Direct measurement of white noise in MOSFETs...Device Lab Inc. Slide 1 Direct measurement of white...
Transcript of Direct measurement of white noise in MOSFETs...Device Lab Inc. Slide 1 Direct measurement of white...
Device Lab Inc. Slide 1
Direct measurement of white noise in MOSFETs
Kenji Ohmori* and Shuhei Amakawa*,**
* Device Lab Inc.** Hiroshima University
11th International MOS-AK WorkshopSilicon Valley, December 5, 2018
Device Lab Inc. Slide 2
One-page introduction of Device Lab Inc.Spin-off company from University of Tsukuba
Ø Founded in April 2017Ø Based in Tsukuba, Japan (1 hour from Tokyo)Ø Focus on electric characterization of devices
ü Entrope™ high-frequency noise probeü Measurement solutions for characterizing
MOSFETs and neuromorphic devices
Device Lab Inc. Slide 3
Outline
Ø Noise in MOSFETsü Low frequency (1/f, RTN)ü High-frequency (thermal, shot)
Ø Measurement systemsü Conventional system (low-frequency)ü High-frequency noise probe
Ø Measurement resultsØ Comparison with modelsØ Summary
Device Lab Inc. Slide 4
Schematic diagram of MOSFET noise
Frequency (Hz)
Noi
se in
tens
ity
f −ν
fc
Capacitive coupling
RTN
Thermal noise, Shot noiseFlicker
Lowfreq.
~100 kHz
SId = 4kBTΔf R
SId = 2qIdΔf
Thermal noise
Shot noise
Noise figure measurement “ac”
Conventional measurement (LNA, fast IV) “dc”
Ø Practical frequency limit in LFN systems: 100 kHzØ Noise figure: complicated, requires noise sourceØ Direct observation of a corner frequency (under DC bias)
is important.
Jitter Phase noise
~1 GHz
Device Lab Inc. Slide 5
OriginsofNoise
Ø 1/f(flicker)noise,randomtelegraphnoiseü trapping/de-trappingbetween
oxidetrapsandchannelelectrons(numberfluctuation)
ü mobilityfluctuation
Ø Thermal(Johnson)noiseü phononscatteringü resistance
Ø Shotnoiseü Discretenatureofconductingelectronnumber
EFEV
EC
SiO2 Si
Inversion carriers
Gate�
Rsd� Rsd�
Rch�
Potential barrier�
Source� Drain�
Trap�
Scattering�
P∝Exp −2d
α−ΔEkT
⎛
⎝⎜
⎞
⎠⎟
d: distance between a trap and an electron
α: localization length of wave-function
∆E: the energy difference between the trap and the electron
Trap/detrapprobability
Device Lab Inc. Slide 6Slide 6
Why 1/f properties?
From von Haartman, Ph.D thesis (2006).
Ø Random telegraph noise: Lorentzian-shaped power spectral density
Ø 1/f noise: superposition of Lorentzians with various time constants
RTN 1/f
Device Lab Inc. Slide 7
Measurement of thermal and shot noiseNoise figure (Y-factor method)
Ø Phenomenawithultrafastrelaxationtime(e.g.,phononscattering)
Ø Y-factormethodonwaferrequiresü anoisesourceandhigh-frequencysetups
(probes,teststructuresforde-embedding)
Agilent Application Note 57-1
Gate�
Rsd� Rsd�
Rch�
Potential barrier�
Source� Drain�
Trap�
Scattering�
Device Lab Inc. Slide 8
Excessnoisefactor@10GHz
G. D. J. Smit et al., EDL 31 (2010) 884.
Ø A significant increase in RF noise compared to PSP prediction was observed for sub-100-nm MOSFETs.
PSP: a compact MOSFET model developed by Arizona State University and NXP Semiconductors Research
Device Lab Inc. Slide 9
Ø Phase noise at high offset frequencies (∝B) is assuming greater importance
C = B log2 1+SN
!
"#
$
%&
Ideal
Actual spectrum with phase noise
Osc
illat
or o
utpu
t po
wer
Frequency
Phase noise originates from devices (thru up-conversion), circuit, and outside
Offset freq.
