PSP 101.0 Update

1
PSP 101.0 Update

• PSP Group
• May 2006

2
Outline

• SWJUNCAP correction (Infineon, Freescale)
• Documentation (CMC)
• Poly depletion improvement (Philips)
• Higher-order derivatives (IBM)
• Infineons comments (Infineon)
• Code (Freescale)
• 1/f noise (TI)
• Conclusion

3
SWJUNCAP issue

• SWJUNCAP2, 3 was introduced for BSIM
  compatibility
• If SWJUNCAP2, the gate-edge length LG is
  currently set to wrong value
• Currently LG 0
• Correct LG WE
• Will be fixed in next PSP release

4
Documentation

• The document entitled From BSIM3/4 to PSP
  Translation of Flicker Noise and Junction
  Parameters has been placed on PSP website
• The document entitled Summary of Improvements in
  PSP101.0 relative to PSP100.1  has been placed on
  PSP website

5
Poly Depletion (Philips)

• Parameter NP represents poly doping level
• If NP 0, poly depletion (PD) is switched off
• Scaling rule NP NPO (1 NPL LEN / LE)
• Commonly NPL lt 0
• NP decreases with decreasing LE
• So, PD effect increases with decreasing LE
• Below certain LE, (1 NPL LEN / LE) becomes
  negative
• NP is clipped to 0
• PD-effect is suddenly switched off
• Issue gives discontinuous behavior as function
  of L

6
Poly Depletion

• Proposed solution adapt scaling ruleCurrently
  NP NPO (1 NPL LEN / LE)Proposed NP
  NPO MAX(10-6, 1 NPL LEN / LE)
• Backwards-compatible in all practical cases
• Avoids unexpected phenomena during parameter
  extraction

7
Poly Depletion

• If NP is very large, the PD effect is (almost)
  negligible
• Currently PD is not calculated for NP lt 1 and
  for NP gt 1028
• Proposed PD is not calculated for NP 0 only
• In new situation the effect of NP is more
  transparent
• Setting NP 0 (local model) or NPO 0 (global
  model) switches PD effect off
• In all other cases PD effect is included, even if
  negligibly small
• Fully continuous behavior as function of L
  (global model) and as function of NP gt 0 (local
  model)

8
Higher-Order Derivatives at VDS0 and AX

• Most compact models (e.g. PSP and BSIM4) use soft
  clamping of VDS to some value Vdsat in the
  saturation region
• Clamping function determines
• Transition from triode to saturation region
  (intentional)
• Behavior around VDS0 (unintentional, but
  unavoidable)
• In PSP see Joardar et al., TED 45(1), 1998
• Best possible solution known today

9
Higher order derivatives and AX

• AX scaling is intended to improve gDS description
• Long devices require larger values (e.g., 12-20)
• Short devices require smaller values (e.g., 3-8)
• AX also influences distortion description around
  VDS0
• AX gt n - 1 ensures that n-th derivative exists at
  VDS0
• PSP forces AX gt 2, so 3rd derivative always
  exists
• Larger AX suppresses oscillations around VDS0
• Only for short devices extra care may be needed
• Higher-order distortion description can be traded
  against gDS description

10
Limiting AX

• If one needs n-th order derivative
• In local model
• Keep AX gt n 1
• In global model
• Keep
• where LE,min is LE for shortest channel
• Scaling rule

11
Gummel Test 3rd derivatives
Vg 1.2V, Vd Vg/2 Vx, Vs Vg/2 Vx, Vb
0V, W/L 10/0.1µm, default parameters, sweeping
AXL
12
Gummel Test 5rd derivatives
Vg 1.2V, Vd Vg/2 Vx, Vs Vg/2 Vx, Vb
0V, W/L 10/0.1µm, default parameters, sweeping
AXL
13
Termination 5th derivatives
Vg 1.2V, Vd Vg/2 Vx, Vs Vg/2, Vb 0V, Rs
250O W/L 10/0.1µm, default parameters,
sweeping AXL
14
DPHIB Scaling

