Access Official Technical PDF

Share Summary:LinkedInEmail

As transistor dimensions shrink, electrostatic control becomes a central design challenge. Stacked silicon nanosheet field-effect transistors (SiNS FETs) are a gate-all-around architecture under study for scaling beyond FinFETs, but their compact dimensions and complex geometry make performance difficult to assess with simplified models alone. A self-consistent NEGF simulation offers a way to evaluate how device geometry and process conditions affect quantum transport and key electrical characteristics.

A demonstrated study of dual SiNS nFETs used the non-equilibrium Green’s function (NEGF) method coupled with the Poisson equation. By changing design variables from a defined baseline structure, the simulations compared their effects on threshold voltage, subthreshold slope and on-current. The results illustrate how physics-based simulation can support design exploration—not a universal prediction of how a manufactured device will perform.

Why use NEGF simulation for stacked nanosheet FETs?

FinFETs improved on planar transistors in part by providing greater effective channel width and stronger gate control. Gate-all-around devices extend that approach: the gate surrounds the channel, improving its electrostatic control. Stacked nanosheets use multiple horizontal channels within one transistor structure, making them a candidate for continued scaling.

The nanoscale dimensions and shape of these devices make quantum and atomistic effects relevant. Design choices may also interact: changing a channel dimension or process condition can affect several performance measures at once. Evaluating variables independently can help identify sensitivities, while a transport model that can represent the device’s physical structure provides a basis for comparing those changes.

NEGF describes carrier transport through the device, while the Poisson equation calculates electrostatic potential across the broader structure. Solving the equations self-consistently links the carrier distribution and potential. In the study, the transport solver covered the silicon region and the Poisson solver covered the full device region.

Models behind the quantum transport analysis

The simulation framework incorporated models for several physical effects that can influence nanoscale transistor behavior. These included:

  • Silicon bandstructure: An effective-mass approximation represented conduction-band valleys, with nonparabolicity correction. Parameters were calibrated against tight-binding results to account more accurately for quantum confinement and strain or stress effects.
  • Electron-phonon scattering: Deformation-potential theory represented intra- and inter-valley scattering. The model parameters were calibrated to silicon mobility, with adjustments near the channel surface to represent observed mobility reduction under strong inversion.
  • Doping and impurity effects: An atomistic doping model represented discrete dopant charges and impurity scattering using random point charges generated from a continuous doping profile.
  • Channel and gate variability: The framework could represent surface roughness and variation in gate-metal grain size and workfunction. Different channels could have distinct roughness configurations.
  • Contacts and electrostatics: The contact treatment included virtual metals with parameters for effective mass, workfunction and band edge. The model also treated contact atoms as independent carrier reservoirs, allowing curved contact surfaces to be represented.

Not every available model was enabled in the sensitivity runs. Atomistic traps, remote Coulomb-scattering dipoles and gate leakage were omitted because they required statistical analysis based on experimental data or were considered to have minor effects for this demonstration. That distinction matters when interpreting results: the values reflect the selected models and simulation conditions, not every possible device effect.

Which design variables affected performance?

The study varied 11 design variables from a base dual-nanosheet nFET structure. The reported results show that sensitivity depends on both the variable and the metric. For example, reducing channel width by 50% was associated with a 9.8% reduction in on-current and a 9.8% reduction in subthreshold slope was not reported; instead, the table gives a 9.8% decrease in SS. The same change increased threshold voltage by 61.2 mV. (Here, SS denotes subthreshold slope.)

Other reported changes help illustrate the trade-offs:

  • Reducing gate length by 14% shifted threshold voltage by -54 mV, increased subthreshold slope by 15.9%, and increased on-current by 4.8%.
  • Increasing channel height by 33% changed threshold voltage by -71 mV and increased both subthreshold slope and on-current by 18.5% and 19.1%, respectively.
  • Increasing equivalent oxide thickness (EOT) by 20% shifted threshold voltage by -17 mV, increased subthreshold slope by 4.6%, and reduced on-current by 2.8%.
  • Reducing source/drain doping from 3 × 1020 to 2 × 1020 cm-3 shifted threshold voltage by 9 mV and reduced on-current by 21.6%.

These are changes relative to the study’s baseline and conditions. They should not be treated as general design rules: the simulated structures and conditions were explicitly set up for a demonstration rather than based on a specific real device.

Using sensitivity results in device development

A sensitivity analysis can help teams prioritize which design variables warrant further investigation. It can reveal where a geometry or process adjustment has a pronounced effect on a target metric, and where performance measures move in different directions. In the reported study, simulations also examined variability from discrete dopants, channel roughness and random gate-metal workfunction. These cases showed variation in off-current and degradation in on-current across configurations.

That information can guide subsequent engineering work, but optimization cannot stop at electrical metrics. The authors note that fabrication difficulty, cost and yield also need consideration. Simulation is most useful when its assumptions, calibration and applicable limits are understood and when predicted trade-offs are considered alongside manufacturing constraints.

Conclusion

Self-consistent NEGF-Poisson simulation provides a physics-based approach for examining transport and electrostatics in stacked silicon nanosheet FETs. The demonstrated sensitivity study shows how changes to channel dimensions, gate geometry and doping can affect threshold voltage, subthreshold slope and on-current in different ways. For device engineers and technical decision-makers, the practical value is a structured way to compare design options and identify trade-offs before selecting candidates for further evaluation. Results remain specific to the simulated structure, models and conditions, so they should inform—not replace—consideration of process feasibility, cost and yield.

Access Official Technical PDF

About The Author