C: Channel capacity in bits/sB: Channel bandwidth in HzS: Signal powerN: Noise power
Shannonʼs equation
Why measure high-frequency noise?
Standard maxB
4G 20MHz
5G 400MHz
WiGig(802.11ad) 2.16GHz
THz(802.15.3d) 69.12GHz
Device Lab Inc. Slide 10
Accurate noise model a must for 5G & beyond
Ø Adverse effects of white noise on mm-wave have experimentally been confirmed [1][2]
Ø High-frequency device noise is not merely a theoretical, academic concern!
Ø Measurement-based predictive device noise model is vital to success of 5G and Beyond 5G
[1] Chen et al., “Influence of white LO noise on wideband communication,” IEEE Microwave Theory Tech., vol. 66, no. 7, pp. 3349–3359, July 2018.[2] Chen et al., “Does LO noise floor limit performance in multi-gigabit millimeter-wave communication?,” IEEE Microwave Compon. Lett., vol. 27, no. 8, pp. 769–771, Aug. 2017.
Device Lab Inc. Slide 11
Low-frequency noise (empirical)
McWhorther modelØ Number fluctuation
Hooge modelØ Mobility fluctuation model
NfIS
d
id
⋅= H
2
α
tgox VVWLCqN −=
Device Lab Inc. Slide 12
SId =
KF gm2
CoxWLeff
1f AF
SId ,1/ f =
q2kTIdµeffLeff2 Cox f
ef 108NOIAln N0+N
*
NL +N*
⎛
⎝⎜⎜
⎞
⎠⎟⎟+NOIB(N0 −NL)+
NOIC2 N0
2−NL2( )
⎡
⎣⎢⎢
⎤
⎦⎥⎥+ ...
Noise models
1/f noise (SPICE2)
SId ,therm =
83kT(gds + gm+ gmb) for Vds >Vdsat
Thermal noise (SPICE2)
1/f noise (BSIM3)
Device Lab Inc. Slide 13
Low-frequency noise measurement systems
Ø ProPlus Design Solutionsü 9812D Advanced
1-f Noise Analyzerü Maximum frequency: 10 MHzü Current LNA noise floor: 3.6×10-23 A2/Hz @ 5 kHz
(Wideband)
Ø Keysight Technologiesü E4727A Advanced Low-Frequency Noise Analyzerü Maximum frequency: 40 MHzü Current LNA noise floor: 1×10-24 A2/Hz @ 10 kHz
Device Lab Inc. Slide 14
10-14
10
-12
10-10
10
-8
10-6
10-4
10-2
Id (A
/µm
)
1.20.80.40.0-0.4-0.8Vg-Vt (V)
Vd: 1 V Vd: 50 mV
Noise floor and frequency limit
Ø Limitationü Amplifier designü Configuration (wiring)
SiO2 3nm L: 240 nm Vt: 0.01 V
Vg-Vt: 1.0 V Vg-Vt: 0.2 V Vg-Vt: 0 V Noise floors
1mA
100µA
10µA
Range:
Demonstration by Keysight B1530A WGFMU (Fast IV)
Device Lab Inc. Slide 15
Concept of high-frequency noise probe
Ø Concept of high-frequency noise probeü Locate the LNA as close to DUTs as possibleü Very wideband LNA
LNA
Four pins (D,G,S,Sub)
K. Ohmori et al., 2013 VLSI Circuits. K. Ohmori et al., 2012 VLSI Technology.
Device Lab Inc. Slide 16
Vdm
Vd
Vg Vs (GND)
Vsub
DUT (nFET)
Vout
D
SSub
G
Probe On wafer
Stabilized DC- power supply
Spectrum analyzer
Rpu
Instruments
Rm
Amp
VVoltmeter
Rg
Rsub
Simplified diagram of the Noise Probe System
Ø Probeʼs built-in amplifier enables high-frequency measurements.
Ø The pull-up resistor, Rpu, often (but not always) dictates floor noise and the permissible Id current.
