Progress in Noise Figure Studies of GaN HEMTs

1
Progress in Noise Figure Studies of GaN HEMTs
CANE Center for Advanced Nitride Electronics

• University of California, Santa Barbara
• Chris Sanabria Hongtao Xu Tomás Palacios Likun
  Shen Arpan Chakraborty Sten Heikman Val
  Zomorrodian Umesh K. Mishra Robert A. York

UCSB VIRGINIA TECH WRIGHT STATE UNIVERSITY
UNIVERSITY OF MISSOURI-ST. LOUIS CALTECH THE
UNIVERSITY OF MICHIGAN THE OHIO STATE UNIVERSITY
2
From Last Review

• Two Differential Oscillator Designs
• First GaN differential oscillator
• Accepted IEEE Microw. Comp. Lett.
• Excellent harmonic performance
• Comparable power to other GaN results

• Phase Noise Measurements
• High 1/f3 corner (this and others work)
• Similar noise performance at off-sets past 1/f3
  corner

3
New Improved

• Redo of previous design
• Larger tank C
• Smaller tank L
• Still NO FIELD-PLATE
• SiN capacitors (No BST)
• Biased lower than previously
• Had thought highest power best phase noise
• Biasing lower ? 4 dBc/Hz phase noise improvement
• Similar power and efficiency to previous work

4
New Improved

• Taking into account measurements being at
  different biasings, is a 16 dB improvement
• Just the tank caused improvement?
• Better material?
• Still working towards oscillator prediction
• But this presentation will be about

5
Outline

• Modeling
• NF with gate leakage
• Some insight on field plate NF
• NF modeling with Zgd
• Small-Signal Parameters vs. Bias
• Noise Measurements
• NF and gate leakage part 2
• AlN interlayer with Sap. SiC substrates study
• Noise of Other Devices
• High frequency (Tomás Palacios)
• Thick GaN cap (Likun Shen)

6
Systems

• Focus Load (Source)-Pull
• Learning system for noise
• NF to 26 GHz, to 40 GHz later
• Cryogenic Source-Pull NF Measurements
• Tuner stands completed
• RF lines replaced, will be back soon
• Maury Load-Pull
• Wider frequency range (3 to 16 GHz)
• New configuration
• No load tuner
• Better repeatability (days instead of weeks)

7
Predicting Noise Figure

• Had Previously Used
• Pucel Model
• Pospieszalski Model
• Concerns
• Can model in a circuit simulator
• -Might be hard to make bias-dependent
• -Both require NF measurements prior to modeling
• -Not easy to explain device noise (ex. Why FP
  better?)
• Different approach ? van der Ziel

8
How To Predict Noise Parameters?

• Usually in very simple form
• Math is easy
• Insightful
• Will derive with gate leakage
• Assumptions
• Rd is negligible ? on output
• Cpgd, Cpgs, Cpds, Cds, Lg, Ls, Ld are negligible
  ? only change matching
• No feedback (for the moment)
• Zgd Rgd 1/(jwCgd) open
• Emitter degeneration
• No correlation between sources
• G 2/3

9
Predicting Noise Parameters

• Even with Igs, equations are manageable
• Insights
• F a Ig
• The drain noise term and input parasitics term
  are comparable in contributing to NFmin at best
  noise biasing
• Need Rs, Rg, Ri, gm, Cgs, Igs, and G to predict
  noise

10
Modeling with Gate Leakage (but Not Zgd)

• G 2/3. Other values not as accurate.
• Works on all previous samples.

11
How Does Gate Leakage Affect the Noise Parameters?

• GaN HEMTs have enough gate leakage to be a
  problem for noise
• Matching may be a little less dependent on gate
  leakage because parasitics were ignored in
  modeling
• Next Get working in ADS

12
How Does Gate Leakage Affect the Noise Parameters?

• A 0.1 0.3 dB degradation for these devices

13
Field Plate Noise Parameters

• Measurements _at_ 5 GHz, from H. Xu
• Field plate changes the noise parameters in a
  complicated way, but for an improvement.
• H. Xu has also seen a similar trend in phase
  noise performance of oscillators with FP
• Is this improvement from improved Rg?

14
Other Parasitics and Noise Parameters

• Table at 5 GHz
• Rg contributes 0.1 dB to NF
• Hard to believe FP improvement of noise is just
  from lower Rg
• Rs contributes 0.2 dB to NF

15
Modeling Including Zgd Current Work

• Theoretical analysis too complex to gain useful
  insight.
• Work to date shows
• Gain decreases ? NF increases
• Changes noise match
• Might have to look for another way to explain
  field plates effects on noise.

16
Modeling Summary

• Reasonably good noise parameter prediction when
  including gate leakage
• ? can be used to understand where noise in device
  is coming from
• Gate leakage important for noise performance of
  GaN HEMTs
• A decrease in gate resistance is not enough to
  explain why field-plating helps the noise
  performance of GaN HEMTs

17
Outline

• Modeling
• NF with gate leakage
• Some insight on field plate NF
• NF modeling with Zgd
• Small-Signal Parameters vs. Bias
• Noise Measurements
• NF and gate leakage part 2
• AlN interlayer with Sap. SiC substrates study
• Noise of Other Devices
• High frequency (Tomás Palacios)
• Thick GaN cap (Likun Shen)

18
Small-Signal Parameters

• 1. For bias-dependent noise modeling, must know
  how the small-signal parameters change with bias
  too
• Easier to model noise in circuit simulator if
  noise sources dont change vs. bias
• Not bias-dependent
• Thermal noise (but the resistance may be
  bias-dep.)
• Bias-dependent
• Shot noise
• Low-frequency noise
• Drain noise might be (G)
• 2. When comparing different devices for noise,
  want to see if any differences that may occur in
  noise is from noise sources or from a difference
  in the device small-signal parameters.
• How are the small-signal parameters different for
• SiC?
• Sap?
• With a AlN interlayer?
• Field-plated devices?

