Ferroelectric Memory and Negative Capacitance T.P. Ma Yale - - PowerPoint PPT Presentation

Ferroelectric Memory and Negative Capacitance

T.P. Ma Yale University

• A class of ionic crystals that exhibit spontaneous

polarization, switchable by external electric fields

• Must be below the ‘Curie’ temperature

Ferroelectrics: The Basics

Example: PZT

Ferroelectric Polarization

• - - -

Electrode Electrode + + + +

• - - -

+ + + +

• P

+P df, f ,P(E) V

P-E hysteresis

Ps Pr Ec

P E

0’

The 1T-1C Memory Cell

based on Metal/Ferro/Metal (MFM) capacitor

(the current commercial products)

Word Line Bit Line

Similar to DRAM Cell Structure

MFM capacitor

The Single-Transistor (FeFET) Memory Cell

Similar to Flash Cell Structure

Operating Principle of FeFET

VD ID n+ n+

P

• - - - -

ID ~0 VD n+ n+

P

+ + + + +

Programming: Read out:

(@ Vg=0) ID0 "0" Large ID "1"

+ + + + +

• - - - -
• +
• +
• +
• +
• +

(FE)

• n

"1"

• +
• +

(FE)

• ff

"0"

Id-Vg Hysteresis of FeFET

• 2
• 1

1 2 3 4 1E-5 1E-4 1E-3 0.01 0.1 1 10

Id(uA)

Vg(V)

5

Pt/SBT/Y

2O3/Si

W/L(um)=40/2

Memory Window

The Obstacles that have so far Prevented the Commercialization of FeFET

• Problems with Previously Available

Ferroelectrics (PZT and SBT)

• There are Tremendous Difficulties in
• CMOS Compatibility
• Thermal Budget; Process Integration
• Scaling Below 0.1 µm Node

Representative Memory Window Retention Characteristics

1 10 100

1.0 1.5 2.0 2.5 3.0 4

log( ) 1.2 10 t





4

log( ) 2.7 10 t





Depolarization Leakage

log( ) 1.15 t log( ) 2.39 t

Vg(V)

Time(s)

PZT FeFET

What Limits the Retention Time?

• Depolarization Field
• Gate Leakage/Trapping

T.P. Ma and Jin-Ping Han, IEEE Electron Device Letters, 23(7), p.386 (2002).

Depolarization Field Comes from Incomplete Charge Compensation

For Ideal M-F-M Structure with Complete Charge Compensation, There is NO Depolarization Field

• Electrode

Electrode + + + +

• + + + +

Compensating Charge Compensating Charge

– P P +

Depolarization Field Comes from Finite MIS Capacitance

Gate voltage V induces polarization P, and voltage across ferroelectric VF = CISV/(CIS + CF) – P/(CIS + CF). When V goes to 0, VF + VIS = (QF – P)/CF + QIS/CIS = 0, which reduces to QF = PCIS/(CIS + CF), leading to a depolarization field Edp = PCF/(CIS + CF).

For FEDRAM, the finite insulator/semiconductor capacitance, CIS, causes incomplete charge compensation.

Gate Leakage/Trapping Causes Memory Loss

Electron (or Hole) Injection Followed by Trapping Leads to Local Charge Compensation and Gradually Diminished Effect of Polarization.

+ + +

• p-Si

M Jf(Ef, t) Ji(Ei) F I + +

• +

+ +

• p-Si

M Jf(Ef, t) Ji(Ei) F I

“ON” “OFF”

FeFET Memory Retention Loss Mechanisms

1 10 100

1.0 1.5 2.0 2.5 3.0 4

log( ) 1.2 10 t





4

log( ) 2.7 10 t





Depolarization Leakage

log( ) 1.15 t log( ) 2.39 t

Vg(V)

Time(s)

PZT FeFET

The Game Changer:

HfO2-Based Ferroelectrics

• CMOS Compatible Materials and Processes
• Desirable Combination of Dielectric

Constant, Remnant Polarization, and Coercive Field

• Scalability: Same as CMOS Gate Dielectric
• J. Müller et al., IEEE IEDM Technical Digest pp.10.8.4-10.8.5. (2013)

Retention Time For HfO2-Based FeFET

• Depolarization Field

Edp(HfO2) < Ec(HfO2) [~1MV/cm]:

Nearly No Rentention Loss Due to Depolarization Field

• Gate Leakage/Trapping

Trap Density in HfO2 is 2 Orders of Magnitude Lower Than those in PZT and SBT

Calculation of Edep

• Polarization, F and CF
• Series comb. CIS
• IL cap. CIL
• Semi cap. CS

CIS=CILCS/(CIL+CS) CILCS_min/(CIL+CS_min)<CIS<CIL

Edep = PCF/[F(CIS + CF)]

• N. Gong, and T.P. Ma, SISC 2016

Parameters affecting Edep

Comparison of Edep/Ec for HfO2, PZT, SBT

(Fixed MW) MWdFEc

• Edep/Ec (HfO2) is smallest
• Least retention loss

[1] W. Shih, et al., J. Appl. Phys. vol. 103, 2008. [2] S. Lee, et al., J. Appl. Phys. vol. 91, , 2002. [3] J. Müller, et al., ECS J. Solid State Sci. and Tech., vol. 4, 2015. [4] J. Müller, et al., ECS Trans., vol. 69 , 2015.

