CEM (on the NumericalAI platform) is not “Palace alone.” It is a three-layer stack — preprocess, Palace FEM, and postprocess — built for superconducting and quantum-device workflows. HFSS remains the broader commercial EM suite. CEM wins where the job is design → validated EM → circuit Hamiltonian → design assessment, not platform-scale antennas.
Quantum Metal design + YAML → Gmsh mesh + Palace config, with input verification gates before the solve.
Open-source FEM: eigenmode (with built-in EPR), driven S-parameters, electrostatic, magnetostatic, transient — GPU/cloud scalable.
E_C/E_J, EPR & mode ID, Hamiltonian (scqubits), collisions, driven κ/Q_ext, sweeps, design-assessment report.
For a superconducting qubit / resonator chip, both tools can answer EM questions. CEM continues past the field solve into quantum-circuit parameters and a pass/fail design report.
GDS import or HFSS modeler; airbox / JJ lumped sheets set up per project, or scripted via qiskit-metal / PyAEDT.
Adaptive FEM eigenmode / driven; workstation or HPC seat.
Native plots, S-params, eigenfrequencies, Q.
Participation & Hamiltonians via external tooling (qiskit-metal, pyEPR) rather than one built-in step.
Report assembled by the engineer from notebooks / scripts.
Quantum Metal .py + YAML → mesh + Palace JSON; JJ ports, materials, enclosure; verify gates.
Electrostatic · Eigenmode · Driven · Magnetostatic · Transient on cloud/GPU compute.
C → E_C/E_J; eig + EPR; S → κ / Q_ext; convergence & EPR↔C gates.
scqubits spectra, α, g/χ; mode ID; collision warnings.
Design-assessment HTML: targets vs results, mode map, readout, sweeps.
Mapping takeaway: HFSS covers steps 1–3 deeply for general EM. CEM owns steps 1–5 as one product for quantum-device design — especially the post-EM Hamiltonian and validation layer that HFSS handles through external tooling (qiskit-metal, pyEPR) rather than one built-in step.
Teal is CEM territory. Violet-labeled bands are where CEM goes beyond a raw EM solver. Amber is HFSS. Empty bands are honest gaps — CEM isn’t the tool there.
CEM’s teal + violet bands are its core coverage: resonators, filters, transmons, CPW readout, capacitance extraction, S-parameters, participation ratios, scqubits spectra, collision checks, and parametric sweeps with a customer-ready assessment report. Platform-scale antennas, RCS, and hybrid IE/SBR+ remain HFSS territory.
A glossary for HFSS users: where each HFSS step lives in CEM.
Replaces much of the HFSS project setup for Quantum Metal designs: geometry variables, chip/substrate, airbox / outer BC, Josephson lumped ports, materials, solver type, and mesh generation via Gmsh (pyPalace). Verification checks catch bad units, missing materials, port/JJ geometry, and attribute sync before burning GPU hours.
.py → .msh + Palace .jsonSame class of answers as HFSS FEM: eigenfrequencies, driven S-parameters, electrostatic capacitance, magnetostatic currents, transient. Open source (Apache-2.0), massively parallel, GPU-capable — no per-seat license, pay for compute.
This is CEM’s largest functional gain over “Palace alone” and over a bare HFSS project. Palace already exports EPR natively; CEM productizes the rest — mode ID, gates, scqubits, sweeps, and reporting — so teams are not stitching pyEPR + notebooks + ad-hoc scripts after each run.
A side-by-side view for teams evaluating CEM for quantum / superconducting hardware work.
| Capability | CEM (NumericalAI) | Ansys HFSS |
|---|---|---|
| License model | CEM No solver license fee · pay-per-compute on platform | HFSS Commercial per-seat licensingPricing not public; varies by bundle and negotiation |
| FEM eigenmode | Competitive FEM Palace eigenmode; frequencies + QOpen-literature Palace results agree with cryogenic measurement to ~0.3% on resonator freq. (couplings less tight; arXiv:2511.09041) — solver-class evidence, not yet a CEM-specific benchmark. | Gold standard Industry sign-off pedigree for FEM eigenmode |
| Driven / S-parameters | Covered Palace driven + postprocess κ / Q_ext from S-params | Covered Native driven solutions, rich port / report UI |
| Capacitance extraction | Covered Electrostatic → Maxwell C → E_C / E_J handoff | Covered HFSS / Q3D-style paths depending on license stack |
| Josephson ports / EPR | First-class JJ lumped ports from Metal junctions; Palace-native EPR (port-EPR) → mode IDEPR post-processing is built into the solver — not a separate pyEPR project |
Manual / external Lumped RLC in modeler; EPR typically via pyEPR or custom scripts |
| Hamiltonian / spectra | Built-in scqubits adapter (f₀₁, α, g/χ, spectrum) | Not native Requires external quantum tooling |
| Design validation gates | Built-in Mesh/config verify · convergence · EPR↔C · collision warnings | Partial Adaptive mesh convergence; quantum consistency left to user |
| Parametric sweeps | Cloud-native L_J (and job) grids → comparison table / HTML · elastic parallel compute without seat math | Covered Optimetrics / parametric; HPC and distributed options available under commercial seats |
| Customer design report | Built-in Design-assessment HTML (targets, modes, gates, readout) | Manual Plots & exports; narrative report is engineer-built |
| Geometry front end | Quantum Metal path Parametric chip designs + YAML sim setup; not general CAD | Integrated Full modeler + industry meshing for arbitrary 3D |
| IE / MoM / FEBI / SBR+ | Not in CEM FEM-only solver class | Covered Core HFSS differentiator for large platforms |
| Circuit / multiphysics | Quantum circuit Hamiltonian / scqubits — not Ansys circuit or thermal link | Ecosystem Circuit, EMIT, thermal / structural co-sim |
| Scale | Cloud / GPU Elastic parallel jobs and GPU compute without per-seat solver licensing | Strong HPC / distributed-memory and domain-decomposition options; commercial seat model for capacity |
| Extensibility | Open solver source Palace Apache-2.0; scriptable preprocess / postprocess APIs | Closed solver · open automation Proprietary solver core; mature MIT-licensed PyAEDT for scripting, geometry, setup, and post-processing |
Is CEM a drop-in replacement for HFSS?