• Discrepancy between model code and documentation
• document DPHIB (DPHIBO DPHIBL (LEN/LE))
  
• code DPHIB DPHIBO (1 DPHIBL (LEN/LE))
  
• First scaling rule is preferred allows for usage
  of DPHIBO0
• Will be changed in next release
• Existing parameter sets can be easily updated
  from existing to new PSP version

DPHIBL DPHIBO DPHIBL
15
More Scattering for PSP101.0 than for PSP100.1
(Infineon)?
AF vs. L, PSP100
PSP101 DPHIB_eff vs. L
Actually, same degree of scattering but for
different parameters
16
Infineon Depletion Capacitance

• Infineon reports too smalldepletion capacitance
• Explanation
• Body effect decreases atlarge negative Vb
• NEFF based on average body effect for all Vb
• Therefore, NEFF too small at Vb0
• Consequently, depletion capacitance too small

Non-uniform doping is a major direction of
research at PSU
17
Code Correction (Freescale)

• In lines 1430, 1528 and 1555 of
  PSP101_module.include 10E-5 will be changed into
  1E-5
• Invisible to the model user as far as we can tell
  at this point

18
1/f-noise issues (TI)

• TI requested 2 modifications to 1/f-noise model
• 1/f noise variations are larger for smaller
  transistors
• EF dependence on L

19
1/f noise variations

• Increased variation of 1/f noise is indeed
  expected based on the theory behind the
  unified1/f noise model (implemented in BSIM3/4
  PSP)
• Large-area devices
• Many traps
• Individual spectra close to 1/f
• Small spread across wafer
• Small-area devices
• Few traps
• Individual spectra non-1/f behavior
  (Lorentzians)
• Even swapping source and drain may change noise
  spectrum
• Spectrum averaged over many devices 1/f
• Large spread across wafer

20
Small Area Devices
Philips results 90-nm CMOS (Scholten et al.,
ICNF 2005)
PMOS
NMOS
confirms microscopic origin of 1/f noise in terms
of traps similar experiments reported in
Brederlow et al.,IEDM1999, p.160, 0.25 µm CMOS
Scholten et al., TED50, p.618, 2003, 0.18 µm
CMOS
21
Spread vs. Area
G. Ghibaudo (ESSDERC 94)
data 0.35 µm CMOS
theory
22
Statistics of 1/f Noise (I)
Infineon results
Brederlow et al.,IEDM1999, p.160
Wirth et al., TED52, p.1576, 2005
23
Statistics of 1/f Noise (II)
Philips results (Scholten et al., ICNF 2005)
PMOS
NMOS
levelling-off also seen by Wirth
24
Statistics of 1/f Noise (III)
Ertürk et al. (IBM), ELECTRONICS LETTERS 27th
October 2005 Vol. 41 No. 22
smaller W ? larger spread
25
Summary - Noise

• Based on theory and experiments, a smaller device
  area leads to
• Larger spread in 1/f noise
• Larger deviation from 1/f behavior (e.g. EF ? 1)
• The core model should describe noise for small
  area devices as 1/f, since the average spectrum
  is 1/f
• Statistical variations should be implemented
  externally (as part of statistical model)
• First step vary noise parameters with s ?
  1/sqrt(area)
• Second step EF variation

26
PSP Update Summary

• PSP update will include
• Correct DPHIB scaling
• 10e-5 ? 1.0e-5
• Fixed SWJUNCAP usage
• Fixed poly doping scaling
• Not backwards compatible, but very easy
  conversion
• Prepared to move fast
• Verilog-A
• C-code in SiMKit
• Naming PSP 102.0
• Should we retain PSP 101.0?

27
Conclusions

• PSP102.0 can be released very soon
• 5th derivative issue is clarified,
  recommendations are made for the model users
• Infineons comments are acknowledged and included
  in the research plans (SRC proposal)
• as a part of general effort on non-uniform
  doping
• 1/f noise description is clarified
• Thank you for your input!