[Intermediate] — The device opened up. How a gate voltage conjures a conducting layer out of silicon that was doped the wrong way, what threshold voltage means, and the three regions that define every use of a transistor.

Before you start

  • You know the MOSFET’s structure — insulated gate over a channel between two doped regions — module 11.
  • You can explain depletion regions and how applied voltage changes their width — module 10.
  • You know majority and minority carriers, and that doping suppresses minorities — module 9.
  • You remember the 60 mV/decade rule — module 10.

After this lesson you will be able to:

  1. Explain accumulation, depletion and inversion, and why inversion is the surprising one.
  2. Define threshold voltage and explain what physically happens at it.
  3. Identify the three operating regions from and , and explain pinch-off.
  4. Explain why the triode region gives exactly the switchable resistance module 3 demanded, and why is the designer’s main knob.

Study route: section 3 is the conceptual leap. Section 5 (pinch-off) is the part most people find genuinely confusing — the diagram is there to help.


1. Why this exists (real-world motivation)

Module 11 said the gate creates a field that “forms a channel”. That sentence was doing a great deal of unexamined work.

Look at what it is claiming. The body is P-type — its mobile carriers are holes, and module 9 established that doping suppresses the minority electrons to around /cm³. To conduct between two N-type regions you need a path made of electrons. There are essentially none available.

So where does the channel come from? Not from the gate — it is insulated and cannot inject anything. Not from the source and drain alone. The answer is genuinely surprising: the gate voltage converts a thin surface layer of P-type silicon into N-type, without changing a single dopant atom.

That process is called inversion, and it is the mechanism at the centre of this course.


2. Terminology

TermPlain-English definitionExample / analogy
MOS capacitorGate, oxide and semiconductor — a capacitor with silicon as one plate—
AccumulationGate voltage attracting majority carriers to the surfaceCrowding the usual residents in
DepletionGate voltage pushing majority carriers away, exposing fixed ionsClearing the room
InversionEnough gate voltage to attract minority carriers, flipping the surface typeNew residents move in
Threshold voltage ()Gate voltage at which strong inversion forms a usable channelThe tipping point
Overdrive ()How far past threshold you are—
Cutoff — no channelSwitch off
Triode / linearChannel present, small — behaves as a resistorSwitch on
SaturationChannel pinched off at the drain — current stops rising with —
Pinch-offThe channel vanishing at the drain end—
Channel width divided by length — the geometric design knob—
Subthreshold swingMillivolts of gate change per decade of current, below threshold—

3. Inversion — the conceptual leap

Consider just the gate, the oxide and the P-type body. Ignore source and drain for a moment. This stack is a MOS capacitor, and its behaviour as gate voltage rises comes in three stages.

Stage 1 — accumulation ( negative)

A negative gate attracts positive charges. The body’s majority carriers are holes, so holes pile up at the surface.

   GATE  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓  (negative)
   OXIDE ░░░░░░░░░░░░░░
   BODY  ○ ○ ○ ○ ○ ○ ○ ○  ← holes accumulate at the surface
     P   ○   ○   ○   ○     surface is MORE p-type than before

The surface is more strongly P-type than the bulk. No use for conduction between N-type regions.

Stage 2 — depletion ( slightly positive)

A positive gate repels holes, driving them down into the bulk. What is left behind is the fixed negative acceptor ions — immobile, exactly as in module 10.

   GATE  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓  (slightly positive)
   OXIDE ░░░░░░░░░░░░░░
   BODY  ⊖ ⊖ ⊖ ⊖ ⊖ ⊖ ⊖ ⊖  ← holes pushed away, fixed ions exposed
     P   ○   ○   ○   ○     a DEPLETION region, no mobile carriers

The surface now has no mobile carriers of either type. Still no channel — but notice the gate is building a depletion region purely by field, with no junction involved.

Stage 3 — inversion ( strongly positive) — the surprising part

Push the gate more positive and the field grows strong enough to attract electrons to the surface.

