TCAD tool:

An under-the-hood perspective
Oves Badami
[Link]@[Link]
Electrical Engineering Dept.
IIT Hyderabad
Past Present and Future
D
R
SRC DRN N
S
R
Bulk Planar Transistor C

DRAIN

SOURCE
SOURCE
SOURCE

6/12/2025 Oves Badami 2

Our activities @ IITH
• Modelling and simulation of electron
devices.
• Developing in-house TCAD tool for exploring
cutting edge research
• Exploring underlying fundamentals of
semiconductor devices
• Logic – Schottky FET/MuGFETs/Power Devices
• Memory – RRAM and 3D NAND Flash
• Role of ML in TCAD tools

6/12/2025 Oves Badami 3

Contents

• Computer modelling – Importance

• Under the hood of a TCAD tool

• Case Study of a Poisson Solver

• Advantage of in-house TCAD – Schottky Barrier FET

• Old wine in new bottle – Machine Learning

6/12/2025 Oves Badami 4

Contents

• Computer modelling – Importance

• Under the hood of a TCAD tool

• Case Study of a Poisson Solver

• Advantage of in-house TCAD – Schottky Barrier FET

• Old wine in new bottle – Machine Learning

6/12/2025 Oves Badami 5

Modelling: A necessity – Ecologically
Generated using Gemini.
• 2 gm silicon chip fabrication requires
• 1.6 kg fossil fuels
• 2 gm of chemicals
• 32 litres of water

• The silicon wafer requires about 160 times

the energy of the raw silicon  Why ???
doi: 10.1021/es025643o

6/12/2025 Oves Badami 6

Modelling: A necessity – Commercially
• A very wide design space Generated using Gemini

• Humans (experts) are costly and inefficient

especially for routine tasks
• Computers don’t need to rest.
• Boosting the productivity

6/12/2025 Oves Badami 7

Modelling: A necessity – Scientifically
• With experiments we generally end Generated using Gemini.

up with devices that are black

boxes.
• All the effects occur at the same
time
• Simulations and modelling provide
a peak into the nature.
• Example of RRAM devices
• Isolation of effects is possible

6/12/2025 Oves Badami 8

Contents

• Computer modelling – Importance

• Under the hood of a TCAD tool

• Case Study of a Poisson Solver

• Advantage of in-house TCAD – Schottky Barrier FET

• Old wine in new bottle – Machine Learning

6/12/2025 Oves Badami 9

How does the TCAD tool works – Equilibrium
• A TCAD tool solves a system of Structure Inputs
equations self-consistently
𝑑 𝑉 𝑞 𝑁 −𝑁 +𝑝 𝑥 −𝑛 𝑥
=− Poisson Initial
𝑑𝑥 𝜀 Solver guess
𝐸 −𝐸 𝑉 𝑛 ,𝑝 ,𝑉

𝑛 ,𝑝 ,𝑉
𝑛 𝑥 = 𝑁 𝑒𝑥𝑝 −
𝑘𝑇 Carrier Conc.
Calculation
𝐸 −𝐸
𝑝 𝑥 = 𝑁 𝑒𝑥𝑝 − 𝑛 ,𝑝
𝑘𝑇
NO Conv. YES END
• EC =-qV is the conduction band ?
(potential energy)
6/12/2025 Oves Badami 10
How does the TCAD tool works – Non-equilibrium
• A TCAD tool solves a system of Structure Inputs
equations self-consistently
𝑑 𝑉 𝑞 𝑁 −𝑁 +𝑝 𝑥 −𝑛 𝑥
=− Poisson Initial
𝑑𝑥 𝜀 Solver guess

𝑑𝐽 𝑑𝑛 𝑉 𝑛 ,𝑝 ,𝑉

𝑛 ,𝑝 ,𝑉
=𝐺−𝑅 𝐽 = 𝑞𝑛𝜇𝑬 + 𝑞𝐷 Carrier Conc.
𝑑𝑥 𝑑𝑥
Calculation
𝑑𝐽 𝑑𝑝
=𝑅−𝐺 𝐽 = 𝑞𝑝𝜇𝑬 − 𝑞𝐷 𝑛 ,𝑝
𝑑𝑥 𝑑𝑥
NO Conv. YES END
𝑑𝑉 ?
𝑬=−
𝑑𝑥
6/12/2025 Oves Badami 11
How does the TCAD tool works – Quantum Correc.
• Electrons are billiard balls Structure Inputs

