12018-11-14

Predictive Maintenance for Plasma Tools

Michael KlickPlasmetrex GmbH

22018-11-14

Content

Industry 4.0 in the semiconductor industry

Smart manufacturing: Reactive → Real-time → Predictive

Typical process risk in plasma processing

Predictive maintenance (PdM) – the first steps – examples

Consequences for infrastructure and business models

The right place to discuss smart manufacturing and PdM

32018-11-14

Abstract

Plasma processes are widely used in the semiconductor industry, they are completely distinct from mechanical manufacturing. Plasma processes are running in vacuum chambers and there are opened every month or quarter for maintenance. Each maintenance measure at a production chamber causes costs in the order of some 10 k€. Therefore, the prediction of the right time for maintenance can reduce manufacturing costs dramatically. On the other hand, plasma processes are usually treated as black box due to their complexity. All important process parameter as uniformity, rate, selectivity, and stability depend of the plasma’s parameters as flux of ions and reactive species. Thus, the main peculiarity of plasma processes can be compressed is one sentence: ‘The plasma is the tool’. Beyond this we have to take into account that plasmas can run in different modes, can oscillate, cause breakdowns at the chamber wall and depend on the state of the chamber wall. In particular the chamber wall changes its surface properties by the deposition of byproducts. So the only realistic approach for the predictive maintenance for plasma tools must be based on the plasma’s properties. It will be shown how plasma parameter can describe plasma and so also the effective chamber state, chamber differences and show undesired instabilities as arcing and wear of chamber parts. The early detection of changes and undesired effects are here the key for predictive maintenance. Examples show the early detection of process faults, real-time process characterization, and preconditions and methods for chamber matching.

42018-11-14

Aspects of Industry 4.0 in the Semiconductor Industry

Smart manufacturing

Predictive maintenance

...Yield Prediction

Virtual Metrology

AugmentingReactive with Real-time & Predictive

J. Moyne, A Roadmap for the Future of Smart Manufacturing in Microelectronics, APC conference, Austin, 2015. IRDS Factory Integration Roadmap

52018-11-14

ISMI 2007

62018-11-14

Parts of Predictive Maintenance for Plasma Processing

Pre-process

faultsdetectionas maskissues

Fast conditioningafter PM and

dry clean, real-timedetection ofproduct mix

issues

Real-timechamber

faultdetection &

preiction

Fast chamber matching

and processtransfer anddevelopment

CriticalDimensions,

yield

Test & conditioningwafer usage

Up-time, maintenance, spare parts & manpower

72018-11-14

Process Groups and Process Understanding

EUV is already atthe edge → extremely expensiveLinear optics → Good process understanding

Lithography Plasma Process

Wet pr.RTP

Diffusion ...

ExpensiveLarge verity of very complex processes:EtchDepositionImplantNitridation→Poor process understanding

82018-11-14

Plasma Equipment is Crucial Equipment

CVD processes is mainly Plasma Enhanced CVD (PECVD).Dry etch stands for plasma etch.

Source: http://www.icknowledge.com/products/equipmentforecast.html

92018-11-14

Conclusions for Predictive Maintenance for Plasma Tools

Predictive Maintenance needs Equipment model Process model including impact of clean both are parts of digital twin

Issues

Process Dynamics Process drift/shift and variability,

e.g. by pre-process Complicated maintenance practices Model portability and maintenance

Nonlinearities Different plasma modes First wafer effect J. Moyne, A Roadmap for the Future of

Smart Manufacturing in Microelectronics, APC conference, Austin, 2015.

10

16. Fachtagung für PlasmatechnologieGreifswald, 18. - 20. Februar 2013

What Means Poor Data Quality ?

