Modelling of plasma physics — the two ends being fusion plasma (hot, magnetically/inertially confined; GENE/XGC/SOLPS/SOL-KiT, mostly free national-lab codes) and low-temperature plasma (LTP) for semiconductor manufacturing (etch/deposition; HPEM/COMSOL/CFD-ACE+). The codes are disjoint ecosystems (a fusion model won’t run a wafer chamber), though the underlying kinetic machinery (Boltzmann/Vlasov, PIC-MCC, non-Maxwellian EEDF) overlaps — the bridge Zenithon’s team exploits via the tokamak scrape-off-layer (see).
The market sub-cluster (consolidation-by-acquisition)
Standalone plasma-sim software is a small, consolidated, acqui-buy-shaped niche — every credible solver asset has been bought by an OEM/EDA player:
- Lam ← Coventor (2017), Esgee/VizGlow plasma sim (2022)
- Applied Materials ← CFD-ACE+ from ESI (2022)
- Silvaco ← Tech-X (2025)
- Synopsys ← Ansys ($35B, 2025)
Quantemol is the cautionary base rate: a standalone plasma-sim/chemistry-data business that stayed tiny and unfunded for ~20 years. The Zenithon Seed To A Strategy thesis is that the ML-surrogate layer on top (neural operators + proprietary fab-calibration data) is unbuilt white space the OEMs haven’t shipped — and is what re-rates the category from acqui-buy tool to venture-scale platform.
Players
- ML-surrogate layer: Zenithon (plasma process), SandBox Semiconductor (etch/dep recipe opt).
- Physics-based / incumbent: COMSOL (plasma module), Lam Semiverse (SEMulator3D — physics-based, not ML), Quantemol (QDB/QEC chemistry data on Kushner’s HPEM).