TDEB Top Bar

Trending
Semiconductor Industry News — The Digital Electronics Blog
Follow Us

TDEB Masthead

The Semiconductor Hub

Semiconductors.  Chips and Systems.  Innovation.

Indexed by leading Search Engines, AI and referenced by Engineers worldwide 500+ Articles 10M+ Pageviews 30+ Reports 50+ Citations

Interview Question: Does Circuit Power Rise with Temperature?

Murugavel Ganesan
by
0
Share Facebook X LinkedIn WhatsApp Email

As temperature increases, does the power consumed by a typical combinational circuit increase, stay the same, or decrease? Briefly justify your answer.

This is a classic low power interview question. The short answer: total power goes up, mainly because leakage current rises steeply with temperature, with a smaller increase in short-circuit power. Switching power, at a fixed clock frequency, stays essentially the same. The original 2007 answer on this page said switching power rises because resistivity rises; that part was wrong, and the corrected reasoning is below.

The three parts of CMOS power

The power drawn by a CMOS logic circuit has three parts:

  • Switching (dynamic) power: the energy to charge and discharge the capacitance of every node that toggles.
  • Short-circuit power: during an input transition, for a short time both the pull-up and pull-down transistors conduct, so current flows straight from supply to ground.
  • Leakage (static) power: current that flows even when nothing switches, mostly subthreshold current through transistors that are nominally off.

Temperature affects each part differently.

Switching power: about the same

Every time a node with capacitance C charges to the supply V and then discharges, the supply delivers C x V^2 of energy, half lost in the pull-up and half in the pull-down. This energy does not depend on the resistance of the transistors or wires: higher resistance makes the charging slower, but the same charge still moves through the same voltage. So at a fixed clock frequency and the same activity, switching power does not change with temperature. A hotter circuit is slower, and if it becomes too slow the design fails timing, but that is a delay problem, not a switching-power increase.

Short-circuit power: up a little

As temperature rises, carriers in the transistor channel scatter more often off the vibrating silicon lattice, so their mobility falls (see the temperature dependence of mobility). Lower mobility means lower drive current, slower gates and slower rising and falling edges. A slower input edge keeps both transistors of the next gate partly on for longer, which increases short-circuit current. This was the reasoning in the original answer, and it holds. Threshold voltage also falls as temperature rises, which partly offsets the slowdown, but for typical supply voltages mobility wins and gates get slower when hot.

Leakage power: up a lot

Subthreshold leakage depends exponentially on the ratio of threshold voltage to the thermal voltage kT/q (see subthreshold conduction). Higher temperature raises kT/q and lowers the threshold voltage, and both push leakage up sharply. In processes from about 90 nm onward, leakage became a large fraction of total power, so in a modern chip this is usually the dominant temperature effect. It can also create a feedback loop: more leakage means more heat, which means more leakage, which is why thermal runaway is checked during design.

The answer to give

Power increases. Leakage power rises strongly, short-circuit power rises slightly because edges are slower, and switching power at a fixed frequency stays about the same.

Deep dive: the numbers and details

This part holds the equations and a worked example. Skip it if the overview is all you need.

Switching energy does not depend on resistance

Charging a capacitor C from 0 to V through any resistor R draws charge Q = C x V from the supply, so the supply delivers E = Q x V = C x V^2. The capacitor stores C x V^2 / 2, and the resistor dissipates the other C x V^2 / 2, regardless of R. R only sets the time constant R x C. Switching power is therefore:

P_switch = alpha x C x V^2 x f

with alpha the activity factor and f the clock frequency. None of these terms depends on temperature directly.

Leakage versus temperature

A simple model of subthreshold current is:

I_sub = I_0 x exp( (V_GS - V_T) / (n x kT/q) ) x (1 - exp(-V_DS / (kT/q)))

with n typically 1.2 to 1.5. At 25 °C, kT/q is about 25.7 mV; at 100 °C it is about 32.2 mV. Take V_T = 0.35 V at 25 °C, falling by about 1 mV per °C to 0.275 V at 100 °C, n = 1.3, and V_GS = 0 for an off transistor. Ignoring the change in I_0:

at 25 °C:  exp(-0.350 / (1.3 x 0.0257)) = exp(-10.48) = 2.8 x 10^-5
at 100 °C: exp(-0.275 / (1.3 x 0.0322)) = exp(-6.57)  = 1.4 x 10^-3

The ratio is about 50. In practice I_0 also rises with temperature while mobility falls, so measured increases vary, but a leakage increase of an order of magnitude or more between room temperature and 100 °C is common for high-performance logic. The numbers above are illustrative values chosen for the example, not data for a specific process.

Worked example

A block draws 100 mW of switching power, 5 mW of short-circuit power and 10 mW of leakage at 25 °C. At 100 °C, assume switching stays at 100 mW, short-circuit rises by 20 percent to 6 mW because of slower edges, and leakage rises 10 times to 100 mW:

Component25 °C100 °C
Switching100 mW100 mW
Short-circuit5 mW6 mW
Leakage10 mW100 mW
Total115 mW206 mW

Total power rises by about 80 percent, almost all from leakage.

A subtlety: temperature inversion

At low supply voltages, close to the threshold voltage, the falling threshold can matter more than the falling mobility, so gates can get faster when hot. This is called temperature inversion, and it is why low-voltage designs must be timed at both cold and hot corners. It changes which corner is slowest, but leakage still rises with temperature.

Common mistakes

  • Saying switching power rises because resistance rises. Resistance changes delay, not the C x V^2 energy per transition.
  • Forgetting leakage. In older long-channel processes it was small; in any process from the last two decades it is the main temperature effect.
  • Assuming hot is always the slow corner. At low voltage, temperature inversion can make cold the slow corner.

Further reading

Murugavel Ganesan
About the Author
Murugavel Ganesan
Product Director | SerDes · IVN · Edge AI · Compute | Ex Intel · Infineon

A semiconductor product leader working across product architecture, definition, strategy, and silicon and software execution, from concept through volume production. He has worked with teams across Asia, Europe and the USA, including at Intel, Infineon and ArrayComm.

His work covers ASICs, SoCs, FPGAs and chiplets for wireless, wireline, automotive, IoT and AI, including PCIe, CXL and DDR interfaces alongside RF and mixed-signal design.

Share Facebook X LinkedIn WhatsApp Email

Post a Comment

0 Comments

Your comments will be moderated before it appears here.

Post a Comment (0)

#buttons=(Ok, Go it!) #days=(20)

Our website uses cookies to enhance your experience. Learn more
Ok, Go it!