Sign Up

PROPRIETARY AI

Inside HeaviMind & DANTE

The two AI engines running inside every Circuit Studio pipeline, and how their predictions hold up against measured silicon.

HeaviMind

Generates ranked candidate circuits from your target specs, then optimizes placement and routing into a tapeout-ready layout.

DANTE

Fits physically precise, simulation-ready transistor models from measured device data, most importantly large-signal measurements, and feeds them into HeaviMind.

Back to Product Overview

HEAVIMIND · VALIDATION

Predicted vs. measured silicon

A good simulation is easy to show. What matters is whether the chip that comes back from the foundry behaves the way the tool said it would.

So we check. These are fabricated amplifiers, measured on the bench and plotted against what HeaviMind predicted for them: gain, input and output match, and output P1dB, across each design's band.

Where measurement and prediction agree, it's evidence that the numbers you see in Circuit Studio carry through to silicon. Where they differ, the measured data goes back into our device models, so the next prediction starts from real silicon. It's how trust in an AI design tool has to be earned: one measured chip at a time.

6–18 GHz · 5 V bias

MeasuredPredicted (HeaviMind)
S-parameters

dB vs frequency (GHz)

Output P1dB

dBm vs frequency (GHz)

HEAVIMIND · END TO END

Months of PA design work, done in hours

One complete power amplifier design, start to finish: from the initial requirements, through each matching network and optimization pass, to the finished PA and its optimized layout.

By hand, each of those stages is its own loop: synthesize a match, simulate, trade gain against efficiency and output power, adjust, then fight the layout until it performs like the schematic and still fits the die. For a multi-stage PA, that typically adds up to months.

HeaviMind runs that whole loop itself, from your requirements to an optimized amplifier and its layout, in a matter of hours. Nothing here is mocked up: it's a real design from a real run, and the plots below are the finished PA's simulated performance.

Join the PA beta

Completed PA design · 17–21 GHz

Psat 40.0–40.6 dBm · PAE at Psat 40–42%

Small-signal S-parametersS21S11S22

dB vs frequency (GHz)

Large-signal gain17 GHz18 GHz19 GHz20 GHz21 GHz

dB vs output power (dBm)

Power-added efficiency17 GHz18 GHz19 GHz20 GHz21 GHz

% vs output power (dBm)

Time comparison is an estimate: a typical manual design-and-layout cycle for a multi-stage power amplifier against a typical pipeline run. Actual times vary with the design and specs.

HEAVIMIND · MODEL CONFIDENCE

Know the limits before you spend a run

Before a pipeline starts, HeaviMind scores how likely your target spec is to be met on the technology you picked. The answer takes milliseconds, not a design run.

  • See how far each technology goes in frequency, gain, P1dB, noise figure and supply voltage.
  • Skip specs that were never going to close, and find the ones you can push further.
  • Compare technologies and devices before you commit: each has its own model in Circuit Studio.

Check a spec

Set your targets and see how likely a design run is to meet all of them.

Centre frequency

Bandwidth

Min. gain

Min. P1dB

Max. noise figure

Technologies combined, each at its best supply voltage. Circuit Studio lets you set both per design.

Confidence

–

Retrained

Circuit Studio always scores with its newest models, so a live score can differ slightly.

Loading model data…

Drag to rotate, click a point to try it

Confidence

0%50%100%

Model updated

HEAVIMIND · ONLINE LEARNING

A model that keeps up with the software

A confidence score is only worth trusting if it matches the design engine you actually run. So these models are never trained once and frozen. A fleet of our servers generates and scores new designs around the clock, and each model retrains whenever something underneath it changes.

  1. 01

    Something changes

    A new design-engine release or an updated technology library is picked up automatically.

  2. 02

    New designs run

    The fleet re-simulates the specs most likely to be affected and probes new ones near each technology’s limits.

  3. 03

    The model retrains

    Each model is retrained on the updated data, with its settings tuned again from scratch.

  4. 04

    Only a better model ships

    The new model replaces the old one only if it scores better on the same fixed test set.

See HeaviMind & DANTE in Action

Both engines run automatically inside every Circuit Studio pipeline - try a design and see the output for yourself.