title laser testing and analysis of see in ddr3 memory ... · heavy ions in previous works1 –may...
TRANSCRIPT
Title
Name
Laser testing and analysis of SEE in DDR3 memory components
1
P. Kohler, V. Pouget, F. Wrobel, F. Saigné, P.-X. Wang, M.-C. Vassal
October 09 2017 2Pierre Kohler
• Dynamic memories (DRAMs) widely used in space applications for mass storage
• DDR3 are now being considered for space applications in the coming years– Clock frequency up to 1066 MHz (2133 Mbps)– Capacity up to 8 Gb
• Interest for identification of a new radiation tolerant component in order to integrate it in space modules– DDR3L : Low power DDR3 interesting for energy saving at module level
Context & Motivation
What about DDR3L radiation effects sensitivity ?
SEE, TID…
Methodology optimization
NSREC 2017 - “Analysis of Single-Event Effects in DDR3 and DDR3L SDRAMs using Laser Testing and Monte-Carlo simulations”
October 09 2017 3Pierre Kohler
• Devices under study
• Experimental setup
• Laser testing results
• Monte-Carlo simulation
• Conclusion
Outline
October 09 2017 4Pierre Kohler
Devices under study
• DUTs specifications
SampleMemory
typeDensity
Max clockfrequency
Supplyvoltage
A DDR3 4 Gb 666 MHz 1.5 V
B DDR3/L 2 Gb 933 MHz 1.35 / 1.5V
C DDR3/L 4 Gb 800 MHz 1.35 / 1.5V
Sensitive area
Irradiation beam
Silicon substrateSilicon substrate e
Removed plastic• Sample preparation
– DUTs are mounted in flip-chip FBGA plastic package
– The plastic is removed by acid attack to provideaccess to the silicon substrate backside
October 09 2017 5Pierre Kohler
Test bench hardware overview
• Test bench specifications
– Controlling DDR3/L SDRAMs under irradiation
– DUTs clock frequency up to 800 MHz
– Up to 8 DUTs
– Parametric tests
• Data retention time test
• Power consumption monitoring
– Functional tests
• Read/write periodic cycles
• Individual current limiting (Delatcher)
Motherboard
System on Chip (SoC)
Processing system (PS)
CPU0 CPU1
Test software
Programmable Logic (PL)
Memory
controller
Latchup
flags IP
User
interface
I2C
UART
Ethernet
Mechanical interface board
Ext. power
supply
SODIMM boardDevice Under Test
Current monitoring
Current limiting
Power supply
management board
October 09 2017 6Pierre Kohler
• Laser testing facility– 2-photon laser facility of the IES lab of
the University of Montpellier
– Estimated TPA induced charge track FWHM : 0.7 µm
• Test procedure
– Tests were performed at 400 MHz
– Standard 64 ms auto-refresh interval was used (except for specific test modes)
– Laser testing methodology• Starting with low pulse energies to observe non-
destructive event
• Gradual increase of laser energy to measure SEE threshold
• Scan along Z axis to find the most sensitive depth
• 2D mapping of SEU to descramble the memory organization
Experimental setup
Asynchronous
laser shots
Initialization
Write pattern
Read/compare
End of the test
SEE ?
Irradiation
complete?
SEE report
Current monitoring
SEE report
Latchup ?
October 09 2017 7Pierre Kohler
• Memory array organization
– All DDR3 SDRAMs are divided in 8 banks
– Banks in DUT A and C are divided in 4 blocks
– Banks in DUT B are divided in 2 blocks
– Each block is divided in sub-blocks
• Data bits organization
– Descrambling : Correspondence between corrupted logical address and laser spot position
– 3 tested devices exhibit very different scrambling strategies
• For DUT A and B : the 8 bits of a same word are distributed into several sub-blocks
• For DUT C, 4 bits of a same word are located in the same sub-block More prone to MBU
Laser testing results
Memory organization
DUTSize X[µm]
Size Y[µm]
Sub-blocks/row
Sub-blocks/col
Estimated bitsize [nm²]
A 100 50 8 33 104
B 68 51 12 33 104
C 72 43 8 27 5.103
Sub-blocks description
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
bank
B4,5,6,7
A0,1,2,3
D4,5,6,7
C0,1,2,3
F4,5,6,7
E0,1,2,3
H4,5,6,7
G0,1,2,3
J4,5,6,7
I0,1,2,3
L4,5,6,7
K0,1,2,3
N4,5,6,7
M0,1,2,3
P4,5,6,7
O0,1,2,3
A4,5,6,7
B0,1,2,3
C4,5,6,7
D0,1,2,3
G4,5,6,7
F0,1,2,3
G4,5,6,7
H0,1,2,3
I4,5,6,7
J0,1,2,3
K4,5,6,7
L0,1,2,3
M4,5,6,7
N0,1,2,3
O4,5,6,7
P0,1,2,3
DUT A
DUT B
DUT C
8 or 12 sub-blocks
27 or 33 sub-blocks
A6 A2 A7 A3 A5 A0 A4 A1
D0
F3
C6
E4
