Improving the semiconductor design process requires understanding and accounting for uncertainties. There is uncertainty in the materials and process used to manufacture the final chip. Even for a perfect process that produced identical chips, each may be placed into different devices and environments introducing uncertainties such as usage and cooling environment.
Determining optimal reliable configurations under such uncertainties is difficult and can require substantial simulation time using traditional methods. For example, modeling the interaction between processing, power use, thermal dissipation, and cooling systems relies on complex physics-based simulations such as CFD and FEA.
Techniques such as sensitivity analysis can be used to understand the major drivers of performance uncertainty, but such techniques tend to require infeasibly large Monte Carlo style samples of data to produce reliable results.
The solution is to first train a machine learning model using data from the process or complex simulation collected by an intelligent sampling plan. Once trained, the machine learning model can rapidly make accurate predictions and can replace the need for running additional physics-based simulations or collecting further physical data. With the roadblock of computational cost removed many otherwise infeasible analyses may be conducted to improve the design process.
Join us for this webinar to learn how artificial intelligence and machine learning can be used to improve the semiconductor design process.