• Peak occurs at oxide semiconductor 𝑛 ,𝑝 ,𝑉

interface Poisson Initial
Solver guess
• Electrons a waves 𝑉 𝑉

𝑛 ,𝑝 ,𝑉 ,𝑉
Quantum
Corrections
𝑉 𝑉
Carrier Conc.
Calculation
𝑛 ,𝑝
NO Conv. YES END
?
6/12/2025 Oves Badami 12
How does the TCAD tool works – Self-Heating
• Bumping into someone is not Structure Inputs

always a pleasant experience 𝑛 ,𝑝 ,𝑉

• Increases the temperature Poisson Initial
Solver guess
• Temperature plays an important 𝑉 𝑇
role in device performance

𝑛 ,𝑝 ,𝑉 ,𝑇
Fourier Heat
Equation
• Mobility
• Intrinsic carrier concentration 𝑇 𝑉
Carrier Conc.
Calculation
𝑛 ,𝑝
NO Conv. YES END
?
6/12/2025 Oves Badami 13
Contents

• Computer modelling – Importance

• Under the hood of a TCAD tool

• Case Study of a Poisson Solver

• Advantage of in-house TCAD – Schottky Barrier FET

• Old wine in new bottle – Machine Learning

6/12/2025 Oves Badami 14

 Poisson Solver – Equilibrium Case Study
• Structure – Diode • Structure – 1D Diode

p-type n-type p-type n-type

(Doping NA) (Doping ND) (Doping NA) (Doping ND)

Neumann Boundary Dirichlet Boundary

• In this case study, we will assume 𝐸 = 0 𝑒𝑉

6/12/2025 Oves Badami 15
 Poisson Solver – Discretization of the structure
• Structure – 1D Diode
• Meshing – Nodes and Elements

p-type n-type
(Doping NA) (Doping ND) Structure – Discretized Domain

Δx
p-type n-type
(Doping NA) (Doping ND)

6/12/2025 Oves Badami 16

 Poisson Equation – Discretization
Δx
𝑑 𝑉 𝜌 𝑁 −𝑁 +𝑝−𝑛 p-type n-type
=− =−
𝑑𝑥 𝜖 𝜖 (Doping NA) (Doping ND)

𝑉 − 2𝑉 + 𝑉 𝑁 , −𝑁 , +𝑝 −𝑛
=−
Δ𝑥 𝜖
• Relationship in electrostatic potential
between the neighbouring nodes
Set of Generic Points
• 𝑝 , 𝑛 depends on the electrostatic Δx
potential at node i (𝑉 )
i-1 i i+1
6/12/2025 Oves Badami 17
 Poisson Equation – Discretization
𝐸 ,−𝐸 Δx
𝑛 = 𝑁 𝑒𝑥𝑝 − p-type n-type
𝑘 𝑇
(Doping NA) (Doping ND)
𝐸 −𝐸 ,
𝑝 = 𝑁 𝑒𝑥𝑝 −
𝑘 𝑇

𝐸 , −𝐸 , =𝐸 𝐸 , = −𝑞𝑉

• Electrons and holes depend non-linearly on

electrostatic potential Set of Generic Points
• This makes the Poisson equation a non- Δx
linear equation
i-1 i i+1
6/12/2025 Oves Badami 18
 Solving the Poisson’s equation
𝑉 − 2𝑉 + 𝑉 𝑁 , −𝑁 , +𝑝 −𝑛
𝐹 ≡ + =0
Δ𝑥 𝜖

Residual
• Electrons (𝑛 ) and holes (𝑝 ) depend non-linearly on electrostatic
potential.
• This makes the equations non-linear.
• To solve the system of equations we need to use Newton-Raphson scheme.

6/12/2025 Oves Badami 19

 Solving the Poisson’s equation –
Newton Raphson Method
• A quick recap of Newton-Raphson method for equation with one
unknown (𝑓(𝑥)=0)

