IAE Academy

FOR CORE AND IT JOBS · CHAPTER 5 OF 7

Fabrication and layout

This chapter follows a CMOS chip from bare wafer to wired transistors, shows how a layout turns into W and L, and explains the wire resistance, wire capacitance and latch-up that the layout decides. An engineer needs it to see why a circuit on paper behaves differently once it is drawn in silicon.

Updated 30 Oct 2026

What you will be able to do

You will be able to list the steps that build a CMOS chip and say what each one makes. You will read a layout as layers and find the transistors in it. You will estimate the resistance, capacitance and delay of a wire from its shape. You will explain why latch-up happens and why contacts to the well and substrate stop it.

Before you start

  1. Chapter 1, The MOSFET as a switch and the CMOS inverter

  2. Resistance of a bar,

  3. Parallel-plate capacitor,

  4. Charging a capacitor through a resistor (exponential rise)

  5. Natural logarithms,

  6. A forward-biased silicon pn junction starts to conduct near 0.6 V

  7. A bipolar transistor has an emitter, base and collector; its collector current is times its base current

Building the chip: the CMOS process

A chip is built in layers, each patterned on top of the last. The table lists each step once, but photolithography and etching repeat for every layer, and implantation comes at two stages: the n-well before the gate, then the n+ and p+ source and drain regions after it. That order matters: the gate blocks the source and drain implants from the channel, so source and drain line up with its edges.

Step

What happens

What it makes

Wafer

Start from a thin disc of p-type silicon

The substrate shared by every transistor

Oxidation

Heat the wafer in oxygen to grow silicon dioxide

Thin gate oxide and thick isolation oxide

Photolithography

Coat with resist (a light-sensitive film), shine light through a mask (a plate carrying one layer's pattern), develop

A stencil showing where the next step acts

Etching

Remove material the resist does not cover

Patterned oxide, polysilicon or metal

Ion implantation and diffusion

Fire dopant ions (impurity atoms such as boron or phosphorus) in, then heat to spread them

n-wells first; after the gate, the n+ and p+ (heavily doped) source and drain regions

Polysilicon gate

Deposit polysilicon (doped silicon made of many small crystals, a conductor) over the thin oxide and etch it

Gates and short local wires

Contacts and metal layers

Etch holes in the oxide, fill them with metal, deposit metal and pattern it into wires

Connections from silicon and polysilicon up to metal wires

Chemical-mechanical polishing

Polish the surface with a slurry (a liquid carrying fine abrasive) and a rotating pad

A flat surface for the next metal layer

The nMOS transistors sit directly in the p-type substrate. A pMOS needs an n-type body, so an n-well is implanted into the substrate and the pMOS is built inside it. Thick oxide between neighbouring devices isolates them. The substrate is tied to ground and the n-well to the supply , so the junction between them is reverse biased and only a tiny leakage current crosses it.

WATCH OUT FOR

The n-well is the body of the pMOS, not an nMOS. Test: the body is the material under the channel, n-type for a pMOS and p-type for an nMOS.

Reading a layout

A layout is the set of mask shapes drawn on a few layers: diffusion (the source and drain material), polysilicon, contact (a square hole in the oxide) and metal. Layers on different levels do not connect unless a contact joins them. The rule that matters most is that a transistor exists wherever polysilicon crosses diffusion. The overlap is the channel, the polysilicon is the gate, and the diffusion on either side is source and drain.

ASPECT RATIO READ FROM A LAYOUT

Valid for a straight polysilicon strip crossing a straight diffusion strip, using drawn sizes. is the polysilicon width where it crosses the diffusion, which is the channel length . is the diffusion width under the gate, which is the channel width . Both in the same unit.

Sizes are written in multiples of , half the minimum drawn gate length, so in this scheme a design can, in principle, move to a finer process by changing alone. The table gives typical values; rule sets differ in the details. Minimum widths keep a shape from breaking during etching. Minimum spacings keep two shapes from merging into a short.

Rule

Typical size

Polysilicon width (the gate length)

2λ

Polysilicon to polysilicon spacing

2λ

Diffusion width

3λ

Diffusion to diffusion spacing

3λ

Metal 1 width and spacing

3λ

Contact hole

2λ by 2λ

Polysilicon extension past diffusion

2λ

An inverter in words, with at the top and ground at the bottom. A p+ diffusion strip (the pMOS) runs left to right inside an n-well in the upper half. An n+ strip (the nMOS) runs left to right in the lower half. One vertical polysilicon strip crosses both near their middles: it is the shared input and both gates. Left of the polysilicon, contacts join the pMOS source to a metal rail along the top and the nMOS source to a ground rail along the bottom. Right of it, one vertical metal strip contacts both drains and is the output. The rail also contacts the n-well, and the ground rail the substrate, each through a tap (a small heavily doped region).