Where do they come from? Two sources: thermal generation in the depletion region, and — far more importantly in a real MOSFET — the N-type source and drain, which are packed with them.

   GATE  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓  (strongly positive)
   OXIDE ░░░░░░░░░░░░░░
   BODY  ● ● ● ● ● ● ● ●  ← ELECTRONS at the surface: an N-type layer
     P   ⊖ ⊖ ⊖ ⊖ ⊖ ⊖ ⊖    inside P-type silicon
         ○   ○   ○

The surface has been inverted. A layer of P-type silicon a few nanometres thick is now behaving as N-type, and it connects the N-type source to the N-type drain.

NOTE

No dopant atom moved. The acceptor ions are still there, still negative, still fixed. What changed is which mobile carriers occupy the surface — the field pulled in enough electrons to outnumber the holes locally.

This is the whole transistor. A voltage on an insulated plate reaches through an insulator and reversibly changes what kind of semiconductor the surface is. Remove the voltage and the electrons disperse; the channel vanishes instantly and completely.

Note also that it is reversible and fast — no chemistry, no diffusion, just carriers redistributing under a field. That is why it can happen billions of times per second.

Threshold voltage is the gate voltage at which inversion becomes strong enough to form a usable channel. It is set at manufacturing time by body doping, oxide thickness, and the gate material’s work function — which is why a process is characterised by its , and why designers treat it as fixed.


4. Predict before reading on

The channel is formed by electrons attracted from the source and drain into a P-type body.

What happens to the channel if you raise the source voltage while keeping the gate fixed?

Check your answer

The channel weakens and may vanish. What matters is — the gate voltage relative to the source — not the gate’s absolute voltage. Raising the source reduces , reducing the overdrive, shrinking the channel.

This has a very practical consequence you will meet in module 13: an NMOS transistor is bad at passing a logic high. As its source terminal rises toward , falls toward and the device chokes itself off — it can only pull the output up to , not to the rail.

That single fact is why CMOS needs both transistor types, and why NMOS-only logic could never deliver module 5’s rail-to-rail outputs.


5. The three regions

With a channel present, applying drives current through it. How it behaves depends on how compares with the overdrive .

Cutoff —

No inversion layer, no channel. Current is nearly zero — see section 7 on why “nearly” matters enormously.

Triode (linear) — and

The channel exists along its whole length. For small the device behaves as a resistor whose value is set by :

This is precisely what module 3 asked for, ten modules ago: a resistance controlled by a third terminal. The shopping list is now filled.

Saturation —

Here is where intuition usually fails, so take it slowly.

The channel’s local strength at any point depends on the gate voltage relative to the channel at that point. Near the source the channel sits at ~0 V, so the full acts on it. Near the drain the channel has been pulled up toward , so the effective gate-to-channel voltage there is only .

Raise until and the channel at the drain end is exactly at threshold. Beyond that it vanishes locally — the channel pinches off.

   Triode: channel full length          Saturation: pinched off at drain

   S ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ D              S ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓╲___ D
     ════════════════════                ═══════════════════
     uniform-ish channel                 channel tapers, then ends
                                         short depletion gap here

Current does not stop. Carriers reaching the pinch-off point are swept across the small depletion gap by the strong field there — exactly the mechanism sweeping minority carriers across a reverse-biased junction in module 10.

But the current stops increasing. Raising further just moves the pinch-off point slightly and drops the extra voltage across the gap. The current becomes almost independent of :

A saturated MOSFET is a voltage-controlled current source — set by , indifferent to . That is what makes it a good amplifier, and it is the region analogue designers live in.


6. — the designer’s knob

Both current equations contain : channel width over channel length.

  • is fixed by the process. “A 7 nm process” refers to a minimum feature size. Designers almost always use minimum , because shorter is faster.
  • is chosen per transistor. It is the one geometric parameter a logic designer routinely varies.

Current scales linearly with :

at V
1.036.00 μA
2.072.00 μA
2.8100.80 μA
4.0144.00 μA

Now the loose end from module 8 ties off. Hole mobility is 2.8× lower than electron mobility, so a PMOS transistor has 2.8× less . To make a PMOS drive as hard as an NMOS, you give it 2.8× the width — which is exactly the 2.8 in the table.

