Power: Dynamic, Short-Circuit and Leakage
Measure where the energy of a CMOS gate goes: charging its load, a short circuit during each edge, and leakage at rest.
You Will Be Able To
- Compute dynamic energy as C·V² per cycle and dynamic power as α·C·V²·f.
- Explain short-circuit current and why slow edges increase it.
- Say which transistor leaks in each input state and why leakage grows with temperature.
- Weigh a lower supply voltage against the delay it costs.
- Lesson
- G9.9
- Time
- 30 min
- Runs on
- ngspice
- Check
- 5 questions
One inverter, one load, two supplies. At 1.8 V one full input cycle cost 38.3 fJ. At 1.2 V it cost 17.2 fJ. A third less voltage saved more than half the energy, and the gate became 2.5 times slower.
- energy per cycle
- mean delay
Read the plot as a table
| Supply voltage | energy per cycle | mean delay |
|---|---|---|
| 900 mV | 9.644 fJ | 384.2 ps |
| 1 V | 11.99 fJ | 202.1 ps |
| 1.1 V | 14.69 fJ | 148.4 ps |
| 1.2 V | 17.39 fJ | 94.75 ps |
| 1.3 V | 20.65 fJ | 80.74 ps |
| 1.4 V | 23.91 fJ | 66.73 ps |
| 1.5 V | 27.17 fJ | 52.72 ps |
| 1.6 V | 31.09 fJ | 47.91 ps |
| 1.7 V | 35.01 fJ | 43.1 ps |
| 1.8 V | 38.92 fJ | 38.29 ps |
Measured with ngspice 42 on SKY130 tt, 27 °C. Energy = supply voltage × charge drawn in one cycle.
Dynamic Energy: C·V² per Cycle
When the output rises, the supply pushes a charge C·V onto the load. It provides that charge at voltage V, so it delivers C·V² of energy. Half is stored on the load. The other half turns into heat in the PMOS.
When the output falls, the NMOS dumps the stored half to ground. Nothing new comes from the supply. So one full cycle of the output costs C·V².
- Power
- Device
- Load
- Power
- Output rises
The supply delivers C·V². Half heats the PMOS; half is stored on the load.
- Output falls
The NMOS turns the stored half into heat. The supply gives nothing.
- During an edge
Both transistors are partly on, and current flows straight from the supply to ground.
- At rest
The transistor that is off still leaks a tiny current, all the time.
Read the diagram as text
- Supply: at voltage V
- PMOS: pull-up
- Load C: output node
- NMOS: pull-down
- Ground: heat ends here
- Supply → PMOS: C·V²
- PMOS → Load C: charge
- Load C → NMOS: discharge
- PMOS → NMOS: short circuit
- NMOS → Ground
Dynamic, short-circuit and leakage energy. The rest of the lesson measures each one.
| Supply | C·V² for the 10 fF load | Measured per cycle | Measured with no load |
|---|---|---|---|
| 1.8 V | 32.4 fJ | 38.3 fJ | 3.9 fJ |
| 1.2 V | 14.4 fJ | 17.2 fJ | 1.9 fJ |
Measured: the unit inverter, 50 ps input edges, energy = supply voltage × charge drawn in one input cycle.
The measured energy is larger than C·V² of the load. The gate's own transistors add capacitance at the output, and the no-load run shows their share: 3.9 fJ at 1.8 V. Here that is gate overlap only. These netlists leave the drain junctions at zero area, so a laid-out cell draws a little more.
A chip has millions of nodes, and most do not switch every cycle. The activity factor α is the fraction of cycles in which a node makes a full cycle. Dynamic power is then P = α · C · V² · f.
Short-Circuit Current During an Edge
While the input passes through the middle of its swing, both transistors are partly on. Current flows straight from the supply to ground. It charges nothing.
- v(if), 50 ps edge
- v(is), 1 ns edge
- i(vf)
- i(vs)
- Time 1.5 nsv(if), 50 ps edge = 1.8 V · v(is), 1 ns edge = 900 mV · i(vf) = -1.061 nA · i(vs) = 6.883 µA
The inputs rise and the outputs fall. With the slow edge both transistors stay partly on for most of a nanosecond, and current flows straight through them.
