Title: Numerical Simulation of Electronic Noise in Si MOSFETs
1Numerical Simulation of Electronic Noise in Si
MOSFETs
EIT4
- C. Jungemann
- Institute for Electronics
- Bundeswehr University
- Munich, Germany
- Acknowledgments B. Neinhüs, B. Meinerzhagen, A.
Scholten, A. Heringa
2Outline
- Introduction
- Theory
- Acceleration Effects
- Noise source modeling
- Noise in NMOSFETs
- Noise in a BJT
- Noise in an IMOS
- Conclusions
3Introduction
4Introduction
- Noise is a fundamental property of electron
transport and cannot be avoided - Fluctuation-dissipation theorems (e.g. Nyquist
theorem) are only valid at equilibrium (Shot and
thermal noise are macroscopic manifestations of
microscopic noise.) - Transport in nanoscale devices is nonlocal and
quasi-ballistic - Physics-based methods required for device level
simulations!
5Introduction
NNN structure at zero bias
- Terminal current fluctuations are due to electron
scattering within the device via displacement and
conduction currents - Noise theory describes the variance and the
correlation of the fluctuations
51017
51017
21015
6Introduction
Power spectral density (PSD)
- PSD vanishes at very high frequencies due to
acceleration effects (finite electron mass means
no real white noise) - nonquasistationary
- Plasma resonance at very high frequencies (gt1THz)
in silicon
7Theory
8Theory
LBE is the basis for LHD and LDD models
9Theory
10Theory
11Theory
- Transport and noise parameters of the LDD are
calculated consistently under homogeneous bulk
conditions based on the single particle LBE - The parameters are generated for a wide range of
doping concentrations, lattice temperatures,
strain conditions, driving fields etc, and stored
in lookup tables for later use.
12Impact of the Acceleration Term
13Acceleration Effects
In the DD approximation the mobility and the PSD
of the velocity fluctuations are assumed to be
frequency independent
Ndop1017/cm3
Up to about 100GHz this is correct for
siliconThe macroscopic relaxation time
approximation fails
14Acceleration Effects
NNN structure (Full LBE)
Up to about 100GHz acceleration effects can be
neglected in silicon
15Acceleration Effects
Undoped silicon at room temperature E(t)
30kV/cm1cos(2pft)
Above 100GHz nonquasistationary effects occur in
silicon
16Noise source modeling
17Noise source modeling
NNN structure
Bulk, ND1017/cm3
Diffusion noise source yields the best results HD
model yields similar good results Device results
strongly deviate from thermal or shot noise
18Noise source modeling
NNN structure biased at 6V
Generation noise due to impact ionization Noise
source is given by microscopic white shot noise
19Noise source modeling
NNN structure biased at 0 and 1V
Terminal current noise is due to cold and warm
electrons Hot electrons can produce noise via
impact ionization
20Noise in NMOSFETs
21Noise in NMOSFETs
180nm Technology, tox 3nm, Vdrain 1.8V,
f2.5GHz
Lgate1mm
Measurements and Tsuprem simulations by Philips
(A. Scholten) Simulation includes quantum
correction for channel DD and HD simulations
performed without any parameter matching
22Noise in NMOSFETs
180nm gate length, tox 3nm, Vdrain 1.8V,
Vgate1.0V
Gate noise
Drain noise
Also in MOSFETs drain noise is not due to hot
electrons
23Noise in NMOSFETs
50nm channel length, 1.3nm oxide, Vdrain0.9V,
f10GHz
Noise specs of small NMOSFETs increase only
moderately
24Noise in a BJT
25Noise in a BJT
1D 50nm Si NPN bipolar transistor
Fano factor of electron collector noise at
VCE0.5V
Doping profile
Overestimation of shot noise is caused by model
failure
26Noise in a BJT
50nm Si bipolar transistor
VCE0.5V, VBE0.65V
Quasiballistic transport leads to model failure
27Noise in an IMOS
F. Mayer et al., TED, Vol. 53, p. 1852, 2006
28Noise in an IMOS
CIMPAT, Vgate/drain-3.5V, Lchannel5.0mm
CMOS has a Fano factor of less than one for
inversion IMOS generates two or more orders of
magnitude more noise
29Conclusions
30Conclusions
- Consistent hierarchy of noise models (DD, HD,
LBE) - Transport and noise parameters are consistently
generated for the DD and HD models by LBE bulk
simulations - The transport and noise parameters are local in
real space and frequency independent - Acceleration effects can be neglected below
100GHz in silicon - Modified noise sources (diffusion noise) give
good results
31Conclusions
- Good agreement of measurements and simulations
for MOSFETs - Terminal current noise is produced by cold or
warm electrons - Hot electrons can produce noise via impact
ionization - No dramatic increase of noise in scaled MOSFETs
- IMOS suffers from huge noise due to avalanche
breakdown