Device Lab Inc. Slide 17
Amplifier calibration for noise measurement
Ø Detailed measurement-based calibrationü Solid theoretical foundationsü No fudge factorsü Amplifier noise accounted for
LNA
Four pins (D,G,S,Sub)
Device Lab Inc. Slide 18
Semiconductor Analyzer (B1500A)
Function Generator
Digital Multimeter
Spectrum Analyzer
Entrope Noise Probe
LNA Power Supply
Monitor
Hardware components
Probe Station
Entrope™ Noise Probe
Entrope™ Software (PC)
Entrope™ system
Should be purchased through a local distributer
Device Lab Inc. Slide 19
Probe type Frequency Floor noise (minimum)
Entrope™ 101A 100k - 100MHz ~3×10-23 (A2/Hz) @ 10 MHz
Entrope™ 102A 100k - 100MHz ~5×10-23 (A2/Hz) @ 10 MHz
Specification of Entrope™ Noise Probe
System specification
Max DUT bias voltage 40 V
Max drain current 5 mA
No. of biasing terminals 3 (Source: GND)
RDUT > 50 Ω
Device Lab Inc. Slide 20
Ø Sampleü n-MOSFETü Technology node: 0.13 µmü Lg: 120/180/240/300 nm
Ø Measurementü dc (Id-Vd, Id-Vg, Id-Vb)ü noiseü at room temperature
Sample and measurement conditions
Id-Vd (Lg: 120 nm)2.5
2.0
1.5
1.0
0.5
0.0
Id (m
A)
0.80.60.40.20.0
Vd (V)
Vg: 0.6 V, Vg: 0.7 V, Vg: 0.8 V Biases for noise measurement
Device Lab Inc. Slide 21
Noise measurement (Vgs: 0.6V, Lg: 120 nm)
Vds (V): 0, 0.02, 0.04, 0.06, 0.08, 0.1, 0.2, 0.3, 0.5, 0.6, 0.7, 0.8
1/f
Device Lab Inc. Slide 22
Noise measurement (Vgs: 0.7/0.8V, Lg: 120 nm)
2.5
2.0
1.5
1.0
0.5
0.0
Id (m
A)
0.80.60.40.20.0
Vd (V)
Vg: 0.6 V, Vg: 0.7 V, Vg: 0.8 V Biases for noise measurement
Vgs: 0.6, 0.7, 0.8 VVds: 0, 0.02, 0.04, 0.06, 0.08, 0.1, 0.2, 0.3, 0.5, 0.6, 0.7, 0.8 V
Device Lab Inc. Slide 23
Variability (Lg: 120 nm)
DUT1 DUT2
DUT3Ø Device-to-device
variability seems to be large in the lower frequency region.
Ø RTN appears even in the higher frequency region.
Device Lab Inc. Slide 24
SId ,therm =
83kT(gds + gm+ gmb) for Vds >Vdsat SPICE2
10-23
2
4
68
10-22
2
4
68
10-21
Sid
@ 3
00M
Hz
(A2 /H
z)
10-232 3 4 5 6 7 8
10-222 3 4 5 6 7 8
10-21
(8kT/3)(gm + gds + gmb) (A2/Hz)
SId extraction in saturation (Lg: 120 nm)
Vgs: 0.6, 0.7, 0.8 VVds: 0.2, 0.3, 0.5, 0.6, 0.7, 0.8 V
Vgs: 0.8V0.7V
0.6V
Model: only thermal noiseConductance (gds, gm, gmb)10
8
6
4
2
0
Con
duct
ance
(mS
)
0.80.60.40.20.0Vd (V)
gm + gds + gmb
gds
gmb
gm
Vgs: 0.8 V
Ø Sid values at 300 MHzØ Saturation region (Vds > 0.2 V)
Excess noise confirmed!
Device Lab Inc. Slide 25
Ø Measurement-based predictive white noise model is vital to success of 5G and Beyond 5G
Ø By using Entrope™ Noise Probe, we demonstrated higher frequency noise measurementü Clear direct observation of white noise ü Simple reliable approach for predictive noise models
Summary of our message