Electromigration reliability: Question bank — Hardware deep dive

Before we start, one shared picture in words: a wire is a corridor of metal atoms; a strong current is a wind of electrons that shoves those atoms along. Nothing breaks unless atoms pile up or drain away somewhere — that "somewhere" is always a spot where the atom flow changes (a divergence). Keep that sentence in mind; most traps below are just it in disguise.

Definition The symbols this page uses (define once, use freely below)

Before any trap mentions them, here is every symbol on this page in plain words and units:

$J$ = current density, the current squeezed through each unit of cross-sectional area (units $A/cm^{2}$ ). Its sign/direction is that of conventional current — the direction a positive charge would flow. Electrons flow the opposite way, so the electron wind pushes atoms opposite to $J$ .
$L$ = wire length (units $cm$ ), the distance current travels along the interconnect. It teams with $J$ to form the Blech product $J \cdot L$ that decides short-line immortality.
$E$ = electric field inside the metal (units $V/cm$ ). It points along $J$ (same direction), because inside a resistive wire the field is what drives the current forward. It is not an outside applied field — it is the internal field that Ohm's law ties to the current.
$ρ$ = resistivity of the metal (units $Ω \cdot cm$ ): how strongly the material resists current. Ohm's law in local form says $E = ρ J$ — a bigger field is needed to push the same current through a more resistive metal.
$q$ = elementary charge ( $1.6 \times 1 0^{- 19}$ coulombs), the size of one unit of electric charge; it converts "field" into "force on a charged ion."
$Z^{*}$ = effective charge, a unit-less lumped number bundling the direct field push and the (dominant) electron-wind push. Its sign is usually set by the wind term, so it points atoms toward the anode.
$k_{B}$ = Boltzmann's constant ( $8.617 \times 1 0^{- 5}$ eV/K), the bridge between temperature and energy — it turns "how hot" ( $T$ in kelvin) into "how much thermal jiggle energy" atoms have.
$E_{a}$ = activation energy (in eV): the energy hill an atom must climb to hop to a new site.
$D$ = diffusivity (how easily atoms wander, units $cm^{2} / s$ ); $N$ = atomic density (atoms per unit volume). More mobile, more crowded atoms → larger atom flux.
$MTTF$ = Mean Time To Failure (units of time, e.g. years) — the average lifetime before a wire fails by EM. This is the quantity Black's equation predicts; see MTTF and FIT rates.
$n$ = current exponent in Black's equation (a unit-less power); $A$ = a process/geometry constant; $T$ = absolute temperature in kelvin.

The three pictures below ground the traps: s01 shows flux divergence, s02 shows void vs. hillock, and s03 shows Blech back-stress. Each figure carries its own label in the top corner so the in-text references ("figure s01", etc.) point unambiguously.

True or false — justify

Metal atoms drift toward the cathode because the electric field pulls the positive ions there.

False. The electron-wind force dominates in good conductors, so net atom motion is toward the anode — the same direction electrons flow, opposite to conventional current

J

A perfectly uniform wire with uniform current density will still eventually fail by EM in its middle.

False in the idealized case. Uniform flux means mass-in equals mass-out everywhere (look at the balanced arrows in figure s01), so nothing accumulates — failure needs a flux divergence (a via, grain triple-point, width step, or barrier), which real wires always have.

Cooling a chip always extends electromigration lifetime.

Not always. The Arrhenius reliability model term does improve, but Joule heating & self-heating can keep local

T

high, and low temperature can worsen Stress migration and thermomechanical stress that create new divergences.

Copper interconnects are effectively immune to electromigration.

False. Copper is more resistant (higher

E_{a}

), not immune; it migrates along surfaces and interfaces (Cu–cap, Cu–barrier), which is why cap layers like CoWP matter — see Copper damascene process.

The exponent

n = 2

in Black's equation is a universal physical constant.

False.

n \approx 2

only when void growth limits life;

n \approx 1

when void nucleation limits it. Picking the wrong

n

can misjudge lifetime by orders of magnitude.

A short enough wire can carry huge current density forever without EM failure.

True — this is the Blech effect (figure s03). When

J \cdot L

(current density times wire length

L

) is below a critical product, back-stress from piled-up atoms cancels the electron wind before a void grows.

In Black's equation the exponential is

e^{+ E_{a} / k_{B} T}

, and a positive sign there means higher temperature gives longer life.

False. Here

k_{B}

(Boltzmann's constant) links

T

to energy; a positive exponent that shrinks as $T$ rises means the Mean Time To Failure (MTTF) drops with temperature — exactly what we want, since hotter atoms diffuse faster (see MTTF and FIT rates).

Voids and hillocks are just two names for the same defect.

False. A void is atom depletion → an open circuit (cathode/upstream side); a hillock is atom pile-up → a short to a neighbor (anode side) — the two ends of figure s02.

