Pass-Transistor and Transmission-Gate Logic
Build logic from switches that pass a signal. See where an NMOS-only gate loses a threshold, what that costs, and how a level restorer or a transmission gate fixes it.
You Will Be Able To
- Build an XOR from two pass devices and dual-rail inputs.
- Explain why an NMOS pass device delivers a weak 1, and what that costs the next gate.
- Restore a weak level with a feedback PMOS, the idea behind CPL.
- Choose between NMOS-only pass logic and transmission gates.
- Lesson
- G9.12
- Time
- 30 min
- Runs on
- ngspice
- Check
- 5 questions
Two NMOS transistors make an XOR, if a' and b' are already at hand. A static CMOS XOR needs eight, and four more for the inverted inputs. The two-transistor version has a catch, and the figure shows it.
- v(b)
- v(yn), NMOS only
- v(yr), with a restorer
- v(yt), transmission gates
- Time 225 psv(b) = 900 mV · v(yn), NMOS only = 577.8 mV · v(yr), with a restorer = 575.7 mV · v(yt), transmission gates = 558.3 mV
b rises. With a = 0, every version passes b to y.
- Time 291.4 psv(b) = 1.8 V · v(yn), NMOS only = 899.8 mV · v(yr), with a restorer = 900.1 mV · v(yt), transmission gates = 1.625 V
The NMOS-only y crosses 0.9 V after 66.4 ps. The transmission-gate y did it after 22.2 ps.
- Time 2.2 nsv(b) = 1.738 V · v(yn), NMOS only = 1.111 V · v(yr), with a restorer = 1.8 V · v(yt), transmission gates = 1.795 V
Two nanoseconds later the NMOS-only y is still at 1.111 V. The restored and the transmission-gate outputs are at 1.800 V and 1.800 V.
Read the plot as a table
| Time | v(b) | v(yn), NMOS only | v(yr), with a restorer | v(yt), transmission gates |
|---|---|---|---|---|
| 0 s | 0 V | 192.3 pV | 1.999 nV | 262.3 nV |
| 225 ps | 900 mV | 577.8 mV | 575.7 mV | 558.3 mV |
| 291.4 ps | 1.8 V | 899.8 mV | 900.1 mV | 1.625 V |
| 500 ps | 1.8 V | 1.01 V | 1.347 V | 1.8 V |
| 1 ns | 1.8 V | 1.07 V | 1.797 V | 1.8 V |
| 1.5 ns | 1.8 V | 1.093 V | 1.8 V | 1.8 V |
| 2 ns | 1.8 V | 1.107 V | 1.8 V | 1.8 V |
| 2.2 ns | 1.738 V | 1.111 V | 1.8 V | 1.795 V |
| 2.5 ns | 0 V | 62.8 nV | -173.3 nV | 716.3 nV |
| 3 ns | 0 V | -770 pV | 98.31 nV | 249.6 nV |
| 3.5 ns | 0 V | 217.8 pV | -23.54 nV | 262.7 nV |
| 4 ns | 0 V | 191.6 pV | 7.983 nV | 262.3 nV |
Measured with ngspice 42 on SKY130, 1.8 V. Pass devices 1 µm wide; restorer PMOS 0.42 µm by 0.5 µm.
Pass Devices and the Threshold Drop
A pass device connects a signal to the output instead of connecting a rail. Lesson G9.3 showed that an NMOS passes a strong 0 but a weak 1. Its gate must stay a threshold above its source, so its source cannot rise to the gate voltage.
| a | b | y, NMOS only | y, transmission gates |
|---|---|---|---|
| 0 | 0 | 0.000 V | 0.000 V |
| 0 | 1 | 1.427 V | 1.800 V |
| 1 | 0 | 1.427 V | 1.800 V |
| 1 | 1 | 0.000 V | 0.000 V |
Measured at rest: the four input rows of the lab.
Every 0 is a clean 0 V. Every 1 through an NMOS stops short. The transmission gate adds a PMOS beside each NMOS, and the PMOS passes a strong 1.
What a Weak 1 Costs, and Restoring It
| Input row | Inverter current behind an NMOS-only XOR | Behind a restored NMOS XOR | Behind a transmission-gate XOR |
|---|---|---|---|
| a = 0, b = 0 | 2.0 pA | 4.3 pA | 2.0 pA |
| a = 0, b = 1 | 116.2 nA | 428.5 pA | 424.6 pA |
| a = 1, b = 0 | 116.2 nA | 428.5 pA | 424.6 pA |
| a = 1, b = 1 | 2.0 pA | 4.3 pA | 2.0 pA |
Measured: the current the receiving inverter draws at rest, 1.8 V.
A weak 1 leaves the next inverter's PMOS partly on. In the rows where y should be 1, that inverter draws 116.2 nA for as long as the input stays. A million such gates would draw 116 mA at rest.
- Device
- I/O
- Datapath
- Data
- Control
- Power
- The weak 1 arrives
The pass network brings y to about 1.427 V. That is enough for the inverter to switch.
- Feedback
The inverter output falls and turns the restorer on. It pulls y the rest of the way: 1.800 V. The inverter current falls to 428.5 pA.