- Time 6.5 nsv(if), 50 ps edge = 0 V · v(is), 1 ns edge = 900 mV · i(vf) = 2.076 pA · i(vs) = 39.71 µA
The inputs fall and the outputs rise. Now the supply charges the 10 fF loads. With the slow edge the NMOS is still partly on, so part of the current goes straight to ground.
Read the plot as a table
| Time | v(if), 50 ps edge | v(is), 1 ns edge | i(vf) | i(vs) |
|---|---|---|---|---|
| 0 s | 0 V | 0 V | 2.046 pA | 2.046 pA |
| 1 ns | 97.93 mV | 4.896 mV | -13.8 µA | -773.6 nA |
| 1.5 ns | 1.8 V | 900 mV | -1.061 nA | 6.883 µA |
| 2 ns | 1.8 V | 1.794 V | 1.282 nA | -1.451 µA |
| 3 ns | 1.8 V | 1.8 V | 426.6 pA | 531.6 pA |
| 4 ns | 1.8 V | 1.8 V | 424.6 pA | 423.6 pA |
| 5 ns | 1.8 V | 1.8 V | 424.6 pA | 424.2 pA |
| 6 ns | 0 V | 1.794 V | -50.64 pA | 721.7 nA |
| 6.5 ns | 0 V | 900 mV | 2.076 pA | 39.71 µA |
| 7 ns | 0 V | 1.688 mV | 2.045 pA | 5.156 µA |
| 8 ns | 0 V | 0 V | 2.046 pA | 886.6 fA |
Measured with ngspice 42 on SKY130, 1.8 V. i(vf) and i(vs) are the currents the two inverters draw.
| Input edge | Charge drawn while the output falls | Charge drawn while the output rises | Energy of the full cycle |
|---|---|---|---|
| 50 ps | -2.1 fC | 23.4 fC | 38.3 fJ |
| 1 ns | -1.5 fC | 25.6 fC | 43.3 fJ |
Measured from the same run, 1.8 V. A negative charge flowed back into the supply.
While the output falls, the supply even takes charge back. That charge is mostly the PMOS's own. As the rising input turns it off, the charge in its channel and its gate-to-source and gate-to-well capacitance flows back. The next falling edge draws it again.
So the charge on each edge mixes this gate charge with the short-circuit charge. Only the difference over a full cycle measures the short circuit.
The slow edge draws more over the cycle: the short circuit. Over a full cycle it costs 4.9 fJ more, about 13 % on top of the fast case.
At a low supply the effect can vanish. At 1.2 V the two transistors are hardly on together. There 1 ns edges cost 16.4 fJ, slightly less than the 17.2 fJ with fast edges.
Leakage: Current When Nothing Switches
An off transistor is not a perfect switch. A small subthreshold current still flows from drain to source. It grows about tenfold for every 0.1 V or so that the threshold is lowered, and it grows quickly with temperature.
| Input at rest | Transistor that leaks | Current at 27 °C | Power at 27 °C | Current at 85 °C |
|---|---|---|---|---|
| 0 V | NMOS (off) | 2.0 pA | 3.7 pW | 8.6 pA |
| 1.8 V | PMOS (off) | 424.6 pA | 764.3 pW | 9.3 nA |
Measured: the unit inverter at rest, 1.8 V, operating point after 1 ns.
The two states differ by a factor of 208: in these models the PMOS leaks far more than the NMOS. At 85 °C the leakage with the input high is 22 times that at 27 °C.
Gate leakage is current tunnelling through a very thin gate oxide. The SKY130 device models used here leave gate tunnelling out, so no run on this site can show it. The 1.8 V devices have a thick oxide, where it is small. In processes with oxide only a few atoms thick it becomes a large share of leakage.
Leakage flows all the time, in every gate. A gate that toggles once per microsecond spends far more on switching. A block that is idle for seconds spends almost everything on leakage. Low Power lessons W1 and later show how chips cut both.