−𝑓(𝑥 ) −𝑓 𝑥
𝑥 =𝑥 + ∆𝑥 =𝑥 −𝑥 = = 𝑓 𝑥 −𝑓 𝑥
𝑓 (𝑥 ) 𝑓 𝑥
𝑑𝑓(𝑥)
𝑓 𝑥 = Jacobian
𝑑𝑥
Residual
• For 𝑘 = 0, 𝑥 is the initial guess.
• We keep on calculating new x until the update (∆𝑥 ) smaller than
our predefined tolerance
6/12/2025 Oves Badami 20
 Solving the Poisson’s equation –
Newton Raphson Method
• Discretization of a differential equation leads to a system of difference
equation.
𝑉 − 2𝑉 + 𝑉 𝑁 , −𝑁 , +𝑝 −𝑛 p-type n-type
𝐹 ≡ + =0 Δx
Δ𝑥 𝜖 (Doping NA) (Doping ND)
2 34 …. N-1
𝑉 − 2𝑉 + 𝑉 𝑁 , −𝑁 , +𝑝 −𝑛
𝐹 ≡ + =0
Δ𝑥 𝜖

𝑉 − 2𝑉 +𝑉 𝑁 , −𝑁 , +𝑝 −𝑛
𝐹 ≡ + =0
Δ𝑥 𝜖
6/12/2025 Oves Badami 21
 Solving the Poisson’s equation –
Newton Raphson Method
• Extension of Newton Raphson method to equations with multiple
variables
Δx
∆𝒙 = 𝑱 𝟏𝑭
p-type n-type
• 𝑭 is the residual similar to the equation (Doping NA) (Doping ND)

𝑉 =𝑉 , +𝑉 ,
𝐹 𝐸
𝐹 𝑞𝑉
𝑉 =𝑉 +𝑉 𝐸
, ,
⋮
𝑭=
𝐹
𝑞𝑉 = − 𝐸 − 𝐸
⋮ 𝐸
𝐹 𝑘 𝑇 𝑛
𝑉 = 𝑙𝑜𝑔
𝑞 𝑁
6/12/2025 Oves Badami 22
 Solving the Poisson’s equation –
Newton Raphson Method
𝑉 − 2𝑉 + 𝑉 𝑁 , −𝑁 , +𝑝 −𝑛
𝐹 ≡ +𝑞 =0
Δ𝑥 𝜖

1 0 0 0 0 0
𝑑𝐹 𝑑𝐹 𝑑𝐹 𝑑𝐹 𝑑𝐹 2 𝑞
… … =− − 𝑛+𝑝 j=i
𝑑𝑉 𝑑𝑉 𝑑𝑉 𝑑𝑉 𝑑𝑉 ∆𝑥 𝑘 𝑇

𝑱= ⋮ 𝑑𝐹 1
⋮ = j=i±1
𝑑𝐹 𝑑𝐹 𝑑𝐹 𝑑𝐹 𝑑𝑉 ∆𝑥
… …
𝑑𝑉 𝑑𝑉 𝑑𝑉 𝑑𝑉 𝑑𝐹
0 0 0 0 0 1 =0 𝑒𝑙𝑠𝑒
𝑑𝑉

6/12/2025 Oves Badami 23

Solving the Poisson’s equation –
Newton Raphson Method
1 0 0 0 0 0
𝑑𝐹 𝑑𝐹 𝑑𝐹 𝑑𝐹 0 0
… … ∆𝑉 −𝐹
𝑑𝑉 𝑑𝑉 𝑑𝑉 𝑑𝑉
⋮ ⋮ ⋮
⋮ ⋮ ⋮
𝑑𝐹 𝑑𝐹 𝑑𝐹 𝑑𝐹
… … ∆𝑉 −𝐹
𝑑𝑉 𝑑𝑉 𝑑𝑉 𝑑𝑉
0 0 0 0 0 1 0 0

Jacobian -Update -Residual

(NxN) (Nx1) (Nx1)

𝑽 = ∆𝑽 +𝑽 𝑽 = 𝑉 0 …0 𝑉
6/12/2025 Oves Badami 24
Complete Flow – Equilibrium simulations
Structure Inputs

Poisson Initial
Solver guess
𝑉 𝑛 ,𝑝 ,𝑉
𝑛 ,𝑝 ,𝑉

Carrier Conc.
Calculation

𝑛 ,𝑝
NO Conv. YES END
?

