WIND memory think tank - June 22, 2010 1
Magnetic memories and embedded applications
Claude CHAPPERT CNRS
"Nanospintronics" group Institut d'Electronique Fondamentale
Université Paris Sud - Orsay
SSA WIND
Mass storage Information Lifecycle Management
WIND memory think tank - June 22, 2010 2
Application
+ automatized migration of data from one pool to the others
Virtually unlimited capacity, extendable at will, technology robust
Memory
Fast disks
High capacity disks
Tape storage
… but not a long term « passive » archival : requires constant rewriting !
Recording : fast data rate
Backup : short recovery time/point objectives
Archival
Con
text
WIND memory think tank - June 22, 2010 3
D3
- Mag
neti
c R
AM
Towards embedded Non Volatile STTRAM
S.H. Kang, Qualcomm, NVM Workshop, UCSD, April 11-13 (2010)
WIND memory think tank - June 22, 2010 4
Mag
neti
c R
AM
S.H. Kang, Qualcomm, NVM Workshop, UCSD, April 11-13 (2010)
5
Spin-RAM specifications “to be achieved”
Products / Demos Predicted Position vs CMOS
Cell size 20 – 80 F2 ~ 10 F2 >> NAND << SRAM
Technology Above IC embedded NVM
Speed < 40 ns (2.7 ns) ≤ ns ? ~SRAM, µP
Endurance 1015 ~ infinite >> NAND
Non volatility > 10 years
Scalability 90 nm 20 nm ? ???
Logic circuits
2009 : - new demos from Toshiba (perpendicular recording), Hitachi - agreement CROCUS-Towers for producing the TA-MRAM - large contract gov./Samsung/Hynix in South Korea to produce STT-RAM - large industry/academic contrats in Japan, France on « Magnetic logic » - many other companies : Avalanche, MagIC+IBM, GRANDIS, NEC…
WIND memory think tank - June 22, 2010
Performance assessment of memory technologies.
WIND memory think tank - June 22, 2010 6
Mag
neti
c R
AM
Kryder+Kim, IEEE Mag45, 3406 (2009)
With CMOS technology only:
Slow communication between logic and memory
- long interconnections -complexity of interconnecting paths
-larger occupancy on wafer - large static dissipation
With hybrid CMOS/magnetic:
Fast communication between logic and memory
- numerous short vias -simpler interconnecting paths -Smaller occupancy on wafer
- instant on/off - extended possibilities for
programmation and reconfiguration
New paradigm for architecture of complex electronic circuits (microprocessors...)
Tighter integration between logic and memories
+ general trend towards programmable memory (FPGA, …)
7 WIND memory think tank - June 22, 2010
WIND memory think tank - June 22, 2010 8
Low standby Power Programmable logic devices
100% 61% 0% 14%
Active Inactive
Non volatile, multi-core logic: are powered only the core that need to operate others preserve state and start « instantly » when powered on
Towards a magnetic FPGA
CLB
CLB
CLB
CLB
CLB
CLB
CLB
CLB
“switches” and “logic blocs” (CLB) are made: non volatile programmable by Spin- MRAM elements
D3
- Mag
neti
c R
AM
WIND mem
ory
9
French Research Project “CILOMAG”
Towards a magnetic FPGA
CLB
CLB
CLB
CLB
CLB
CLB
CLB
CLB
“switches” and “logic blocs” (CLB) to be made: non volatile (run-time) programmable by integrating Spin- MRAM elements
WIND mem
ory
10
Configuration Memory of a e-FPGA
Conf_in
CLB
TILE
SBOX
Local configuration Memory (Volatile)
Non volatile external memory to store the logic
configuration
L. Torrès, Y. Guilleminet L. Rougé
WIND mem
ory
11
Magnetic -TILE Design
160 µm
L. Torrès, Y. Guilleminet L. Rougé
Thermally assisted TA- MRAM (CROCUS)
New in 2010: French Research Project “SPIN”
WIND memory think tank - June 22, 2010 12
Mag
neti
c R
AM
EETimes / June 9th, 2010 : Menta SAS and LIRMM tape out the world’s first MRAM-based FPGA enabling compact integration of MRAM and embedded-FPGA solutions :
Menta’s eFPGA Core programmable logic architecture CROCUS and CEA-LETI magnetic technology, CMOS 130 nm ST Microelectronics technology 120 nm magnetic junctions Thermally assisted writing mode (CROCUS )
Objective : customizable to use as domain specific-FPGA (dsFPGA) in a SoC with target applications features.
Same test chip: magnetic LUT, Flip-Flop, e-MRAM, by IEF and SPINTEC
First chips expected by Q4-2010
WIND memory think tank - June 22, 2010 13
D3
- Mag
neti
c R
AM
Example of STT-RAM impact in SoC
S.H. Kang, Qualcomm, NVM Workshop, UCSD, April 11-13 (2010)
Many companies are scheduling e-MRAM powered MCU : - Renesas (2012) - Qualcomm/TSMC - Grandis/NEC/Hynix - CROCUS - …
45nm CMOS, Logic compatible e-STTRAM produced by Qualcomm/TSMC (IEDM 2009)
WIND memory think tank - June 22, 2010 14
D3
- Mag
neti
c R
AM
S.H. Kang, Qualcomm, NVM Workshop, UCSD, April 11-13 (2010)
Summary for MRAM
WIND memory think tank - June 22, 2010 15
D3
- Mag
neti
c R
AM
Everspin’s MRAM is very successful, demonstrates potential and reliability of MRAM, but only on a niche market for low density/low capacity and technology not much downscalable
CROCUS’ TA-MRAM shows higher density, speed, better downscalability (but not so much), but still remains complex. Propect: embedded MRAM, partial cache memory replacement in SoCs.
STT-RAM technology is rapidly getting mature,BUT: to reach high densities:
- today’s current densities still too high by a factor of 3 to 5
- downscaling a uniform, high enough value of magnetic anisotropy on a large array is a problem (Toshiba: perp. magnetization, Hitachi: synthetic AF free layer, CROCUS: exchange bias to AF layer)
STT-RAM has potential for 3D stacking ! (controller area << RAM area)
STT-RAM getting adapted for mass storage ?
MLC STT-RAM has been demonstrated (Hitachi, VLSI 2010)