A modern chip loses more time and energy moving a bit across its flat surface than it does computing that bit — because a wire's delay grows with its length squared . 3D stacking fixes this by turning a long horizontal trip into a short vertical hop through a metal-filled tunnel (a TSV), and the whole parent note is just the physics of that tunnel and why the vertical trip is so much cheaper.
This page assumes you know nothing . Every letter, ratio, and squiggle used in the parent topic is built here from the ground up, in the order that lets each one lean on the one before it.
Definition How to read the "::=" sign on this page
Whenever you see X ::= Y , read it as ==" X is defined to mean Y "==. The "::=" is just a tidy "is defined as" arrow — the left side is the new word, the right side is its plain-words meaning. It is not maths you compute; it is a dictionary entry.
Before any symbol, hold this image: a chip is a flat square of silicon covered in tiny switches (transistors) and thin metal wires connecting them. 3D stacking puts a second such square directly on top , and drills tunnels straight down to connect them.
A die ::= one rectangular piece of silicon cut from a wafer, carrying transistors and wiring. Think of it as one floor of a building . "2D" = everything on one floor; "3D stacking" = several floors, one above the other.
Everything below is the vocabulary needed to describe how the elevator between floors works and why the elevator wins .
Definition The micrometre
μ m (micrometre, "micron") ::= one millionth of a metre, i.e. 1 μ m = 1 0 − 6 m . The Greek letter μ ("mu") is the standard shorthand for "micro" = one millionth.
Common mistake The letter "m" means two different things — watch the context
Why it trips people: in 1 mm = 1000 μ m the first "m" is the prefix milli (a multiplier, 1 0 − 3 ) and the second "m" is the unit metre (the thing being measured). So "mm" literally reads "milli-metre." Rule of thumb: a lone "m" as a prefix in front of another unit = milli; a lone "m" standing as the unit itself = metre. On this page m (upright) always means metre and m as a prefix always means milli.
Intuition What a micron looks like
A human hair is about 70 μ m thick. A TSV tunnel is about 5 μ m wide and 50 μ m tall — thinner than a hair, roughly as tall as a hair is wide. A cross-chip wire runs millimetres : a typical cross-die route is about 5 mm = 5000 μ m . Compared with a 50 μ m TSV that is 5000/50 = 100× shorter for the vertical hop. (Even a modest 1 mm wire is already 20 × longer.) Hold that ratio; the whole topic pays off from it.
Powers of ten you will meet, smallest (most negative exponent) first :
Symbol
Name
Meaning
Where it shows up
f
femto
1 0 − 15
capacitance (fF)
p
pico
1 0 − 12
capacitance (pF), delays
n
nano
1 0 − 9
ordinary metal vias, transistor sizes
μ
micro
1 0 − 6
TSV width & height
m
milli
1 0 − 3
full chip width L
These prefixes are just stickers on the front of a number telling you which power of ten to multiply by. Nothing more mysterious.
The parent note repeatedly says one thing "grows like" another. That phrase means: if I make one quantity bigger, how does the other respond? Three response-shapes appear.
Definition "Proportional to" — the symbol
∝
y ∝ x ::= "y is proportional to x " = if you double x , y doubles too. It is a straight line through the origin . It hides the constant multiplier because we only care about the shape of the response.
x 2 "
y ∝ x 2 ::= double x and y becomes four times bigger (because 2 2 = 4 ). The little raised 2 means "multiply the thing by itself once": x 2 = x × x . This is the upward-curving line in the figure. Wire delay does this — that curve is the villain the whole topic defeats.
Definition The natural logarithm —
ln
ln ( z ) ::= the very slow-growing curve: to make ln ( z ) go up by 1, you must multiply z by about 2.718. So ln answers "how many times did I multiply?" It appears in the TSV capacitance formula as ln ( b / a ) , and because it grows so slowly, ==the exact ratio b / a barely matters== — a comforting fact when you build real tunnels.
Recall Why do we care that
ln grows slowly here?
Because it sits in the denominator of the capacitance formula. A slowly-changing denominator means capacitance is forgiving of small manufacturing variation in the oxide thickness. ::: Slow-growing denominator → stable, predictable capacitance.
Every symbol in the physics derivations comes from four everyday electrical ideas. Picture water in pipes.
Definition The four basics
Charge Q (or a line-density λ ) ::= how much electricity, measured in coulombs. Picture a bucket of water . λ ("lambda") is charge spread along a line — buckets per metre.
Voltage V ::= electrical pressure — the push that moves charge. Picture water pressure (height of a tank). Measured in volts.
Current ::= flow of charge, buckets per second. Picture water rushing through the pipe.
Electric field E field ::= how steeply voltage changes across space — the slope of the pressure . Picture how fast the water surface tilts. Strong field = steep tilt = charge feels a big shove.
E field (field) vs E (energy) — do not confuse them
Why it trips people: later we meet energy per switch written E = 2 1 C V 2 . That E is energy , a totally different quantity from the electric field in this section. On this page the field always carries a subscript, E field , and a bare E always means energy. The parent note reuses the plain letter E for both — now you know which is which by context (a field has a direction and a slope ; an energy is a single number of joules ).
Intuition Why the topic needs these
A TSV carries a signal, which is just voltage V switching between "high" (a 1) and "low" (a 0). To switch it you must pour charge in and out . How much charge each switch costs is set by capacitance — the next and most important symbol.
C ::= how much charge you must pour in to raise the voltage by one volt: C = Q / V . Picture the cross-sectional area of a water tank : a wide tank needs lots of water to raise its level by a bit (big C ); a thin tank fills fast (small C ). Measured in farads (F); real chip values are femtofarads, fF = 1 0 − 15 F.