6/12/2025 Oves Badami 25

Contents

• Computer modelling – Importance

• Under the hood of a TCAD tool

• Case Study of a Poisson Solver

• Advantage of in-house TCAD – Schottky Barrier FET

• Old wine in new bottle – Machine Learning

6/12/2025 Oves Badami 26

 The Quest for sub-60 transistors
• The physical dimensions of the transistors
have been scaled but not the applied biases
• MOSFETs have a lower limit on SS
• Power density crisis a major hurdle to scaling
• Solution: Devices which remove the SS
limitation.
• Tunnel FET, Ferroelectric FET, I2FET

𝑙𝑜𝑔 𝐼
𝑉

6/12/2025 Oves Badami 27

 Intriguing Experiments
Schottky Schottky
Contact Top gate Contact

Backgate

Bhattacharjee et. al. Applied Physics Letters 2017

• “Theoretically” this should not be possible [Solomon Elec. Dev. Letts. 2010,
Vandenberghe et. al. APL 2013]

6/12/2025 Oves Badami 28

 Proposed Methodology

• Drain bias is small in experiments

leading to equilibrium like carrier
concentration in the channel

G. A. Gauhar et. al. Advanced Theory and Simulations 2024

6/12/2025 Oves Badami 29
Results

Expt.: Bhattacharjee et. al.

App. Phys. Lett. 2017

G. A. Gauhar et. al. Advanced Theory and Simulations 2024

6/12/2025 Oves Badami 30
Results

G. A. Gauhar et. al. Advanced Theory and Simulations 2024

6/12/2025 Oves Badami 31
Contents

• Computer modelling – Importance

• Under the hood of a TCAD tool

• Case Study of a Poisson Solver

• Advantage of in-house TCAD – Schottky Barrier FET

• Old wine in new bottle – Machine Learning

6/12/2025 Oves Badami 32

 An Epiphany
Structure Inputs
• No matter what in all the structures
we are solving the same set of
equations Poisson Initial
• Poisson’s eq. Solver guess

• Carrier concentration 𝑉 𝑛 ,𝑝 ,𝑉

𝑛 ,𝑝 ,𝑉
• Lattice temperature, Quantum Carrier Conc.
corrections Calculation

• This allows us to postulate that there 𝑛 ,𝑝

must be a single equation which NO Conv. YES END
describes the device – Given a ?
Boundary condition
6/12/2025 Oves Badami 33
 An Epiphany
Structure Inputs 𝑉 𝑉
𝑉 𝑉
𝑚∗
𝑇 𝑉
Poisson Initial 𝑉
Solver guess
DATA
𝑉 𝑛 ,𝑝 ,𝑉
𝑛 ,𝑝 ,𝑉

Carrier Conc. 𝑉 𝑛
Calculation 𝑉 𝑛
𝑛 ,𝑝 𝑚∗
𝑇 𝑛
NO Conv. YES END
𝑛
?

6/12/2025 Oves Badami 34

 Why are Neural Networks expected to work ?
𝒗 = 𝒙𝟏 𝒘𝟏 + 𝒙𝟐 𝒘𝟐
𝑉 x1 w1
𝑉 𝒇(𝒗)
𝑉 𝑉 v
∑ 𝒇
𝑚∗ w2
𝑇 𝑉
𝑉 x2 𝑓 is an activation function

• Neural networks as Universal function approximators

• Neural Networks are essentially expansion in terms of the input
variables

6/12/2025 Oves Badami 35

 Problem Statement
𝑑 𝑑
𝜀 𝑦 𝑉 = −𝜌 = −𝑞 𝑁 − 𝑁 + 𝑝 − 𝑛 Y
𝑑𝑦 𝑑𝑦
X
𝑑 −ℏ 𝑑
𝜓 𝑦 +𝑈 𝑦 𝜓 𝑦 =𝜀 𝜓 𝑦
𝑑𝑦 2𝑚∗ 𝑦 𝑑𝑦 𝐸 Oxide Oxide 𝐸
𝑈 𝑦 = −𝑞𝑉 𝑦 + ∆𝐸 ∆𝐸 S.C ∆𝐸
𝐸
𝑛 = 𝑛 , 𝜓 (𝑦)
𝐸
𝐸
• T is varied between 3 nm to 40 nm 𝐸 𝐸
• Why double gate structure ?
• FinFET
• Stacked Nanosheet
6/12/2025 Oves Badami 36
Results – Poisson solver

6/12/2025 Oves Badami 37

Results – Carrier concentration

6/12/2025 Oves Badami 38

Results – Eigen Energy
Ec

6/12/2025 Oves Badami 39

Conclusion
• TCAD tools have played and are expected to play even more
critical role in the future
• This necessitates the us to have a firm understanding of the
working of the TCAD tool
• Developing your own certainly helps
• Machine learning is a promising way to overcome the increasing
issues of computation time (computational burden), non-
convergence.

6/12/2025 Oves Badami 40