Surrogate models predict RF performance 1000x faster than electromagnetic simulation
Their competitor had just announced a Wi-Fi 7 module with specs that matched their 'next-generation' product — the one still in design. If they couldn't accelerate the design cycle, they'd launch a year late into a market that had already moved on. The engineering team was maxed out, and RF designers were commanding $200K+ salaries with 6-month notice periods.
Designing RF front-end modules for 5G and Wi-Fi 7 requires optimizing linearity, power efficiency, noise figure, bandwidth, thermal management, and die area simultaneously. Each electromagnetic simulation takes 2-4 hours. Engineers could evaluate maybe 50 design candidates per week. The design space had millions of viable configurations.
We built surrogate models (neural networks) trained on thousands of historical EM simulation results that predict RF performance metrics in milliseconds instead of hours. Multi-objective Bayesian optimization navigates trade-offs between competing specs and presents Pareto-optimal designs. Our overnight RF engineering team runs full EM verification on AI-recommended designs and feeds results back for continuous model improvement.
The AI suggested an impedance matching topology we'd never considered. Our senior designer said it wouldn't work. We simulated it anyway. It outperformed every manual design by 2dB.— VP of RF Engineering
Let's talk about what AI + a supplemental engineering team can do for your business.
Talk to a Dev Lead →