A fact about valence-band structure, established in module 8, has become a number in a layout tool. That is the course’s method working end to end.

The cost is not free: a wider transistor presents proportionally more gate capacitance to whatever drives it. Sizing is therefore a genuine optimisation, not a free win — a theme that recurs throughout digital design.


7. Cutoff is not actually off

The square-law model says current is zero below threshold. It is not, and the discrepancy governs modern chip design.

Below threshold the channel is weakly inverted, and current falls exponentially rather than to zero:

where is the subthreshold swing in volts per decade — typically 70–100 mV/decade.

has a hard floor:

That is module 10’s 60 mV rule, returning as promised. It is the same Boltzmann statistics — the fraction of carriers with enough energy to cross a barrier — and it is not an engineering limitation that better fabrication could remove. It is thermodynamics.

Why this ended voltage scaling

Follow the chain:

  1. To turn a transistor properly off, must be comfortably above 0, or leakage is unacceptable.
  2. To turn it properly on, must be comfortably above , or drive current collapses.
  3. So has a floor set by , which has a floor set by the 60 mV/decade swing.

Below about 1 V the arithmetic stops working. You cannot lower without exponentially worse leakage, and you cannot lower without losing the drive to switch.

And from module 2, — with pinned, the quadratic discount that had paid for every frequency increase since the 1970s was gone.

Trace the whole chain: Boltzmann statistics → 60 mV/decade → floor → floor → no more savings → clock frequencies plateau → multicore processors.

The number of cores in your laptop is set by the distribution of thermal energy among electrons. Modules 2, 10 and 12 have been building to that sentence.


8. Worked example — runnable

Save as mosfet_lab.py and run python3 mosfet_lab.py.

import math
 
# A simplified but self-consistent process
K_PRIME = 200e-6      # mu*Cox, amperes per volt^2 (process transconductance)
V_TH    = 0.4         # threshold voltage, volts
V_DD    = 1.0         # supply, volts
SUBTHRESHOLD_SWING = 0.070   # volts per decade of current
 
def region(v_gs, v_ds, v_th=V_TH):
    """Which of the three operating regions is the device in?"""
    if v_gs <= v_th:
        return "cutoff"
    return "triode" if v_ds < (v_gs - v_th) else "saturation"
 
def drain_current(v_gs, v_ds, w_over_l=1.0, v_th=V_TH, k=K_PRIME):
    """Long-channel MOSFET current, all three regions."""
    v_ov = v_gs - v_th                     # overdrive voltage
    if v_ov <= 0:                          # cutoff: exponential subthreshold leak
        return leakage_current(v_gs, w_over_l, v_th, k)
    if v_ds < v_ov:                        # triode / linear
        return k * w_over_l * (v_ov * v_ds - v_ds**2 / 2)
    return 0.5 * k * w_over_l * v_ov**2    # saturation
 
def leakage_current(v_gs, w_over_l=1.0, v_th=V_TH, k=K_PRIME, swing=SUBTHRESHOLD_SWING):
    """Below threshold, current falls exponentially -- not to zero."""
    i_ref = 0.5 * k * w_over_l * (0.1 ** 2)     # small reference current at v_ov = 0.1
    decades_below = (v_th + 0.1 - v_gs) / swing
    return i_ref * 10 ** (-decades_below)
 
def on_resistance(v_gs, w_over_l=1.0, v_th=V_TH, k=K_PRIME):
    """Small-signal resistance in the triode region -- the 'switched resistance'."""
    v_ov = v_gs - v_th
    if v_ov <= 0:
        return float("inf")
    return 1.0 / (k * w_over_l * v_ov)
 
if __name__ == "__main__":
    print(f"process: k' = {K_PRIME*1e6:.0f} uA/V^2, V_th = {V_TH} V, V_DD = {V_DD} V")
    print()
 