The figure below is this definition drawn as two water tanks — the narrow tank (small C , fills with little charge) versus the wide tank (big C , guzzles charge for the same rise). Read it left-to-right before continuing; every later "big tank / tiny tank" phrase points back to this picture.
R
R ::= how much a wire fights the flow of charge — a narrow, long pipe resists more. For a wire, R grows with its length ℓ : longer wire, more fight.
Definition The RC product and delay
τ
When resistance R feeds a capacitance C , the tank fills on a timescale τ = R C ("tau", the Greek t , used for "time constant"). Big R (slow flow) or big C (big tank) → slow. This product R C is the delay of the wire.
Recall If a wire's length triples, how does its RC delay change?
× 9 . ::: Because delay ∝ ℓ 2 and 3 2 = 9 .
The TSV is modelled as three nested cylinders. Naming their radii lets the formulas be exact.
Definition The tunnel's letters
a ::= radius of the copper core (the wire itself).
b ::= radius out to the edge of the oxide liner (insulating sleeve). So the oxide fills the ring between a and b .
h ::= the height of the TSV = how deep it goes through the silicon (this is the ℓ of §5 for the vertical path).
L ::= the side length of the whole square die (millimetres).
p ::= the pitch = centre-to-centre spacing between neighbouring connections. Small pitch = tightly packed = more connections.
a and b appear as a ratio b / a
Physics of a coaxial (nested-cylinder) capacitor only cares about the proportion of outer to inner radius, not their absolute size — doubling both changes nothing about the field shape. That is why the formula holds ln ( b / a ) , a ratio inside a log : thicker oxide (bigger b / a ) → smaller capacitance → faster, cheaper tunnel.
ε
ε ("epsilon") ::= how easily the insulating material lets an electric field pass — its "electrical stretchiness." Written ε = ε r ε 0 : a material factor ε r (for oxide ≈ 3.9 ) times a universal constant ε 0 = 8.85 × 1 0 − 12 F/m. More ε → bigger capacitance (roomier tank). It is the material knob in the capacitance formula.
Definition Perimeter and area
Perimeter of the square die = 4 L (four edges) — a 1-dimensional border, grows like L .
Area of the die = L 2 — a 2-dimensional surface, grows like L 2 (the same curve as §2).
Intuition The counting that makes HBM possible
A flat chip sends signals out through its edge only, so the number of connections N 2 D = 4 L / p — it grows like L . Stack a die on top and you can put tunnels over the whole face , so N 3 D = L 2 / p 2 — grows like L 2 . For a 10 mm die at 40 μ m pitch that is 1000 edge links versus 62 , 500 face links: ~60× more parallel wires . Wide parallel buses at modest speed = huge bandwidth = exactly how High Bandwidth Memory (HBM) works.
Definition Fourier heat-flow letters
Q heat ::= heat power flowing (watts). We subscript it because Q elsewhere meant charge ; here it is heat .
k ::= thermal conductivity — how well a material carries heat (silicon is good, oxide is poor).
A ::= the cross-section area heat flows through.
t ::= the thickness the heat must cross (a taller stack = bigger t ).
Δ T ::= the temperature rise (Δ , "delta", means "the change in"). This is how hot the buried layer gets.
Units and prefixes um fF pF
Capacitance C equals Q over V
Growth shapes proportional square and ln
RC delay grows like length squared
Switching energy half C V squared
Tunnel geometry a b h ratio
TSV coax capacitance formula
Perimeter 4L vs Area L squared
Bandwidth area beats edge
Fourier heat flow delta T
Each foundation box feeds the topic exactly where the parent note uses it: growth-shapes + R + C make the delay argument, geometry + permittivity make the capacitance formula, perimeter-vs-area makes the bandwidth argument, and heat is the counterweight that limits it all.
The delay story continues in Interconnect RC Delay ; the "why did wires stop shrinking" backstory is Moore's Law ; the side-by-side cousin of stacking is Interposers and 2.5D Integration ; mixing dies from different processes is Chiplets and Heterogeneous Integration ; the heat wall is detailed in Thermal Management in ICs ; and how it all gets sealed into a product is Chip Packaging .
Self-test: can you say each aloud before revealing?
How to read the "::=" sign "is defined as" — the left side is the new word, the right side its plain-words meaning.
1 μ m in metres1 0 − 6 m (one millionth of a metre).
The two meanings of the letter "m" prefix milli (1 0 − 3 ) when in front of a unit; the unit metre when standing alone.
What y ∝ x 2 means for doubling x y becomes 4× larger.
Why ln appears in the TSV formula and why we don't sweat its exact value it grows very slowly, so capacitance is forgiving of oxide-ratio variation.
Capacitance in one sentence charge needed per volt of rise, C = Q / V ; the "tank's cross-section."
The difference between E field and E E field is the electric field (a slope with direction); bare E is energy (a number of joules).
Energy to switch a capacitor E = 2 1 C V 2 ; the 2 1 is the average back-pressure discount.
The TSV capacitance formula C TSV = ln ( b / a ) 2 π ε h — from nested cylinders of radii a , b and height h .
Why wire delay grows like length squared both R ∝ ℓ and C ∝ ℓ , and τ = R C ∝ ℓ 2 .
What a , b , h are on a TSV copper-core radius, oxide-edge radius, tunnel height.
Why radii enter as the ratio b / a coaxial physics depends only on the proportion, not absolute size.
Permittivity ε in words how easily an insulator passes an electric field; the material knob for C .
Perimeter vs area scaling edge ∝ L (few links), face ∝ L 2 (many links) → bandwidth win.
Fourier temperature rise Δ T = Q heat t / ( k A ) ; thicker/hotter stacks get hotter → the stacking limit.