has resistive memory found its place within the internet...
TRANSCRIPT
Has Resistive Memory
Found Its Place Within
the Internet of Things?
Michael N. KozickiArizona State University
Narbeh DerhacobianAdesto Technologies
•Market
•Technology
•Manufacturability
•Now what?
•Market
•Technology
•Manufacturability
•Now what?
https://www.apple.com/pr/products/apple-watch/Apple-Watch.html
Progress through scaling
Worldwide Device Shipments by Segment (Millions of Units)http://www.gartner.com/newsroom/id/3088221
The rise of mobile
Device Type 2014 2015 2016 2017
Traditional PCs (Desk-Based
and Notebook) 277 251 243 233
Ultramobiles (Premium) 37 49 68 89
PC Market 314 300 311 322
Ultramobiles (Tablets and
Clamshells) 226 214 228 244
Computing Devices Market 540 514 539 566
Mobile Phones 1,879 1,940 2,007 2,062
Total Devices Market 2,419 2,454 2,546 2,628
Market evolution
Desktops
Gaming
Servers
Routers
Laptops
Cell
Phones Smart
Phones
1990’s
Computing MobilityCommunicationEra of2010’s 2000’s
Wearables
Sensors
IoT
M2M
Autonomy
2020’s
The vast market potential of the Internet of Things (IoT)…“The Internet of things and the technology ecosystem surrounding it are
expected to be a $8.9 trillion market in 2020, according to IDC.” http://www.zdnet.com/internet-of-things-8-9-trillion-market-in-2020-212-billion-connected-things-7000021516/
Market evolution = new requirements
Then Now Future
Speed & density Low power, integrated,
connected, secure
GHz + Gb pJ + nW
Technology requirements• Very low power devices
–A CR2032 button cell only holds around 2 kJ» Operations must be low energy (<pJ)
–Device, and circuit limitations» Low voltage (<1V), low current (tens of A)
» Small periphery/high array efficiency (for low density)
• Small form factor
–Small geometry devices and compact design
–Thin and possibly even flexible chips
• Low cost
–Manufacturability is key, small area helps!
• Radiation tolerance
–Medical devices, wearables, mil-aero, soft errors
Are we there yet?
• Logic devices can already operate under these conditions
–Might not be ideal but they work!
• There is a huge problem with existing memory and storage
–Voltages are high (Flash, SSD)
–Detection of state is difficult in highly scaled devices (DRAM, Flash, SSD)
–Currents are high (HDD, MRAM, PCRAM)
–Form factor is not ideal (HDD)
IBM: Materials Will Spur Next Wave of
Chip Innovation Electronic News, 11/4/2005
“Innovation in materials has replaced scaling as
the main source of performance and feature
improvements in leading edge CMOS chips.”
•Market
•Technology
•Manufacturability
•Now what?
R. Waser, R. Dittmann, G. Staikov, and K. Szot., “Redox-Based Resistive Switching Memories – Nanoionic Mechanisms, Prospects, and Challenges”, Adv. Mater., vol. 21, 2632–2663
(2009).
Resistance change memory taxonomy (circa 2009)
Glassy solid electrolyte
high resistanceMobile ions added during
processing or via
electroforming
Inert
electrode
Oxidizable
electrode
+
-
M M+ + e-
M+ + e- M
e-
e-
Reverse bias dissolves electrodeposit via oxidation/reduction
Metallic electrodeposit
low resistance M+
M+M+
M+
M+
M+
M+
M+M+
M+
Programmable Metallization Cell (PMC)
Cryo-TEM
image of
filament
within solid
electrolyte 15nm
Note: Programmable Metallization Cell (PMC) is a platform technology for a variety of mass transport
applications. Conductive Bridging Random Access Memory (CBRAM) is the term generally applied to
memory applications of PMC.
Oxidation
Applied bias
Reduction
Ion
cu
rren
t
+
-
Programming
U. Russo, D. Kamalanathan, D. Ielmini, A.L. Lacaita, and M.N. Kozicki, “Study of Multilevel
Programming in Programmable Metallization Cell (PMC) Memory,” IEEE Transactions on Electron
Devices, Vol. 56, 1040 – 1047 (2009).
