Future Perspective of
Future Perspective of
Non
Non
-
-
Volatile Memory
Volatile Memory
Stefan Lai, PhD, Fellow IEEE
Intel Corporation
Vice President, Technology and Manufacturing Group
Why Non
Mobile Lifestyle: Data Anytime, Anywhere
Mobile Lifestyle: Data Anytime, Anywhere
Enterprise Enterprise
Richest Richest Experience Experience
Greatest Mobility Greatest Mobility
Internet
Computing is Going Mobile
Computing is Going Mobile
Saturday
Sunday Friday
Thursday Wednesday
Tuesday
Plaza hotel
School
Memories for Mobility
Memories for Mobility
y
y NonNon--volatile = low power: power down to save volatile = low power: power down to save power
power
y
y Low cost: one memory for program and data Low cost: one memory for program and data storage
storage
y
y ≠≠ magnetic hard disk: need solid state, low magnetic hard disk: need solid state, low power, rugged
power, rugged
y
Moore’s Law in NV Memory
Moore’s Law in NV Memory
y
y Moore’s Law will continue through innovation Moore’s Law will continue through innovation
–
– Process complexity will increase to address Process complexity will increase to address fundamental limits of physics
fundamental limits of physics y
y To maintain Moore’s Law cost learning curve, To maintain Moore’s Law cost learning curve, difficult to do it by transistor technology alone
difficult to do it by transistor technology alone
–
– New opportunity for new memory structures and new New opportunity for new memory structures and new materials
materials y
y Current mainstream NV memory technologies Current mainstream NV memory technologies of ETOX and NAND will continue to be the key
of ETOX and NAND will continue to be the key
technologies for more than 5 years out
technologies for more than 5 years out
y
Moore’s Law will Continue with
Moore’s Law will Continue with
ETOX
ETOX
Flash
Flash
y
y Currently shipping 8Currently shipping 8thth generation of ETOXgeneration of ETOX®®
flash memory in high volume
flash memory in high volume
–
– ~50% cell size reduction per generation~50% cell size reduction per generation
y
y Good visibility into 90 nm and 65 nm generationGood visibility into 90 nm and 65 nm generation
y
y Current projection shows scaling continues at Current projection shows scaling continues at 45 nm node but challenged to meet 50% goal
45 nm node but challenged to meet 50% goal
y
y Further innovation required to maintain cost Further innovation required to maintain cost learning curve
learning curve
y
y Industry trend: complex transistor structure to Industry trend: complex transistor structure to meet scaling challenges
1986 / 1.5µm 1988 / 1.0µm 1991 / 0.8µm 1993 / 0.6µm
1996 / 0.4µm 1998 / 0.25µm 2000 / 0.18µm 2002 / 0.13µm
5.4X
1986 / 1.5µm 1988 / 1.0µm 1991 / 0.8µm 1993 / 0.6µm
1996 / 0.4µm 1998 / 0.25µm 2000 / 0.18µm 2002 / 0.13µm 1986 / 1.5µm
1986 / 1.5µm 1988 / 1.01988 / 1.0µµmm 1991 / 0.81991 / 0.8µµmm 1993 / 0.61993 / 0.6µµmm
1996 / 0.4µm
1996 / 0.4µm 1998 / 0.251998 / 0.25µµmm 2000 / 0.182000 / 0.18µµmm 2002 / 0.13µm
5.4X
ETOX
ETOX
®
®
Technology Scaling
Technology Scaling
y
y 18 years and 8 Generations of ETOX18 years and 8 Generations of ETOX®® to 0.13 to 0.13 µµmm
Volume Production Year / Technology Generation
Example of Future Transistor
Example of Future Transistor
Source
Source
Drain
Drain
Gate
Gate
Source: Intel
Source: Intel
Gate
Gate
Silicon
Silicon
Drain
Drain
Source
Source
Example of
Example of
Current Flash
Current Flash
Memory Cell
New Materials in Silicon
New Materials in Silicon
Technology
Technology
y
y The semiconductor industry has been The semiconductor industry has been addressing the performance and cost
addressing the performance and cost
issues by introducing new materials
issues by introducing new materials
–
– Tantalum pentoxide for DRAM storage dielectricTantalum pentoxide for DRAM storage dielectric
–
– Cobalt and Nickel for S/D formationCobalt and Nickel for S/D formation
–
– Copper and low k dielectric for interconnectCopper and low k dielectric for interconnect
–
– High K dielectric for transistor gateHigh K dielectric for transistor gate
y
y For nonFor non--volatile memories, new memory volatile memories, new memory materials provide new opportunities for
materials provide new opportunities for
further memory cost reduction
Review of New Memories
Review of New Memories
y
y Explosion of research and development on new Explosion of research and development on new memory technology and material
memory technology and material
y