    print("operating region as a function of the two applied voltages:")
    print(f"  {'V_gs':>6s} {'V_ds':>6s}   region")
    for v_gs, v_ds in [(0.2, 0.5), (1.0, 0.1), (1.0, 0.3), (1.0, 0.9), (0.6, 0.5)]:
        print(f"  {v_gs:6.2f} {v_ds:6.2f}   {region(v_gs, v_ds)}")
    print()
 
    print("drain current vs gate voltage (V_ds = V_DD, W/L = 2):")
    for v_gs in (0.0, 0.2, 0.4, 0.6, 0.8, 1.0):
        i = drain_current(v_gs, V_DD, w_over_l=2.0)
        print(f"  V_gs = {v_gs:.1f} V -> I_d = {i*1e6:10.4f} uA   ({region(v_gs, V_DD)})")
    print()
 
    # The switchable resistance module 3 asked for
    print("on-resistance in triode (W/L = 2) -- module 3's switched resistor:")
    for v_gs in (0.5, 0.7, 1.0):
        print(f"  V_gs = {v_gs:.1f} V -> R_on = {on_resistance(v_gs, 2.0):9.0f} ohm")
    print(f"  V_gs = 0.0 V -> R_on = infinite (cutoff)")
    print()
 
    i_on  = drain_current(V_DD, V_DD, w_over_l=2.0)
    i_off = drain_current(0.0,  V_DD, w_over_l=2.0)
    print(f"on current  = {i_on*1e6:.3f} uA")
    print(f"off current = {i_off*1e12:.3f} pA")
    print(f"  on/off ratio = {i_on/i_off:.2e}")
    print()
 
    # W/L: the designer's only geometric knob
    print("current scales linearly with W/L (V_gs = V_ds = V_DD):")
    for wl in (1.0, 2.0, 2.8, 4.0):
        print(f"  W/L = {wl:4.1f} -> I_d = {drain_current(V_DD, V_DD, wl)*1e6:8.2f} uA")
 
    print()
    print(f"subthreshold swing = {SUBTHRESHOLD_SWING*1000:.0f} mV/decade")
    print(f"  the room-temperature floor is ln(10)*kT/q = 59.5 mV/decade")
    print(f"  turning off {V_TH} V below threshold buys"
          f" {V_TH/SUBTHRESHOLD_SWING:.1f} decades of current reduction")
 
    assert region(0.2, 0.5) == "cutoff"
    assert region(1.0, 0.1) == "triode"
    assert region(1.0, 0.9) == "saturation"
    assert i_on / i_off > 1e5
    assert abs(drain_current(V_DD, V_DD, 4.0) / drain_current(V_DD, V_DD, 2.0) - 2.0) < 1e-9
    print()
    print("mosfet_lab: passed")

Expected output:

process: k' = 200 uA/V^2, V_th = 0.4 V, V_DD = 1.0 V

operating region as a function of the two applied voltages:
    V_gs   V_ds   region
    0.20   0.50   cutoff
    1.00   0.10   triode
    1.00   0.30   triode
    1.00   0.90   saturation
    0.60   0.50   saturation

drain current vs gate voltage (V_ds = V_DD, W/L = 2):
  V_gs = 0.0 V -> I_d =     0.0000 uA   (cutoff)
  V_gs = 0.2 V -> I_d =     0.0001 uA   (cutoff)
  V_gs = 0.4 V -> I_d =     0.0746 uA   (cutoff)
  V_gs = 0.6 V -> I_d =     8.0000 uA   (saturation)
  V_gs = 0.8 V -> I_d =    32.0000 uA   (saturation)
  V_gs = 1.0 V -> I_d =    72.0000 uA   (saturation)

on-resistance in triode (W/L = 2) -- module 3's switched resistor:
  V_gs = 0.5 V -> R_on =     25000 ohm
  V_gs = 0.7 V -> R_on =      8333 ohm
  V_gs = 1.0 V -> R_on =      4167 ohm
  V_gs = 0.0 V -> R_on = infinite (cutoff)

