"JEDEC发布2016年最新版JEP122H规范：半导体器件故障机制与模型"

版权申诉

5星 · 超过95%的资源 36 浏览量更新于2024-02-25 收藏 2.19MB PDF 举报

The JEDEC JEP122H 2016.pdf publication discusses failure mechanisms and models for semiconductor devices. This document, which is a revision of the previous version JEP122G from October 2011, was released in September 2016 by the JEDEC Solid State Technology Association. JEDEC standards and publications are carefully prepared, reviewed, and approved by the JEDEC Board of Directors and legal counsel to ensure accuracy and reliability. These standards are created to serve the public interest by providing guidelines and best practices for the semiconductor industry. The JEP122H publication covers a range of topics related to failure mechanisms in semiconductor devices, including the causes of failures, diagnostic techniques, and modeling approaches. It aims to help engineers and technicians understand the root causes of device failures and develop effective strategies for preventing and mitigating these issues. By providing in-depth insights into failure mechanisms and models, this document equips industry professionals with the knowledge and tools they need to enhance the reliability and performance of semiconductor devices. It serves as a valuable resource for manufacturers, designers, and researchers working in the semiconductor field. Overall, the JEDEC JEP122H 2016.pdf publication is an essential guide for anyone involved in the design, manufacturing, or testing of semiconductor devices. It offers a comprehensive overview of failure mechanisms and models, helping to ensure the continued advancement of technology in this critical industry.

JEDEC Publication No. 122G

Page 8

3 Inclusions, deliberate omissions, and resources (cont’d)

3.3 JEDEC resources (cont’d)

JESD90 A Procedure for Measuring P-Channel MOSFET Negative Bias Temperature

Instabilities

JESD92 Procedure for Characterizing Time-Dependent Dielectric Breakdown of Ultra-Thin Gate

Dielectrics,

JESD201 Environmental Acceptance Requirements for Tin Whisker Susceptibility of Tin and Tin

Alloy Surface Finished

JEP119 A Procedure For Executing Sweat

JEP139 Guideline For Constant Temperature Aging To Characterize Aluminum Interconnect

Metallizations For Stress-Induced Voiding

J-STD-020 Joint IPC/JEDEC Standard For Moisture/Reflow Sensitivity Classification For

Nonhermetic Solid State Surface-Mount Devices

JP002 Current Tin Whiskers Theory and Mitigation Practices Guideline

IPC 610 PCB Inspection Criteria

4 The basic thermal acceleration equation

When estimating acceleration due to temperature, it is customary to use the expression in equation (2.1).

It is an adaptation of the Arrhenius equation and has been used to express both a single failure

mechanism's sensitivity to temperature (i.e., its acceleration factor) and also a product's thermal

acceleration factor. By measuring parametric change caused by temperature, activation energy in its

physical sense can be estimated. When used to estimate the reliability of a product, as shown in

equation (4.1), it is also being used to express that product's reliability with respect to temperature and as

a function of time.

Top

= λ

Tst

/ A

= λ

Tst

/ exp[(–E

/ k)(1 / T

– 1 / T

)] (4.1)

where

Top

is the observed failure rate at an operating temperature T

–1

)

Tst

is the observed failure rate at stress temperature T

–1

)

is the apparent activation energy in electronvolts (eV)

k is Boltzmann’s constant (8.62 × 10

–5

eV/kelvin)

is the absolute temperature of the stress in kelvins

is the absolute temperature of an operating condition in kelvins

Equation (4.1) (same as 2.1) must be used with caution because this relationship holds only for systems

in which the failure rate is constant. However, very few practical situations exist in which the failure rate

is truly constant. Nevertheless, the assumption of constant failure rate is still commonly used. An

"apparent" activation energy, E

, may be used for product failure rate estimates in the absence of

extensive failure analysis based on root cause and responsible failure mechanisms.

JEDEC Publication No. 122G

Page 9

5 Models for common failure mechanisms

The failure mechanisms described in the several sections of this standard constitute commonly accepted

industrial models, validated by a team of reliability experts (from the SEMATECH/ISMI Reliability Council)

and buttressed by citations to the most cogent published literature.

5.1 FEoL Failure Mechanisms – Time-Dependent Dielectric Breakdown (TDDB) – Gate Oxide

Time-Dependent Dielectric Breakdown (TDDB) is an important failure mechanism in ULSI devices. The

dielectric fails when a conductive path forms in the dielectric, shorting the anode and cathode. It should

be noted that TDDB for intermetal diectric is discussed in 5.7. It is important to make a distinction

between the empirical models and the physical models developed to explain these empirical models.

The experimental data can generally be described by one of four models (see citations [5.1.1 to 5.1.12]

for historically significant papers):

1) E model or constant field/voltage acceleration exponential model

2) 1/E model or, equivalently, anode hole injection model

3) V model, where failure rate is exponential with voltage

4) Anode hydrogen release for the power-law model

The physical models used to explain the four empirical models are: A) thermo-chemical model, B) anode

hole injection, C) bulk trap generation, and D) anode hydrogen release model. While the

thermo-chemical and V models adopt the assumption of field-driven mechanism, both anode hole

injection and anode hydrogen release models assume a current-driven mechanism in addition to the role

of applied voltage or oxide fields depending on the specific conditions.

