draft – work in progress - not for publication 13 july 2005 pids preliminary results and key...
TRANSCRIPT
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
PIDS Preliminary Results and Key Issues: 2005 ITRS
Peter M. Zeitzoff for PIDS Technology Working Group
ITRS Public MeetingSan FranciscoJuly 13, 2005
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
PIDS RosterJapan•T. Sugii (Chair)•S. Sawada•J. Ida•S. Oda•T. Hiramoto•S. Takagi•K. Shibahara•A. Hori•D. Hisamoto•T. Nakamura•Y. Tadaki•N. Nagashima•Y. Takeda•S. Tahara•K. Imai•Y. Akasaka•R. Shirota•M. Yoshimi•T. Sasaki
Taiwan•R. Liu (Chair)•Y. J. Mii•C. Diaz•W. T. Shiau•M. J. Tsai
Korea•I. S. Yeo
Europe•T. Skotnicki (Chair)•K. Schruefer•S DeLeonibus•K. De Meyer•R. Lander•M. Jurczak
US•P. Zeitzoff (Chair)•J. Chung•J. Brewer•L. Tran•M. Rodder•Q. Xiang•T-J. King•T. Ning•G. Yeap•M. Duane•T. Dellin•W. Tsai
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Outline
• Scope and Subcategories
• Non-Volatile Memory
• DRAM
• Logic
• Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
PIDS Scope• PIDS = Process Integration,
Devices, and Structures• Main concerns
– MOSFET, memory, and passive devices and structures• Device physical and electrical characteristics
and requirements– Broad issues of device and circuit
performance, density, and power dissipation, particularly as they drive overall technology requirements
– Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
PIDS Subcategories
• Logic
– High-performance
– Low-power for mobile applications
• Memory
– DRAM
– Non-volatile memory (NVM)
• Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Outline
• Scope and Subcategories
• Non-Volatile Memory
• DRAM
• Logic
• Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
• NOR Flash– Technology node (F = minimum poly half pitch) remains unchanged
from 2004 update.– Tunnel oxide scaling slightly faster in near term.– Interpoly EOT scaling faster in both near term and long term
(deployment of high-k).– Control gate Lg scaling is considerably faster than 2004 update.– Cell size scaling is faster than 2004 update in both μm2 and F2.– Advances result from good engineering, not fundamental
breakthroughs.• NAND Flash
– F moves past DRAM in both near term and long term.– Cell size area factor, a = cell size/F2 remains the same (but F moves
with node).– W/E voltages remain the same as 2004 update (not moving with
node).– Tunnel oxide thickness remains the same as 2004 update.– Interpoly dielectric thickness remains the same as 2004 update.– Endurance and retention remain the same as 2004 update.
Non-Volatile Memory (NVM)
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Long-term
Near-term
2004 2005 2006 2007 2008 2009
04 NAND/AND Flash 90 70 65 55 50 45
03Flash Roadmap 90 80 70 65 55 5003DRAM1/2pitch 90 80 70 65 57 50
2010 2011 2012 2013 2014 201504 NAND/AND Flash 45 40 35 32 28 25
03Flash Roadmap 50 39 35 2803DRAM1/2pitch 45 35 32 25
Technology Node(nm)
Proposal of NAND/AND Flash Roadmap
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
• FeRAM– Technology node still lags substantially behind DRAM/NAND Flash– Ferroelectric capacitor area scaling slows down – 3D is major challenge– Endurance improvement slows down starting from 2004.
• SONOS/NROM– 2-bit/cell due to localized charge storage.– Scalability expected similar to floating gate device.– No floating gate coupling issues.
• MRAM– Technology node pulls back by 1-2 years in near term and long term.– Both switching field and switching energy are aggressively scaled after 2008
compared to 2004 update. Much of this acceleration is labeled “red” indicating breakthroughs are needed for MRAM to stay competitive.
• PCRAM– New in 2005 PIDS: mainstream production expected soon.– For near term years, technology node similar to DRAM.
Non-Volatile Memory (NVM) (con’t.)