WORKED EXAMPLE

In an inverter layout . The polysilicon is wide where it crosses the nMOS diffusion, and that diffusion is wide. The process transconductances () satisfy , and the nMOS threshold equals the pMOS threshold magnitude, . What pMOS diffusion width under the same polysilicon puts the switching threshold (the input at which the output equals the input) at ?

Solution
  1. The channel length is the polysilicon width, because the gate spans the channel: .

  2. The channel width is the diffusion width under the gate: , which meets the minimum. So .

  3. From Chapter 1, with , when , because the two saturated currents must match there. So .

  4. The pMOS uses the same , so , which is .

Answer: (), twice the nMOS width.

Wires and latch-up

Every wire is a thin conductor lying over an insulator, so it has resistance along its length and capacitance to what lies under and beside it. Neither is in the schematic. A long wire is a resistor driving a capacitor, and it slows the gate that drives it.

WIRE RESISTANCE IN SQUARES

Valid for a uniform strip with current along its length. is resistivity (), length, width, thickness (m). is the sheet resistance in ohms per square. counts how many squares the strip holds.

WIRE CAPACITANCE TO THE LAYER BELOW

Parallel-plate estimate for a wire over a conducting layer, ignoring fringing fields at the edges, so real wires have more. for silicon dioxide; , and the oxide thickness in m give in F.

DELAY OF A WIRE AS AN RC

Valid for a step of height from an ideal driver into one lumped and one lumped starting at 0 V. is the time to reach half of . A real wire spreads its and along its length and is faster, so this is an upper estimate.

DELAY GROWS WITH THE SQUARE OF LENGTH

Valid for a uniform wire of length (µm), lumped as above, so an upper estimate. (, in µm) and (fF/) are the resistance and capacitance per unit length, so and . fF gives fs.

WORKED EXAMPLE

A metal wire is 1 mm long and 0.2 µm wide, with and fF/µm. Find its resistance, capacitance and .

Solution
  1. The length is 1000 µm, so the strip holds squares, which is what the squares formula needs.

  2. .

  3. The capacitance adds up along the length: fF.

  4. The delay formula needs fs ps, so ps. The length form agrees: and fs.

Answer: , fF, ps (an upper estimate).

Latch-up is a failure the layout itself creates. The n-well, substrate and diffusions form two parasitic bipolar transistors (parasitic: made by accident from regions built for other jobs). The npn has the nMOS n+ source as emitter, the p-substrate as base and the n-well as collector. The pnp has the pMOS p+ source as emitter, the n-well as base and the p-substrate as collector. So each collector is the other's base. A stray current into the npn base gives a collector current ; that is the pnp base current, so the pnp returns to the npn base.

LATCH-UP LOOP CONDITION

Valid once a stray current has turned both parasitic junctions on, ignoring the well and substrate resistance that diverts part of it. Each trip round the loop then returns more current than it started with, so the current grows until the pair joins to ground. , : the parasitic transistors' current gains.

CURRENT NEEDED TO TRIGGER THE LOOP

(Ω) is the resistance from where the stray current enters to the nearest contact tying that region to its rail (substrate to ground, n-well to ). Its drop biases a parasitic base-emitter junction, which turns on near 0.6 V. Contacts close to the transistors keep low, so only a large current can trigger the loop.

Check yourself

CHECK YOURSELF 1

In a layout, where does a transistor exist?

  1. (A)

    Where polysilicon crosses diffusion

  2. (B)

    Where metal 1 crosses polysilicon

  3. (C)

    Wherever a contact hole is drawn

  4. (D)

    Wherever two diffusion strips touch

Answer and solution

Answer: (A)

  1. A transistor needs a gate over a channel between a source and a drain.

  2. Polysilicon crossing diffusion makes that: the overlap is the channel, and the diffusion on either side is source and drain.

  3. Metal and contacts only connect layers, so they make no channel.

CHECK YOURSELF 2One or more options are correct

Which of the following reduce the chance of latch-up?

  1. (A)

    A contact from the n-well to close to each pMOS

  2. (B)

    A contact from the substrate to ground close to each nMOS

  3. (C)

    Raising the gain product above 1

  4. (D)

    Shortening the path from the transistor to its well or substrate contact

Answer and solution

Answer: (A), (B), (D)

  1. A nearby contact to the well or the substrate lowers , so rises.

  2. A shorter path to the contact lowers the resistance, which does the same.

  3. A gain product above 1 is what lets the loop sustain itself, so raising it makes latch-up more likely.

CHECK YOURSELF 3

A polysilicon wire is 50 µm long and 0.25 µm wide, with sheet resistance 5 . Find its resistance in ohms.

Answer and solution

Answer: (accepted range )

  1. Squares: .

  2. .

CHECK YOURSELF 4

A wire 100 µm long and 1 µm wide runs over the substrate, separated from it by oxide 0.5 µm thick, with F/m. Ignoring fringing, use the parallel-plate estimate to find its capacitance in fF.

Answer and solution

Answer: (accepted range )

  1. Area: .

  2. .

  3. F fF.