VLSI · Power Delivery

Power Integrity

Power Integrity (PI) is the practice of verifying that every transistor on a chip receives stable, sufficient voltage to operate at its intended performance and reliability level, across the full path from voltage regulator to PCB to package to on-die routing.

Overview

A power-delivery network (PDN) is the complex chain of interconnect — regulator, board, package, and on-die metal — that delivers current from the source to every transistor. As process nodes shrink, wire widths shrink and resistance rises, making supply voltage drop and noise one of the most critical signoff concerns in modern chip design.

Power integrity is increasingly treated as a chip-package-system (CPS) problem: accurately assessing PI requires modeling the electrical interaction between the die, the package, and the board together, not the chip in isolation.

Power Delivery Network (PDN)

⤢ Click to zoom Power ring Straps Rails Die / core area Rails — zoomed view (standard-cell power/ground rails) VDD VSS

Simplified on-die PDN: a peripheral power ring and cross-die straps carry bulk current at low resistance; fine-grain rails (zoomed) deliver current directly to each standard-cell row.

The PDN is a hierarchical grid of metal layers connected by vias, distributing power from package pins/bumps down to individual standard cells. It typically consists of:

Rings & Straps

Wide upper-metal power rings around the core and straps running across the die to carry bulk current with low resistance.

Rails

Fine-grain lower-metal power rails that run alongside standard cell rows, delivering current to individual gates.

IR Drop

IR drop is the voltage loss that occurs as current flows through the resistive PDN. When the voltage at a cell's supply pin falls below its expected level, the cell's drive strength and switching speed degrade, which can miss timing or switching thresholds and potentially cause functional failure. IR drop analysis is considered the most critical part of power integrity signoff.

Ohm's law (per-segment IR drop) ΔV = I · Rwire  —  and for a full path, ΔVtotal = ∑ Ik · Rk across each resistive segment from source to cell

A cell far from the source sees the accumulated drop of every segment along its path — which is why cells in the middle of a large block, farthest from the power ring, are typically the worst IR-drop offenders.

⤢ Click to zoom VDD source Lowest voltage Distance from power source Voltage V nominal (1.0) V min (functional)

Left: voltage sags (blue → red) with distance from the power source across the die. Right: the same sag plotted as voltage vs. distance — cells farthest from the source risk dropping below the minimum functional voltage.

Static IR Drop

Caused by average voltage deviation across the resistive power grid under steady-state current draw. Easier to compute, used for early and broad signoff checks.

Dynamic IR Drop

Caused by localized, time-varying switching activity and current surges. Requires time-domain analysis that models grid resistance, capacitance, and inductance together with switching current sources — a more accurate but more expensive analysis.

Decoupling Capacitors (Decaps)

A decoupling capacitor (decap) is placed as a small local charge reservoir directly at or near a standard cell's supply pins. When a gate switches, it draws a burst of current; without a nearby decap, that current has to travel all the way back through the resistive/inductive PDN to the source, sagging the local supply. With a decap in parallel, most of that instantaneous current is supplied locally, so the rail barely moves.

⤢ Click to zoom Without Decoupling Cap VDD VDD − δVDD PMOS A (sags — no local charge) NMOS VSS VSS + δVSS C₀ load With Decoupling Cap VDD VDD (stable) PMOS A (held steady by Cₑ) NMOS VSS VSS (stable) C₀ load Cᴇ decap charge flows to A during switching

Without a nearby decap, switching current must travel back through the resistive PDN, sagging the local VDD/VSS rails at node A. A decoupling cap (Cᴇ) placed close to the cell supplies that charge locally, holding the supply steady.

Decap sizing (first order) Cdecap ≥ ΔI · Δt ⁄ ΔVallowed  —  equivalently, target impedance Ztarget = ΔVallowed ⁄ ΔImax

Where ΔI is the current step during switching, Δt is the response time before the PDN/regulator can react, and ΔVallowed is the maximum tolerable voltage ripple. Designers budget decap so the PDN's impedance stays below Ztarget across the relevant frequency range.

⤢ Click to zoom Switching current demand Time Voltage Nominal VDD Without decap With decap (smoothed)

Decoupling capacitors supply local charge during a switching current spike, keeping the voltage dip shallow (teal) compared to an undecoupled supply (red) that can sag below the functional minimum.

Decoupling capacitors are placed throughout the design to keep transistors supplied with stable voltage despite switching noise and long-distance power routing. They act as local charge reservoirs that:

Smooth voltage spikes

Absorb sudden current demand from nearby switching logic before it propagates as noise.

Reduce IR drop & Ldi/dt

Supply localized charge during peak switching activity, reducing both resistive drop and inductive transients.

Electromigration (EM)

Electromigration is the gradual displacement of metal atoms in interconnect caused by sustained high current density: the "electron wind" physically pushes metal ions along the wire over time, eventually opening a void on one end and piling up a hillock on the other. Left unchecked this leads to opens, shorts, or increased resistance — a slow, cumulative failure mode rather than an instant one, which is why it is budgeted for the chip's expected lifetime (often 10 years) rather than caught by a single functional test.

Black's equation (median time to failure) MTTF = A · J−n · e(Ea ⁄ kT)

J is current density, Ea is the activation energy of the metal (material-dependent), T is temperature, and A, n are process-derived constants. Higher current density and higher temperature both shorten expected wire lifetime — which is why EM limits tighten sharply near hot spots, and why modern signoff increasingly folds in self-heating rather than assuming a flat ambient temperature.

Power and signal EM analysis are required for foundry signoff at every advanced node, and increasingly include self-heating effects and statistical EM budgeting rather than fixed worst-case margins.

Analysis & Signoff

CheckPurpose
Static IR DropVerifies average voltage drop across the PDN under nominal current draw
Dynamic IR DropVerifies voltage drop under realistic, time-varying switching scenarios (vectored or vectorless)
Electromigration (EM)Verifies current density in power/signal wires stays within foundry-qualified limits
Decap BudgetingEnsures sufficient decoupling capacitance is placed to meet dynamic voltage targets
Chip-Package-System (CPS) Co-analysisExtends PI analysis across die, package, and board together

Subtopics

EMIR Analysis Ready

Static & dynamic analysis, Power/Signal EM, package co-analysis, Sigma-AV/DVD, ROM, and multi-die signoff, in depth.

UPF Coming soon

Unified Power Format — describing power intent, domains, and low-power design constraints.

Sources