Lumees predictive PDKs

Five kits. 40 nm planar to 2 nm nanosheet.

Five process design kits, from 40 nm planar to 2 nm nanosheet, built from public data, with every number traceable to its source.

The five Lumees kits: each transistor across its width, at the same scale
The five Lumees kits: each transistor across its width, at the same scale

A process design kit (PDK) is everything a chip designer needs to build on a technology: transistor models, standard-cell libraries, timing and power data, wiring rules and resistance and capacitance of every metal layer. Foundry PDKs are confidential, tied to one process and usually unavailable for research and tool development.

Lumees Lab builds its own predictive PDKs: complete, realistic kits for five technology generations, derived from published roadmaps and process papers. They are the technology foundation for our tools, including Lumina Spice (circuit simulation). We use them to develop, test and calibrate those tools on full designs without depending on any foundry.

The five kits span the four transistor architectures of the last fifteen years: planar bulk, fully depleted silicon-on-insulator, FinFET and gate-all-around nanosheets. Together they let us study how design behaviour changes from one generation to the next, using one method throughout.

Gate pitch, tightest metal pitch, cell height and supply across the family
Gate pitch, tightest metal pitch, cell height and supply across the family

How we build them

One cited description per kit. Each kit starts from a single technology description. Every value in it (supply voltage, gate pitch, oxide thickness, metal pitches, transistor targets) carries its origin:

  • cited: taken from a roadmap (ITRS, IRDS), a published process paper, or a public research kit;
  • derived: computed from cited values by a numbered, documented rule;
  • open: not yet sourced, shown as such and tracked until it is.

Nothing is a silent default.

Share of cited, derived and open values per kit
Share of cited, derived and open values per kit

Generated, not hand-drawn. Transistor models are fitted to the cited targets. Standard cells are placed and routed by our own generator, which follows each kit's layout rules. Timing and power libraries are characterized from the netlists of the generated layouts, including their wiring parasitics.

Checked against independent references. Each kit is compared with published data it was not fitted to. Every library is checked cell by cell: logic function extracted from the drawn layout, layout rules, and timing against transistor-level simulation of real logic paths. Every error these checks have found is documented with its cause and fix.

Honest by design. These are predictive kits: realistic and internally consistent, but not calibrated to any foundry's silicon, and not a model of any specific commercial process.

What each kit contains

  • Device models: BSIM4 (planar), BSIM-IMG (FD-SOI, with the back gate) and BSIM-CMG (FinFET, nanosheet), with process corners
  • Standard-cell libraries: logic, sequential and physical cells, several threshold-voltage flavours and cell heights, drive strengths X1–X8
  • Timing and power libraries (Liberty) across process, voltage and temperature corners
  • Interconnect stack: every metal layer with size-effect resistivity and capacitance tables
  • Design views: GDS layouts, LEF abstracts, technology LEF, CDL and SPICE netlists, Verilog models, layer maps
  • A technical sheet per kit, generated from the kit's own data

Why wires matter more every generation

Resistance of the narrowest routing metal per node
Resistance of the narrowest routing metal per node

As wires shrink, their resistance per micrometre rises steeply: the metal cross-section falls and electrons scatter at the surfaces and grain boundaries. At 6 and 2 nm, many paths are limited by wires rather than transistors. Our kits model this size effect for every layer, so timing on the kits sees it too.

Research built on the kits

Each kit carries one featured analysis:

KitFeatured analysis
LM40PB-LAging: transistor wear-out and aged timing libraries
LM22FD-BBack-bias compensation of aging and variation
LM12FF-GSelf-heating and temperature inversion
LM06FF-HVariability (statistical timing) and wire-dominated paths
LM02NS-HThe 2 nm kit, its method, and backside power delivery

Privacy settings

Choose what this site may use. You can change your mind at any time from the Privacy settings link at the bottom of every page.

What the site needs to work, and the record of the choice you make here, kept in your browser. No cookies are set.

Always on

Would let us count visits and see which pages are useful, without advertising or cross-site tracking. We do not use any analytics today, so nothing is loaded whichever you choose.

Cookie policy · Privacy policy