➢ Example:Real process power in plasma etching

➢ Power delivered bythe tool (bias generator power)

➢ Real processpower depending on chamber hardware

➢ The real power in the process chamber is less than 50% →

Reason for chamber mismatching

R. Wagner, M. Klick, APCM 2012,Grenoble, France, 2012.

By courtesy of

112018-11-14

Potential Process Risks in Plasma Processing

Chamber stateByproducts / process mixConditioning / dry cleanIdle time, Gas temperature

Wafer propertiesOpen areaHard / resist maskSurface temperature

Deposition/etch rate

uniformityselectivityRF power in plasma

Power losses in RF sub-circuit / Match boxElectrode system / coil

Power loss caused by increased contact resistance trough heating

Erosion of anodization at:

Destroyed structures at semi-conductor wafer due to wafer arcing

D e s c r i p t i o n o f t h e I m a g e :P r o d u c t : 1 2 8 M S 1 7L o t : 3 A 1 4 6 1 0 4W a f e r : 2 3D i e : ( - 1 4 , 6 )X , Y : ( - 9 5 4 1 6 , 4 7 7 4 9 )T o o l : O R B O T W FD e f e c t C o d e : 3 : M M m i t a u f l i e g e n d e m M a t e r i a lD a t e : 2 9 - A U G - 0 1O p e r a t o r : U N K N O W N

C o m m e n t s :

D e s c r i p t i o n o f t h e I m a g e :P r o d u c t : 1 2 8 M S 1 7L o t : 3 A 1 4 6 1 0 4W a f e r : 2 3D i e : ( - 1 4 , 6 )X , Y : ( - 9 5 4 1 6 , 4 7 7 4 9 )T o o l : O R B O T W FD e f e c t C o d e : 3 : M M m i t a u f l i e g e n d e m M a t e r i a lD a t e : 2 9 - A U G - 0 1O p e r a t o r : U N K N O W N

C o m m e n t s :

Destroyed structures caused by particles

Gas distribution plate holes

slit valve liner door

122018-11-14

Plasma's Can Run in Different Modes

Main etch

unstable Change of chemistry, here to

SF6 / HBr / O2, can drive plasma processes into a unstable state (E-H mode transition).

Depends on:

– Lack of RF power control– Chamber state

Not visible in tool data !

Maskopen

stable

Plasma etcher, ICP/CCP, Coil at ceramic dome

By courtesy of

E. Chasanoglou et al., TI Germany, E-H-Mode transition and its detection in SF

6 plasma during Si trench etch, APCM 2013, Dresden, Germany, 2013.

132018-11-14

After PM

Before PM

Undefined Wall of Plasma Chamber

By courtesy of

142018-11-14

From Reactive to Predictive

Not well defined boundary conditions

Lack of processunderstanding and visibility.

The plasma is the tool

Plasma process understandingby plasma models

Plasma sensors formodel parametrization

152018-11-14

Model-based Understanding of Conditioning Procedures

The plasma is the tool→ Plasma parameters

Conditioning processes are always needed to drive the chamber to the right state.

Test of plasma processes in order to understand and control conditioning procedures.

Joint project:Samsung - Plasmetrex

K. H. Baek et al., Journal of Vacuum Science & Technology A 35, 021304 (2017)

Chamber wall completely coated

162018-11-14

Model-based Understanding of Conditioning Procedures

The plasma is the tool→ Plasma parameters

Higher power enhances the chamber wall conditioning

Less than one wafer needed now

Joint project:Samsung - Plasmetrex

K. H. Baek et al., Journal of Vacuum Science & Technology A 35, 021304 (2017)

Chamber wall completely coated

172018-11-14

Knowing Already During Process Why it Gets Faster

Chamber A

Wafer 35 - 40

Wafer 40 - 75

High asymmetry low asymmetry→ short process length → longer process.

An increased asymmetrycauses a higher ion energy(etch rate) and therefore ashorter process time (EPD).

Trend ofprocess length

Asym

metry

Plasma Parameters used for Process Characterization, Oh Sang Hun et al., apcm Europe Dresden, 2018.

