Skip to content
Close
Talk to an expert
Japanese
Talk to an expert
Japanese

Bond-wire radiation pattern prediction

Predicting 3D bond-wire radiation patterns via full-wave sweeps & surrogate models

Bond-wire radiation pattern prediction
Electromagnetic waves

The challenge

Bond wires are small, but at millimeter-wave frequencies their geometry measurably changes how much unwanted radiation is emitted, and in which direction.

Mapping that behavior means exploring a six-parameter geometry space plus frequency. A uniform five-level grid over those inputs is 5⁷ = 78,125 full-wave solves, an estimated 5.4 days of solver time.

Approach with Quanscient Allsolve

Quanscient Allsolve generated the training data: 160 bond-wire geometries sampled by Latin hypercube, each solved at five frequencies, as 16 concurrent sweeps scripted through the Allsolve SDK.

Each 3D radiation pattern was compressed from 4,376 directions into 25 spherical-harmonic coefficients, and Gaussian process models were fitted to predict them for unseen geometries.

Key results

  • Dataset in 81 minutes: 800 full-wave solves (160 geometries × 5 frequencies, 0.7–1.4M DoF each) including meshing, with passivity of 0.9998–1.0000.
  • Validated accuracy: On 32 held-out designs, the median RMS error over the main beam was 0.350 dB and the median peak-direction error 1.56°.
  • Instant exploration: The surrogate rebuilds a full 3D pattern in ~1.5 ms and reports its own uncertainty, directing the next solves where data is thinnest.
Full-wave FEM and surrogate predictions in the XY, XZ, and YZ planes for representative held-out geometries.