrom's law to size transistors for a mismatch target.
  • ✓Apply common-centroid placement and dummy cells to cancel systematic errors.

1. Why this circuit matters

Current mirrors, differential pairs and bandgaps work only if two transistors are identical. On silicon they never are. Atomic-scale randomness and slow variations across the wafer make neighbouring transistors differ by millivolts. Good layout is how analog designers fight back, and it is often the difference between a chip that meets its specification and one that fails.

💡 Think of it like this

Two runners race on a track that slopes slightly. If one of them always runs the uphill half, they lose, however evenly matched the runners are. Give each of them equal parts uphill and downhill (common centroid), and the slope no longer decides the race.

2. Definition

Matching is how closely nominally identical devices behave alike. Mismatch has a random part, set by device area (Pelgrom's law), and a systematic part, caused by gradients and edge effects. Layout techniques such as common-centroid placement and dummy devices cancel the systematic part.

3. Meet the components

Device A and device B

The two transistors that must match, for example the two sides of a current mirror.

Unit cells in a common-centroid pattern

Each device is split into identical fingers, arranged (e.g. ABBA) so both devices share the same centre.

Dummy cells

Unconnected cells at the ends of the array, so every active cell has the same neighbours.

4. How it works, step by step

  1. 1

    Split into unit cells. Build each transistor from identical fingers of the same W and L.

  2. 2

    Recognise the gradient. Parameters such as V_th change slowly across the die. Cells far apart see slightly different V_th.

  3. 3

    Side by side is unfair. If A sits on one side and B on the other, the gradient shifts them by different amounts: a systematic error.

  4. 4

    Share the centre. Place the cells so A and B have the same centre of gravity (common centroid, such as ABBA). A linear gradient then raises A exactly as much as B.

  5. 5

    Even out the ends. Cells at the end of an array see different etching and stress. Dummy cells at the ends make every active cell see the same surroundings.

  6. 6

    What layout cannot fix. The random part remains: σ(ΔV_th) = A_VT/√(W·L). Only more gate area reduces it: four times the area halves it.

🔑 The one thing to remember

Layout cancels systematic mismatch; only gate area reduces random mismatch.

Layout: click a cell to swap it between device A and device B

A+0.25A-0.55A-0.35A-0.15B+0.05B+0.25B+0.45B+1.65A+0.35A-0.45A-0.25A-0.05B+0.15B+0.35B+0.55B+1.75centroid Acentroid BCell colour = local V_th shift (blue low, red high). Numbers in mV.

Process and device

Key formulas

  • σ(ΔV_th) = A_VT / √(W·L)

    Pelgrom's law for random mismatch

  • ΔV_th,sys = mean V_th(A) − mean V_th(B)

    Zero for any linear gradient when the centroids coincide

  • ΔI / I ≈ 2·ΔV_th / V_OV

    Current error of a mirror in saturation

Mismatch budget

Cells A : B8 : 8
End cells A / B2 / 2
Centroid A (x, y)-20.0, 0.0 µm
Centroid B (x, y)20.0, 0.0 µm
Systematic ΔV_th-0.800 mV
Random σ(ΔV_th)0.884 mV
Systematic ΔI/I0.800 %
Random ΔI/I (3σ)2.652 %

Layout can drive the systematic error to zero. The random error only falls with more gate area: four times the area halves it.

Things to try

  • Start with Side by side, then Interdigitated, then Common centroid (2-D): watch the systematic error drop to zero.
  • Load Common centroid (1-D) with dummies off: the gradient cancels but the end cells cause an error. Tick dummies to fix it.
  • Click individual cells to build your own pattern and line the two centroids up.
  • Double the unit-cell W = L (4× area) and check that the random σ halves.
  • Raise V_OV from 100 mV to 300 mV: the same ΔV_th now causes a third of the current error.