TCAD stands for technology computer-aided design. In practice it means simulating a semiconductor device on a computer, solving the physics of carriers, electric field, light and heat, so that you can predict how the device behaves before you build it. This guide explains what TCAD does, where it is used and how to get reliable results from it.
What does TCAD actually compute?
At the core of most TCAD studies is the drift-diffusion model. It combines the Poisson equation for the electrostatic potential with continuity equations for electrons and holes. Given a device geometry, doping profile, materials and contacts, the solver returns the carrier densities, the potential and the current at every bias point.
From these fields you obtain the quantities engineers care about:
- current-voltage and capacitance-voltage curves
- threshold voltage, subthreshold slope and saturation current of a transistor
- breakdown voltage and avalanche behavior of diodes and SPADs
- responsivity and dark current of photodetectors
- light output, threshold current and efficiency of VCSELs and other lasers
Where TCAD is used
TCAD is used wherever building and measuring devices is slow or expensive. Typical cases are CMOS and power devices, photodetectors and infrared pixels, single-photon avalanche diodes, tunnel junctions and vertical-cavity surface-emitting lasers. In each case the simulation answers questions such as which doping profile gives the lowest dark current, or how much a laser slows down when it heats itself.
Why electro-thermal and optical coupling matter
Many devices cannot be understood with drift-diffusion alone. A VCSEL needs the optical field, because the light changes the carrier density, and it needs the temperature, because self-heating shifts the gain peak and the cavity resonance. In our own simulations of a quantum-well VCSEL, self-heating lowers the slope of the light-current curve and red-shifts the emission by about 7 nm at 10 mA. Coupling transport, optical and thermal solvers is what turns a simulation into a prediction.
How to trust a TCAD result
- Compare with a reference. Run the same device, physics and mesh in a second solver and compare the curves point by point. We did this for a MOSFET, where Aurora agrees with a commercial drift-diffusion solver within 1.5% over eight decades of drain current, and for an infrared II-VI pixel, where the responsivity matches within 1% up to the cutoff.
- Calibrate on measurements. A model that is fitted to your data tells you which parameters really matter.
- Check the mesh. Refine it near junctions and interfaces until the result stops changing.
- Keep the inputs. A study you can rerun is a study you can defend.
You can see the comparison curves on the Flow page.
Doing it faster with Flow
Setting up a TCAD campaign by hand means writing input files, launching dozens of runs and collecting the results. Flow, our workspace for multiphysics simulation, does this for you. You describe the study in plain language, a local AI agent writes the workflow and the input files, and the runs and results stay on your own machine.
When to ask for help
If you need an answer about a device and your team has no time to build the model, we can run the study with you: simulation, calibration on your measurements, optimization or a custom physical model. See our TCAD simulation and device modeling services, or contact us with a short description of your device.
Leave a Reply