The Supply Voltage Trade
| Supply | Energy per cycle | Mean delay | Energy × delay |
|---|---|---|---|
| 0.9 V | 9.6 fJ | 384.2 ps | 3705 fJ·ps |
| 1 V | 12.0 fJ | 202.1 ps | 2423 fJ·ps |
| 1.2 V | 17.4 fJ | 94.7 ps | 1648 fJ·ps |
| 1.5 V | 27.2 fJ | 52.7 ps | 1432 fJ·ps |
| 1.8 V | 38.9 fJ | 38.3 ps | 1490 fJ·ps |
Measured: the unit inverter into 10 fF, driven by a unit inverter at the same supply.
Lowering the supply saves energy as V², but the transistors carry less current, so the gate slows. From 1.8 V to 0.9 V the energy fell to 0.25 of it, and the delay grew 10.0 times.
Near the threshold voltage the delay rises steeply. Designs run each block at the lowest supply that still meets its timing. Low Power lesson W1.4 covers changing the supply at run time.
Asked in interviews
Where does the power of a CMOS gate go?
Dynamic power α·C·V²·f to charge and discharge capacitance. Short-circuit power while both networks conduct during an edge. Leakage power from off transistors all the time.
Why does lowering VDD save so much power?
Dynamic energy goes with V², so a supply 30 % lower saves about half the energy. The cost is delay, since the transistors carry less current.
What makes leakage worse?
Lower threshold voltages, higher temperature, and very thin gate oxides. Leakage flows whether the gate switches or not.
Try It
q_a and q_b are the charge each inverter drew in one input cycle. Times its supply that is 38.3 fJ at 1.8 V and 17.2 fJ at 1.2 V.
The delays at 1.2 V are longer: 69.4 ps and 102.2 ps against 37.3 ps and 45.7 ps.
Remove the line Ca and run: the energy at 1.8 V falls to the share of the gate's own transistors. Then change both 50p of VIa to 1n and compare.
Run the netlist to draw its plots here.
q_a = 2.13043e-14 from= 5.00000e-10 to= 8.50000e-09 q_b = 1.43264e-14 from= 5.00000e-10 to= 8.50000e-09 fall_a = 3.733268e-11 targ= 1.062333e-09 trig= 1.025000e-09 rise_a = 4.573065e-11 targ= 5.120731e-09 trig= 5.075000e-09 fall_b = 6.943944e-11 targ= 1.094439e-09 trig= 1.025000e-09 rise_b = 1.022361e-10 targ= 5.177236e-09 trig= 5.075000e-09Check Yourself
Common Mistakes
Wrong Scale the energy with the supply voltage: two thirds of the voltage, two thirds of the energy.
The energy goes with V². From 1.8 V to 1.2 V the measured energy per cycle fell from 38.3 fJ to 17.2 fJ, to 0.45 of it.
Fix Use C·V² per cycle. Two thirds of the voltage gives about four ninths of the energy.
Wrong Count only the half of C·V² that is lost in the PMOS while the output rises.
The other half is stored on the load and lost in the NMOS when the output falls. Over a full cycle the supply provides all of C·V².
Fix Per full cycle, a node that rises and falls once costs C·V² from the supply.
Wrong Ignore the slope of a gate's input when estimating its power.
With 1 ns input edges both transistors conduct together for longer. One full cycle cost 43.3 fJ instead of 38.3 fJ.
Fix Keep edges sharp. Timing tools limit the slew of every net partly for this reason.
Wrong Assume a gate at rest draws no current.
An off transistor still leaks. The unit inverter with its input high draws 424.6 pA at 27 °C and 9.3 nA at 85 °C.
Fix Count leakage for every gate, whether it switches or not. It matters most in large, mostly idle blocks.
Remember
- One full output cycle draws C·V² from the supply. Dynamic power is α·C·V²·f.
- Short-circuit current flows while both networks conduct during an edge. Slow edges waste more.
- Off transistors leak all the time, and leakage grows quickly with temperature.
- A lower supply saves energy as V² but costs delay. Use the lowest supply that meets timing.