19
Small-Signal Parameters

• Extraction (Instead of H. Xus large
  small-signal model)
• S-parameters taken during noise measurements at
  each bias
• Small signal circuit constructed for each bias in
  ADS
• Checked against measured S-parameters
• Applied to
• Sapphire no AlN interlayer
• Sapphire with AlN interlayer
• SiC no AlN interlayer
• SiC with AlN interlayer
• Devices are not field-plated
• Extrinsic parasitics assumed to be the same
• Will cause error, but should be the same error
  across the different samples
• Will only show two samples for clarity

20
Small-Signal Parameters Capacitances gm

• Versus Samples
• Only difference is Cds smaller for SiC samples
  (not shown)

• Versus Bias
• Ids? Cgd ? (25 fF ? 50 fF)
• Ids? Cgs ? (0.22 pF ? 0.24 pF)
• gm as expected for DC I-V
• t and Cds relatively flat

21
Small-Signal Parameters Resistances

• Versus Bias
• Ids? Ri ? (10 W ? 18 W)
• Ids? Rds ? (1100 to 650 W
  ? 400 to 200 W)
• Rgd hard to determine, probably changes a little
  as with Cgd

• Versus Samples
• SiC samples have better Rds

22
Small-Signal Parameters Summary

• Parameters not different with AlN layer
• True when looking at similar ft, fmax devices
• AlN layer still gives better devices (ft and fmax
  higher)
• Only Cds and Rds are different between SiC and
  Sapphire samples.
• Not important for noise measurements and modeling
• But the SiC samples will have better fmax because
  of better Rds
• Some of the parameters important for noise (Ri,
  Cgs, extrinsic resistances) might be able to be
  modeled with an average value, or a linear fit
  vs. bias.
• No bias-dependence for field-plated devices, but
• From H. Xu LFP vs. Rg
• With LFP ? , Cgd ? and Rgd ?

23
Outline

• Modeling
• NF with gate leakage
• Some insight on field plate NF
• NF modeling with Zgd
• Small-Signal Parameters vs. Bias
• Noise Measurements
• NF and gate leakage part 2
• AlN interlayer with Sap. SiC substrates study
• Noise of Other Devices
• High frequency (Tomás Palacios)
• Thick GaN cap (Likun Shen)

24
SiC/Sap., AlN Study Reminder

• Showed previously that AlN layer helps noise
  performance
• Now trying to
• Reconfirm with new samples
• See if AlN interlayer on SiC gives even better
  performance
• Was having problems measuring because

25
Why the Difference?
Very similar

• An example of what we want to see for
  measurements

26
Why the Difference?
Noisier than expected
Almost 1 dB variation

• Why such a change? ft fmax are similar

27
A Check on Measurement Accuracy

• At 5GHz, still a 0.5dB Nfmin difference at 100mA
• Easier to see at 10GHz
• 0.7 dB _at_ 90 mA
• 1 dB _at_ 100 mA

28
Answer Gate Leakage

• Devices biased at Vds 5 V, Ids 10 mA, at 10 GHz

• A good HEMT with Igs 5mA has a NF 1.9 dB
• Must monitor ft, fmax, and Ig if comparing NF of
  different devices.
• If we use van der Ziel model with typical
  small-signal parameters and gate leakage to
  predict these values

29
SiC/Sap., AlN study

• No field-plates
• Same trend with new samples
• Max Igs of 6mA

• SiC performs better at higher biasing compared to
  Sap.
• SiC performance probably the same with/without
  AlN layer

30
Outline

• Modeling
• NF with gate leakage
• Some insight on field plate NF
• NF modeling with Zgd
• Small-Signal Parameters vs. Bias
• Noise Measurements
• NF and gate leakage part 2
• AlN interlayer with Sap. SiC substrates study
• Noise of Other Devices
• High frequency (Tomás Palacios)
• Thick GaN cap (Likun Shen)

31
High Frequency Devices

• Tomás Palacioss new standard high-frequency
  devices
• Newer devices are field-plated
• Better than a year ago by
• 0.3 dB
• NFmin 0.4 dB _at_ 10GHz
• Believed to be the record for the gate length
• Will measure with Focus load-pull later

32
Thick GaN Cap Devices

• Newer devices from Likun Shen
• High leakage at low bias (68 mA at Ids
• 10 mA)
• Lower NF at high Ids than even standard HEMTs on
  SiC
• Thick cap repressing noise?
• Measure before passivation for newer standard
  HEMT samples

33
Summary

• New differential oscillator has good phase-noise
  performance even without FP, high-Q capacitors,
  or optimizing for phase noise.
• Noise Figure of non-FP devices can be predicted
  without prior measurements.
• Gate leakage of GaN HEMTs is a problem for noise.
• Gate resistance alone does not explain why a FP
  device has better noise performance, nor does a
  feedback explanation.
• A thin AlN interlayer helps the NF performance at
  larger biasings.
• SiC gives better NF performance than Sapphire
  (even with an AlN interlayer) at large
  drain-source current biasings.
• UCSB may have the NF record in GaN (more to
  come).
• Thick GaN cap devices show low noise even at a
  high bias.

34
Looking Forward

• First
• Cryogenic NF measurements
• Try to understand FP
• NF measurements
• Look at bias-dependence
• Try different modeling
• Correlation?
• Other?
• Try to derive G?
• Comparison with GaAs HEMTs
• Then
• Low-frequency noise setup
• Noise modeling in ADS working with large
  small-signal model to predict phase noise