on current  = 72.000 uA
off current = 0.144 pA
  on/off ratio = 5.00e+08

current scales linearly with W/L (V_gs = V_ds = V_DD):
  W/L =  1.0 -> I_d =    36.00 uA
  W/L =  2.0 -> I_d =    72.00 uA
  W/L =  2.8 -> I_d =   100.80 uA
  W/L =  4.0 -> I_d =   144.00 uA

subthreshold swing = 70 mV/decade
  the room-temperature floor is ln(10)*kT/q = 59.5 mV/decade
  turning off 0.4 V below threshold buys 5.7 decades of current reduction

mosfet_lab: passed

Note the on-resistance at full drive: 4167 Ω for a of 2. That is the same order as the 5 kΩ used in module 4’s RC timing example — the two modules were describing the same transistor from different ends.


9. Common pitfalls and traps

  1. Thinking the gate injects carriers. It is insulated. It attracts carriers that come from the source and drain.
  2. Believing inversion changes the doping. Dopant ions never move. Only the mobile carrier population at the surface changes.
  3. Using absolute gate voltage instead of . Everything depends on gate voltage relative to the source. This is the root of the NMOS-can’t-pull-up problem.
  4. Thinking pinch-off stops the current. Current continues; it merely stops increasing with .
  5. Treating cutoff as zero current. Subthreshold leakage is exponential, not zero, and at a billion transistors it dominates idle power.
  6. Forgetting costs capacitance. Widening a transistor strengthens it and loads its driver. There is an optimum.
  7. Trusting the square-law model at small geometries. It is a long-channel approximation. Below ~100 nm, velocity saturation makes current closer to linear in than quadratic.

10. Check your understanding

  1. Why can an NMOS transistor pull an output down to 0 V cleanly, but not up to ?

    Answer Pulling down, the source is at ground, so $V_{gs} = V_{DD}$ stays at full overdrive all the way — the device conducts hard right down to 0 V.
    Pulling up, the source terminal is the rising output. As it rises, $V_{gs} = V_{DD} - V_{out}$ shrinks. When $V_{out}$ reaches $V_{DD} - V_{th}$, the overdrive hits zero and the transistor cuts off. It cannot deliver the last $V_{th}$ volts.
    PMOS has the mirror-image problem. This complementary weakness is exactly why CMOS pairs them: NMOS pulls down, PMOS pulls up, each doing the job it is good at.
  2. Why does saturation current depend on but barely on ?

    Answer $V_{gs}$ sets how strongly the channel is inverted — how much charge is available to carry current. $V_{ds}$ beyond pinch-off does not add charge; it only lengthens the depletion gap at the drain slightly, and the extra voltage is dropped across that gap.
    The current is limited by charge supply, not by the accelerating voltage. That makes a saturated MOSFET a voltage-controlled current source. (In real short-channel devices, channel-length modulation gives a small residual $V_{ds}$ dependence.)
  3. A process reduces from 0.4 V to 0.2 V, with a 70 mV/decade swing. What happens to leakage, and why might a manufacturer do it anyway?

    Answer Leakage rises by $10^{0.2/0.07} = 10^{2.86} \approx$ 720×.
    They do it because drive current depends on $V_{ov} = V_{DD} - V_{th}$. At $V_{DD}$ = 1.0 V, lowering $V_{th}$ from 0.4 to 0.2 raises the overdrive from 0.6 to 0.8 V — and since saturation current goes as $V_{ov}^2$, that is $(0.8/0.6)^2 = 1.78×$ the current, hence a faster circuit.
    This is the fundamental speed-versus-leakage tradeoff. Real chips resolve it by using multiple threshold voltages on the same die: low-$V_{th}$ devices on critical paths, high-$V_{th}$ everywhere else.
  4. Why is the 59.5 mV/decade subthreshold floor a thermodynamic limit rather than an engineering one?