1 pJ operating point
Model is based on a Ag/Ag-Ge-S/W 1T-1R cell and
includes transistor load
D. Mahalanabis, H. J. Barnaby, Y. Gonzalez-Velo, M.N. Kozicki, S. Vrudhula, P. Dandamudi, “Incremental Resistance
Programming of Programmable Metallization Cells for Use as Electronic Synapses,” Solid State Electronics, in press.
Incremental programming
Pulse programming• 100 µs, 1.5 V write pulses• 200 µs, -1.5 V erase pulses• 5 kΩ current limiting resistance• 100 mV read pulses
Program Erase
On-state resistance vs.programming current
I. Valov, R. Waser, John R. Jameson and M.N. Kozicki, “Electrochemical metallization memories—
fundamentals, applications, prospects,” Nanotechnology, vol. 22 (2011) doi:10.1088/0957-
4484/22/25/254003
“NVM”
range
“Volatile”
range
Why?
17
Example: Ag12As35S53
Mostly As-S bonds, Ag-S bonds
Techniques:
Full DFT, 500 atom system
X-ray diffraction, neutron
scattering, EXAFS
Jaakko AkolaUniversity of Jyväskylä and
Tampere Technological University,
Finland
Bob JonesJülich Research Center, Germay
Tomas WagnerUniversity of Pardubice, Czech
Republic
Where do the metallic filaments form?
Cavities comprise 24% of the volume of Ag12As35S53, (SiO2 is 32% but cavities are more dispersed?)
Filament morphology
J.R. Jameson, N. Gilbert, F. Koushan, J. Saenz, J. Wang, S. Hollmer, M. Kozicki, and N. Derhacobian., “Quantized
Conductance in Ag/GeS2/W Conductive-Bridge Memory Cells,” IEEE Elec. Dev. Lett., vol. 33, 256-259 (2012).
µ
Data show resistance quantization (12.9kW) in programmed devices
µ
µ
µ
Continuous filament forms for iprog > VtGo
= 20 to 50 µA
Filament
branch
S. Rajabi, M. Saremi, H.J. Barnaby, A. Edwards, M.N. Kozicki, M. Mitkova, D. Mahalanabis, Y. Gonzalez-Velo, A. Mahmud,
“Static impedance behavior of programmable metallization cells”, Solid-State Electronics, vol. 106, 27–33 (2015).
Impedance spectroscopy
Ion source/sink
High ρ, low D
Electrolyte-oxide layered structure
“A dual-layered electrolytic resistance memory has been demonstrated
for the first time. Complete nanosecond switching of all cells in the
4kbit array, satisfactory retention, scalability down to 20nm, endurance
up to 1E7 cycles, and preliminary 4-level operation ...”
Cu-Te electrolyte
on
GdOx dielectric
K. Aratani et al., “A Novel Resistance Memory with High Scalability and Nanosecond
Switching,” IEDM Tech. Digest, 2007.
P. Dandamudi, H. J. Barnaby, M. N. Kozicki, Y. Gonzalez-Velo, K. E. Holbert, “Total Ionizing
Dose Tolerance of the Resistance Switching of Ag-Ge40S60 based Programmable
Metallization Cells”, 2013 Conference on Radiation Effects on Components and Systems
(RADECS), September 23rd – 27th, 2013, Oxford, UK
Gamma exposure to 10 Mrad
No impact of TID on the LRS or HRS
of Ge-S devices
Cumulative Distribution Function
•Market
•Technology
•Manufacturability
•Now what?
Forming Switching
Cross-
point
device
SiO2 cap(400nm)
Cu-SiO2-W devices, 15 nm deposited SiO2, 500 °C anneal
Cu-SiO2 devices
Rser
R1=355ohm, L1=6.06 x 10-6H
Impedance spectroscopy
frequency
Off state
frequency
On state
Incremental programmingProgramming from
0.45 to 0.75 V
Erase from -0.40 to -1.00 V
Integrated diode isolation - write
n+ Si
Sarath C. Puthentheradam, Dieter K. Schroder, and Michael N. Kozicki, “Inherent diode isolation in
programmable metallization cell resistive memory elements,” Appl. Phys. A (2011) 102: 817–826.
Cu top electrode - 35 nm
Cu doped SiO2 - 15 nm
Al - 200 nm
Dielectric
n+ Si
Cu top electrode - 35 nm
Cu doped SiO2 - 15 nm
Al - 200 nm
Dielectric
Integrated diode isolation - erase
Sarath C. Puthentheradam, Dieter K. Schroder, and Michael N. Kozicki, “Inherent diode isolation in
programmable metallization cell resistive memory elements,” Appl. Phys. A (2011) 102: 817–826.
Rfilament
Relectrolyte
On state
Saturation current scales
with programming current
- depends on filament area
Diode device characteristics
Sarath C. Puthentheradam, Dieter K. Schroder, and Michael N. Kozicki, “Inherent diode isolation in
programmable metallization cell resistive memory elements,” Appl. Phys. A (2011) 102: 817–826.
Gamma exposure to 7.1 Mrad
Multi-level resistance
programming is maintained
following gamma exposure
W. Chen, H. J. Barnaby, M. N. Kozicki, A. H. Edwards, Y. Gonzalez-
Velo, R. Fang, K. E. Holbert, S. Yu, W. Yu, “ Study of Gamma –Ray
Exposuure of Cu-SiO2 Programmable Metallization Cells”, IEEE
Trans. Nuc. Sci., in press.
The CBRAM Advantage
Conductive Bridging RAM (CBRAM®) is a
breakthrough technology platform that
overcomes a critical power barrier to
widespread innovation. Adesto’s Mavriq™
memory, built on CBRAM technology, is
adaptable for a broad range of
applications–from medical devices and
appliances to wearables and
smartphones. It enables 100 times less
energy consumption than today’s leading
memory technologies without sacrificing
performance and reliability.
Memory for the
Internet of Things
CBRAM in battery operated wearable device
Heart monitor:
Recording of an
ECG on a serial
NVM device.
Low energy memory allows
10x longer operational life
CBRAMStd. EEPROM #1Std. EEPROM #2
CBRAMStd. EEPROM #1Std. EEPROM #2
After 1 hour
Memory for medical applications
Storage of code and data in medical equipment
and devices.
Examples: Orthopedics, blood bags, catheters, glucose
meters, wireless patient monitoring
Data integrity of serial non
volatile memory devices after
gamma and e-beam irradiation
Pass – Full Data Integrity and Functionality Preserved
Fail – Data Loss
Tests performed by several medical device companies and
Typical
Dose for
Medical
Sterilization
• Minimal resistance change with time, independent of initial resistance value
• Minimal resistance change with respect to temperature, independent of initial resistance value
R vs. time at 200C R vs. temperature at 60mins
CBRAM® retention
•Market
•Technology
•Manufacturability
•Now what?
VLSI Symposia, Kyoto, June 2013
0.6 V, 1 pJ operation!
Embedded ReRAM product
With 3D XPoint, Micron Technology,
Inc. Is Predicting Memory RevolutionBY NEHA GUPTA · OCTOBER 9, 2015
According to Micron’s Todd Farrell, stagnation in new
memory development has resulted in multiple
computing challenges, which have been
compounded by the rapid processor advancements.
…the ideal memory design needs to fit into the
existing manufacturing infrastructure. In some cases,
Farrell noted, memory designs have perished
because they cannot be produced in scale using
existing manufacturing systems.
HP and SanDisk partner on new
memory chips to counter Intel-Micron
alliancePublished: Oct 8, 2015 12:15 p.m. ET
Hewlett-Packard Co. and SanDisk Corp. are collaborating on a
new breed of memory chips... The companies… predicted that
their forthcoming chips will be 1,000 times faster than flash
memory. The new chips also will be able to replace the widely
used chips known as DRAMs at much lower cost...
The goals they described mirror some of those laid out in late
July by chip makers Intel Corp.and Micron Technology Inc. which
announced a new memory technology they call 3D Xpoint.
Final thoughts
• The semiconductor industry is embracing
the requirements of low energy mobile
and autonomous systems.
• PMC/CBRAM is a low voltage/low energy
technology that is manufacturable.
• Products are already in the marketplace
and momentum is growing.
• The Internet of Things (IoT) will likely
drive further growth in the applications of
low energy resistive memories.
Thank you!
Support provided by:
Axon Technologies (memory)DTRA and AFRL (radiation research)
And thanks to Adesto for all the cool toys!