y Scopes range from full scale manufacturing Scopes range from full scale manufacturing development to industrial labs to new startup
development to industrial labs to new startup
companies to university research
companies to university research
y
y Time frames range from product now to nano Time frames range from product now to nano technology research 10
technology research 10--20 years out20 years out
y
y No perfect memory, each has its own unique No perfect memory, each has its own unique challenge
Examples of new memory
Examples of new memory
technologies
FeRAM
FeRAM
yy Operation Operation
y
y Selected PZT crystalline Selected PZT crystalline materials have bi
materials have bi--stable stable center atom
center atom
y
y Data is stored by Data is stored by
applying an voltage to applying an voltage to polarize the internal polarize the internal dipoles “Up” or “Down” dipoles “Up” or “Down”
y
y Reading by sensing the Reading by sensing the displacement current displacement current
y
y Fast read/write < 100 Fast read/write < 100 nSec
nSec
y
y Destructive read, limited Destructive read, limited number of read cycles number of read cycles due to fatigue
due to fatigue
y
y Very low power Very low power consumption consumption
Applied Electric Field Moves Center Atom
Perovskite Crystal Unit Cell
PZT (PbO,ZrO2, TiO2) Lead-Zirconate-Titanate
Di/Monovalent Metal Atoms Tetra/Pentavalent Atom
Oxygen Atoms
PZT Film Polarized Top Electrode
Bottom Electrode
EFilm EApp
ENet = EApp- EFilm
y
y OperationOperation
y
y Cell is 1 MJT + 1 TransistorCell is 1 MJT + 1 Transistor
y
y Electric current switches the Electric current switches the magnetic polarity of sense layer magnetic polarity of sense layer
y
y Change in magnetic polarity Change in magnetic polarity sensed as resistance change in sensed as resistance change in tunnel junction
tunnel junction
y
y Non destructive readNon destructive read
y
y Very fast read/write Very fast read/write performance, < 10
performance, < 10 nSec nSec reported
reported
y
y Unlimited number of R/W Unlimited number of R/W endurance
endurance
y
y Relatively high write currentRelatively high write current
MRAM
Ovonics Unified Memory
Ovonics Unified Memory
Data Storage Region
Resistive Electrode Amorphous Chalcogenide Phase Change Material Heater Poly Crystalline y
y OperationOperation
y
y Chalcogenide material Chalcogenide material alloys used in re
alloys used in re-
-writable CDs and DVDs writable CDs and DVDs
y
y Electrical energy (heat) Electrical energy (heat) converts the material converts the material between crystalline between crystalline (conductive) and (conductive) and amorphous (resistive) amorphous (resistive) phases phases y
y Cell reads by Cell reads by
measuring resistance measuring resistance
y
y Non-Non-destructive readdestructive read
y
y ~10~101212 write/erase cycleswrite/erase cycles y
y Medium write powerMedium write power
y
y Relatively easy Relatively easy
Complex Metal Oxide RRAM
Complex Metal Oxide RRAM
Poly Crystalline y
y OperationOperation y
y PCMO material, PCMO material,
Complex metal oxide
Complex metal oxide
studied for high temp
studied for high temp
superconductivity
superconductivity
y
y Change in resistance Change in resistance with applied electric field
with applied electric field
y
y Low resistance with Low resistance with forward bias, high
forward bias, high
resistance when reverse
resistance when reverse
electric field applied
electric field applied
y
y Low field read by Low field read by
measuring resistance
Programmable Metallization Cell
Programmable Metallization Cell
yy OperationOperation y
y Silver “dissolved” in Silver “dissolved” in chalcogenide
chalcogenide
y
y Change in resistance Change in resistance
with applied electric field
with applied electric field
driving silver to form low
driving silver to form low
resistance path
resistance path
y
y Reversible with reverse Reversible with reverse field
field
y
y Low field read with no Low field read with no disturb
disturb
y
y Very fast write and Very fast write and relatively low power
Resistance Polymer Memory
Resistance Polymer Memory
Poly Crystalline y
y OperationOperation
y
y Polymer material with Polymer material with special formulation
special formulation
y
y Change in resistance Change in resistance due to ionic transport
due to ionic transport
with applied electric field
with applied electric field
y
y Low resistance when Low resistance when ionic conductance paths
ionic conductance paths
formed, high resistance
formed, high resistance
when process is
when process is
reversed when paths
reversed when paths
broken
broken
y
y Low field read with no Low field read with no read disturb
Carbon Nanotube Switches
Carbon Nanotube Switches
y
y OperationOperation
y
y Carbon nanotube Carbon nanotube
suspended in crossbar
suspended in crossbar
architecture
architecture
y
y Electrostatic field attracts Electrostatic field attracts the nanotube and held
the nanotube and held
together by van der
together by van der
Waals force
Waals force
y
y Reverse bias repulse the Reverse bias repulse the wires
wires
y
y Reading by low vs high Reading by low vs high resistance
resistance
y
Molecular Memory
Molecular Memory
y
y OperationOperation
y
y Molecules in cross Molecules in cross point array
point array
y
y Change in electronic Change in electronic states in Redox
states in Redox
process with applied
process with applied
voltage
voltage
y
y Memory state sensed Memory state sensed by charge
by charge
displacement
displacement
y
y More than 1 state can More than 1 state can be stored by design of
be stored by design of
molecule (up to 8
Molecular Tunnel Memory
Molecular Tunnel Memory
y
y
Operation
Operation
y
y Molecules in cross Molecules in cross point array
point array
y
y Change in Change in
resistance with
resistance with
applied electric field
applied electric field
y
y Memory state Memory state sensed by
sensed by
measured
measured
resistance change
Ferroelectric Polymer Memory
Ferroelectric Polymer Memory
CMOS Base Wafer
Interface Logic Sense Amps/ Interconnect Charge Pump Charge Pump Word line Polymer Layer
Bit line Bit line Bit line
Polymer Layer Word line Word line Insulator (polymer) Word line Polymer Layer
Bit line Bit line Bit line
Polymer Layer
Word line Word line
y
y OperationOperation
y
y Polymeric Ferroelectric RAM Polymeric Ferroelectric RAM (PFRAM)
(PFRAM)
y
y Polymer chains with a dipole Polymer chains with a dipole moment
moment
y
y Data stored by changing the Data stored by changing the polarization of the polymer
polarization of the polymer
between metal lines
between metal lines
y
y Zero transistors per bit of storageZero transistors per bit of storage
y
3D One Time Program Memory
3D One Time Program Memory
y
y OperationOperation
y
y MultiMulti-layer one time -layer one time
programmable diodes at programmable diodes at
minimum lithography minimum lithography
dimensions dimensions
y
y Simplified architecture Simplified architecture with minimum number of with minimum number of
metal lines metal lines
y
y Zero transistor per cellZero transistor per cell
y
y Standard CMOS Standard CMOS
transistors under and transistors under and
outside array, process outside array, process
cost amortized over cost amortized over
multiple memory cells multiple memory cells
y
Millipede Memory
Millipede Memory
y
y OperationOperation
y
y Basic concept similar to punch Basic concept similar to punch cards
cards
y
y Thermally assisted, rewritable Thermally assisted, rewritable displacement media, PMMA displacement media, PMMA
y
y Large array of independently ZLarge array of independently Z- -axis controlled tips
axis controlled tips
y
y Low moving mass Low moving mass
y
y Direct point to point motion, no Direct point to point motion, no rotational latency
rotational latency
y
y Low power, no motors, actuators Low power, no motors, actuators
y
y High data transfer rates, High data transfer rates,
concurrent data transfer to/from concurrent data transfer to/from multiple tips
multiple tips
y
Spin Polarized E Beam Magnetic Memory
Spin Polarized E Beam Magnetic Memory
y
y OperationOperation
y
y Basic concept similar to Basic concept similar to Magnetic Hard Drive Magnetic Hard Drive
–
– Mechanical arm -Mechanical arm -> spin > spin polarized E Beam
polarized E Beam –
– Spinning Magnetic Media Spinning Magnetic Media - -> steered E Beam
> steered E Beam
y
y Much lower first read latency, Much lower first read latency, first access limited by E Beam first access limited by E Beam steering; fast read with E Beam steering; fast read with E Beam
y
y Lower power, no motor, no arm Lower power, no motor, no arm actuator
actuator
y
y No moving part, rugged, No moving part, rugged, vacuum space required vacuum space required
y
y Not as small or compact as all Not as small or compact as all solid state memories
A Simplified View
A Simplified View
y
y All above memories can be divided into two All above memories can be divided into two broad categories:
broad categories:
1.
1. XX--Y addressable, limited by lithography and Y addressable, limited by lithography and follows Moore’s Law
follows Moore’s Law
2.
2. Seek and scan memory, no lithography, Seek and scan memory, no lithography,
limited by scan mechanism, size and density
limited by scan mechanism, size and density
of storage areas in storage medium
Breaking the scaling barrier
Breaking the scaling barrier
y
y Memory technologies requiring transistor switch will be Memory technologies requiring transistor switch will be limited in scaling
limited in scaling --> FeRAM, MRAM, diode switch scales > FeRAM, MRAM, diode switch scales better
better --> OUM, RRAM> OUM, RRAM
Ä
Ä Follows MooreFollows Moore’’s Law limited by transistor/diodes Law limited by transistor/diode
y
y Simple cross point switch with no transistor or diode Simple cross point switch with no transistor or diode more scalable: molecular memories
more scalable: molecular memories
Ä
Ä Continue MooreContinue Moore’’s Law beyond transistor ages Law beyond transistor age
y
y MultiMulti--layer memories have the potential to be lowest layer memories have the potential to be lowest cost for litho defined memory
cost for litho defined memory
Ä
Ä Drop below historical MooreDrop below historical Moore’’s Laws Law
y
y Seek and scan can be the lowest cost but involves new Seek and scan can be the lowest cost but involves new memory storage and sense mechanism
memory storage and sense mechanism
Ä
No transistor switch better than
No transistor switch better than
transistor switch
transistor switch
Big
Small is not that small
Small is not that small
y
y X-X-Y addressable memory is Y addressable memory is always limited by the limiting always limited by the limiting lithography steps: have to lithography steps: have to connect to the real world connect to the real world circuits
circuits
y
y Either self assembled circuits at Either self assembled circuits at the smaller geometry or new the smaller geometry or new innovative architecture
Breaking the scaling barrier
Breaking the scaling barrier
y
y Memory technologies requiring transistor switch will be Memory technologies requiring transistor switch will be limited in scaling
limited in scaling --> FeRAM, MRAM, diode switch scales > FeRAM, MRAM, diode switch scales better
better --> OUM, RRAM> OUM, RRAM
Ä
Ä Follows MooreFollows Moore’’s Law limited by transistor/diodes Law limited by transistor/diode
y
y Simple cross point switch with no transistor or diode Simple cross point switch with no transistor or diode more scalable: molecular memories
more scalable: molecular memories
Ä
Ä Continue MooreContinue Moore’’s Law beyond transistor ages Law beyond transistor age
y
y MultiMulti--layer memories have the potential to be lowest layer memories have the potential to be lowest cost for litho defined memory
cost for litho defined memory
Ä
Ä Drop below historical MooreDrop below historical Moore’’s Laws Law
y
y Seek and scan can be the lowest cost but involves new Seek and scan can be the lowest cost but involves new memory storage and sense mechanism
memory storage and sense mechanism
Ä
Multi
Multi
-
-
layer is smaller
layer is smaller
Level N+1: Antifuse, P-, P+, conductor, P+
(fill and antifuse made transparent) Matrix Confidential Level N+1: Antifuse, P-, P+, conductor, P+
(fill and antifuse made transparent) Matrix Confidential
Single Layer cell area =
Single Layer cell area =
4
4 λλ22
Multi Layer cell area =
Multi Layer cell area =
4
4 λλ22/ N (number of / N (number of
layers)
Breaking the scaling barrier
Breaking the scaling barrier
y
y Memory technologies requiring transistor switch will be Memory technologies requiring transistor switch will be limited in scaling
limited in scaling --> FeRAM, MRAM, diode switch scales > FeRAM, MRAM, diode switch scales better
better --> OUM, RRAM> OUM, RRAM
Ä
Ä Follows MooreFollows Moore’’s Law limited by transistor/diodes Law limited by transistor/diode
y
y Simple cross point switch with no transistor or diode Simple cross point switch with no transistor or diode more scalable: molecular memories
more scalable: molecular memories
Ä
Ä Continue MooreContinue Moore’’s Law beyond transistor ages Law beyond transistor age
y
y MultiMulti--layer memories have the potential to be lowest layer memories have the potential to be lowest cost for litho defined memory
cost for litho defined memory
Ä
Ä Drop below historical MooreDrop below historical Moore’’s Laws Law
y
y Seek and scan can be the lowest cost but involves new Seek and scan can be the lowest cost but involves new memory storage and sense mechanism
memory storage and sense mechanism
Ä
Seek and Scan has best scaling
Seek and Scan has best scaling
potential
Market Potential
Value of polymer or seek and
Value of polymer or seek and
scan Memory
scan Memory
Log Read/Write Time (faster>)
Log Read/Write Time (faster>)
Cost (expensive>)
Disk Drives
Semi-Conductor
Memories
SRAM DRAM Polymer or seek
Value of New NV RAM
Value of New NV RAM
Log Write Time (faster >)
ROM Space ROM EPROM
Not in-system Changeable
Volatile Memory
Space
SRAM DRAM
New NV RAM Space
FeRAM OUM MRAM
Limited Unlimited Read Read Cycles Cycles
Flash Space
ETOX NAND
Unique Challenges of Non
Unique Challenges of Non
-
-
Volatile
Volatile
Memories
Memories
y
y Commonly accepted NV memory retention Commonly accepted NV memory retention time spec is 10 years
time spec is 10 years
y
y Retention is Retention is NEVERNEVER limited by the typical bits: limited by the typical bits: it is always limited by defect mechanisms
it is always limited by defect mechanisms
y
y Storage mechanisms must have high energy Storage mechanisms must have high energy barrier for long term stable retention
barrier for long term stable retention
y
y In most cases, the write mechanism may cause In most cases, the write mechanism may cause degradation to either the write mechanism
degradation to either the write mechanism
itself or to retention: giving limited cycling
itself or to retention: giving limited cycling
capability
System Consideration
System Consideration
y
y The new memory technologies are not going to The new memory technologies are not going to be direct plug and play replacement of
be direct plug and play replacement of
conventional memory
conventional memory
y
y They may operate in different modes and may They may operate in different modes and may have higher error rate or higher defect density
have higher error rate or higher defect density
y
y To take best advantage, system architecture To take best advantage, system architecture should comprehend memory operation
should comprehend memory operation
y
y Error detection and error correction will be an Error detection and error correction will be an integral part of the future memories
Summary
Summary
y
y Moore’s Law will continue through innovation Moore’s Law will continue through innovation
–
– Process complexity will increase to address Process complexity will increase to address fundamental limits of physics
fundamental limits of physics y
y To maintain Moore’s Law cost learning curve, To maintain Moore’s Law cost learning curve, difficult to do it by transistor technology alone
difficult to do it by transistor technology alone
–
– New opportunity for new memory structures and new New opportunity for new memory structures and new materials
materials y
y Current mainstream memory technologies of Current mainstream memory technologies of ETOX and NAND will continue to be the key
ETOX and NAND will continue to be the key
technologies for more than 5 years out
Summary (continued)
Summary (continued)
y
y Many potential new memory technologiesMany potential new memory technologies
y
y Simple cross point memory scales better than Simple cross point memory scales better than transistor switch but will always be litho limited
transistor switch but will always be litho limited
y
y MultiMulti--layer cross point provides lowest cost layer cross point provides lowest cost opportunities following Moore’s Law
opportunities following Moore’s Law
y
y Seek and scan memories can break through the Seek and scan memories can break through the litho barrier and scales more than Moore’s Law
litho barrier and scales more than Moore’s Law
y
y System architecture integral part of memory System architecture integral part of memory system