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Outline
• Scope and Subcategories
• Non-Volatile Memory
• DRAM
• Logic
• Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
• Scaling unchanged from 2003– DRAM half-pitch (F) : 3 year cycle
• 65nm in 2007, 45 nm in 2010
• Other results– a=(cell area)/F2: a=8 through 2007, a=6
thereafter
– Area size factor (% of chip area taken up by storage cells): 63% through 2007, 56% thereafter
– STC storage node dielectric: the same as 2003 ITRS
DRAM
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
First Year of IC Production200580nm
200670nm
200765nm
200855nm
200950nm
201045
2011 201235nm
201332
2014 201525nm
201622
2017 201818nm
2019 2020
DRAMsHigh-k dielectric Al2O3 & Ta2O5, MIS structure (k ~ 10–25) MIM Structures (k ~ 20–50) BST or other materials (k>100)
Three dimensional array device
Emerging research memory devices
Potential Solutions: DRAM
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Scaling of DRAM Storage Dielectric
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2005 2010 2015 2020
0
5
10
15
20
25
30
35
40
45
50
DRAM storage node celldielectric: equivalent physicalthickness, EOT nm
DRAM storage node capacitorvoltage, V
Electricfield of capacitordielectric, MV/cm
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Outline
• Scope and Subcategories
• Non-Volatile Memory
• DRAM
• Logic
• Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Logic: 2005 Scaling Approach and Categories• MASTAR (detailed analytic device model from STM and
corporate partners) was used – MASTAR has been extensively verified against literature and other
data
– Initial choice of scaled MOSFET parameters is made
– Using MASTAR, MOSFET parameters are iteratively varied to meet ITRS targets
• Types of Logic– High Performance (HP) (e.g., MPU): target is historical 17%/year
transistor performance increase
– Low Power (for mobile applications): target is specific, low level of leakage current
• Low Standby Power (LSTP): very low leakage; for lower performance, consumer applications (e.g., cellphone)
• Low Operating Power (LOP): low dynamic power, rel. high performance (e.g., notebook computer)
• 2005 scaling results changed somewhat from 2003 ITRS
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Intrinsic Transistor Delay, CV/I
(lower delay = higher speed)
Leakage Current(HP: standby power dissipation issues)
2005 ITRS: Low Power & High Performance (HP)
Preliminary Results
1.E-06
1.E-05
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
2005 2007 2009 2011 2013 2015 2017 2019
Calendar year
Isd
,lea
k (
uA
/um
)
HP Target:17%/yr, historical rate
LOP LSTP
LSTP Target: Isd,leak ~ 10 pA/um
LOP
HP
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
The “CMOS Change Crunch” Multiple, Big Changes Over Next 7 Years
2000 2005 2010 2015 2020First Year of “Volume Production”
High k Gate Dielectric HPLP
LPHP = High Performance Applications
= Low Power Applications
Driver:
Others
Fully Depleted SOI HP LP
Multiple Gate MOSFET HP LP
HPStrained Si LP
Preliminary Results
HPMetal Gate Electrode
LP
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
LSTP: EOT and Gate Leakage Current Density (Jg) Scaling
Preliminary Results
Jg,sim (SiON)
EOT
Jg,limit
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Parallel Paths for High-Performance Logic in 2005 ITRS
Preliminary Results
Year in Production Units 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020Number of years from 2005 years 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15Technology Generation hp65 hp45 hp32 hp22 hp16
Physical Lgate (High Performance) nm 32 28 25 22 20 18 16 14 13 11 10 9 8 7 6 5
MG
Planar Bulk
FDSOI[MG = Multiple Gate (e.g., FinFET)]
•Approach
–Extend planar bulk as long as possible
–Implement FDSOI in parallel with planar bulk
–Establish MG later; MG is the ultimate, scaled CMOS
[FDSOI = Fully Depleted, Ultra-thin Body SOI]
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Outline
• Scope and Subcategories
• Non-Volatile Memory
• DRAM
• Logic
• Reliability
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
• High-k Gate Dielectrics– Dielectric breakdown; Transistor instability
• Metal Gate– Ion drift, VTH stability, oxidation; thermal-
mechanical• Cu/ Low k
– Electromigration and voiding; stability of interfaces; TDDB
– Impact of porous, weaker, less thermally conductive dielectrics
• Packaging– Solder bumps; fracture; EM in packaging; CTE
mismatch• Design & Test for Reliability
– Reliability simulation; Reliability screens
Reliability: Top 5 Near-Term Challenges
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
• Reliability risk is growing– New materials (e.g., high k/metal gate; low k) and new
devices (e.g., FINFET) and new packaging• Introduce new and/or modified failure mechanisms• Mechanisms need to be identified, modeled and controlled• Have less-than-historic time and resources to ensure
reliability– Difficult tradeoffs may require reduced reliability
margins– Need new Design for Reliability tools and reliability
screens• Need to sustain current high reliability levels in
spite of unprecedented changes
Reliability: Key 2005 Issues
Preliminary Results
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
• Memory– DRAM: Rapid scaling continuing– Numerous different types of NVM, with unique attributes
and scaling scenarios• Logic
– High-performance logic: performance increases by 17%/yr. ratehigh leakage current
– Low-power logic: low leakage currentreduced performance
– Numerous and rapid technology innovations required– 2005 ITRS: re-evaluation of scaling scenarios, timing and
sequence of technology innovations• Reliability
– Ensuring reliability for numerous and rapid technological innovations is a critical challenge
– Nevertheless, need to sustain current high reliability levels
Summary
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
BACK UP
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Parallel Paths for LSTP in 2005 ITRS
Preliminary Results
[MG = Multiple Gate (e.g., FinFET)]
•Approach
–Extend planar bulk as long as possible
–Implement FDSOI and MG in parallel
–Slowed Lg scaling for FDSOI in latter years
[FDSOI = Fully Depleted, Ultra-thin Body SOI]
DRAFT – Work In Progress - NOT FOR PUBLICATION 13 July 2005
Parallel Paths for LOP in 2005 ITRS
Preliminary Results
[MG = Multiple Gate (e.g., FinFET)]
•Approach
–Extend planar bulk as long as possible
–Implement FDSOI and MG in parallel
–FDSOI ends in 2016
[FDSOI = Fully Depleted, Ultra-thin Body SOI]