NC STATE UNIVERSITY
The State of ZettaRAM The State of ZettaRAM
Center for Embedded Systems Research (CESR)Department of Electrical and Computer Engineering
North Carolina State Universitywww.tinker.ncsu.edu/ericro
Eric Rotenberg and Ravi K. Venkatesan*
* Now at Intel Bangalore
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20062
Goal of TalkGoal of Talk
High level (casual) talk about ZettaRAMConsolidate papers and patents
– Core technology– Three different embodiments– Key novel properties– Implications
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20063
Core TechnologyCore Technology
New memory from ZettaCore– Genesis in DARPA Moletronics– Molecule stores 1 charge (0, +1)– Some molecule types store
multiple charges (0, +1, +2)
Long term– 1 molecule = 1 bit
Near term– Use molecules in aggregate– Molecular capacitor
Porphyrin Molecule
ZettaRAM Molecular Capacitor
Metal
Electrolyte
Metal
Molecules
Linkers
linker
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20064
Molecular CapacitorMolecular Capacitor
Metal (working electrode)
Molecules
Linkers
Electrolyte
Metal (counter electrode)
Vox e- e- e- e- e- e-
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20065
Linker & ElectrolyteLinker & Electrolyte
Neither conduct electronsElectrons tunnel across linkerElectrolyte ions form aligned dipoles
– Electrically interface counter electrode to molecules, yet charged molecules isolated
– Also provide critical charge shielding, prevent huge electric field across short linkers
Intrinsic retention times of 10s of seconds to minutes
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20066
Three EmbodimentsThree Embodiments
Transistor-free CrossbarTwo hybrid molecule/silicon devices
– Flash-like MoleFET– 1T-1C DRAM cell with molecular capacitor
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20067
CrossbarCrossbar
single memory cell
electrolytemoleculeslinkers
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20068
Crossbar FeaturesCrossbar Features
No explicit patterning of cells– Cell forms implicitly between electrodes– Density only limited by wire pitch– Easy path to minimum DRAM density
Silicon free– Easy 3D stacking (no silicon growth)– Deposit on arbitrary surfaces? Flexible memory?
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/20069
Crossbar FateCrossbar Fate
Crossbar earliest embodiment 2x2 in the lab Repeatability issues?
– Disturbs due to floating electrode voltages– Better control with transistor switches at intersections
Meanwhile– Moletronics 2nd phase brought transistor fab engineers– Shift towards hybrid molecules/silicon– Leverage predominance of silicon fabrication
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200610
Conventional Flash MemoryConventional Flash Memory
thin oxide
metal floating gate
thin oxide
metal gate
source diffusion
drain diffusion
silicon channel
to wordline
to bitline
Charge state of floating gate modulates Vt
hence Ids
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200611
MoleFET Flash MemoryMoleFET Flash Memory
thin barrier oxide
electrolyte
metal gate
source diffusion
drain diffusion
silicon channel
to wordline
linkers++ + + + + + + + + + +molecules to bitline
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200612
MoleFET FeaturesMoleFET Features
Fixed charge provides discreteness– Remarkable accuracy compared to “continuous” floating gate
– Robust, work within tighter noise margins
– Path to smaller devices
Discreteness especially benefits multi-bit storage– Multi-bit successful in Flash domain
– Molecules with multiple discrete charge states makes multi-bit much easier
00
01
10
11
0
+1
+2
+3
Vprog > Vox1
Vprog > Vox2
Vprog > Vox3
Flash MoleFET
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200613
MoleFET IssuesMoleFET Issues
Striving for larger Vt shift
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200614
Conventional 1T-1C DRAM CellConventional 1T-1C DRAM Cell
Word Line
Bit
Lin
e
capacitor
access transistor
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200615
1T-1C DRAM Derivative1T-1C DRAM Derivative
source diffusion
drain diffusion(working electrode)
silicon channel
WORDLINE
BIT
LIN
E
metal (counter electrode)
electrolyte
oxide
gate
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200616
DRAM Voltage Scaling LimitsDRAM Voltage Scaling Limits
Q
V
Qcrit
Vox
Q
VVmin
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200617
ZettaRAM’s Power Scaling AdvantageZettaRAM’s Power Scaling Advantage
Qcrit nearly constant due to noise sources– Sense amp margins– Bitline imbalances– Leakage– Radiation
Conventional DRAM– Charge-voltage coupling, Q = CV– Charge constrains voltage
ZettaRAM derivative– Charge-voltage decoupling– Fixed charge independent of voltage– Charge does not constrain voltage
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200618
DRAM Voltage Scaling LimitsDRAM Voltage Scaling Limits
Q
V
Qcrit
Vox
Q
VVmin
• Hard to increase C• Even harder when reducing 2D area
• Engineer new molecules with lower thresholds
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200619
Key PropertiesKey Properties
Flexibility and Precision Self-Assembly Charge-Voltage Decoupling Speed/Energy Tradeoff Multiple Discrete States Admixtures
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200620
Flexibility and PrecisionFlexibility and Precision
Hundreds of molecules synthesized Significant flexibility in customizing molecular attributes
– Design of organic molecules– Design of attachment groups– Influence surface concentration (density), threshold voltage
(power), electron transfer rate (speed, retention time)
Semiconductors also flexible– But attributes (e.g., threshold voltage) depend on bulk properties– Sophisticated “recipes” required– High cost to achieve precision– Contrast bulk properties with intrinsic chemical properties of
molecules
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200621
Self-AssemblySelf-Assembly
Auto arrangement of molecules in single, uniform, dense monolayer– Autonomous and parallel– Efficient fabrication
Reconsider possibilities thought impractical with conventional tech. Mixed logic/DRAM chips (DRAM embodiment)
– Conventional logic and DRAM processes too different due to stacked capacitors
– Self-assembled monolayers yield high charge density without elaborate stacked capacitor structures
– Apply thin oxide concept of MoleFET to reduce leakage (speed tradeoff) 3D stacking (crossbar embodiment)
– Molecules self-assemble on any compatible surface– Easy path to 3D memory stacking
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200622
Charge-Voltage DecouplingCharge-Voltage Decoupling
Fixed charge independent of voltage Charge does not constrain voltage Power-scalable DRAM derivative extends
roadmap of this important memory technology
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200623
Speed/Energy TradeoffSpeed/Energy Tradeoff
Voltage padded with respect to Vox
– Molecule speed slower as voltage approaches Vox
– Pad write voltage for competitive latency
Opportunity for architectural management Apply “fast voltage” for critical requests and “slow
voltage” for non-critical requests
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200624
Intelligent ManagementIntelligent Management
Hybrid Write Policy
FastQ = 4F: 1.2 V
WB: 1.2 V
SlowQ = 4F: 1.0 V
WB: 1.0 V
HybridQ = 4F: 1.2 V
WB: 1.0 V
HybridQ = 64F: 1.2 V
WB: 1.0 V
HybridEWB, Q = 4
F: 1.2 VWB: 1.0 V
0
0.2
0.4
0.6
0.8
1
no
rma
lized
bit
line
en
erg
y
FastQ = 4F: 1.2 V
WB: 1.2 V
SlowQ = 4F: 1.0 V
WB: 1.0 V
HybridQ = 4F: 1.2 V
WB: 1.0 V
HybridQ = 64F: 1.2 V
WB: 1.0 V
HybridEWB, Q = 4
F: 1.2 VWB: 1.0 V
0
0.2
0.4
0.6
0.8
1
no
rma
lized
bit
line
en
erg
y
critical fetches
Vfast (high energy)
non-critical writebacks
Vslow (low energy)
FastQ = 4F: 1.2 V
WB: 1.2 V
SlowQ = 4F: 1.0 V
WB: 1.0 V
HybridQ = 4F: 1.2 V
WB: 1.0 V
HybridQ = 64F: 1.2 V
WB: 1.0 V
HybridEWB, Q = 4
F: 1.2 VWB: 1.0 V
0
0.5
1
1.5
2
no
rmal
ized
exe
cuti
on
tim
e
FastQ = 4F: 1.2 V
WB: 1.2 V
SlowQ = 4F: 1.0 V
WB: 1.0 V
HybridQ = 4F: 1.2 V
WB: 1.0 V
HybridQ = 64F: 1.2 V
WB: 1.0 V
HybridEWB, Q = 4
F: 1.2 VWB: 1.0 V
0
0.5
1
1.5
2
no
rmal
ized
exe
cuti
on
tim
e
hybridslow
Energy
Time
hybridfast
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200625
Multiple Discrete StatesMultiple Discrete States
Easier multi-bit storage due to discrete states Discreteness reduces variability
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200626
AdmixturesAdmixtures
Different molecules can be mixed in same chip Although hybrid technologies not new:
– Nanotechnology offers new twist– Different molecules can occupy same physical space
admixture
bipartite
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200627
Dual MoleculesDual Molecules
NN
NN
ZnHS
NN
NN
Zn
HO
primary storage molecule
secondary storage molecule
F
F
F
F
F
F F
F
FF
F
F
F
F
F
NN
NN
ZnHSNN
NN
Zn
primary storage molecule secondary storage molecule
F
F
F
F
F
F F
F
FF
F
F
F
F
F
Fe
HO
HO
Fe
O
CH3
O
CH3
primary storage molecule
secondary storage molecule
E. Rotenberg and J. Lindsey.Variable-Persistence Molecular Memory Devices and Methods of Operation Thereof. US Patent #6,944,047.
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200628
PossibilitiesPossibilities
Unusual memory hierarchy– Memories with different attributes (speed, power,
volatility) cohabit same space– New challenges and opportunities for optimizing data
“placement” for power and performance
Admixtures enable different business models– Ship product with multiple molecules but only one
configured– Multiple virtual products in one physical product
NC STATE UNIVERSITY
Eric Rotenberg NANONET 09/14/200629
The State of ZettaRAMThe State of ZettaRAM
Contribution– Consolidated and distilled papers and patents– Unified discussion of core technology, embodiments, key
properties, and implications ZettaRAM has signs of disruptive technology
– Cheap fabrication of high perf. memory (by all metrics)– Practical mixed logic/DRAM– Practical 3D memory– Exceeds DRAM power scaling limits– Intelligent power management– Efficient multi-bit storage– Memory hierarchies cohabiting same space– Multiple virtual products in one physical product