A3
C3
B0
E0
B5
D5
A7
F7
D1
F2
C7
E5
A2
C2
B1
E1
A4
F4
B6
D6
F0
D3
E6
C4
B3
E3
A0
C0
A5
F5
B7
D7
F1
D2
E7
C5
B2
E2
A1
C1
B4
D4
A6
F6
October 09 2017 8Pierre Kohler
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0
5
10
15
20
25
30
DUT A DUT B DUT C
SE
U T
hre
shold
(n
J²)
SEU Threshold pattern 0x00 - 1.5VSEU Threshold pattern 0x00 - 1.35VSEU Threshold pattern 0xFF - 1.5VSEU Threshold pattern 0xFF - 1.35VSEFI Threshold - 1.5V
• SEU and SEFI laser energy threshold measurement
– SEUTH for 0x00 pattern and the SEFITH
are roughly the same for the 3 DUTs
– SEUTH for the 0xFF pattern differs by a factor 3 between DUT A and B
– In low power mode SEUTH is substantially lower than in normal power mode (≈ 60%)
Laser testing results
SEE laser energy thresholds
October 09 2017 9Pierre Kohler
Bitline
Stacked capacitor
CC CCCBCB
VREF
N+N+bWL
N+N+bWL
P
Metal
screen
Dielectric
screen
STI
• Unexpected result : Important ratio between SEU and SEFI thresholds
– Contrary to what was observed under heavy ions in previous works1
– May be explained by the buried word-line (bWL) technology cell architecture
– The small pitch of the bWLs (56 to 80 nm) may act as a partially reflecting virtual metal layer
1 M. Herrmann et al, "New SEE and TID test results for 2-Gbit and 4-Gbit DDR3 SDRAM devices RADECS, 2013
Laser testing results
SEU/SEFI thresholds comparison
October 09 2017 10Pierre Kohler
1
10
100
1000
0.80 1.00 1.20 1.40 1.60 1.80
MC
U m
ult
ipli
cit
y
E2/ETH2
0x00 patternA
B
C
• Laser pulse energy increase MCU multiplicity increase
• 2 phenomena may explain the significant rise of the multiplicity
– The effective laser spot size expands with energy
– Total amount of charge generated by photoelectric effect increases with laser energy diffusion towards adjacent cells
• These results may constitute behaviors under heavy ions with high LET or at grazing incidence
• Specific behavior of DUT C : with pattern 0xFF, damage threshold and correctable SEU threshold are nearly the same
1
10
100
0.80 1.00 1.20 1.40 1.60 1.80 2.00
MC
U m
ult
ipli
cit
y
E2/ETH2
0xFF patternA
B
Laser testing results
MCU multiplicity vs E²/ETH²
October 09 2017 11Pierre Kohler
• Observation : Even above SEE threshold, sporadic events detection
– Read cycle range : 1 to 10s (depending on the address range)
– Laser pulse period : 100 ms(much larger than the refresh period)
Test cycle
Laser pulse
SEE
detection
SEE detection
period
Laser testing results
Atypical upset occurrence - Observation
• Definition : SEE detection period is the SEE average time between two SEE observations
October 09 2017 12Pierre Kohler
• Hypothesis : Existence of a critical time window for generating SEUs
• From previous works1 2
– Bit-line upset : bit-line disturbance occurring during a specific read or refresh cycle state when bit-lines are placed in a floating state
1G. Schindlbeck, "Types of soft errors in DRAMs," RADECS, 20052A. Bougerol, F. Miller and N. Buard, "SDRAM Architecture & Single Event Effects Revealed with Laser“, IOLTS 2008
Equalizer
VCC/2
Precharge
VCC/2
Memory cell Sense amplifier
VCC/2
VCC/2
0 1 2 3
VREF
VREFEQPRE WL /SA SA
1
VREF+
VREF-
VREF- 0
VREF+ 1BL
/BL
Comp. DQ
1
Critical time window for bit-line upset
Laser testing results
Atypical upset occurrence - Hypothesis
0 2 3 3
October 09 2017 13Pierre Kohler
• Highlight of the critical window for SEUs in the memory array using laser testing tool
– SEU detection period vs laser pulse period
(at TREFRESH = 32 and 64 ms)
Max of TSEE when TLASER = k x TREFRESH
Existence of a vulnerable time window within each refresh cycle
SE
E d
etec
tio
np
erio
d(s
)
Laser pulse period (ms)
0
50
100
150
200
250
0 20 40 60 80 100 120 140
Refresh period : 32 ms
0
100
200
300
400
0 20 40 60 80 100 120 140
Refresh period : 64 ms
• SEU threshold vs laser pulse period supports this hypothesis
– SEUTH remains constant, regardless of the laser pulse period
These upsets are not induced by charge integration during several consecutive pulses
Laser testing results
Evidence for bit-line upsets
8
9
10
11
12
20 40 60 80
SE
U t
hre
sho
ld (
nJ²
)
Laser pulse period (ms)
0xFF pattern
0x00 pattern
October 09 2017 14Pierre Kohler