5.1.1 Constraints and limitations

These models are intended for application to SiO

gate dielectric over a large range of oxide thicknesses.

Models and parametric values to effectively model “high-k” oxides (based on materials or composites

other and silica) are unknown so these materials are not yet treated in this document.

Controversies regarding these models still exist, hence users are encouraged to review literature before

they make their own decisions for their unique technologies.

JEDEC Publication No. 122G

Page 10

5.1.2 Models

5.1.2.1 E model [5.1.3 to 5.1.16]

In the E model for gate oxide thickness greater than 4 nm, the cause of low electric field (<10 MV/cm)

TDDB is due to field-enhanced thermal bond breakage at the silicon-silica interface. The E-field serves to

reduce the activation energy required for thermal bond breakage and therefore exponentially increases

the reaction rate for failure. The time-to-failure (TTF), inverse to reaction rate, decreases exponentially

with temperature:

TTF = A

* exp(E

/ kT) = TTF = A

* exp(–γ(Τ) E

) * exp(E

/ kT) (5.1.1)

where

= arbitrary scale factor, dependent upon materials & process details

γ = field acceleration parameter in cm/MV, is temperature dependent, γ (T) = a / kT where a

is the effective dipole moment for the molecule (see also the note at the end of this

clause)

= externally applied electric field across the dielectric in MV/cm. The value must be

voltage compensated for thin t

(otherwise the correction is insignificant) band bending if

an accumulation layer is formed, but no compensation is needed if an inversion layer is

formed. E

is the quotient of the compensated voltage & the oxide thickness, t

. Note

that t

should be electrically or physically measured.

= apparent activation energy, typically expressed in electronvolts (eV) = (ΔH)

k = Boltzmann’s constant

T = temperature in kelvins

Previous work [5.1.20] at very large electric field shows that γ has a 1 / T dependence, which affects TTF

equivalently to an activation energy that decreases linearly with the electric field,

= (ΔH)

– a E

(5.1.2)

where

= apparent activation energy, typically expressed in electronvolts (eV); E

may be nearly

temperature independent if several types of disturbed bonding states are present in the

dielectric and the reaction rates are mixed during high field and/or high temperature

TDDB testing

(ΔH)

= the enthalpy of activation for bond breakage in the absence of external electric field

(~2.0 eV in [5.1.12, 5.1.24, 5.1.25])

a = effective molecular dipole-moment for the breaking bonds (~7.2 eÅ and a-range is 7 to

13 eA in [5.1.12, 5.1.24, 5.1.25])

JEDEC Publication No. 122G

Page 11

5.1.2 Models (cont’d)

5.1.2.1 E model [5.13 to 5.1.16] (cont'd)

For intrinsic failures in SiO

dielectrics of thickness ~<10 nm, γ = ~2.5 to 3.5 Naperians per MV/cm

(~1.1 to 1.5 decades per MV/cm) and E

= 0.6 to 0.9 eV. For extrinsic defects, effective oxide thickness

can be quite thin, and therefore, the effective field can be very large, which may reduce temperature

sensitivity.

Long-term TDDB studies [5.1.17] showed that TDDB data were described effectively by the E model

[5.1.13 to 5.1.16] while TDDB data published in [5.1.18] followed the 1/E model [5.1.19 to 5.1.23]. All

models are now based on fundamental, physical parameters (not empirically fitted parameters) and fit

TDDB data quite well [5.1.17, 5.1.24, 5.1.25]. The key issue is whether any given model can also obey

the Poisson area scaling, which is universally accepted.

The good fit of the physics-based E model to the low field/long-term TDDB data strongly suggests electric

field is the dominant degradation driver at low stresses characteristic of customer applications and that

constant current stress is NOT relevant to customer application.

5.1.2.1.1 TDDB numerical example using E model

Objective: Calculate the acceleration factor (AF) for gate oxide failure by TDDB for an office

environment vs. a test structure accelerated environment.

Assumptions:

Office conditions are: 50 °C chip temperature (inside a desktop PC) & an electric field of 4 MV/cm

Accelerated condition (for a test structure) is: 300 °C & 8 MV/cm

effective molecular dipole-moment a: 7.2 eÅ

Field acceleration parameter, γ(50°C) : 2.586 Naperians (natural logarithm) per cm/MV

Field acceleration parameter, γ(300°C) : 1.4577 Naperians (natural logarithm) per cm/MV

Apparent activation energy, E

: 0.75 eV (approximate center of the nominal range for E

)

AF, the ratio of the time-to-failure (TTF) values will be:

AF (ratio of TTF

office

to TTF

accel

values) = exp[–(γ(T

office

– γ(T

accel

)] * exp[(E

/ k)(1 / T

office

– 1 / T

accel

)]

AF = exp[–(2.586 cm/MV*4 MV/cm – 1.4577 cm/MV*8 MV/cm)] * exp[(0.75 eV / 8.62 x 10

–

eV/K)(1 / (273 + 50)K –

1 / (273 + 300)K)]

AF = 3.7344 * 1.27 x 10

= 4.7462 x 10

Conclusion: So, moving from the accelerated test structure environment to the office environment will

increase TTF value to 2 x 10

times the accelerated stress value, of which 160,000 X is

due to electric

field and 127,000 X is due to temperature. One must consider whether the same failure mechanisms are

active under such a heavily accelerated stress vs ordinary operation.