On the shared FEM territory, CEM (via Palace) is competitive on eigenmode and driven work — with elastic cloud/GPU compute and no per-seat licensing. In the open literature the Palace solver class agrees with cryogenic measurement on resonator frequencies (couplings less tight) — solver-class evidence, not yet a CEM-specific benchmark. With preprocess + postprocess, CEM productizes what most teams still assemble by hand: geometry setup gates, Palace-native EPR → Hamiltonian, consistency checks, and a design report. Outside FEM — installed antennas, RCS, IE/SBR+ hybrids — HFSS remains the right tool.
◇ Position: coexist for breadth · lead for quantum FEM→HamiltonianSBR+, MoM, FEBI remain HFSS-only. CEM isn’t intended for platform-scale antenna or RCS work — use HFSS there.
Use HFSS for aerospace, auto radar, defense EMPreprocess covers Quantum Metal → mesh/config deeply. It is not HFSS-style freeform 3D modeling for arbitrary assemblies.
Thermal / structural / circuit harmonic-balance co-sim is outside CEM’s current stack.
Regulated aerospace/defense EM procurement may still require HFSS regardless of FEM competitiveness.
Open-source prior art exists for GDS→mesh→eigen/S→Hamiltonian stacks. CEM’s edge is a hosted, gated, reported product: Palace-native EPR, mode ID, scqubits, validation gates, sweeps, and a design-assessment report — not a claim that the workflow is first of its kind.
Not in bare HFSS or DIY open-sourcePre-solve verify, p-refinement convergence, EPR↔C consistency, and collision warnings reduce silent wrong answers.
Not native to HFSSTargets vs results, mode map, readout κ/Q_ext, sweeps — one HTML artifact for partners and design reviews.
Not native to HFSSNo solver seat tax; cloud-GPU parallelism for early-design exploration while HFSS can remain the sign-off tool.
Works alongside HFSSHow to read this comparison. CEM wraps preprocess + Palace + postprocess into a high-throughput, quantum-aware FEM→Hamiltonian workbench. It’s built to work alongside HFSS (or measurement) — which remain the tools for general EM breadth and regulated sign-off — not to replace them.
| If you know HFSS… | In CEM look for… | Layer |
|---|---|---|
| Project / design setup | Quantum Metal design + palace_parameters.yaml |
Preprocess |
| GDS / geometry import | Metal .py path (GDS+manifest path expanding) |
Preprocess |
| Lumped RLC / FakeJunction | JJ LumpedPort from Metal junctions + YAML L/C |
Preprocess |
| Airbox / radiation BC | enclosure.* + outer_boundary (PEC / Absorbing / PMC) |
Preprocess |
| Adaptive mesh passes | Gmsh sizing + optional Palace Model.Refinement (AMR) |
Preprocess / Palace |
| Eigenmode solution | Palace Eigenmode → eig.csv / port-EPR.csv |
Palace |
| Driven / S-parameter sweep | Palace Driven → port-S.csv → κ / Q_ext |
Palace + Post |
| Capacitance matrix | Palace Electrostatic → terminal-C.csv → E_C/E_J |
Palace + Post |
| pyEPR participation analysis | EPR ingest + mode ID + EPR↔C gate | Postprocess |
| Manual scqubits / QuTiP notebooks | scqubits adapter (spectra, α, g/χ); QuTiP path roadmap | Postprocess |
| Optimetrics / parametrics | Cloud job sweeps + L_J comparison tables | Platform + Post |
| Slide deck for design review | design_assessment_report.html |
Postprocess |
| IE / SBR+ / FEBI | No CEM equivalent Keep HFSS | — |
Geometry → validated EM → circuit Hamiltonian → design-assessment report, on cloud GPUs — no per-seat solver license.