- Pulling y back to 0
When the pass network passes a 0, it must overpower the restorer. So the restorer is made long and weak.
Read the diagram as text
- NMOS pass network: delivers a weak 1
- Node y: the weak level
- Receiving inverter: its output goes low
- Restorer PMOS: weak, from VDD
- NMOS pass network → Node y: weak 1
- Node y → Receiving inverter
- Receiving inverter → Restorer PMOS: gate low: on
- Restorer PMOS → Node y: pulls to VDD
Complementary pass-transistor logic (CPL) uses the same idea twice: it computes both polarities with NMOS networks and cross-couples two such PMOS.
The restorer costs some speed when y falls, since the pass device must fight it. The fall took 23.6 ps against 13.5 ps without it.
The Transmission-Gate XOR
- a = 0: the upper gate passes b
N1 has a' = 1 at its gate and P1 has a = 0: both are on. The lower gate is off. So y = b.
- a = 1: the lower gate passes b'
Now N2 and P2 are on and the upper gate is off. So y = b': the XOR.
- Both rails
In each gate the NMOS passes a strong 0 and the PMOS a strong 1. Drop the PMOS and the 1 stops a threshold below VDD.
Read the circuit as text
- N1 NMOS transistor: gate →
n1, drain →n2, source →n3 - P1 PMOS transistor: gate →
n4, source →n2, drain →n3 - N2 NMOS transistor: gate →
n5, drain →n6, source →n3 - P2 PMOS transistor: gate →
n7, source →n6, drain →n3 - b port: terminal →
n2 - b' port: terminal →
n6 - a' port: terminal →
n1 - a port: terminal →
n4 - a port: terminal →
n5 - a' port: terminal →
n7 - y port: terminal →
n3
Four transistors with dual-rail inputs. With only a and b available, two inverters make a' and b'.
With both devices in each gate, every row reaches a rail, and the receiving inverter draws only leakage. It is also the fastest of the three: 22.2 ps rising and 18.9 ps falling.
The price is the PMOS: twice the transistors of the NMOS-only XOR, and both polarities of a on every gate. Multiplexers are the other classic use: lesson G9.3 builds one.
Asked in interviews
Why does an NMOS pass a weak 1?
It conducts only while its gate is a threshold above its source. As the output rises towards the gate voltage the NMOS turns off, a threshold below VDD.
What is CPL?
Complementary pass-transistor logic. NMOS pass networks compute each function and its complement, and cross-coupled PMOS restore the weak high levels before the output inverters.
Why use a transmission gate instead of a single NMOS?
The PMOS passes a strong 1 and the NMOS a strong 0, so the output reaches both rails. The next gate then has no static current and full noise margins.
Try It
t_y1 and t_y2 are the transmission-gate outputs in the two rows where y = 1: 1.800 and 1.800 V. n_y1 and n_y2 are the NMOS-only outputs: 1.427 and 1.427 V.
n_i1 is the current the inverter behind the NMOS-only XOR draws in row 01: 116.2 nA.
Problem G9.12.1 asks for an XOR that reaches both rails without that current.
Run the netlist to draw its plots here.
t_y0 = 2.623001e-07 t_i0 = 2.046233e-12 t_y1 = 1.799999e+00 t_i1 = 4.246337e-10 t_y2 = 1.799999e+00 t_i2 = 4.246338e-10 t_y3 = 2.623001e-07 t_i3 = 2.046228e-12 n_y0 = 1.923019e-10 n_i0 = 2.046224e-12 n_y1 = 1.426535e+00 n_i1 = 1.161954e-07 n_y2 = 1.426535e+00 n_i2 = 1.161954e-07 n_y3 = 1.923030e-10 n_i3 = 2.046247e-12Check Yourself
Common Mistakes
Wrong Pass a 1 through an NMOS and treat it as a full logic 1.
It stops a threshold below the supply: 1.427 V at rest. Two nanoseconds after the input switched it had only reached 1.111 V.
Fix Use a transmission gate, or restore the level before it reaches anything that cares.
Wrong Assume a weak 1 is harmless because the next inverter still reads it as 1.
The next inverter's PMOS is not fully off. With a 1.427 V input it drew 116.2 nA at rest, against 424.6 pA behind a transmission gate.
Fix Count the static current of every gate that a pass network drives, or restore the level.
Wrong Chain many pass devices in series without a buffer.
Each pass device adds resistance and each node adds capacitance. The delay of the chain grows like an RC ladder: with the square of its length (lesson G9.8).
Fix Insert an inverter every few stages to restore the edge and the level.
Wrong Drive the two gates of an XOR with the same polarity of a.
Then both pass devices are on together, or both are off. y is shorted between b and b', or it floats.
Fix One device needs a and the other a'. In a transmission gate, the NMOS and PMOS of one gate also take opposite polarities.
Remember
- A pass device passes its input instead of a rail. An XOR needs only two of them with dual-rail inputs.
- An NMOS passes a strong 0 and a weak 1, a threshold below VDD.
- A weak 1 makes the next inverter draw static current. A feedback PMOS restores the level; CPL builds on that.
- A transmission gate passes both rails, at the cost of a PMOS and both polarities of the control.