Spot the error

"To double a wire's lifetime, just double its width so current density halves — since MTTF is proportional to

1/ J

The proportionality is

J^{- n}

, not

J^{- 1}

. With

n \approx 2

, halving

J

multiplies life by

2^{2} = 4

, not

2

— the person quietly assumed

n = 1

"EM current limits and RC-delay limits are unrelated design rules, so optimize them separately."

They share the same wire geometry. Narrowing a wire raises

J

(worse EM) and raises resistance (worse Interconnect RC delay); a width choice trades both at once — see Design rules & current density limits.

"Since

E = ρ J

, the driving force

F = Z^{*} q ρ J

shows the force comes from resistivity, so a lower-resistivity metal always has less EM force."

Force per ion drops with

ρ

, but lifetime depends on the whole flux law including diffusivity

D

, atomic density

N

E_{a}

, and the failure geometry. Low

ρ

helps the field term, yet interface diffusion can still dominate (as in copper).

"Because atoms flow toward the anode, we should worry about hillocks (pile-ups) first — that's where the wire snaps open."

A snap-open is a void, which forms on the cathode/upstream side where atoms drain away. Hillocks cause shorts, a different failure mode.

"The effective charge

Z^{*}

is literally the ionic charge of the metal atom."

No —

Z^{*}

is a lumped effective factor bundling the direct field force and the (dominant) electron-wind momentum transfer. It is often large and its sign is set by the wind term, not the true ionic charge.

"Blech immortality means a short wire feels no EM force at all."

The force is still there; it is balanced by the growing back-stress before a void can nucleate. Remove the current and both relax — immortality is an equilibrium, not an absence of force.

Why questions

Why does electromigration get harder to manage at each smaller technology node?

Wires narrow while current stays similar or rises, so current density $J$ climbs, and since MTTF

\propto J^{- n}

, lifetime collapses steeply.

Why is a divergence in atomic flux the true villain, not the flux itself?

Uniform flux moves the same mass in and out of every region, so nothing accumulates; only where the flux changes do atoms build up (hillock) or deplete (void) — figure s01.

Why does resistivity

ρ

appear in the EM driving force at all?

Because designers set current density

J

, not the internal field

E

; Ohm's law

E = ρ J

translates one into the other, so

ρ

is the conversion factor that lets

F = Z^{*} q ρ J

be written in terms of

J

Why does Black's equation multiply an Arrhenius temperature term by a power of

J

instead of using only temperature?

The electron wind (current-driven force) sets how hard atoms are pushed, while temperature (via

k_{B} T

) sets how easily they diffuse — EM needs both, so the law combines a

J^{- n}

mechanical factor with the

e^{E_{a} / k_{B} T}

thermal factor.

Why do vias, grain-boundary triple points, and diffusion barriers show up as EM hot-spots?

Each is a place where the atom flux cannot continue unchanged — flow enters at one rate and leaves at another — producing the accumulation or depletion that seeds hillocks and voids.

Why can't we simply lower the current forever to guarantee no EM?

Lowering current also lowers performance and may not be an option; instead designers combine current caps, short-line (Blech) routing, temperature control, and material choices — no single knob is free.

Edge cases

What happens to EM risk when the current density

J

approaches zero?

The driving force

F = Z^{*} q ρ J \to 0

, so the electron wind vanishes and EM effectively stops — though residual Stress migration from thermal stress can still move atoms without any current.

At exactly the critical Blech product

(J \cdot L)_{cr i t}

, is the wire immortal or failing?

It sits at the marginal boundary — back-stress just balances the wind (figure s03), so no net void growth in theory; in practice designers stay safely below

(J L)_{cr i t}

because real geometry and temperature vary.

A jumper is short enough to be Blech-immortal, but connects to a long line — is the whole path safe?

No. Immortality is per-segment; the long line can still fail at its own divergences. A short piece protects only itself, not its neighbors.

If two identical wires differ only in that one runs

2 0^{\circ} C

hotter, which fails first and by roughly how much sooner?

The hotter wire, because the Arrhenius term

e^{E_{a} / k_{B} T}

makes diffusion faster; near

350

K with

E_{a} = 0.7

eV the ratio is a bit over three-fold shorter life.

When void nucleation dominates (

n \approx 1

), does doubling current hurt more or less than in the void-growth regime (

n \approx 2

Less — with

n = 1

doubling

J

halves life, whereas with

n = 2

it quarters life. The lower exponent means current density is a weaker lever in the nucleation-limited case.

Recall One-line self-test before you leave

Which single sentence explains the most traps on this page? ::: "Nothing fails unless the atom flux changes somewhere, and the electron wind — not the field — sets the direction toward the anode." Every void/hillock, Blech, and copper trap is a corollary of that line.