    Answer It comes from the Boltzmann distribution of carrier energies. The fraction of carriers with enough energy to surmount a barrier falls as $e^{-E/kT}$, so changing the barrier by $\ln(10)\,kT/q$ changes the current tenfold. That is a property of thermal equilibrium, not of materials or lithography.
    The only escapes are to lower $T$ (impractical for consumer devices) or to use a device that does not rely on thermal injection at all — which is exactly what tunnel FETs and negative-capacitance FETs attempt, both still research topics. It is the same physics that fixes the diode's 60 mV/decade in module 10.

11. Practice — independent task

Task: You are sizing transistors for a standard cell using the module’s process ( μA/V², V, V, swing 70 mV/decade). PMOS is 2.8× lower than NMOS.

  • (a) Compute the saturation current of an NMOS with at .
  • (b) What must a PMOS have to deliver the same current? Verify with drain_current() using the reduced .
  • (c) Compute the triode on-resistance of both at . Are they equal? Should they be?
  • (d) A designer proposes halving both widths to save area. Compute the new currents and on-resistances. Using from module 4, and assuming the load capacitance is dominated by wiring rather than by the gates, what happens to the delay?
  • (e) Now assume the load is dominated by the next stage’s gate capacitance, which also halves. What happens to the delay now? Explain why the two answers differ.
  • (f) Compute total leakage for 10⁹ transistors at , at = 0.4 V and again at = 0.25 V. Convert both to watts at 1.0 V.
  • (g) Write leakage_power(n_transistors, v_th, supply) and use it to find the at which a billion-transistor chip leaks 1 W.

Done when: you can state the PMOS width ratio and justify it from mobility, and explain why halving transistor widths helps or hurts depending on what dominates the load capacitance.

Hint for (d) and (e), only if stuck Delay goes as $\tau = R_{on}C_{load}$. Halving $W$ doubles $R_{on}$.
If $C_{load}$ is fixed wiring capacitance, delay doubles — a clear loss.
If $C_{load}$ is the next gate's input capacitance, it halves too, so $\tau = 2R \times C/2 = RC$ — delay is unchanged, and you saved area and switching energy for free.
This is why "make everything minimum size" is often correct in dense logic and badly wrong when driving long wires — and why buffer sizing is a real optimisation problem.

12. Tradeoffs and limits of this model

  • The square-law equations are long-channel approximations. In modern short-channel devices, carriers reach velocity saturation and current becomes closer to linear in . The qualitative picture — three regions, threshold, pinch-off — survives; the exponents do not.
  • Short-channel effects are omitted. Drain-induced barrier lowering (DIBL) makes depend on ; the gate loses electrostatic control as shrinks. Combating this drove the move to FinFETs and gate-all-around structures, which wrap the gate around the channel on several sides.
  • The body is assumed grounded. A non-zero body-source voltage shifts — the body effect — which matters in stacked transistors like the series NMOS chains you will meet in module 15.
  • Temperature dependence is ignored here. Mobility falls and leakage rises with temperature, so a hot chip is both slower and leakier.

Before moving on

  • Explain accumulation, depletion and inversion, and why inversion is surprising.
  • Explain what threshold voltage is and what physically happens at it.
  • Identify the operating region from and , and explain pinch-off without saying current stops.
  • Explain why the triode region is module 3’s switchable resistance.
  • Explain the knob and derive the 2.8× PMOS widening from module 8.
  • Trace the chain from Boltzmann statistics to multicore processors.

Recap: A gate voltage across an insulator drives the silicon surface through accumulation, depletion and finally inversion, where attracted minority carriers make a P-type surface behave as N-type and connect source to drain. Threshold voltage is where that channel becomes usable. Below it, current falls exponentially at best 59.5 mV/decade; above it, the device is a voltage-controlled resistor at low and a voltage-controlled current source once pinched off. Current scales with , which is why PMOS devices are drawn 2.8× wider.

Next: Module 13 — CMOS puts an NMOS and a PMOS together. Module 3’s voltage-divider table becomes a real circuit, the complementary weaknesses cancel, and you get a gate that delivers module 5’s rail-to-rail output while burning almost nothing.


↑ Back to How Computers Work