• Comparison of heavy ion SEU data obtained on DUT A in previous work1 and Monte-Carlo simulation– Simulation of only cell upsets performed with
MC-Oracle
– Bits size and organization estimated from values obtained with IR camera and laser irradiation tool
– We observe a good fit above 20 MeV.cm²/mg
• Lower part of the heavy ion data seems to confirmed the occurrence of bit-lines upsets– Lower threshold due to larger collecting nodes
– Low cross-section due to the critical time window and refresh period ratio
Monte-Carlo Simulation
1E-17
1E-16
1E-15
1E-14
1E-13
1E-12
1E-11
1E-10
1E-09
1E-08
0 10 20 30 40 50 60 70 80 90
Cro
ss s
ecti
on
(cm
²/b
it)
LET (MeV.cm2/mg)
MC-Oracle
Heavy Ion data
1K. Grürmann et al., "Heavy Ion Sensitivity of 16/32-Gbit NAND-Flash and 4-Gbit DDR3 SDRAM," 2012 REDW
October 09 2017 15Pierre Kohler
• Laser testing and analysis of SEE in DDR3 memory components from 3 different manufacturers using two-photon laser testing
• Specific test bench developed and validated during experimental TID, HI and laser test campaigns
• Laser testing results• Extraction of physical and technological information (bit, array organization)• SEU/SEFI thresholds measurement and comparison• MCU multiplicity vs laser energy• Demonstration of the occurrence of bit-line upsets
• Laser good complementarity tool before HI experiments and for extracting important information for SEE assessment
Conclusion
Thank you for your attention
October 09 2017 16Pierre Kohler
Backup slides
October 09 2017 17Pierre Kohler
Standard DRAM architecture
DRAM principles
DRAM Cell• DRAM : Dynamic Random Access Memory
• 1T/1C Cell
• Binary information stored in a capacitor connected to an access transistor
• Capacitor leakage current induces the need to periodically refresh information
• Volatile : Information are lost after power cycling
Memory array
Inputs / Outputs
Peripheral circuitry
Memory physical organization• Common standard (JEDEC)
• Manufacturer dependent elements:
– Technology node
– Memory cells scrambling (physical distribution of the cells on the chip)
– Capacitor technology
– Peripheral circuitry
VREF
Access transistor
Storage capacitor
Wo
rd li
ne (
WL)
Bit line (BL)
October 09 2017 18Pierre Kohler
Motherboard
Zynq-7000 System-on-Chip
Programmable Logic (PL)
Processing system (PS)
CPU0 CPU1
Bus
SEL flags IP
DDR3 controller
Refreshcontrol IP
SODIMM socketLPC connector
Current limiters DUTs
8 flags
UART
Ethernet
I2C
I2C
FIFO (Events reports)
Hardware design architecture
Refresh control IP• User control of the Refresh
command• Aim : Data retention time
measurement• Configurable time interval
SEL flags IP• Stores latchup flags in a register• “OR” logic gate asserts an output
signal when a latchup is detected
FIFO• Software CPU shared memory• Stores events reports
October 09 2017 19Pierre Kohler
Software Architecture
Functionaltesting
Parametrictesting
Driver INA209
DDR3 controller
I2C Driver
I2C
Refreshcontrol IP
SEL Flags IP
Refreshcontrol
Currentlimiting
Memory write and
read
Currentmonitoring F
I
F
O
User interface
HardwareSoftware
Driver Refresh control IP
Driver SEL Flags IP
Application Drivers
October 09 2017 20Pierre Kohler
Single-Event Effects
• Algorithm proposed to classify Single-Event Effects :
– SEL monitoring runs in parallel
– Identify SEU, Multiple upsets, stuck bits and different types of SEFIs
– Reveal errors in control registers
• The algorithm should evolvedepending on the test results
October 09 2017 21Pierre Kohler
• Use of the two processors available in the Zynq module
– CPU0 functions• Parse ethernet user commands
• Control test program sequences
• Unqueue events reports stored in a FIFO
– CPU1 functions• Manage DUTs read/write accesses
• Check/correct SEUs
• Report events in a FIFO
• CPU1 pre-processing
– In order to prevent FIFO flooding• Add MBUs detection and correction
functions
• Add SEFIs detection functions
• Shared memory stores test parameters and commands sent fromCPU0 to CPU1
Software development
Processing system (PS)
CPU1
FIFOEvents reports
CPU0
DD
R3
co
ntr
olle
r
256 kB shared memory (OCM)
UART
Ethernet
Test
Pre-processing
Debug
Report
Test commands
Programmable Logic (PL)