182018-11-14

Particles on Wafer – Arcing

„Bubbles“:Particles with size > 1µmEDX result: AlSome particle trajectories exhibit interesting traces of indentations on the wafer surface – „like stones over water“

Wall contamination:Increased surface roughness Images taken with an optical microscope reveal molten wall areas

18

Particle traces on wafer

EDX Analysis of particles: Al

Al is ejected from chamberwall„Hot“ Al drops land on wafer

Conclusion Breakdwon at chamber wall !

Al is ejected from chamberwall„Hot“ Al drops land on wafer

Conclusion Breakdwon at chamber wall !

Microscope images of the white chamberwall area (right: higher magnification)

192018-11-14

Real-time Arcing Detection at SPEED Chamber

Arcing is detected by plasma parameters:Location of breakdown in chamber determines pattern: Classical peak in collision rate

→Particles from chamber wall to wafer

Drop in asymmetry shows direct effect on electric fields at wall and wafer.→ No particles in wafer

Arcing Detection and Root Cause Analysis in Low Pressure PECVD, K. S. Siegert et al.,apc|m 2018, Dresden, Germany. By courtesy of

202018-11-14

Dennis Föh, WAC improvement, apc|m Europe 2018

MTBC – Mean Time Between Clean must always be better... often limited by polymer build up that comes back to you as particles

ESC lifetime must always be better... limited due to attack during waferless auto clean (WAC) high costsIncreased surface roughness Less aggressive dry clean

Lower physical impact by lower ICP (source) power Special dry clean setup if complicated tool design

Improvement of WAC necessary for better cleaning performance and less ESC erosion

Less Cleaning Procedures and Longer Lifetime

By courtesy of

212018-11-14

Dual Frequency Tool

Dennis Föh, WAC improvement, apc|m Europe 2018

By courtesy of

222018-11-14

Goal: Extended lifetime of ElectroStatic Chuck (ESC) & longer Mean Time Between Clean (MTBC)

Standard way of plasma clean (WAC) modification – trial and error think about it & modify run it for several month and cleaning cycles to see if MTBC is better and check if ESC lasts longer

Way out – Plasma model and plasma parameter usage of plasma parameters for a fast and efficient identification of needed

modification in the WAC recipe try to predict estimated ESC lifetime ASAP to evaluate the new WAC

The RF current is an excellent measure for Exelan tool type: the higher the RF current, the higher the plasma density is or the wider the plasma is reaching through the confinement ring

system in direction of the chamber walls

C)

How to Improve a Plasma Clean

By courtesy of

Dennis Föh, WAC improvement, apc|m Europe 2018

232018-11-14

Trial & error WAC improvement is out of fashion

Usage of plasma parameters allow a fast and efficient way to improve and change plasma processes according to specific needsBy using the Hercules plasma monitor it was possible to modify the WAC directly to the final process parameters, no iterations were done.The linear increase of He BSC Flow (leakage) allows a very fast evaluation of new WACs regarding the ESC lifetime to expect.

He BSC-flow vs. life time Dennis Föh, WAC improvement, apc|m Europe 2018

242018-11-14

Breakdowns as Early Fault Indicator

Breakdowns at Lam etch chamber show wear and upcoming fault !

Break down

Clamping Voltage to drop→ loss of clamping force

By courtesy of

252018-11-14

PdM as Part of Smart Manufacturing Needs New Ways

It needs additional subject matter expertise (SME) with new business models.

We need to focus on improving data quality Making data stores “prediction-ready” Implementation data quality

improvement best practices

It need other approaches as digital twin as complementary parts of the solution.

J. Moyne, A Roadmap for the Future of Smart Manufacturing in Microelectronics, APC conference, Austin, 2015. IRDS Factory Integration Roadmap

19th European apc|m Conference

apc|m 2019 in Villach

26

Main focus similar to APC Conference.Motto: Sensor Integration for Production ImprovementLast year’s conference in Dresden: 171 participants from 70 companies and 14 countries.

Close to an Infineon site, the Alps, and the Mediterranean Sea

19th European apc|m Conference