Answers are grounded in this paper and cite their sources.
Quantum confinement becomes significant once a transistor channel is thinner than about 10 nm, so technology computer-aided design (TCAD) models of such devices must include it. Nearly all machine-learning surrogates reported for gate-all-around junctionless field-effect transistors (GAA-JLFETs) cover n-type devices only, although complementary logic also requires p-type devices. In this work, we calibrated a quantum-corrected drift-diffusion TCAD model against measured data from an n-type GAA-JLFET and then adapted it to a p-type device. Six design parameters were sampled by Latin hypercube sampling, and the transfer characteristic of each design was simulated at drain voltages of −0.1 and −1 V. Runs were retained on the basis of simulation reliability rather than device performance, and the noise floor was set from current samples with a physically impossible sign. The resulting 217 devices were used to train a dilated residual one-dimensional convolutional network that predicts the full transfer curve directly from the design parameters. On 26 held-out test devices, the network reached a median R 2 of 0.9994 and a worst-case R 2 of 0.9959, and it predicted the ON current with a mean error of 1.6%. Each prediction requires a single forward pass instead of a TCAD solve, which removes the simulation bottleneck from the design of p-type GAA-JLFETs.
No takes yet. Share an insight, caveat, or question.
Das et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: