TCAD Simulation for ePCM
Technology Development
Elisa Petroni
Advanced Simulation for NVM, LETI Innovation Days – Grenoble
28/06/2019
Outline
• Introduction
• embedded PCM for automotive applicationsMaterials and architecture optimization
• TCAD device modelingtechnology needs and simulation calibration
• TCAD test casepredictivity and architecture optimum assessment
• Different approaches & outlooks
• Conclusions
2
IntroductionST Product Family Focus
3
The leading provider of products and solutions for Smart Mobility and the Internet of Things
Portfolio delivering complementarity for target end markets, and
synergies in R&D and manufacturing
Dedicated
Automotive ICs
Discrete &
Power
Transistors
Digital
ASICs
MEMS &
Specialized
Imaging Sensors
Analog, Industrial &
Power Conversion
ICs
General Purpose &
Secure MCUs
EEPROM
IntroductionPCM Working Principles
4
Phase Change Materials:
High r logic 0 Low r logic 1
Amorphous Crystalline
SET pulse
RESET pulse
time, nsec
Te
mp
era
ture
, °C
Tmelt
Tc
• Semiconductor ternary alloy (GST), usually Ge2Sb2Te5
• Two (or more) stable phases @ Tamb amorphous vs. crystalline phases
• Reading mechanism resistance change of the GST
• Writing mechanism self-heating due to current flow (Joule effect)
GST 225 22% Ge 22% Sb 55% Te
.1
.1
.1
.2
.2
.2
.3
.3
.3
.4
.4
.5
.5
.6
.6
.7
.7
.8
.8
.9
.9
.9
.1
.1
.1
.2
.2
.2
.3
.3
.3
.4
.4
.5
.5
.6
.6
.7
.7
.8
.8
.9
.9
.9
Sb
Ge Te
.4
.5
.6
.7
.8
ePCM for AutomotiveWhy and How
5
embedded-PCM for automotive:
• Easy integration with advanced logic
• Low voltage
• No impact on CMOS front-end
• Cost-effective
• Few additional masks
• Versatility & Scalability
• Unipolar programming MOS or BJT selector depending on specific
application requirements
• Programming current scaling with cell area reduction
• Reliability
• Extrinsic reliability assessed both after soldering and for 10y @150C HTDR
ePCM cell
ePCM for AutomotiveMaterial Engineering
6
Ge-rich chalcogenide:
• Germanium content is the key to improve Tc
of Ge-Sb-Te compounds
• Tc ~ 350 °C compliant with temperature
soldering profile for automotive
TCAD ePCM ModelingTechnology Needs
7
Why ePCM simulation is needed?
Predictable info in order to simplify process trials and integration speeding up technology development!
• Scalability efficiency check for dimension scaling (isotropic, non-
isotropic, etc…)
• Cell architecture functionality of different cell geometries and working
concepts
• Material properties optimization process improvement of
surrounding steps (i.e. for thermal confinement increasing …)
Russo et al., IEEE Trans. Electron Devices, 55, 2 (2008)
Synopsys Inc., Application Notes (2008)
8TCAD ePCM Modeling
Transition Rates (1/2)Phase Transition Modeling:
TCAD approaches differ for selected phase transition model
Transition Rate Equations:
SDevice by Synopsys® (→ Poisson-continuity solver)
• Each phase i is represented by volume fraction σi si = 1
• Temporal variations of si are defined by ordinary differential
equations ሶsi = σj≠iσtϵTijcij sj − eijsi
• Initial guess for si from equilibrium conditions energy, temperature
capture transition rate = c
emission transition rate = e
Tmelting = 888 K
crystalline
amorphous
Zi = giexp(−βEi)
Two phase model crystalline vs. noncrystalline
Melted state amorphous state with
physical properties for
T > Tm = melted state properties
TCAD ePCM ModelingTransition Rates (2/2)
9
Transition rate:
definition of one transition rate, tipically cA Ccapture rate c
Tg
For emission transition rates detailed balance principle:
e
c=sj∗
si∗
C A
• Arrhenius law
• Peng growth
• Simplified rate (accounting remaining crystallization
process from Tg to Tm) Theta function
TCAD ePCM ModelingCell Structure
10
3D structure
Wall architecture
• optimized heater dimensions enhancing joule
heating effect and programming efficiency
• Self-aligned structure TE + GST + Heater
Heater
Bottom electrode
Top electrode
Bottom electrode
Top electrode
c-GST
Y view direction X view direction
M0 CD
TCAD ePCM ModelingEmulated Cell
11
Calibration flow:
contact (BL)
conta
ct
(GN
D)
2D slice
std metals and insulators (ρ, kth, cV …):
surrounding materials, fundamental in ePCM
ACTIVE CELL:
RCELL = RGST + Rheater + RTE
RGST
RTE
Rheater
TE
BE
GST
Heater
Heater & TE – metals (ρ, kth, cV …)
GST semiconductor! Dirichlet boundary conditions
TCAD ePCM ModelingGST Material
12
GST is modeled as p-doped semiconductor
linking RGST to microscopic and transport properties
RGST = ρGST ∙ geometrical factor
In a semiconductor:
ρGST = ρ∞ eEACTkBT
ρGST =1
q(nμn + pμp)
ρGST ~1
qpμp=
1
qNVDOSμp
eEACTkBT
𝛍 calibration
Crystalline Amorphous = Melted
μn 2 150
μp 425 ൝150 at T = 300Kμpcrys
at T = 650K
𝐄𝐀𝐂𝐓 calibration
by tuning band energy structure in the two phases
(values to be considered as “effective”, that
take account of experimental 𝛒𝐆𝐒𝐓 = ρ∞)
TCAD ePCM ModelingBand Structure
13
from Adler (1978):
𝐄𝐆𝐀𝐏 = 0.5 eV → 0.715 eV
𝐄𝐓𝐑𝐀𝐏(−)
= 0.09 eV
Both crystalline and amorphous GST result in
p-doped semiconductors
EACT results consistent with experimental data
TCAD ePCM ModelingGST Calibration
14
• experimental data
simulated data
𝐑𝐒𝐄𝐓 = 𝐑∞ 𝐞𝐄𝐀𝐂𝐓𝐤𝐁𝐓
Right calibration of
ρGST = ρ∞ eEACTkBT
… but using “effective” or arbitrarily fixed values!
Experimental measurements are needed for
realistic ρGST calibration:
1. Hall measurements for μ and NVDOS
determination
2. Ellipsometry measurements for EGAP
estimation
TCAD ePCM ModelingSET State
15
SET characteristic:
• experimental data
• simulated data
RSET
RON
RGST
RTE
Rheater
for
𝐓𝐂𝐄𝐋𝐋 > 𝐓𝐦𝐞𝐥𝐭
RGST ≪ Rheater + RTE
→ RCELL = 𝐑𝐡𝐞𝐚𝐭𝐞𝐫 + 𝐑𝐓𝐄 = 𝐑𝐎𝐍
TCAD ePCM Test CasePredictivity
16
Scalability:
WCELL1
WCELL2
WCELL3
WCELL4
• experimental data
• simulated data
Sensitivity:
WCELL (= M0 CD)
• experimental data
• simulated data
Hheater
Hheater 3
Hheater 2
Hheater 1
WCELL1 > WCELL2 > WCELL3 > WCELL4Hheater1 > Hheater2 > Hheater3
TCAD ePCM Test CaseSET Sensitivity
17
RON sensitivity: RSET sensitivity:
Hheater
HGST
RON ∝ Hheater
𝐑𝐎𝐍 ~ 𝐑𝐡𝐞𝐚𝐭𝐞𝐫 + 𝐑𝐓𝐄
→ less influenced by HGST
RSET ∝ Hheater
𝐑𝐒𝐄𝐓 𝐯𝐬. 𝐇𝐆𝐒𝐓
→ analytical formula?
• vs. Hheater
• vs. HGST
• vs. Hheater
• vs. HGST
18
RGST = ρGST ∙ 𝐠𝐞𝐨𝐦𝐞𝐭𝐫𝐢𝐜𝐚𝐥 𝐟𝐚𝐜𝐭𝐨𝐫RGST = ρGSTන
0
HGST dr
S(r)
RGST = ρGSTන0
HGST dr
WCELL(t + 2αr)
RGST = ρGSTln(1 +
2αHGSTt )
2αWCELL𝛂
𝟐𝛂𝐫 + 𝐭
𝐭
𝐫
𝐇𝐆𝐒𝐓
TCAD ePCM Test CaseRSET Sensitivity
α = α(V,WCELL, 𝐇𝐆𝐒𝐓)
second order effects
TCAD ePCM Test CaseImelting Optimization
19
Optimizing cell architecture in order to minimize melting current (𝐈𝐦𝐞𝐥𝐭𝐢𝐧𝐠)
888 K
𝐈𝐦𝐞𝐥𝐭𝐢𝐧𝐠
for T > Tmelting = 888 K
→ melted GST
for present range of RGST
→ Heater Heating regime
𝐈𝐦𝐞𝐥𝐭𝐢𝐧𝐠 ∝ ൗ𝟏 𝐑𝐡𝐞𝐚𝐭𝐞𝐫
vs.
𝐑𝐒𝐄𝐓 ∝ 𝐑𝐡𝐞𝐚𝐭𝐞𝐫
Lowering Imelting
decreases readout
performance
TRADE OFF is needed
T = Tamb + I2RheaterRheaterth
TCAD ePCM Test CaseGeometrical Optimum
20
𝐈𝐦𝐞𝐥𝐭𝐢𝐧𝐠 @ fixed 𝐑𝐒𝐄𝐓: when Hheater is increased, HGST is decreased and vice versa
𝐑𝐒𝐄𝐓𝟏
𝐑𝐒𝐄𝐓𝟐𝐑𝐒𝐄𝐓𝟑
𝐑𝐒𝐄𝐓𝟒
minimum of Imelting varying cell geometry
Slight increase Hheater/HGST ratio on RSET3
RSET1 < RSET2 < RSET3 < RSET4
ST CELL
increasing Hheater/HGST ratio
geometrical optimum
Trade off between heating efficiency and
hot spot position
c𝜕T
𝜕t− 𝛻 ∙ (kth𝛻T) ∝ Jn,p
Different TCAD Approaches & Outlooks 21
Beyond Transition Rate Modeling …
… other TCAD approaches:
• Phase Field Method → dynamic of transition parameter and free energy
• Monte Carlo techniques → randomness of nucleation process
… Atomistic & molecular dynamics simulations
→ speeding up material engineering and understanding A. Glière et al., Proc. SISPAD Conference, 63 (2011)
G. Novielli et al., IEDM Technical Digest, 589 (2013)
Long-term Outlook:
coupled system of TCAD and ab-initio simulations …
… in an industrial environment!
Ge TeSb
FORMING EFFECT
Conclusions 22
• PCM solution is well positioned to match specific eNVM target
• TCAD simulation is a powerful tool yet for obtaining predictable info in terms of geometrical
scaling & sensitivity
• Correct calibration of phase change material requires deep knowledge of phase transition
physics and experimental measurements of energetic and transport properties
• Developed GST modeling is able to predict scaling efficiency and sensitivity of SET
resistivity
• Thanks to TCAD simulation, we are able to assess the geometrical optimum of our
technology, very close to present ST working point