White papers, technical articles, and tutorials from Anasim on power delivery for AI silicon.
Foundational work from over two decades of Anasim power integrity research — books, peer literature, PI-FP simulation studies, and a 40+ patent portfolio.
Raj Nair and Donald Bennett
Prentice-Hall PTR Signal Integrity Series, 2010. The first comprehensive treatment of power integrity from chip to system, covering droop theory, continuum simulation, and active noise regulation.
Amazon →
Masanori Hashimoto and Raj Nair
McGraw Hill, 2014. Advances the field into nanoscale processes covering on-chip resonances, advanced PDN modeling, and power integrity for 3D ICs and TSV assemblies.
Amazon →R. Mahajan, Raj Nair et al. — Intel Technology Journal, 2002
Raj Nair — Intel Assembly & Test Technology Journal, 2001
Raj Nair and Donald Bennett, Anasim Corp. — EETimes, March 2008
Raj Nair, Anasim Corp. — EDN / SOCcentral, 2008–2010
Raj Nair, Anasim Corp. — 2008
Clock skew splits a load spike into two peaks; true-physical simulation reveals counterintuitive 66% noise increase.
Continuum simulation demonstrating frequency-dependent noise amplification of ~20% due to local resonances.
Load vector phase differences impact DVD by up to 8%. A critique of vectorless analysis limitations.
40+ US and international patents across power delivery, packaging, IC design, and active noise regulation.
Original, rigorous studies of power integrity in modern AI silicon — built on PDNLab simulation.
Two-scenario power-integrity analysis of an NVIDIA® H100 SM: lockstep sub-core firing vs. an outer-then-inner spatial slice excitation experiment, showing how workload allocation reshapes on-die dynamic voltage droop.
A complete full-stack PDNLab™ model of the NVIDIA® H100 — board through package to all 144 SM cores — visualizing, pre-silicon, how cumulative voltage droops form and propagate through a multi-core 3D power delivery network under realistic workloads.
Silent data corruption in AI data centers traces back to voltage droop — and traditional lumped PDN models miss the physics that produces it. Why a Maxwell-accurate, full-stack simulation changes what you can predict before tapeout.
Traditional power optimization — macro placement, aspect ratios, interconnect-length minimization, and thermal islands — minimizes energy without ever evaluating dynamic supply noise. Where floorplan-stage power integrity fits.
How power planning for a modern GPU or accelerator is actually thought about — from a spatial power map of cores, tensor engines, cache and HBM PHYs, through voltage domains and the impedance budget, to hardening the supply against the coherent di/dt that defines accelerator silicon.
Read articleThe full path from an 18-layer copper fabric to a two-layer transmission-line model — the physical hierarchy, the Innovus power-planning flow that builds it, and every Π-FP parameter mapped to a real power-planning knob, with the manual’s own equations.
Read articleA real on-chip power distribution network is a dozen-layer mess. To make full-chip dynamic simulation tractable, the power grid is reduced to an equivalent two-layer symmetric transmission-line grid —…
Read articleBetween the bare silicon die and the printed circuit board sit two layers that rarely get named in a block diagram but determine how a modern chip is powered, connected, and held together: the package…
Read articleA large language model's computation reduces almost entirely to matrix multiplication, and a frontier accelerator's compute units do one thing: multiply tiles of two matrices and accumulate the result…
Read articleA systolic array is a grid of identical processing elements through which data is pumped one step per clock cycle, with each element performing a single multiply-accumulate as the data passes.
Read articleThe systolic array is the dominant compute structure in AI silicon because it maximizes arithmetic per unit of memory traffic.
Read articleA large language model serving a request is not running one computation. It is running two — and they could hardly be more different. The first reads the prompt; the second writes the answer.
Read articleAn AI core has no single droop. Its droop surface is the convolution of the physical power grid with a specific workload's spatiotemporal current, and it changes with every workload the core executes.…
Read articleWhat you can actually tune in a power-grid model — wire width, pitch, sheet resistance, inductance, and capacitance — what each one physically does, and why damping turns the grid into an R–L–C system rather than a resistor.
Read articleAt modern switching speeds the grid stops acting like wiring and becomes a continuous electromagnetic medium — storing energy in fields, carrying disturbances as waves, and coupling every conductor to its neighbors. A first-principles explanation.
Read articleFrom single aluminum traces to upper-metal meshes, flip-chip feed, on-die decap, and backside power delivery — how on-chip power distribution became one of the hardest problems on the die.
Read articleA recap of Raj Nair’s 2008 EE Times argument — behind the Power Wall sits a Power Integrity Wall, with inductive L·di/dt noise doubling every process generation. The Roots-of-Two derivation, and why AI silicon walked into it.
Read articleA walkthrough of Jiang et al. (HPCA 2026, UT Austin & AMD): how LLM inference excites GPU PDN resonance, their ladder RLC model, and the result where a 10 W swing at resonance beats a 100 W swing off it.
Read articleEvery chip is full of capacitors — the local charge reservoir that holds the supply voltage up the instant millions of transistors switch. What on-die capacitance is, the types, and how it delivers charge.
Read articleInside a modern chip program — the team, the lifecycle, and the experienced power-delivery engineer who dictates how current reaches every transistor and asks the questions that keep a superchip’s power alive.
Read articlePower integrity is a pipeline, not one tool — and each EDA tool owns one stage. A breakdown vendor by vendor: RTL power, extraction, on-die IR, package PDN impedance, SPICE, electrothermal — and where the methodology stops.
Read articleA direct breakdown of modern advanced packaging — CoWoS, EMIB, InFO, TSVs, microbumps, hybrid bonding, Foveros, SoIC, HBM, and UCIe. What each is, how it connects dies, and who uses it.
Read articleHow a die connects to its package, and the key to keeping it straight: which way the active face points. A diagram-led explanation of face-up vs face-down, the C4 bump, and why the chip got flipped over.
Read article~25 kW into one 21.5 cm silicon square — about 25,000 A at 1 V. Why edge delivery is impossible at wafer scale, how power is fed vertically across the whole wafer, the CTE-mismatch connector, and why it forces the continuum view of PI.
Read articleFor modern AI chips the limiter is rarely compute — it's memory. Why moving data sets the pace: bandwidth vs capacity, bandwidth-bound decode, the KV cache, the SRAM scaling cliff, HBM and DRAM — and why it's an energy problem.
Read articleA progressive ten-lesson journey through the physics and engineering of on-chip power delivery — from IR-drop fundamentals through rogue-wave phenomena to sign-off methodology. Everything an engineer needs to understand why modern chips fail and how to prevent it.
Working on a power integrity problem? We scope engagements around a specific engineering question.
Talk to Anasim →