Specify the electrical behavior you want. The platform designs the semiconductor device that produces it.
Driffusion is an inverse design platform for semiconductor devices. Instead of manually designing a device and simulating its behavior, you describe the behavior you want — as a target I-V (current-voltage) curve — and the platform automatically discovers a device structure that produces it.
The platform uses a differentiable physics simulator under the hood. It solves the drift-diffusion equations (the standard semiconductor transport model), computes gradients through the entire simulation, and uses gradient-based optimization to iteratively refine the device geometry and doping profile until the simulated I-V matches your target.
Access is via invite codes distributed to research groups. Enter your code, email, and name to get started. No complex setup required.
Specify the electrical behavior you want the device to exhibit. There are three ways to input your target:
Configure the physical device parameters for the optimization:
Click "Start Optimization" and the platform submits a job to the compute backend. The optimizer runs gradient-based updates on the device parameters, comparing the simulated I-V to your target at each step. Advanced settings like step count, learning rate, and total-variation regularization weight are available for fine-tuning.
The job detail page shows live updates as the optimization runs:
When the optimization completes, you get the final simulated I-V overlaid on your target, the log-MSE error metric, and the full optimization configuration for reproducibility. Completed devices are added to your device catalog for later reference.
Beyond the dashboard, batch optimization and pipeline integration are supported through a programmatic interface.
# Programmatic access on request — contact [email protected]
2-terminal device optimization. The optimizer can discover PN junctions, graded doping profiles, and non-standard structures from scratch.
3-terminal device optimization. Define on-state and off-state targets, and the optimizer will find a gate-modulated structure with the desired Ion/Ioff ratio.
Specify arbitrary current-voltage characteristics. The optimizer matches your target in log-space, treating all decades of current with equal weight.
Implant parameterization constrains the optimizer to produce designs compatible with standard fabrication processes (ion implantation profiles).
The underlying simulator is validated against analytical solutions (Shockley equation, built-in voltage, depletion width) and cross-validated against DEVSIM TCAD, with comprehensive analytical, TCAD-comparison, and gradient-correctness coverage.
Stochastic optimization under process variations. Design devices that maintain performance across fabrication uncertainty and process corners.
The platform is in active development. Here is what it currently supports and what is planned:
The physics solver operates on 2D cross-sections. 3D simulation is planned for a future release.
The solver computes DC operating points. Transient (time-domain) and AC (small-signal) analysis are not yet supported.
Carrier mobility is treated as a constant. Field-dependent and doping-dependent mobility models are planned.
The differentiable solver supports grids up to 30x30 voxels. Larger grids are a roadmap item.
Join the waitlist for early access, or reach out directly if you're a researcher interested in inverse device design.
Join Waitlist