Fixed, the same MW

PZT SBT FE-HfO2 Ec 31 kV/cm 90 kV/cm 1 MV/cm

HfO2 largest Ec

HfO2 thinnest  Scaling adv. Id Time

Fixed MW Less Retention Loss

HfO2 is most scalable & has longest retention

Why Edep/Ec (HfO2) is smallest? What about retention of scaled PZT/SBT?

[1] W. Shih, et al., J. Appl. Phys. vol. 103, 2008. [2] S. Lee, et al., J. Appl. Phys. vol. 91, 2002. [3] J. Müller, et al., ECS J. Solid State Sci. and Tech., vol. 42015. [4] J. Müller, et al., ECS Trans., vol. 69, 2015.

(Edep/Ec)min~ 2P/[MW*(CIL+CF)]

Why Edep/Ec is Small for HfO2

• Largest Ec

smallest dF largest CF smallest Edep/Ec

(Edep/Ec)max/ (Edep/Ec)min ~ CIL/(CF+CIS_min)

Scaling Consideration for FEDRAM

PZT & SBT Require Too Thick FE Layer – Not Suitable for Production

Scaling Consideration for FEDRAM

• J. Müller et al., IEEE IEDM Technical Digest pp.10.8.4-10.8.5. (2013)

HfO2-Based FeFET < 10nm in Thickness

Demonstration of 28nm FeFET Cell

Fabricated at Our Foundry Partner

Excellent Memory Characteristics

Sizable Memory Windows Have Been Measured

10-Year Retention of FeFET Cell

Endurance of HfO2-Based FeFET

Trade-off Between Retention and Endurance

High Endurance but Short Retention – DRAM Like

Excellent Manufacturability

Integrating FE-Gate with CMOS HKMG

Any advanced CMOS foundry can manufacture without new equipment.

Proposed 3-D FeFET Memory Arrays

US Patent Issued

Summary of FeFET Memory

• FeFET memory has numerous advantages over

conventional Flash

• Either Flash-like or DRAM-like memory can be

realized by tuning the programming strength

• Test memory cells have been demonstrated with

foundry partner’s 28nm technology

• It’s most suitable for eFlash applications
• It has great potential for low-cost, high-density

storage technology

Negative Capacitance: Science Fact or Science Fiction?

A Close Examination of Salahuddin/Data's Original Paper in Nano Letters

Negative Capacitance (NC) Was Proposed

11/15/2018 33

Salahuddin claimed that the ferroelectric capacitor can be negative!

• S. Salahuddin, S. Datta, Nano Letter (2008)

Ginzburg-Landau theory



11/15/2018 34

Phase transition from paraelectricity to ferroelectricity

NC Theory



11/15/2018 35

• S. Salahuddin, S. Datta, Nano Letter (2008)

Negative Capacitance! But unstable!

Quasi-Static NC (QSNC) theory



11/15/2018 36

“Quasi-static negative capacitance” (QSNC) theory

Amazing Predictions of the QSNC Theory

11/15/2018 37

• J. Van Houdt, et.al EDL

(2018)

What’s Wrong with the QSNC Theory?

11/15/2018 38

Mistake 1



11/15/2018 39

In equilibrium, only local minimum states exist! We need

Mistake 1 (Continued)



11/15/2018 40

“Negative Capacitance” region cannot exist in equilibrium

Mistake 2



11/15/2018 41

Therefore, the P-E (or P-V) “S Curve” cannot be translated directly to the Q-V curve

It’s well known that P ≠ Q in a FeFET

Why is it P ≠ Q in a MOSFET or any other circuit like the following?

Answer: Due to Incomplete Charge compensation

P Q Q

Vg

Ψs

P ≠ Q

Whenever there is a capacitor In series with the ferroelectric capacitor,

P ≠ Q,

which causes a depolarization field that hurts the retention time This is well known in the ferroelectric memory community*.

*IEEE Electron Device Letters, vol.23(7), 386 (2002)

Mistake 2 (Continued)

Mistake 2 (Continued)

11/15/2018 43

Actual Case

NC theory

Mistake 3

 Ignoring the strong

Coupling of FE/DE Layers

11/15/2018 44

• J. Kittl, et.al APL (2018)

Mistake 3 (Continues)

11/15/2018 45

Experimental Verification

 Experimental MFM Capacitor

11/15/2018 46

C-V Measurements

11/15/2018 47

The FE + DE Series Capacitance Always Shows Lower Total Capacitance

V

APP-VDE Measurements

11/15/2018 48

Sweep rate: 56V/ms Sweep rate: 56V/ms Sweep rate: 80V/ms Sweep rate: 80V/ms

VAPP-VDE Measurements (Continued)

11/15/2018 49

Sweep rate 14V: 56V/ms 8V: 32V/ms

f = 1kHz f = 1kHz f = 1kHz

Sweep rate 20V: 80V/ms 16V: 64V/ms 12V: 48V/ms 8V: 32V/ms

Summary of Negative Capacitance

• Salahuddin’s Quasi-Static Negative

Capacitance (QSNC) theory has been closely examined

• Several major mistakes have been

identified

• Experimental results did not support the 3

major predictions of the QSNC theory

Quasi-static Negative Capacitance: Fact or Fiction?

Thank You!