Upload
ray-garrison
View
217
Download
0
Tags:
Embed Size (px)
Citation preview
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 11
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Circuit-level modelling of carbon nanotube field-effect
transistors
Tom J KazmierskiSchool of Electronic and Computer ScienceUniversity of Southampton, United Kingdom
[email protected], http://www.syssim.ecs.soton.ac.uk
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 22
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Outline
o Introduction New efficient methodology for numerical CNT FET modelling based on
piece-wise non-linear approximation o PNL modelling of non-equilibrium mobile charge density
Two PNL approximations leading to closed-form solution of self-consistent voltage equation
o Drain current calculationo Equivalent circuito Simulation experiments demonstrating speed up and modelling
accuracyo Conclusion: what next?
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 33
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Introduction
o CNT FET theory and operation are gradually better understood.o Early CNT FET models simply used MOS equations – no good.o Now a physical theory of ballistic CNT transport exists.o Circuit-level models have been developed based on theory but they
are very complex in terms of computational intensity.o Recently fast models appeared, based on numerical approximation.o Focus of this talk: new, efficient piecewise non-linear approximation
of mobile charge three orders of magnitude faster than evaluation of physical equations,
but still maintaining high accuracy.
o Important for circuit design where very large numbers of CNT devices will need to be simulated.
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 44
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Non-equilibrium mobile chargeo Non-equilibrium mobile charge is injected into CNT when drain-source voltage
is applied:
o State densities are determined by Fermi-Dirac probability distribution:
VSC – self-consistent voltage
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 55
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Self-consistent voltage equationVSC - recently introduced concept
Strongly non-linear, requires Newton-Raphson iterations and calculation of integrals – standard approach to CNT FET modelling
Total charge at terminal capacitances
Total terminal capacitance
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 66
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Standard approaches to evaluate charge density
o Newton-Raphson technique and finite integrationo Non-equilibrium Green’s function (NEGF)o Recently piece-wise linear and piece-wise non-linear
approximations have been proposed to obtain closed-form symbolic solutions The aim is to eliminate the need for computationally intensive
iterative calculations in development of models for circuit simulators
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 77
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Total drain currentIf VSC is known, total drain current can be
obtained form Fermi-Dirac statistics directly:
Closed-form solution for Fermi-Dirac integral of order 0 exists:
hence:
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 88
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Circuit model of a top-gate CNT FET
If equal portions of the equilibrium charge qN0 are allocated to
drain and source, non-equilibrium charges at drain and
source can be modelled asnon-linear capacitances.
A hypothetical inner node can be created to represent
the self-consistent potential
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 99
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
New technique to accelerate VSC calculationModel 1: 3-piece non-linear approximation of charge density:
solid line: theory
dashed-line: approximation
Linear and quadratic pieces
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1010
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
New technique to accelerate VSC calculationModel 2: 4-piece non-linear approximation:
solid line: theory
dashed-line: approximation
Region boundaries are optimised for best fit
Linear, quadratic and 3rd order pieces
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1111
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Speed-up due to PNL approximation
FETToy – reference theoretical model implemented in MATLAB
CPU times for PNL Model 1 and Model 2 obtained also from a MATLAB script
Model 1 runs 3500 faster and Model 2 – 1100 times
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1212
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Loss of accuracy due to PNL approximation
Model 1 – dashed, FETToy - solid
Typical parameters: T=300K, Ef = -0.32eV
Model 2 – dashed, FETToy - solid
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1313
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
RMS errors for Ef=-0.32eV
Model 2 accurate within 2%, Model 1 – 4.6%, at T=300K
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1414
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Accuracy at extreme temperatures and Fermi levels
Model 1 – dashed, FETToy - solid
Extreme parameters: T=150K, Ef = 0eV
Model 2 – dashed, FETToy - solid
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1515
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Accuracy at extreme temperatures and Fermi levels (2)
Model 1 – dashed, FETToy - solid
Extreme parameters: T=450K, Ef = -0.5eV
Model 2 – dashed, FETToy - solid
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1616
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
RMS errors for Ef=-0.5eV
Across T and EF ranges - Model 2 is accurate within 2.8%, Model 1 – 4.8%
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1717
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
RMS errors for Ef=0eV
T.J. KazmierskiT.J. KazmierskiCircuit-level modelling of carbon nanotube field-effect transistors Circuit-level modelling of carbon nanotube field-effect transistors 1818
MOS-AKMOS-AKMunichMunich
14 September 200714 September 2007
School of Electronics and Computer ScienceSchool of Electronics and Computer ScienceSouthampton UniversitySouthampton University
Conclusion
o New, fast, numerical CN FET model has been proposedso Suitable for a direct implementation in SPICE-like circuit-level
simulatorso Further evidence to support suggestions that costly Newton-
Raphson iterations and Fermi-Dirac integral calculations can be avoided leading to a substantial speed-up.
o Two models proposed and tested in simulationsso Future work will involve CN FET analysis of speed and modelling
accuracy of circuit structures built of CN FETs.