2. Why study SOI MOSFETs nonlinearities ?
Distortion
Silicon-on-Insulator (SOI)
MOSFET SOI 
Simplified process
Low parasitic capacitances
Low leakage current
Low Vth
=> promising for RF ICs f
Non-idealities of “linear circuits”
Amplifiers
Active filters
Used in some applications
Mixers
Oscillators
Frequency multipliers
Inherent to the physics of semiconductors
ghj

3. FD vs PD SOI MOSFETs ID [mA] - : Fully Depleted (FD)G
S/D
Burried Ox
ID [mA]
- : Fully Depleted (FD) G
Burried Ox
S/D
body
--: Partially Depleted (PD)
with floating body
VD [V]
DuTs: FD and PD SOI MOSFETs, 12x6.6 µm/0.25 µm (0.25 µm LETI technology)

4. What happens inside? - gd kink -
High E field near the drain:
Impact Ionization current +
=> impact ionization => creates e--h+pairs +
=> injection of holes inside the body
=> body potential increase up to Vtsb
=> Parallel path for Id and Id increase
=> Vt lowering and Id increase
n ++ p
n ++
Body region
Depletion region
SiO2

5. The simplest is the best
Linearity of SOI MOSFETs using Integral Function Method and Volterra modeling
=> Does the kink influence the linearity ?
=> Which methods to characterize the linearity of MOSFETs ??
The simplest is the best
DC based characterization methods
Taylor series
Integral Function Method (IFM)
Comparison with Large-Signal Network Analyzer (LSNA)
measurements
HF MOSFET model based on Volterra series
Frequency limitation of DC based methods
Third order intermodulation
Conclusions
Devices performances
Outline

6. Method : Taylor analysis
Consider the memoryless nonlinear system:
VG(t)
ID(t) VG ID t f t f
If the circuit is excited by a sine wave,
Taylor series:

7. Methods: quid for large amplitude?VG
Taylor: add terms => too complicated !
IFM: good approximation of HD at LF
further advantage: less sensitive too measurement noise
[CerdeiraSSE02, CerdeiraSSE04, CerdeiraICSICT04]

8. [CerdeiraSSE02, CerdeiraICSICT04]
IFM: How does it work ?
2. Observe that Area1-Area2 is
proportionnal to the THD
3. Define the D function
Area2
Area1
1. Normalize the characteristics
Out In
HD3 is obtained by computing the D function of Ir = I(V)-I(-V)
=> even harmonics eliminated
! HD of order higher than 3 are neglected
[CerdeiraSSE02, CerdeiraICSICT04]

9. IFM takes the influence of the amplitude of the applied signal into account
HD3 [dB]
VG [V]
Not by a scale factor as from Taylor approach, cfr.

10. Comparison with LSNA measurements
full-wave (magnitude and phase) RF (900 MHz) characterization
(V,I fundamental and harmonics at input and output) in single take
=> Real RF nonlinear behavior
[VerspechtMTTS95]
900 MHz, 50 Ω, A = 0.2 V
Good agreement before the minimum
Minimum located at :
- IFM : max. of gm
- LSNA : max. of power gain
Nonlinearity of gm and gd
LSNA
HD2 [dB] gm
and gd
gm only
VG [V]

11. Harmonic distortion af HF
DC method vs 900 MHz measurements in agreements
=> Which frequency limit ??
=> Answer this question with the help of a Volterra series based model: RG
Cgd
Cgs
Gm Gd Cds YL
Vin
Gm=gm1+gm2VG+gm3VG²
Gd=gd1+gd2VD+gd3VD²

12. HD from Nonlinear current method
freq
fp fz
[ParvaisGAAS04]

13. Poles of HD2 and HD3 as a function of ZL
Pole Voltage Gain Av
26 GHz
9 GHz
5 GHz
Pole HD2
Pole HD3

14. Characterization methods
Good agreements between results calculated using IFM and using Fourier coefficients.
IFM: advantages = amplitude dependent, no derivatives.
Frequency validity range cfr. Volterra model (several GHz).

15. HD: PD ~ FD transistors

16. Frequency analysis of the kink effect
frequency limitation caused by RC body impedance
iii Vb Rb Cb
[SinitskyIEDL97]
As Vd  Rb  => fc 
Rb and Cb= body
resistance and
capacitance

17. Third order intermodulation
Volterra model:
ZL↑
Kink effect !

18. PD SOI: Floating body or not ?
PD, from ST Microelectronics
60x 1µm/0.12 µm, f=2 GHz
FB: higher fT, fmax than BC and isoc.
BC/Isoc.: parasitic C, gm degradation

19. Conclusions
At 900 MHz, when the polarization voltage is varied, HD is dominated by the DC I-V characteristics.
Frequency validity range of DC methods provided by a Volterra model.
PD versus FD:
HD > idem (gm dominates)
IMD > depends on the tone separation
cfr. Kink effect
Thanks to
FRIA for financial support