The whole topic is one promise and its collapse: if you shrink a transistor and lower its voltage by the SAME factor, the chip runs faster and packs more devices while burning the same heat per square millimetre. This note builds every letter — L , V , C , I , τ , k , V t h — from a picture, so that when the parent note writes τ = C V / I you already see the triangle of charge, current, and time behind it.
Before any symbol, look at the object we keep shrinking.
A MOSFET (the switch our chips are built from) is a tiny sandwich. Current wants to flow between two terminals called the source and the drain . Between them sits a strip of silicon called the channel . Sitting on top — separated by a thin glassy insulator — is the gate , the control knob. Put voltage on the gate and it invites charge into the channel, opening a path for current; remove it and the path closes.
Everything in Dennard scaling is about making this sandwich smaller while keeping it working. See MOSFET Operation and Square-Law Current for how the current actually flows.
L = channel length
The horizontal distance the charge must travel from source to drain — the length of the red path in the figure above.
Picture: the gap the current crosses.
Why we need it: shorter L means charge arrives sooner → faster transistor. Shrinking L is the point of every new chip generation.
t o x = oxide thickness
The thickness of the thin insulating glass (o x = oxide) between the gate and the channel.
Picture: the gap between the two plates of a capacitor (the gate is one plate, the channel the other).
Why we need it: a thinner insulator lets the gate grip the channel more tightly — a stronger capacitor. This is why thinning t o x fights the shrink of area (Step 1 in the parent).
V = supply voltage
The "push" we apply — how hard we drive electricity through the device. Measured in volts.
Picture: the height of a waterfall; higher V = harder push.
Why we need it: more V = faster switching, but also more heat. The genius of Dennard is scaling V down along with size.
E = electric field, and why E ≈ V / L
The force per unit charge felt inside the channel. Squeeze a voltage V across a length L and the field is roughly E ≈ L V .
Picture: the steepness of the slope the charge rolls down. Same drop V over a shorter run L = a steeper, more violent slope.
Why the topic needs it: if you shrink L but keep V , the slope gets steeper by k and the transistor gets fried. Keeping E constant — "constant-field scaling" — is the whole trick. See Constant-Field vs Constant-Voltage Scaling .
k = scaling factor (k > 1 )
A single number, bigger than 1, that says how much smaller we make things this generation. Length becomes L / k , voltage becomes V / k , and so on.
Picture: a shrink-ray dial. Set k = 1.4 and every length shrinks to about 71% of before.
Why we need it: it lets us track one dial through the whole derivation and read off how speed, power, and heat respond.
k k ompresses. Anything that gets smaller is divided by k ; anything that gets more numerous or faster is multiplied by k .
C = capacitance
How much charge the gate must store to switch the channel on. The gate + oxide + channel form a parallel-plate capacitor : C o x = t o x ε o x W L .
Picture: a bucket. Bigger plates (W L ) or thinner gap (t o x ) = a bigger bucket that holds more charge.
ε o x (Greek "epsilon") is just the material constant of the oxide — how good that particular glass is at storing charge. It does not change when we shrink, so it rides along untouched.
Why we need it: switching = filling and emptying this bucket. A bigger bucket is slower to fill, so C directly controls speed.
I = current
The rate of charge flow — how many electrons per second cross the channel. Measured in amps (coulombs per second).
Picture: litres of water per second down the highway.
Why we need it: current is what fills the capacitor bucket . More current = faster switching but more power. The parent's square-law I ≈ μ C o x ′′ L W ( V − V t h ) 2 is how I depends on all the other symbols.
τ = gate delay (Greek "tau")
The time to switch one gate : fill the bucket C up to voltage V using current I . τ = I C V .
Picture: time = (water needed) ÷ (flow rate). Charge needed is C × V ; flow rate is I ; divide.