NOTE Many papers in the literature may use and plot base 10 (rather than base e) when expressing the field

acceleration factor. One should be careful to note whether base 10 or natural base e is being used. Some authors,

for clarity reasons, will write the field acceleration as decades per MV/cm to emphasize that the base 10 is being used

or Naperians per MV/cm to emphasize that the natural base e is being used. Many authors, however, may not

emphasize this distinction to the reader, so the reader must be cautious. The conversion factor between base 10 and

base e is 2.3:1, i.e., γ

base e

= 2.3 * γ

base 10

. In this document, the natural base e is assumed.

JEDEC Publication No. 122G

Page 12

5.1.2 Models (cont’d)

5.1.2.2 1/E model [5.1.19 to 5.1.23]

The cause of TDDB (even at low fields) is postulated to be due to current through the dielectric by

Fowler-Nordheim (F-N) conduction. F-N injected electrons (from the cathode) cause impact ionization

damage of the dielectric due to as they accelerate through the dielectric. Additionally, when these

accelerated electrons reach the anode, hot holes may be produced that can tunnel back into the dielectric

causing damage (hot-hole anode injection mechanism). The time-to-failure is expected to show an

exponential dependence on the inverse of electric field, 1 / E

TTF = τ

(T) * exp(G(T) / E

) (5.1.3)

where

(T) = temperature dependent pre-factor (~1 x 10

–11

G(T) = field acceleration parameter (~350 MV/cm with a weak temperature dependence)

= externally applied electric field across the dielectric in MV/cm. It must be voltage

compensated for band bending if an accumulation layer is formed, but no

compensation is needed if an inversion layer is formed. E

is the quotient of the

compensated voltage & the oxide thickness, t

. Note that t

should be electrically or

physically measured.

5.1.2.3 V model [5.1.26 to 5.1.27]

Two papers at 2000 IRPS reported data that showed the E model was no longer valid for gate oxide

thickness values <4 nm. It was found that a model exponential with voltage rather than electric field

represented the reliability performance well. Thus, the time-to-failure is expressed as:

TTF = A

* exp(–β V) * exp(E

/ kT) (5.1.4)

where

= arbitrary scale factor, dependent upon materials & process details

β = voltage acceleration parameter

V = applied voltage

= apparent activation energy, typically expressed in electronvolts (eV)

k = Boltzmann’s constant

T = temperature in kelvins

5.1.2.4 Power-law model [5.1.28 to 5.1.37]

In recent years for gate oxide thickness <2 nm, a power-law for voltage dependence of oxide breakdown

is proposed based on extensive experimental database including long-term module stress [5.1.28].

Several research groups independently experimentally confirmed this model [5.1.29 to 5.1.34]. Work

reported at 2006 IRPS demonstrates the power-law voltage dependence remains valid over twelve orders

of magnitude from micro second time frames to hundred hours time span [5.1.32].

The power-law model was originally proposed based on the experimental work on ultra-thin oxides below

5 nm [5.1.28]. Most importantly, the power-law model preserves important breakdown characteristics,

namely Poisson random statistics and weakest link property, which are universally accepted [5.1.28].

This model can be extended to describe a large range of oxide thickness up to 10 nm and at stress

voltages as large as 12 V [5.1.35]. This model is also consistent with the current-driven breakdown

property as demonstrated by independent experimental evidence using substrate carrier injection

technique [5.1.36]. The physical interpretation of power-law voltage dependence has also been

advanced by several groups by comparing experimental data with the first-principle theory and STM

experiments of hydrogen desorption phenomenon [5.1.29, 5.1.30, 5.1.35 to 5.1.37].

剩余113页未读，继续阅读

std86021

粉丝: 109
资源: 2841

"JEDEC发布2016年最新版JEP122H规范：半导体器件故障机制与模型"

JEDEC JEP122H：2016 半导体器件的失效机制和模型 - 完整英文版（111页）.rar

JEP-122E-2009 Failure Mechanisms and Models for Semiconductor Devices

【最新版可复制文字】 JEDEC JEP174-2016.pdf

【最新版可复制文字】 JEDEC JEP157-2009.pdf

【最新版可复制文字】 JEDEC-JEP164-2022.pdf

【最新版可复制文字】 JEDEC JEP172A- 2015.pdf

【最新版可复制文字】 JEDEC JEP150.01 2013-06.pdf

【最新版可复制文字】 JEDEC JEP106BF01-2022 Standard Manufacturer’s.pdf

【最新版可复制文字】 JEDEC JEP95 DESIGN GUIDE 4.7 (DSBGA) DG-4.7F.pdf

【最新版可复制文字】 JEDEC-JEP190-2022 Guideline for Evaluating dv_

最新资源