Why we need it: τ is inverse speed . Halve τ and the chip clocks twice as fast. This one formula is the engine of Step 3 in the parent.
f = frequency
How many switching steps per second — the clock speed (GHz). It is simply the inverse of delay: faster switching (smaller τ ) means higher f , so f ∝ 1/ τ .
Picture: heartbeats per second of the chip.
Why we need it: "the chip got faster" means f went up. When Dennard broke, f stopped climbing — that's the headline everyone felt.
P = power and P / A = power density
Power is energy burned per second — the heat the device makes. Two useful forms: P = I V (current times push) and dynamic power P d y n = 2 1 C V 2 f .
Power density P / A divides that by the device's area A = W L — heat per square millimetre.
Picture: P = the wattage of one bulb; P / A = how hot a panel packed with bulbs gets.
Why we need it: the single quantity Dennard scaling holds constant is P / A . When it stopped staying constant, chips hit the power wall .
V t h = threshold voltage
The minimum gate voltage that turns the transistor ON. Below V t h it is (mostly) off; above it, current flows.
Picture: the lip of a dam . Raise the water (gate voltage) past the lip and it flows.
Why we need it: the drive term is ( V − V t h ) 2 — the headroom between how hard you push and the lip. When V t h couldn't shrink (thermal-noise floor ≈ 26 mV), V couldn't shrink either, and the free lunch ended. See Subthreshold Leakage and Static Power .
k T / q ≈ 26 mV = thermal voltage
A tiny voltage set by temperature (T ), Boltzmann's constant (k here, not the scaling factor!), and electron charge (q ). It measures the random jiggle of electrons from heat.
Picture: the fuzzy waterline of a jittering pond — you can't build a dam lip below the ripples or water spills over anyway.
Why we need it: it is the hard floor under V t h . Push V t h toward it and subthreshold leakage I l e ak ∝ e − V t h / ( n k T / q ) explodes exponentially — the transistor leaks even when "off". This exponential is static power 's villain.
k 's
Why it confuses: the same letter k means the scaling factor (k > 1 , a design choice) AND Boltzmann's constant (in k T / q , a constant of nature).
Fix: context decides. If it multiplies dimensions it's the shrink dial. If it sits next to T over q , it's the physics constant.
Electric field E = V over L
Thermal voltage kT over q
Read it top-down: the geometry and voltage feed the field, capacitance, and current; those feed delay (hence speed) and power (hence power density); and the thermal-voltage floor under V t h is what eventually breaks the chain.
Cover the right side and test yourself — you are ready for the parent note when each is instant.
What do L , W , and t o x physically measure on the transistor? Channel length (source-to-drain distance), channel width (into the page), and oxide-insulator thickness.
Why is the field E ≈ V / L , and why must it stay constant? Voltage V dropped over length L ; keeping it constant while shrinking L forces V to shrink too, so the tiny device isn't fried.
What is k , and what happens to a length vs a device count when we scale by k ? The shrink factor ( k > 1 ) ; lengths divide by k , device density multiplies by k 2 .
Why does capacitance C = ε o x W L / t o x scale by 1/ k , not 1/ k 2 ? Area W L gives 1/ k 2 , but the thinner oxide t o x / k multiplies back by k , netting 1/ k .
What does τ = C V / I mean in plain words? Time to switch = charge to move ( C V ) divided by the rate you move it ( I ) .
How are τ and f related? Inversely — smaller delay means higher clock frequency, f ∝ 1/ τ .
Give the two forms of power used in the topic. P = I V (per device) and P d y n = 2 1 C V 2 f (dynamic).
What is V t h and why does the ( V − V t h ) headroom matter? The gate voltage that turns the transistor on; drive current depends on ( V − V t h ) 2 , so voltage can't fall below a floor above V t h .
What is k T / q and why does it end Dennard scaling? Thermal voltage ≈ 26 mV; it floors V t h because lower V t h causes exponential leakage I l e ak ∝ e − V t h / ( n k T / q ) .
Ready? Head back to the parent note and the derivation will read like sentences, not symbols.